WO2019227581A1 - Interest point recognition method, apparatus, terminal device, and storage medium - Google Patents

Abstract

Disclosed in the present application are an interest point recognition method, an apparatus, a terminal device, and a storage medium. The method comprises: acquiring a pre-set training corpus, using an N-gram model to analyze the training corpus, to obtain word sequence data, upon reception of voice information to be recognized, parsing said to-be-recognized voice information, to obtain M pronunciation sequences of said to-be-recognized voice information, with regard to each of the pronunciation sequences, according to the word sequence data, calculating the occurrence probability of each pronunciation sequence, to obtain the occurrence probabilities of the M pronunciation sequences. By selecting, from the occurrence probabilities of the M pronunciation sequences, the pronunciation sequence corresponding to the occurrence probability reaching a pre-set probability threshold, as a target pronunciation sequence, and acquiring, from an interest point information base, interest point information corresponding to the target pronunciation sequence, as an interest point recognition result of said to-be-recognized voice information. The present invention accurately recognizes the meaning of voice information, improving the accuracy and efficiency of recognizing an interest point.

Description

Interest point recognition method, device, terminal equipment and storage medium

This application is based on a Chinese invention patent application filed on May 29, 2018 with the application number 201810529490.2 and entitled "Interest Point Recognition Method, Device, Terminal Equipment and Storage Medium" and claims its priority.

Technical field

The present application relates to the field of computer technology, and in particular, to a method, a device, a terminal device, and a storage medium for identifying a point of interest.

Background technique

With the progress of society and economic development, many people will travel frequently due to business needs, and some people will use their spare time to travel. In strange places, it is often necessary to search for some addresses or points of interest through smart devices. People provide convenience, and many smart devices provide voice recognition for point of interest recognition.

Most of the current speech recognition functions provided by smart devices use general models to convert the acquired natural language information into speech and text to identify the preset interest points contained in them. However, there are often many natural languages that interfere with the preset interest points. Vocabulary, and because of everyone's expression, accent and other issues, the recognition accuracy of interest points in natural language speech information is not high and the efficiency is low.

Summary of the Invention

The embodiments of the present application provide a method, an apparatus, a terminal device, and a storage medium for identifying interest points, so as to solve the problems of low recognition accuracy and low recognition efficiency of interest points in natural language voice information.

In a first aspect, an embodiment of the present application provides a method for identifying a point of interest, including:

Obtain a preset training corpus;

An N-gram model is used to analyze the preset training corpus to obtain word sequence data of the preset training corpus, wherein the word sequence data includes a word sequence and a word sequence frequency of each of the word sequences. degree;

If the speech information to be identified is received, the speech information to be identified is parsed to obtain M pronunciation sequences of the speech information to be identified, where M is a positive integer greater than 1;

For each of the pronunciation sequences, calculating the occurrence probability of each pronunciation sequence according to the word sequence data, thereby obtaining the occurrence probability of M pronunciation sequences;

Selecting the pronunciation sequence corresponding to the occurrence probability reaching a preset probability threshold from the occurrence probabilities of M said pronunciation sequences as a target pronunciation sequence;

The point of interest information corresponding to the target pronunciation sequence is obtained from the point of interest information database as a point of interest recognition result of the speech information to be recognized.

In a second aspect, an embodiment of the present application provides a device for identifying a point of interest, including:

A training corpus acquisition module for acquiring a preset training corpus;

A training corpus analysis module is configured to analyze the preset training corpus using an N-gram model to obtain word sequence data of the preset training corpus, wherein the word sequence data includes a word sequence and each Word sequence frequency of the predicate sequence;

A voice information parsing module, configured to parse the voice information to be recognized if the voice information to be recognized is received, to obtain M pronunciation sequences of the voice information to be recognized, where M is a positive integer greater than 1;

An occurrence probability calculation module, configured to calculate an occurrence probability of each pronunciation sequence for each of the pronunciation sequences and according to the word sequence data, so as to obtain an occurrence probability of M pronunciation sequences;

A pronunciation sequence confirmation module, configured to select the pronunciation sequence corresponding to the occurrence probability that reaches a preset probability threshold from the occurrence probability of M said pronunciation sequences as a target pronunciation sequence;

The recognition result obtaining module is configured to obtain the point of interest information corresponding to the target pronunciation sequence from the point of interest information database as a point of interest recognition result of the speech information to be recognized.

According to a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, where the processor executes the computer may The steps of the method for identifying points of interest are implemented when the instruction is read.

According to a fourth aspect, embodiments of the present application provide one or more non-volatile readable storage media storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more A plurality of processors execute the steps of the point of interest recognition method.

Details of one or more embodiments of the present application are set forth in the accompanying drawings and description below, and other features and advantages of the present application will become apparent from the description, the drawings, and the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to explain the technical solutions of the embodiments of the present application more clearly, the drawings used in the description of the embodiments of the application will be briefly introduced below. Obviously, the drawings in the following description are just some embodiments of the application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without paying creative labor.

FIG. 1 is an implementation flowchart of an interest point identification method according to an embodiment of the present application; FIG.

FIG. 2 is a flowchart of implementing step S4 in the method of identifying a point of interest according to an embodiment of the present application; FIG.

3 is an implementation flowchart of obtaining a training corpus in a method of identifying interest points provided by an embodiment of the present application;

4 is an implementation flowchart of constructing a point of interest information database in a method of identifying a point of interest provided by an embodiment of the present application;

FIG. 5 is an implementation flowchart of generating a supplementary corpus in the method of identifying interest points provided by an embodiment of the present application; FIG.

FIG. 6 is a schematic diagram of a point of interest identification device according to an embodiment of the present application; FIG.

FIG. 7 is a schematic diagram of a terminal device according to an embodiment of the present application.

Detailed ways

In the following, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

Please refer to FIG. 1, which illustrates a flowchart of implementing an interest point identification method provided by an embodiment of the present application. The point of interest recognition method is applied to a scene of recognition of points of interest in speech information of natural language. The identification scenario includes a server and a client, where the server and the client are connected through a network, and the user sends voice information in natural language through the client. The client may specifically but not limited to various personal computers and notebooks For computers, smart phones, tablet computers, and portable wearable devices, the server can be implemented by an independent server or a server cluster composed of multiple servers. The method for identifying a point of interest provided in the embodiments of the present application is applied to a server, as follows:

S1: Obtain a preset training corpus.

Specifically, the training corpus is used to evaluate the speech information in natural language, and is a corpus obtained by training using related corpora. The content in the training corpus in the embodiment of the present application includes, but is not limited to, points of interest information and a general corpus. Wait.

Among them, Corpus refers to a large-scale electronic text library that has been scientifically sampled and processed. Corpus is the basic resource of linguistic research and the main resource of empirical language research methods. It is used in dictionary compilation, language teaching, traditional language research, statistical or example-based research in natural language processing, corpus, that is, language material, Corpus is the content of linguistic research and the basic unit of corpus.

S2: An N-gram model is used to analyze the preset training corpus to obtain the preset word sequence data of the training corpus, where the word sequence data includes the word sequence and the word sequence frequency of each word sequence.

Specifically, by using the N-gram model to statistically analyze each corpus in the preset training corpus, the number of times that one corpus H appears after another corpus I in the preset training corpus is obtained, and then “corpus I + corpus” is obtained. H "word sequence data appears.

The word sequence refers to a sequence composed of at least two corpora in a certain order. The frequency of the word sequence refers to the proportion of the number of times that the word sequence appears in the entire corpus. The word segmentation here It is a word sequence obtained by combining consecutive word sequences in a preset combination manner. For example, if a word sequence "love tomatoes" appears 100 times in the entire corpus, and the total number of occurrences of all participles of the entire corpus is 100,000 times, the word sequence frequency of the word sequence "love tomatoes" is 0.0001. .

Among them, the N-gram model is a language model commonly used in large vocabulary continuous speech recognition. Using the collocation information between adjacent words in the context, when it is necessary to convert continuous non-space pinyin into Chinese character strings (that is, sentences), Calculate the sentence with the highest probability, so as to realize automatic conversion to Chinese characters, without manual selection by the user, and avoiding the problem of recoding of many Chinese characters corresponding to the same pinyin.

By using the N-gram model to analyze each word sequence data of the preset training corpus, it is only necessary to directly use these word sequence data when calculating the probability of occurrence, which saves calculation time and improves the efficiency of interest point recognition.

S3: If the speech information to be identified is received, the speech information to be identified is parsed to obtain M pronunciation sequences of the speech information to be identified, where M is a positive integer greater than 1.

Specifically, each Chinese pronunciation corresponds to one or more Chinese characters. After receiving the to-be-recognized voice information input by the user on the client, the server decodes the to-be-recognized voice information through an acoustic decoder, and converts into multiple pronunciation sequences. .

Wherein, the pronunciation sequence refers to a text sequence including at least two word segmentations obtained by converting speech information.

For example, in a specific embodiment, the pronunciation sequence obtained by acoustically decoding this to-be-recognized voice information "woxixihuanchichizhonggugumeimeishi" can be a pronunciation sequence A: "I", "like", " "Eat", "China", "Food" can also be the pronunciation sequence B: "I", "Like", "Chi Zhong", "Guomei", "Food", and can also be the pronunciation sequence C: "I", "Western ring", "hold", "China", "all right" and so on.

S4: For each pronunciation sequence, according to the word sequence data, calculate the occurrence probability of each pronunciation sequence, thereby obtaining the occurrence probability of M pronunciation sequences.

Specifically, according to the word sequence data obtained in step S2, a pronunciation probability calculation is performed for each pronunciation sequence to obtain the occurrence probability of M pronunciation sequences.

The Markov hypothesis theory can be used to calculate the occurrence probability of the pronunciation sequence. The occurrence of the Y word is only related to the first Y-1 words, and it is not related to any other words. The probability of the entire sentence is the probability of occurrence of each word. product. These probabilities can be obtained by directly counting the number of simultaneous occurrences of Y words from the corpus. which is:

P (T) = P (W ₁ W ₂ ... W _Y ) = P (W ₁ ) P (W ₂ | W ₁ ) ... P (W _Y | W ₁ W ₂ ... W _Y-1 )Formula 1)

Among them, P (T) is the probability of the entire sentence appearing, and P (W _Y | W ₁ W ₂ ... W _Y-1 ) is the probability that the _Y- th participle appears after the word sequence composed of Y-1 participles.

For example: after speech recognition of the sentence "Chinese nation is a nation with a long history of civilization", a pronunciation sequence divided is: "Chinese nation", "Yes", "one", "has", "long-term" "," Civilization "," history "," of "," nation ", a total of 9 participles appear. When n = 9, it is calculated that the part of" nation "appears in the" Chinese nation is a civilization with a long history. " Probability after the word sequence "historical".

S5: From the occurrence probabilities of M pronunciation sequences, a pronunciation sequence corresponding to the occurrence probability that reaches a preset probability threshold is selected as the target pronunciation sequence.

Specifically, for each sounding sequence, an occurrence probability is obtained through the calculation of step S4, and a total of M sounding sequence occurrence probabilities are obtained. The occurrence probabilities of the M sounding sequences are compared with a preset probability threshold, respectively, and are selected to be greater than Or the occurrence probability equal to the preset probability threshold is used as the effective occurrence probability, and then the pronunciation sequences corresponding to the effective occurrence probability are found, and these pronunciation sequences are used as the target pronunciation sequences.

By comparing with the preset probability threshold, the pronunciation sequences whose occurrence probability does not meet the requirements are filtered, so that the selected target pronunciation sequence is closer to the meaning expressed in natural speech, and the accuracy of interest point recognition is improved.

It should be noted that if the calculated probability of M pronunciation sequences is less than a preset probability threshold, a reminder message will be pushed to the user, for example, "No target location was found, please confirm your pronunciation specifications and try again" At the same time, the voice message record is collected and sent to the background management staff. If the number of target pronunciation sequences is greater than the preset number, sort them in the order of their corresponding probability of occurrence, and select the preset number of pronunciation sequences before sorting as the target pronunciation sequence, for example, the preset number is 5 Then, after the effective occurrence probabilities are sorted, the first five effective occurrence probabilities are selected, and then the pronunciation word order corresponding to the five occurrence probabilities is used as the target pronunciation sequence.

S6: Obtain the point of interest information corresponding to the target pronunciation sequence from the point of interest information database as the result of the point of interest recognition of the speech information to be recognized.

Specifically, after obtaining the target pronunciation sequence, the point of interest information contained in the target pronunciation sequence is obtained from the point of interest information database, and the point of interest information is pushed to the user as a result of the point of interest recognition of the voice information.

In the embodiment corresponding to FIG. 1, the preset training corpus is obtained, and then the N-gram model is used to analyze the preset training corpus to obtain the word sequence data of the preset training corpus. All the words are calculated and analyzed in advance. Sequence data, so that the word sequence data can be used directly in the subsequent calculation of the probability of occurrence, which saves the time of calculating the probability and improves the efficiency; when the voice information to be recognized is received, the voice information to be recognized is parsed to obtain the voice information to be recognized. M pronunciation sequences, for each pronunciation sequence, according to the word sequence data, calculate the occurrence probability of each pronunciation sequence, and select the pronunciation sequence corresponding to the occurrence probability of the preset probability threshold from the obtained occurrence probability of M pronunciation sequences As the target pronunciation sequence, and further obtain the point of interest information corresponding to the target pronunciation sequence from the point of interest information database, and as the result of the point of interest recognition of the speech information to be recognized, this method calculates the probability of the pronunciation sequence and selects the Probability as the result of the screening method, can achieve the speech information The meaning is accurately identified, thereby improving the accuracy of interest point recognition.

Next, on the basis of the embodiment corresponding to FIG. 1, a specific embodiment is used below to calculate the occurrence probability of the pronunciation sequence for each pronunciation sequence mentioned in step S4 according to the word sequence data. The specific implementation method will be described in detail.

Please refer to FIG. 2, which illustrates a specific implementation process of step S4 provided by an embodiment of the present application, which is detailed as follows:

S41: For each pronunciation sequence, obtain all the participles a ₁ , a ₂ , ..., an _n-1 , an _n within the pronunciation sequence, where n is a positive integer greater than 1.

It should be noted that the word segmentation in the pronunciation sequence is obtained in the order of the word order from front to back. For example, for a pronunciation sequence "I love China", the word segmentation is performed in order from the word order to the first. One participle "I", the second participle "Love", and the third participle "China".

S42: According to the word sequence data, use formula (2) to calculate the probability that the _nth participle a _{n of the n} participles appears after the word sequence (a ₁ a ₂ ... a _n-1 ), and use this probability as the pronunciation sequence Probability of occurrence:

After probability _{| (a 1 a 2 ... a} n-1 a n) for the n th word n-th word appears in a _n word sequence _{(a 1 a 2 ... a n} -1) where, P , C (a ₁ a ₂ ... an _n-1 an _n ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 an _n ), C (a ₁ a ₂ ... a _n-1 ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 ).

Specifically, it can be known from step S2 that the word sequence frequency of each word sequence is obtained through the analysis of the training corpus by the N-gram model, and only calculation is required according to formula (2) here.

It is worth noting that because the training corpus used by the N-gram model is relatively large, the data is sparse and serious, and the time complexity is high. .

Wherein, bigram is calculated by using the formula (2) each word a present participle ₂ illustrating a probability after ₁ A _1, a word _{participle. 3} illustrating a probability after ₂ A _2, ..., a _n word The probability A _n-1 that appears after the word segmentation a _n-1 , and then uses formula (3) to calculate the probability of occurrence of the entire word sequence (a ₁ a ₂ ... a _n-1 a _n ):

P (T ') = A ₁ A ₂ ... A _n-1

In the embodiment corresponding to FIG. 2, for each pronunciation sequence, all the participles in the pronunciation sequence are obtained, and the probability that the last participle appears after the combination of all the previous participles is calculated to obtain the probability of the entire sentence. , And then evaluate whether the sentence is reasonable, so as to recognize the semantics contained in the speech information of natural language, and obtain relevant information such as the name of the point of interest to be obtained, which effectively improves the accuracy of the point of interest recognition.

On the basis of the embodiment corresponding to FIG. 1 or FIG. 2, before acquiring the preset training corpus mentioned in step S1, a training corpus may also be constructed. As shown in FIG. 3, the method for identifying interest points further includes:

S71: Construct a point of interest information database.

Specifically, before performing the point of interest recognition, in order to ensure the accuracy of the point of interest recognition, a comprehensive point of interest information database containing points of interest needs to be constructed. The point of interest information database contains the point of interest information of each point of interest. Use the points of interest contained in the existing general model to generate a point of interest information base. You can also construct a point of interest information base by manually collecting points of interest, or use a web crawler to obtain points of interest to build a point of interest information base. , The specific method is not specifically limited here.

Preferably, the method adopted in the embodiment of the present application is to obtain a point of interest by using a web crawler to build a point of interest information database.

The point of interest information includes, but is not limited to, the name of the point of interest, the category to which the point of interest belongs, and the address of the point of interest.

S72: Generate a supplementary corpus based on the interest point information database.

Specifically, the point of interest information in the point of interest information database is extracted, and all the obtained point of interest information is processed according to a preset processing method as a supplementary corpus.

The specific processing method may be segmentation of interest points, or semantic statistics of interest point information, etc., which may be specifically selected according to actual needs, and is not limited here.

S73: Combine the supplementary corpus with a preset basic corpus to obtain a training corpus.

Specifically, because the N-gram model is used to analyze the training corpus, the training corpus must have a huge corpus so that it can evaluate whether a sentence is reasonable, so a preset corpus and supplement that will have sufficient corpus need to be used The corpus is combined to obtain the training corpus.

Among them, the preset basic corpus is selected according to actual needs. For example, the news of Sohu's financial, sports, and current affairs in the past three years is selected, and the corpus generated by text cleaning and collation is used as the basic corpus.

In the embodiment corresponding to FIG. 3, by constructing a point of interest information base and generating a supplementary corpus based on the point of interest information base, the supplementary corpus is combined with a preset basic corpus to obtain a training corpus for N- The training corpus analyzed by the gram model not only has the ability to evaluate whether a sentence is reasonable, but also contains information about points of interest, so that it can accurately evaluate whether a sentence contains points of interest, which is conducive to improving the accuracy of natural language speech information recognition. And the accuracy rate of interest point information identification.

On the basis of the embodiment corresponding to FIG. 3, a specific embodiment is used to describe in detail a specific implementation method of constructing a point of interest information base mentioned in step S71.

Please refer to FIG. 4, which illustrates a specific implementation process of step S71 provided by an embodiment of the present application, which is detailed as follows:

S711: Classify the preset basic interest points according to a preset classification method to obtain a basic classification of the interest point information database.

Specifically, the basic interest points are classified according to a preset classification method of the interest points, the classification is used as the basic classification of the interest point information database, and the interest point information contained in each basic classification is stored in the information point information database. The corresponding position and classification method can be set according to actual needs, and there is no limitation here.

Among them, the basic interest points refer to each small category of the points of interest, and the basic classification refers to the large categories of points of interest. For example, a basic category contained in the information point information base is "cuisine", and the basic points of interest contained below the basic category There are "breakfast," "fast food," "hot pot," "buffet," and "hotel."

S712: For each basic classification, through the web crawling method, obtain the interest point information of all the basic interest points in the administrative region of the country that includes the basic classification, and obtain the interest point information of the basic classification in each administrative region of the country.

Specifically, for each basic classification in the information point information database, a web crawler is used to sequentially crawl each administrative region in the country to obtain the information of the administrative region containing all the basic interest points under this basic classification, thereby obtaining the Point of interest information of basic classifications in each administrative region of the country. According to this method, information of points of interest of all basic classifications in each administrative region of the country is obtained.

Among them, the web crawler is also called Scalable Web Crawler. The crawling object is expanded from some seed URLs (Uniform Resource Locator, Uniform Resource Locator) to the entire Web (World Wide Web, global wide area network), which is mainly a portal site search engine. And data from large web service providers. The crawling scope and number of web crawlers are huge, and the requirements for crawling speed and storage space are high. The order of crawling pages is relatively low. At the same time, because there are too many pages to be refreshed, parallel work is usually used. It is divided into page crawl module, page analysis module, link filtering module, page database, URL queue, and initial URL collection. To improve work efficiency, general web crawlers will adopt certain crawling strategies. Common crawling strategies are: depth-first strategy, breadth-first strategy.

Among them, the basic method of the depth-first strategy is to visit the next level of web links in order from the lowest to the highest depth, until it cannot be further deepened. After completing a crawling branch, the crawler returns to the previous link node to further search for other links. When all links have been traversed, the crawling task ends.

Among them, the breadth-first strategy is to crawl pages according to the depth of the content directory level of the web page, and the pages at the shallower directory level are crawled first. After the pages in the same level are crawled, the crawler goes deeper and continues to crawl. This strategy can effectively control the crawling depth of the page, avoid the problem that crawling cannot be ended when encountering an infinite deep branch, and it is easy to implement without storing a large number of intermediate nodes.

Preferably, the crawling strategy adopted in the embodiment of the present application is a breadth-first strategy.

For each basic classification, through a web crawler, crawl each administrative region of the country in order to obtain the information of the administrative region containing all the basic points of interest under this basic classification, so as to obtain the specific points of interest information of the basic classification in each administrative region of the country. The implementation process includes steps A to E, which are detailed as follows: Step A: Obtain information on administrative regions at all levels throughout the country and the latitude and longitude corresponding to each administrative region.

Specifically, the administrative district information of each city-level unit in the country is obtained, and then the county-level administrative district information included in the municipal-level administrative district is obtained, and then the information of the district, street, township and other office information included in the county-level administrative district is obtained.

The administrative region information includes, but is not limited to, administrative region name, administrative region code, higher administrative region information, and lower administrative region information. For example, as shown in Table 1, the obtained administrative region information is 440300.

Table I

Further, the latitude and longitude information corresponding to each administrative area is obtained.

Among them, longitude and latitude is a coordinate system composed of longitude and latitude. Called geographic coordinate system, it is a spherical coordinate system that uses the spherical surface of three degrees to define the space on the earth, and can mark any position on the earth.

The longitude and latitude coordinate systems commonly used in China include, but are not limited to: WGS84 coordinate system (World Geodetic System 1984, World Geodetic Coordinate System), Beijing 54 coordinate system (BJZ54), Xi'an 80 coordinate system (XIAN80).

Preferably, the latitude and longitude coordinate system used in the embodiment of the present application is a WGS84 coordinate system.

Step B: For each administrative area K, divide the administrative area according to the latitude and longitude according to the preset division side length to obtain n rectangular lists of the same size.

Specifically, the national administrative region list includes several administrative regions, and the sizes of different administrative regions are different. By obtaining the latitude and longitude range of the administrative region, four pole coordinates are obtained, and the coordinates of the four latitudes and longitudes are used as the coordinates of the four vertices of a large rectangle, and then a large rectangle is obtained. Long division results in n rectangles.

It is worth noting that due to the inconsistency in the prosperity of different administrative regions, there are some administrative regions with dense points of interest and some administrative regions with sparse points of interest. Therefore, for different administrative regions, the preset length of the division can be selected according to the actual situation. For administrative regions with dense points of interest, a smaller segmented edge length can be preset. For administrative regions with sparse points of interest, a larger segmented edge length can be preset to facilitate subsequent crawling of points of interest to improve crawling. Speed, thereby improving the efficiency of point of interest acquisition.

Further, the obtained vertices of the four vertices of the large rectangle are converted into space rectangular coordinates and divided according to a preset rectangle length and rectangle width, for example, the coordinates of the lower left corner are (lat_1, lon_1), and the coordinates of the upper right corner are (lat_2, lon_2), set the length of the split side to len, the coordinates of the lower left corner of the first rectangle are lat_1, lon_1, and the coordinates of the upper right corner are lat_1 + len, lon_1 + len; , Lon_1 + len, the upper right corner coordinates are lat_1 + len, lon_1 + len. The number of rectangles generated is:

(int ((lat_2-lat_1) / len) +1) × (int ((lon_2-lon_1) / len) +1).

Among them, int is a rounding function. For example: int (1.334) = 1.

For example, in a specific embodiment, the latitude and longitude range obtained from Shenzhen is: 113 ° 46 '~ 114 ° 37' east longitude, 22 ° 27 '~ 22 ° 52' north latitude, and the space rectangular coordinates are converted into the lower left corner ( 22.45, 113.769444), upper right corner (22.86667, 114.619444). In actual requirements, the side length can be set to 0.04. According to the above method, the first rectangular coordinate is the lower left corner (22.45, 113.769444), the upper right corner (22.49, 113.809444), and the second rectangular coordinate is the lower left corner (22.53, 113.809444), top right corner (22.53, 113.809444).

Step C: For the basic classification J, according to the rectangular list of the administrative area K, generate a URL list of the administrative area.

Specifically, assuming that the current basic classification is J and the current administrative area is K, after the n rectangular lists of administrative area K are generated, the rectangular list is traversed by a web crawler, and each rectangular list is crawled to contain any tasks under the basic classification J. A URL of a basic point of interest to generate a URL list.

For example, in a specific implementation, using a web crawler to crawl a Baidu map with a bottom left corner coordinate (22.53, 113.809444) and a top right corner coordinate (22.53, 113.809444) in a rectangular area containing a basic interest point as "secondary", may Use the following code:

url = ’http: //api.map.baidu.com/place/v2/search? query = ’secondary’ & bounds = ’+

22.53 ’+’, ’+’ 113.809444 ’+’, ’+’ 22.53 ’+’, ’+’ 113.809444 ’+’, ’+’ & page_size = 20 &

page_num = '+ str (page_num) +' & output = json & ak = 9s5GSYZsWbMaFU8Ps2V2VWvDlDlqGaaO '.

Among them, "page_size" refers to the preset number of content contained in each page, page_num refers to the number of pages, and "ak" (Apiconsole Key (AK)) is the developer's Baidu Maps API console key.

Step D: Determine the distribution information of the interest points of the basic classification J in the administrative area K by analyzing the URL list, and obtain the information of the interest points belonging to the basic classification J contained in the administrative area.

Specifically, by performing a webpage parsing on the URL list obtained in step C to obtain the basic point of interest information contained in each URL, thereby obtaining the point of interest information contained in each administrative area.

For example, in a specific implementation manner, the obtained URL list includes 26 URLs, and each URL includes 20 points of interest information, one of which is as follows:

After parsing the address on the URL, the result is: the name of the point of interest is "Kunming Eighth Middle School", and its specific address is "No. 628 Longquan Road, Wuhua District, Kunming City, Yunnan Province", and its administrative area is "Wuhua District ", whose street number is" 35debf29e6063d3aa7da399b ".

Step E: Store the acquired POI information into a corresponding location in the POI database.

Specifically, the acquired point of interest information is classified according to the base to which it is acquired, and stored in a corresponding position in the point of interest information database.

Taking the point-of-interest information obtained in step D as an example, the point-of-interest information whose name of the point of interest is "Kunming Eighth Middle School" is stored in the basic point of interest "medium" which is basically classified as "school".

In the embodiment corresponding to FIG. 4, the preset basic interest points are classified according to a preset classification method to obtain the basic classification of the interest point information database, and then for each basic classification, a web crawling method is used to obtain Each administrative region of the country contains the point of interest information of all the basic points of interest of the basic classification, and the information of the points of interest of the basic classification in each of the administrative regions of the country is obtained, so as to obtain all the information of the points of interest in each administrative region of the country. When identifying points of interest, it can provide accurate and comprehensive points of interest information, which is conducive to improving the accuracy of point of interest recognition.

Based on the embodiment corresponding to FIG. 3, a specific embodiment is used to describe in detail a specific implementation method of generating a supplementary corpus based on the point of interest information base mentioned in step S72.

Please refer to FIG. 5, which illustrates a specific implementation process of step S72 provided by an embodiment of the present application, which is detailed as follows:

S721: Extract interest point information in the interest point information database.

Specifically, the basic interest points in each basic classification and the interest point information contained in the basic interest points are extracted from the interest point information database.

S722: Perform word segmentation processing on the POI information to obtain the POI word segmentation.

Specifically, for each extracted point of interest information, Chinese word segmentation is performed to obtain the point of interest segmentation of the point of interest information.

Among them, Chinese word segmentation refers to cutting a sequence of Chinese characters into individual characters. Word segmentation is the process of recombining consecutive word sequences into word sequences in accordance with certain specifications. Existing word segmentation algorithms can be divided into three categories: word segmentation methods based on string matching, word segmentation methods based on understanding, and word segmentation methods based on statistics. According to whether it is combined with the part-of-speech tagging process, it can be divided into a simple word segmentation method and an integrated method combining word segmentation and labeling.

Preferably, the word segmentation algorithm adopted by the embodiment of the invention is an understanding-based word segmentation method.

For example, in a specific embodiment, the obtained point-of-interest information is "basic classification-food, basic point of interest-fast food, name of point of interest-Yanjin pot, point of interest address-Bagua II, Luohu District, Shenzhen, Guangdong Province You can get “Food”, “Fast Food”, “Guangdong Province”, “Shenzhen”, “Luohu District”, “Bagua Erlu” and “Yanjin Pot” according to the participle.

S723: Establish a mapping relationship between the point-of-interest segmentation and the corresponding point-of-interest information, and save the point-of-interest segmentation, the point-of-interest information, and the mapping relationship in the supplementary corpus correspondingly.

Specifically, after segmenting the point of interest information, each acquired point of interest segmentation is associated with the point of interest information to form a mapping, and the point of interest segmentation, point of interest information, and mapping relationship are correspondingly stored in a supplementary corpus. , So that when a certain point of interest segmentation is identified, the corresponding point of interest information can be found. At the same time, both the point of interest segmentation and the point of interest information are put into the supplementary corpus, which can improve the appearance of relevant information of the point of interest in the training corpus frequency.

Taking the POI segmentation obtained in step S722 as an example, the POI information is "basic classification-cuisine, basic POI-fast food, POI name-Yanjin pot, POI address-Bagua Second Road, Luohu District, Shenzhen, Guangdong Province The Yanjin pot's point of interest contains the set of participles: {"Cuisine", "Fast food", "Guangdong Province", "Shenzhen", "Luohu District", "Bagua Erlu", "Yanjin pot"}.

In the embodiment corresponding to FIG. 5, by extracting the point of interest information in the point of interest information database and performing word segmentation processing on the point of interest information, the point of interest segmentation is obtained, and then a mapping between the point of interest segmentation and the corresponding point of interest information is established. Relationship, and save the interest point segmentation, interest point information, and mapping relationship to the supplementary corpus, so that the supplementary corpus contains interest point information, interest point segmentation, and their mapping relationships, so that in the subsequent interest point detection, it can be based on the corresponding The POI segmentation directly finds the corresponding POI information, thereby improving the efficiency of POI recognition.

On the basis of the embodiment corresponding to FIG. 3, after the point of interest information database mentioned in step S71 is constructed, more points of interest information area can be updated, and the method of identifying points of interest further includes:

If the update instruction is received, the point of interest information base is updated in real time, or the point of interest information base is automatically updated according to a preset condition.

Understandably, the point of interest information will change over time. After some point of interest changes, if the point of interest database is not updated accordingly, the point of interest will be unrecognized when these points of interest are identified. The identification or identification information is incorrect. Therefore, the point of interest information database needs to be updated.

Specifically, the embodiments of the present application provide two ways of updating the point of interest information database, which are to update according to preset conditions, and to perform real-time update when an update instruction sent by a user is received.

Among them, updating according to a preset condition refers to triggering an automatic update procedure to perform an automatic update after the preset condition is reached. The preset condition may be a preset update period. For example, the preset update period is 7 days. It can also be detected that the crawled URL list has changed in step S712. For example, under the same crawling condition, the crawling result has changed from the previous 16000 to 17600. At this time, the interest point The information database is updated. The specific preset conditions can be set in various and flexible settings according to the actual situation, and there is no specific limitation here.

It should be noted that, for the update process of the interest point information database, please refer to the description of steps C to E in S711. To avoid repetition, details are not described here.

In the embodiment of the present application, when an update instruction is received, the point of interest information database is updated in real time or automatically, so that the point of interest information contained in the point of interest information database is always maintained in an accurate state, so that in the subsequent point of interest identification , Can provide accurate and comprehensive point of interest information, which is conducive to improving the accuracy of interest point identification.

It should be understood that the size of the sequence numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

Corresponding to the method for identifying points of interest in the above method embodiment, FIG. 6 shows a point of interest identification device that corresponds one-to-one to the method of identifying points of interest provided by the above method embodiment. For ease of description, only the implementation with this application is shown. Example related parts.

As shown in FIG. 6, the point of interest recognition device includes a training corpus acquisition module 10, a training corpus analysis module 20, a voice information analysis module 30, an occurrence probability calculation module 40, a pronunciation sequence confirmation module 50, and a recognition result acquisition module 60. The detailed description of each function module is as follows:

A training corpus acquisition module 10, configured to acquire a preset training corpus;

A training corpus analysis module 20 is configured to analyze a preset training corpus using an N-gram model to obtain preset word sequence data of the training corpus, where the word sequence data includes a word sequence and a word sequence frequency of each word sequence degree;

The voice information analysis module 30 is configured to parse the voice information to be recognized if the voice information to be recognized is received, to obtain M pronunciation sequences of the voice information to be recognized, where M is a positive integer greater than 1;

The occurrence probability calculation module 40 is configured to calculate the occurrence probability of each pronunciation sequence according to the word sequence data for each pronunciation sequence, thereby obtaining the occurrence probability of M pronunciation sequences;

The pronunciation sequence confirmation module 50 is configured to select, from the occurrence probabilities of M pronunciation sequences, a pronunciation sequence corresponding to an occurrence probability reaching a preset probability threshold as a target pronunciation sequence;

The recognition result obtaining module 60 is configured to obtain the point of interest information corresponding to the target pronunciation sequence from the point of interest information database as a point of interest recognition result of the speech information to be recognized.

Further, the occurrence probability calculation module 40 includes:

Segmentation sequence extraction unit 41 is configured to obtain, for each pronunciation sequence, all the segmentations a ₁ , a ₂ , ..., an _n-1 , an _n within the pronunciation sequence, where n is a positive integer greater than 1;

The occurrence probability calculation unit 42 is configured to calculate the probability that the nth participle a _{n of the n} participles appears after the word sequence (a ₁ a ₂ ... a _n-1 ) according to the word sequence data using the following formula. Probability as the probability of occurrence of the pronunciation sequence:

After probability _{| (a 1 a 2 ... a} n-1 a n) for the n th word n-th word appears in a _n word sequence _{(a 1 a 2 ... a n} -1) where, P , C (a ₁ a ₂ ... an _n-1 an _n ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 an _n ), C (a ₁ a ₂ ... a _n-1 ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 ).

Further, the interest point recognition device further includes:

The point of interest information base construction unit 71 is configured to construct a point of interest information base;

A supplementary corpus acquisition unit 72, configured to generate a supplementary corpus based on the point of interest information base;

The training corpus generating unit 73 is configured to combine a supplementary corpus with a preset basic corpus to obtain a training corpus.

Further, the point of interest information base construction unit 71 includes:

A classification division subunit 711, configured to classify a preset basic interest point according to a preset classification method to obtain a basic classification of an interest point information database;

An information acquisition subunit 712 is configured to obtain the interest point information of all the basic points of interest in each administrative region of the country containing the basic classification in each administrative region of the country by using a web crawling method for each basic classification to obtain the basic classification in each administrative region of the country Point of interest information.

Further, the supplementary corpus acquisition unit 72 includes:

An information extraction subunit 721, configured to extract interest point information in a point of interest information database;

An information segmentation subunit 722, configured to perform word segmentation processing on the point of interest information to obtain the word segmentation of the point of interest;

A corpus acquisition subunit 723 is configured to establish a mapping relationship between a point of interest segmentation and corresponding point of interest information, and correspondingly save the point of interest segmentation, point of interest information, and mapping relationship in a supplementary corpus.

Further, the interest point recognition device further includes:

The information base update module 80 is configured to update the point of interest information base in real time if an update instruction is received, or automatically update the point of interest information base according to a preset condition.

For the process of implementing each function of each module in the point of interest recognition device provided in this embodiment, reference may be made to the description of the foregoing method embodiment for details, and details are not described herein again.

This embodiment provides one or more nonvolatile readable storage media storing computer readable instructions. The nonvolatile readable storage medium stores computer readable instructions, and the computer readable instructions are When the processors execute, the point of interest identification method in the foregoing method embodiment is implemented, or when the computer-readable instructions are executed by one or more processors, the functions of each module / unit in the point of interest identification device in the foregoing device embodiment are implemented. To avoid repetition, we will not repeat them here.

Understandably, the non-volatile readable storage medium may include: any entity or device capable of carrying the computer-readable instruction code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, Read-Only Memory (ROM), Random Access Memory (RAM), electric carrier signals and telecommunication signals.

FIG. 7 is a schematic diagram of a terminal device according to an embodiment of the present application. As shown in FIG. 7, the terminal device 90 of this embodiment includes a processor 91, a memory 92, and computer-readable instructions 93 stored in the memory 92 and executable on the processor 91, such as a point of interest recognition program. When the processor 91 executes the computer-readable instructions 93, the steps in the foregoing embodiments of the method for identifying points of interest are implemented, for example, steps S1 to S6 shown in FIG. Alternatively, when the processor 91 executes the computer-readable instructions 93, the functions of each module / unit in the foregoing device embodiments are implemented, for example, the functions of the modules 10 to 60 shown in FIG. 6.

For example, the computer-readable instructions 93 may be divided into one or more modules / units, and the one or more modules / units are stored in the memory 92 and executed by the processor 91 to complete the present application. One or more modules / units may be instruction segments of a series of computer-readable instructions capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer-readable instructions 93 in the terminal device 90. For example, the computer-readable instructions 93 may be divided into a training corpus acquisition module, a training corpus analysis module, a voice information analysis module, an occurrence probability calculation module, a pronunciation sequence confirmation module, and a recognition result acquisition module. The specific functions of each module are as shown in the device embodiment. To avoid repetition, details are not described here.

The terminal device 90 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server. The terminal device 90 may include, but is not limited to, a processor 91 and a memory 92. Those skilled in the art can understand that FIG. 7 is only an example of the terminal device 90, and does not constitute a limitation on the terminal device 90. The terminal device 90 may include more or fewer components than shown in the figure, or some components may be combined or different components For example, the terminal device 90 may further include an input / output device, a network access device, and a bus.

The so-called processor 91 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), Ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory 92 may be an internal storage unit of the terminal device 90, such as a hard disk or a memory of the terminal device 90. The memory 92 may also be an external storage device of the terminal device 90, such as a plug-in hard disk provided on the terminal device 90, a Smart Memory Card (SMC), a Secure Digital (SD) card, and a flash memory card (Flash Card) and so on. Further, the memory 92 may include both an internal storage unit of the terminal device 90 and an external storage device. The memory 92 is used to store computer-readable instructions and other programs and data required by the terminal device 90. The memory 92 may also be used to temporarily store data that has been output or is to be output.

Those skilled in the art can clearly understand that, for the convenience and brevity of the description, only the above-mentioned division of functional units and modules is used as an example. In practical applications, the above functions can be assigned by different functional units, Module completion, that is, dividing the internal structure of the device into different functional units or modules to complete all or part of the functions described above.

The above-mentioned embodiments are only used to describe the technical solution of the present application, but not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still implement the foregoing implementations. The technical solutions described in the examples are modified, or some of the technical features are equivalently replaced; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the application, and should be included in Within the scope of this application.

Claims (20)

1. A point of interest recognition method, characterized in that the point of interest recognition method includes:

   Obtain a preset training corpus;

   An N-gram model is used to analyze the preset training corpus to obtain word sequence data of the preset training corpus, wherein the word sequence data includes a word sequence and a word sequence frequency of each of the word sequences. degree;

   If the speech information to be identified is received, the speech information to be identified is parsed to obtain M pronunciation sequences of the speech information to be identified, where M is a positive integer greater than 1;

   For each of the pronunciation sequences, calculating the occurrence probability of each pronunciation sequence according to the word sequence data, thereby obtaining the occurrence probability of M pronunciation sequences;

   Selecting the pronunciation sequence corresponding to the occurrence probability reaching a preset probability threshold from the occurrence probabilities of M said pronunciation sequences as a target pronunciation sequence;

   The point of interest information corresponding to the target pronunciation sequence is obtained from the point of interest information database as a point of interest recognition result of the speech information to be recognized.
2. The method of claim 1, wherein, for each of the pronunciation sequences, calculating the occurrence probability of each pronunciation sequence according to the word sequence data comprises:

   For each of the pronunciation sequences, obtaining all the participles a ₁ , a ₂ , ..., an _n-1 , an _n within the pronunciation sequence, where n is a positive integer greater than 1;

   According to the word sequence data, the following formula is used to calculate the probability that the nth participle a _{n of the n} participles appears after the word sequence (a ₁ a ₂ ... a _n-1 ), and the probability is used as the pronunciation sequence Probability of occurrence:

   

   After probability _{| (a 1 a 2 ... a} n-1 a n) for the n th word n-th word appears in a _n word sequence _{(a 1 a 2 ... a n} -1) where, P , C (a ₁ a ₂ ... an _n-1 an _n ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 an _n ), C (a ₁ a ₂ ... a _n-1 ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 ).
3. The method of identifying a point of interest according to claim 1 or 2, wherein before the acquiring a preset training corpus, the method of identifying the point of interest further comprises:

   Building a point of interest database;

   Generating a supplementary corpus based on the point of interest information base;

   Combining the supplementary corpus with a preset basic corpus to obtain the training corpus.
4. The point of interest identification method according to claim 3, wherein the constructing the point of interest information database comprises:

   Classifying a preset basic interest point according to a preset classification method to obtain a basic classification of the interest point information database;

   For each of the basic classifications, through a web crawling method, the point of interest information of all the basic points of interest in the basic classification in each administrative region of the country is obtained, and the information of the points of interest of the basic classification in each administrative region of the country is obtained.
5. The point of interest recognition method according to claim 3, wherein the generating a supplementary corpus based on the point of interest information database comprises:

   Extracting the point of interest information in the point of interest information database;

   Performing word segmentation processing on the point of interest information to obtain the point of interest segmentation;

   Establishing a mapping relationship between the point of interest segmentation and the corresponding point of interest information, and correspondingly storing the point of interest segmentation, the point of interest information, and the mapping relationship in the supplementary corpus.
6. The method of identifying a point of interest according to claim 3, wherein after the constructing the point of interest information database, the method of identifying the point of interest further comprises:

   If an update instruction is received, the point of interest information base is updated in real time, or the point of interest information base is automatically updated according to a preset condition.
7. An interest point recognition device, characterized in that the interest point recognition device includes:

   A training corpus acquisition module for acquiring a preset training corpus;

   A training corpus analysis module is configured to analyze the preset training corpus using an N-gram model to obtain word sequence data of the preset training corpus, wherein the word sequence data includes a word sequence and each Word sequence frequency of the predicate sequence;

   A voice information parsing module, configured to parse the voice information to be recognized if the voice information to be recognized is received, to obtain M pronunciation sequences of the voice information to be recognized, where M is a positive integer greater than 1;

   An occurrence probability calculation module, configured to calculate an occurrence probability of each pronunciation sequence for each of the pronunciation sequences and according to the word sequence data, so as to obtain an occurrence probability of M pronunciation sequences;

   A pronunciation sequence confirmation module, configured to select the pronunciation sequence corresponding to the occurrence probability that reaches a preset probability threshold from the occurrence probability of M said pronunciation sequences as a target pronunciation sequence;

   The recognition result obtaining module is configured to obtain the point of interest information corresponding to the target pronunciation sequence from the point of interest information database as a point of interest recognition result of the speech information to be recognized.
8. The point of interest recognition device according to claim 7, wherein the occurrence probability calculation module comprises:

   Segmentation sequence extraction unit, for each said pronunciation sequence, to obtain all the segmentations a ₁ , a ₂ , ..., an _n-1 , an _{n in} the pronunciation sequence, where n is a positive integer greater than 1 ;

   The occurrence probability calculation unit is configured to calculate the probability that the nth participle a _{n of the n} participles appears after the word sequence (a ₁ a ₂ ... a _n-1 ) according to the word sequence data, and The probability is taken as the occurrence probability of the pronunciation sequence:

   

   After probability _{| (a 1 a 2 ... a} n-1 a n) for the n th word n-th word appears in a _n word sequence _{(a 1 a 2 ... a n} -1) where, P , C (a ₁ a ₂ ... an _n-1 an _n ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 an _n ), C (a ₁ a ₂ ... a _n-1 ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 ).
9. The point of interest recognition device according to claim 7 or 8, wherein the point of interest recognition device further comprises:

   Point-of-interest information base construction unit, which is used to construct a point-of-interest information base;

   A supplementary corpus acquisition unit, configured to generate a supplementary corpus based on the point of interest information base;

   A training corpus generating unit is configured to combine the supplementary corpus with a preset basic corpus to obtain the training corpus.
10. The point of interest recognition device according to claim 9, wherein the point of interest information base construction unit comprises:

    A classification division subunit, configured to classify a preset basic interest point according to a preset classification method to obtain a basic classification of the interest point information database;

    An information acquisition subunit is configured to obtain, for each of the basic classifications, a point of interest information of all the basic interest points of the basic classification in each administrative region of the country through a web crawling method, and obtain the basic classification in each Point of interest information for each administrative district.
11. A terminal device includes a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, and is characterized in that the processor implements the computer-readable instructions as follows step:

    Obtain a preset training corpus;

    An N-gram model is used to analyze the preset training corpus to obtain word sequence data of the preset training corpus, wherein the word sequence data includes a word sequence and a word sequence frequency of each of the word sequences. degree;

    If the speech information to be identified is received, the speech information to be identified is parsed to obtain M pronunciation sequences of the speech information to be identified, where M is a positive integer greater than 1;

    For each of the pronunciation sequences, calculating the occurrence probability of each pronunciation sequence according to the word sequence data, thereby obtaining the occurrence probability of M pronunciation sequences;

    Selecting the pronunciation sequence corresponding to the occurrence probability reaching a preset probability threshold from the occurrence probabilities of M said pronunciation sequences as a target pronunciation sequence;

    The point of interest information corresponding to the target pronunciation sequence is obtained from the point of interest information database as a point of interest recognition result of the speech information to be recognized.
12. The terminal device according to claim 11, wherein, for each of the pronunciation sequences, calculating the occurrence probability of each pronunciation sequence according to the word sequence data comprises:

    For each of the pronunciation sequences, obtaining all the participles a ₁ , a ₂ , ..., an _n-1 , an _n within the pronunciation sequence, where n is a positive integer greater than 1;

    According to the word sequence data, the following formula is used to calculate the probability that the nth participle a _{n of the n} participles appears after the word sequence (a ₁ a ₂ ... a _n-1 ), and the probability is used as the pronunciation sequence Probability of occurrence:

    

    After probability _{| (a 1 a 2 ... a} n-1 a n) for the n th word n-th word appears in a _n word sequence _{(a 1 a 2 ... a n} -1) where, P , C (a ₁ a ₂ ... an _n-1 an _n ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 an _n ), C (a ₁ a ₂ ... a _n-1 ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 ).
13. The terminal device according to claim 11 or 12, wherein before the obtaining a preset training corpus, the processor further implements the following steps when the processor executes the computer-readable instructions:

    Building a point of interest database;

    Generating a supplementary corpus based on the point of interest information base;

    Combining the supplementary corpus with a preset basic corpus to obtain the training corpus.
14. The terminal device according to claim 13, wherein the constructing a point of interest information database comprises:

    Classifying a preset basic interest point according to a preset classification method to obtain a basic classification of the interest point information database;

    For each of the basic classifications, through a web crawling method, the point of interest information of all the basic points of interest in the basic classification in each administrative region of the country is obtained, and the information of the points of interest of the basic classification in each administrative region of the country is obtained.
15. The terminal device according to claim 13, wherein the generating a supplementary corpus based on the interest point information base comprises:

    Extracting the point of interest information in the point of interest information database;

    Performing word segmentation processing on the point of interest information to obtain the point of interest segmentation;

    Establishing a mapping relationship between the point of interest segmentation and the corresponding point of interest information, and correspondingly storing the point of interest segmentation, the point of interest information, and the mapping relationship in the supplementary corpus.
16. One or more non-volatile readable storage media storing computer readable instructions, characterized in that when the computer readable instructions are executed by one or more processors, the one or more processors are caused to execute The following steps:

    Obtain a preset training corpus;

    An N-gram model is used to analyze the preset training corpus to obtain word sequence data of the preset training corpus, wherein the word sequence data includes a word sequence and a word sequence frequency of each of the word sequences. degree;

    If the speech information to be identified is received, the speech information to be identified is parsed to obtain M pronunciation sequences of the speech information to be identified, where M is a positive integer greater than 1;

    For each of the pronunciation sequences, calculating the occurrence probability of each pronunciation sequence according to the word sequence data, thereby obtaining the occurrence probability of M pronunciation sequences;

    Selecting the pronunciation sequence corresponding to the occurrence probability reaching a preset probability threshold from the occurrence probabilities of M said pronunciation sequences as a target pronunciation sequence;

    The point of interest information corresponding to the target pronunciation sequence is obtained from the point of interest information database as a point of interest recognition result of the speech information to be recognized.
17. The nonvolatile readable storage medium according to claim 16, wherein, for each of the pronunciation sequences, calculating the occurrence probability of each pronunciation sequence according to the word sequence data comprises:

    For each of the pronunciation sequences, obtaining all the participles a ₁ , a ₂ , ..., an _n-1 , an _n within the pronunciation sequence, where n is a positive integer greater than 1;

    According to the word sequence data, the following formula is used to calculate the probability that the nth participle a _{n of the n} participles appears after the word sequence (a ₁ a ₂ ... a _n-1 ), and the probability is used as the pronunciation sequence Probability of occurrence:

    

    After probability _{| (a 1 a 2 ... a} n-1 a n) for the n th word n-th word appears in a _n word sequence _{(a 1 a 2 ... a n} -1) where, P , C (a ₁ a ₂ ... an _n-1 an _n ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 an _n ), C (a ₁ a ₂ ... a _n-1 ) is the word sequence frequency of the word sequence (a ₁ a ₂ ... an _n-1 ).
18. The non-volatile readable storage medium according to claim 16 or 17, wherein before the obtaining the preset training corpus, the computer-readable instructions are executed by one or more processors, so that all the The one or more processors also perform the following steps:

    Building a point of interest database;

    Generating a supplementary corpus based on the point of interest information base;

    Combining the supplementary corpus with a preset basic corpus to obtain the training corpus.
19. The non-volatile readable storage medium of claim 18, wherein the constructing a point of interest information database comprises:

    Classifying a preset basic interest point according to a preset classification method to obtain a basic classification of the interest point information database;

    For each of the basic classifications, through a web crawling method, the point of interest information of all the basic points of interest in the basic classification in each administrative region of the country is obtained, and the information of the points of interest of the basic classification in each administrative region of the country is obtained.
20. The non-volatile readable storage medium of claim 18, wherein the generating a supplementary corpus based on the point of interest information base comprises:

    Extracting the point of interest information in the point of interest information database;

    Performing word segmentation processing on the point of interest information to obtain the point of interest segmentation;

    Establishing a mapping relationship between the point of interest segmentation and the corresponding point of interest information, and correspondingly storing the point of interest segmentation, the point of interest information, and the mapping relationship in the supplementary corpus.

Images