CN102722499B - Search engine and implementation method thereof - Google Patents

Abstract

The present invention provides a search engine, which is used to receive an original query searched by a user; analyze the original query to obtain the original word existing in the original query and a synonym of the original word, and replace the synonym in the original query the original word to obtain the synonymous query; search and obtain the original query result webpage collection and the synonymous query result webpage collection according to the original query and the synonymous query result; calculate the coincidence degree of the webpage in the original query result and the synonymous query result ; Merge the result web page collections of the original query and the synonym query according to a predetermined merging strategy corresponding to the coincidence degree, and generate a search result list. The search engine determines the probability of escaping in synonymous query results by judging the coincidence degree of original query results and synonymous query results, and suppresses synonymous query results when the escaping probability is high, so as not to meet user search needs The results appear in the forefront of the search result list, so as to ensure that users have a good experience.

Description

Search Engine and Its Realization Method

technical field

The invention relates to search engine technology, in particular to a search engine capable of expanding synonymous search queries and its implementation method.

Background technique

The rapid development of the Internet provides people with a new carrier for information storage, processing, transmission and use, and network information has quickly become one of the main channels for people to obtain knowledge and information. While such a scale of information resources incorporates almost all the knowledge possessed by human beings, it also brings the problem of how to fully develop and utilize resources for users of resources. The search engine came into being precisely under this demand, and it assists network users to find information on the Internet. Specifically, search engines use specific computer programs to collect information from the Internet according to certain strategies, organize and process the information, provide users with search services, and display relevant information to users.

The online search service provided by the search engine is usually a keyword-based search, that is, the user enters a query expression through the input box of the search engine, and the search engine performs the query and returns the result webpage containing these keywords. Due to the different knowledge backgrounds or usage habits of different users, the keywords used to search for the same thing may be different. In addition, there are many synonyms or near synonyms in natural language, so it is wrong to search only based on the keywords provided by users. not enough. At present, many search engines have the function of expanding query, such as synonymously expanding query. When the search engine receives the original query expression input by the user, it will perform a word segmentation operation on it, and identify whether there are potential synonym pairs in the entry set after word segmentation. Specifically, the search engine will match the segmented entries with a predetermined thesaurus of synonyms, judge whether there are synonyms in these entries, and if so, expand the search query on the basis of synonyms, and use the expanded The query result is combined with the original query result and returned for display to the user. Thus, more relevant search results are provided for the user.

However, the same word may have different meanings in different semantic environments, so its synonyms are only synonymous or near-synonymous in a certain semantic environment, but in a different semantic environment, the synonym cannot be applied. Therefore, in this case, the result obtained by using the synonym expansion query may not be the result that the user wants, and thus, it will bring a poor user experience to the user. For example, the original query entered by the user is "how to make fish-flavored pork shreds". Subsequently, the search engine obtains the potential synonym pair of "how to do" {"how to do", "recipe"} through word segmentation of the original query and matching with the thesaurus, and replaces "how to do" with "recipe" to execute Expand the synonym query and get the corresponding query result. But if the original query provided by the user is "how to make a bedside table", obviously, the user's demand at this time is to understand the production of furniture, and the search engine still uses "recipe" to replace "how to do" to expand the synonymous query, it is Get escaped results that the user didn't intend, so the user will question the accuracy of the search.

In view of this, it is necessary to improve the existing search engines to solve the above problems.

Contents of the invention

The purpose of the present invention is to provide a search engine that adjusts the ranking of synonymous query results in the entire search results by identifying the escape probability of synonymous extended query results, thereby avoiding escape results appearing in the front row of search results, and then Make sure users have a good experience.

The object of the present invention is also to provide a method for realizing the above search engine.

In order to achieve one of the purposes of the above invention, a method for implementing a search engine of the present invention is characterized in that the method includes the following steps:

Receive the raw query that the user searches for;

analyzing the original query to obtain an original word present in the original query and synonyms of the original word, and replacing the original word in the original query with the synonyms to obtain a synonymous query;

Searching and obtaining the original query result webpage set and the synonymous query result webpage set according to the original query and the synonymous query;

calculating the coincidence degree of web pages in the original query result and the synonymous query result;

Merge the result webpage collections of the original query and the synonym query according to a predetermined merging strategy corresponding to the coincidence degree, and generate a search result list.

As a further improvement of the present invention, the calculation of the coincidence degree includes calculating the number |U1∩U2| of overlapping webpages in the original query result and the synonymous query result.

As a further improvement of the present invention, the calculation of the degree of coincidence also includes determining the lesser Min(|U1|, |U2|) in the number of webpages of the original query result and the number of webpages of the synonymous query result; I(U1, U2)=|U1∩U2|/Min(|U1|, |U2|).

As a further improvement of the present invention, the calculation of the degree of coincidence also includes calculating the sum |U1∪U2| of the number of webpages of the original query result and the number of webpages of the synonymous query result; U1∩U2|/|U1∪U2|.

As a further improvement of the present invention, the merging strategy includes: when the value of the coincidence degree is less than a predetermined threshold in the predetermined coincidence degree interval, the predetermined merging strategy is to perform suppression processing when merging synonymous query results, Described suppression processing comprises:

reduce the relevancy weight of web pages in the synonymous query results; or

Insert synonym query results after a particular page of the search results list; or

Adjust the synonym query results behind the original query results.

As a further improvement of the present invention, the merging strategy includes: when the coincidence degree value is greater than a predetermined threshold in a predetermined coincidence degree interval, merge the original and the result of a synonym query.

To achieve the above-mentioned another purpose, a search engine of the present invention includes a search component, and the search component includes:

a query analysis module, configured to receive an original query searched by a user; analyze the original query to obtain the original word in the original query and a synonym of the original word, and replace the original word in the original query with the synonym to obtain synonym query;

A search module, configured to search and obtain the original query result webpage collection and the synonymous query result webpage collection according to the original query and the synonymous query;

A coincidence degree calculation and result merging module, used to calculate the coincidence degree of webpages in the original query results and synonymous query results; and merge the original query and synonymous query result webpages according to a predetermined merging strategy corresponding to the coincidence degree collection, and generate a list of search results.

As a further improvement of the present invention, the calculation of the coincidence degree includes calculating the number |U1∩U2| of overlapping webpages in the original query result and the synonymous query result.

As a further improvement of the present invention, the calculation of the degree of coincidence also includes determining the lesser Min(|U1|, |U2|) in the number of webpages of the original query result and the number of webpages of the synonymous query result; I(U1, U2)=|U1∩U2|/Min(|U1|, |U2|).

As a further improvement of the present invention, the calculation of the degree of coincidence also includes calculating the sum |U1∪U2| of the number of webpages of the original query result and the number of webpages of the synonymous query result; U1∩U2|/|U1∪U2|.

As a further improvement of the present invention, the merging strategy includes: when the value of the coincidence degree is less than a predetermined threshold in the predetermined coincidence degree interval, the predetermined merging strategy is to perform suppression processing when merging synonymous query results, Described suppression processing comprises:

reduce the relevancy weight of web pages in the synonymous query results; or

Insert synonym query results after a particular page of the search results list; or

Adjust the synonym query results behind the original query results.

As a further improvement of the present invention, the merging strategy includes: when the coincidence degree value is greater than a predetermined threshold in a predetermined coincidence degree interval, merge the original and the result of a synonym query.

Compared with the prior art, the beneficial effect of the present invention is: the search engine determines the probability of escaping in the synonymous query results by judging the coincidence degree of the original query result and the synonymous query result, and when the escaping probability is relatively high , to suppress synonymous query results, so as to avoid the results that do not meet the user's search needs from appearing in the forefront of the search result list, so as to ensure that users have a good experience.

Description of drawings

Fig. 1 is the working principle block diagram of the first embodiment of the search engine of the present invention;

Fig. 2 is the workflow diagram of the search engine mining synonymous context shown in Fig. 1;

Fig. 3 is a workflow diagram of the search engine shown in Fig. 1 performing synonymous expansion query;

Fig. 4 is the working principle block diagram of the second embodiment of the search engine of the present invention;

Fig. 5 is the working flow chart that the search engine shown in Fig. 4 executes synonymous expansion query;

Fig. 6 is a working principle block diagram of the third embodiment of the search engine of the present invention;

Fig. 7 is the workflow diagram of the search engine shown in Fig. 6 executing synonymous expansion query;

Fig. 8 is a working principle block diagram of the fourth embodiment of the search engine of the present invention;

Fig. 9 is a workflow diagram of the search engine shown in Fig. 8 performing synonymous expansion query;

Fig. 10 is a working flow chart of a specific embodiment in which the search engine shown in Fig. 8 judges the similarity level of synonyms and marks the synonyms accordingly.

Detailed ways

The present invention will be described in detail below in conjunction with various embodiments shown in the drawings. However, these embodiments do not limit the present invention, and any structural, method, or functional changes made by those skilled in the art according to these embodiments are included in the protection scope of the present invention.

FIG. 1 is a block diagram of the working principle of the first embodiment of the search engine 100 of the present invention. In this embodiment, the search engine 100 collects web pages from the Internet according to a certain strategy, and after organizing and processing the web pages, it can respond to the request of the browser 21 of the client 20 to provide search query services. Wherein, the search engine 100 may include one or more network server entities for storing and managing data and responding to search requests. Client 20 may include one or more user terminal devices, such as personal computers, notebook computers, wireless telephones, personal digital processing (PDA), or other computing devices and communication devices.

These servers and terminal devices all include some basic components in architecture, such as a bus, a processing device, a storage device, one or more input/output devices, and communication interfaces. A bus may include one or more conductors for communication between components of a server or terminal device. Processing devices include various types of processors or microprocessors for executing instructions, processing processes or threads. The storage device may include dynamic memory such as random access memory (RAM) storing dynamic information, static memory such as read only memory (ROM) storing static information, and mass storage including magnetic or optical recording media and corresponding drives. The input device is for the user to input information to the server or terminal equipment, such as keyboard, mouse, stylus, voice recognition device, or biometric device. The output device includes a display, a printer, a speaker, etc. for outputting information. The communication interface is used to enable the server or terminal equipment to communicate with other systems or devices. The communication interfaces can be connected to the network through a wired connection, a wireless connection, or an optical connection, so that the search engine 100 and the client 20 can communicate with each other through the network. The network may include a Local Area Network (LAN), a Wide Area Network (WAN), a telephone network such as the Public Switched Telephone Network (PSTN), an intranet, the Internet, a combination of these networks, or the like. Both the server and the terminal equipment contain operating system software used to manage system resources and control the operation of other programs, as well as application software or program instructions used to realize the functions of specific functional modules.

As shown in FIG. 1 , the search engine 100 can perform synonymous expansion query, which can be divided into an offline part and an online part as a whole. In the offline part, the search engine 100 includes a data repository 12 that can store web page data and synonym pair information, an indexer 13, a web crawler 14, a user query log database 16 that records user query information, and analyzes the user query log. Log Analyzer 17.

The webpage crawler 14 is a program that crawls webpages one by one through the hyperlink relationship between webpages according to a certain strategy. In a specific embodiment, the web crawler 14 selects the URL to be grabbed from the initial URL (Universal Resource Locator, Uniform Resource Locator) library according to a certain scheduling policy, resolves the web server address indicated in the URL, and then Establish a connection, send a request, and receive data, store the obtained webpage data in the webpage storehouse 122 of the data repository 12 and set up a local document collection, and then extract links therefrom to carry out the next step of grabbing, and so on until All URLs are crawled. The scheduling strategy based on which the webpage crawler 14 selects URLs may include breadth-first strategy, depth-first strategy, reverse link number strategy, etc.; the crawling method may be cumulative crawling or incremental crawling. The indexer 13 is used for analyzing and indexing the local document collection. For example, extract entries from the full text of the document through word segmentation, then filter out high-frequency words or low-frequency words to obtain a set of index words, and finally convert the mapping from web pages to index words into mappings from index words to web pages to form a list containing index words and the posting file of the posting list are stored in the index library 121 of the data storage library 12. Methods for segmenting webpage documents include dictionary-based word segmentation methods, comprehension-based word segmentation methods, and statistics-based word segmentation methods. Among them, the more common dictionary-based word segmentation methods include forward maximum segmentation method, reverse maximum segmentation method, and minimum segmentation method.

In the present invention, synonyms refer to entries with different names but with the same or similar meanings, that is, if multiple entries express the same or similar meanings, these entries are mutually synonymous. In this embodiment, the thesaurus 123 includes a synonym correspondence table 1231 and a synonymous context database 1232 . The synonyms correspondence table 1231 pre-designates correspondences between different words and their synonyms, such as a correspondence table between original words and their synonyms obtained through pre-statistics. The correspondence table can also be continuously updated by analyzing the user's historical query and click data. For example, when the title of the clicked query result webpage contains a synonym of an original word but does not appear the original word, and this situation occurs frequently, the original word and the synonym are determined as a synonym pair and added to the synonym Correspondence table 1231.

FIG. 2 shows the workflow of a specific implementation of the search engine 100 mining the synonymous context of synonym pairs. In the present invention, the synonym context refers to the semantic environment in which the original word appears in the synonym pair, which is used to indicate the semantic environment to which the synonym pair is applicable, that is, in this semantic environment, the synonym is suitable for replacing the original word for synonymous expansion Inquire. In this embodiment, the synonymous context is obtained by analyzing user query logs. The user query log database 17 is used to record the user's query and click data after each search, such as query expression, search time, returned result list, and clicked result webpage. Referring to FIG. 2 and shown in FIG. 1 , the log analyzer 17 analyzes historical user queries and click data contained in the user query log database 16 (step 411), including analyzing historical queries and responding to specific queries. The query result web page returned and clicked to visit. Next, the log analyzer 17 will identify whether there is a synonymous context of a synonym pair in these data, and if so, record and store it in the synonymous context database 1232 .

Specifically, the log analyzer 17 first judges whether a historical query contains the original word based on the synonym correspondence table 1231, and if so, obtains a synonym pair including the original word and the corresponding synonym. For example, the historical query formula is "how to make fish-flavored pork shreds", and the log analyzer 17 judges based on the synonym correspondence table 1231 that there is an original word of "how to do it" in the query formula (segmenting "how to do fish-flavored pork shreds" into The two entries of "fish-flavored shredded pork" and "how to do it", and then match these two entries with the original words in the synonym correspondence table, so as to find the original word of "how to do it"), and obtain the corresponding synonyms Right {"How To", "Recipe"}. Subsequently, the log analyzer 17 judges whether, for the query formula, whether the title of the web page accessed by the user contains synonyms but does not contain the original word, and if so, records the synonymous context of the synonym pair. For example, for the query "how to make fish-flavored shredded pork", if the user has clicked on a webpage titled "recipe for fish-flavored pork shreds", the log analyzer 17 will perform the operation of recording the synonymous context. The synonymous context contains at least the historical query formula, such as "how to make fish-flavored pork shreds"; it can also contain the close words of the original word in the historical query formula, such as "fish-flavored pork shreds"; or both are recorded as synonyms Synonymous context for {"how to", "recipe"}. Wherein, the immediately adjacent word can be located before or after the original word; the immediately adjacent word can also be an empty entry, that is, the original query only contains the original word, and there is no immediately adjacent word.

In the above embodiments, the synonymous context is obtained through historical user behavior, but in other embodiments, the synonymous context can also be determined according to the anchor text of the webpage. Anchor text is the text information contained in the hyperlink of the web page. For example, if the hyperlinked text of the website www.sina.com.cn is quoted includes "Sina.com homepage", "Sina homepage", "sina homepage", then these text fields will be recorded as synonym pairs {"Sina Net", "Sina"} synonymous context. In addition, synonymous context can also be determined from the parallel paragraph in the title of the web page. For example, price.mycar168.com/search.asp? factoryid=135 The title of this URL is "BMW Brilliance Quotation, BMW Brilliance Price·Shenzhen Auto World Network". Then, the title can be divided into multiple juxtaposed entry segments "BMW Brilliance Quotation", "BMW Brilliance Price", "Shenzhen Auto World Network" through separators, and the first two segments contain synonym pairs {"price", "price" and "quote" in "quote"}, then these two fragments can also be used as the synonymous context of the synonym pair.

Referring to Figure 2, in the process of synonymous context mining, the user's click behavior is not necessarily completely reasonable, that is to say, the user may unintentionally click some irrelevant results in the process of browsing the search results , in which case the synonymous context recorded would not be accurate. Therefore, in order to eliminate the negative impact caused by this situation, the log analyzer 17 also counts the recorded frequency of the synonymous context, and only when the frequency is greater than or equal to a predetermined frequency threshold, the synonymous context will be Keep the synonymous contexts determined as the corresponding synonymous word pairs, that is, filter out the synonymous contexts with low frequency (step 413).

As shown in FIG. 1 , the online portion of the search engine 100 mainly includes a search component 11 and a user interface 15 . Wherein the user interface 15 is presented by the browser software 21 of the client 20, and is used for inputting the query formula for the user, and displays the search result list according to a predetermined display mode; in addition, after the search is finished, it is also used for recording the user's query information, and It is stored in the user query log database 16. The search component 11 is used to respond to the search request of the client 30 and return the search result to the client 20 . In this embodiment, the search component 11 includes a search module 111 , a query analysis module 112 , and a result synthesis module 113 . For ordinary original queries (excluding extended queries), the query analysis module 112 is generally used to perform word segmentation on the currently received original query, obtain a set of query words, and generate a query vocabulary. After receiving the query word list, the search module 111 matches the index word list in the data index database 121, finds the corresponding index words and the posting list corresponding to each index word, and obtains a collection of webpage documents related to the query words . The result synthesis module 113 arranges the searched webpage documents according to the predetermined correlation weight between each document and the query word, and then returns the result list to the client through the user interface 15 .

The detailed steps for the search engine 100 to perform synonymous expansion query online according to the synonymous context will be described below in conjunction with the workflow shown in FIG. 3 . The query analysis module 112 receives the original query searched by the current user through the user interface 15 (step 421 ), and then analyzes the query formula (step 422 ), including performing word segmentation on the original query. It should be noted that the word segmentation method in this embodiment is based on the forward maximum segmentation method of the dictionary, and the dictionary is constructed from the fragments of entries contained in the synonymous context. As mentioned above, the historical query will be recorded as a synonymous context, and the fragment length of the historical query is greater than the length of the segmented entry of the query. Therefore, the forward maximum segmentation method can ensure that Once the current original query contains a historical query fragment, the fragment will be segmented first, thereby improving the accuracy of subsequent calculations. For example, in the synonymous context mining stage, the historical query type is "how much is Nokia today", then when recording the synonymous context of the synonym pair {"how much money", "price"}, the historical query type "how much is Nokia today" Money" and the adjacent word "Nokia" will be recorded as synonymous context. And the current original query is "who knows how much is Nokia today", according to the forward maximum segmentation method, the length of the longest segment "how much is Nokia today" in the synonym context dictionary is 8, then the query analysis module 112 starts from the left Scan the current original query to the right to determine whether the phrase with a length of 8 appears in the synonymous context dictionary. When it finds a match for "how much is Nokia today", it will be segmented first, so that "Nokia" is It will not be segmented as a separate keyword. In step 422, the query analysis module 112 will also match the entry set obtained after the original query is segmented with the thesaurus 123 to obtain a potential synonym pair and the synonymous context of the synonym pair. The potential synonym pair contains the existing The original term contained in the original query, and the corresponding synonyms for the original term.

Next, the query analysis module 112 judges whether the synonymous context matches the original query (step 423). In this embodiment, the query analysis module 112 will calculate the matching degree of the synonymous context and the original query, and when the value of the matching degree is within a predetermined matching degree interval, it is determined that the synonymous context and the original query match, that is Indicates that the semantic environment of the current original query is suitable for performing extended queries by replacing the original words with synonyms. The calculation of the matching degree can be determined according to the length of the original query formula after removing the original word, and the length of the synonymous context. The following is the formula for calculating the matching degree M in this embodiment when the length of the original query is greater than the length of the original word (that is, q≠orig):

$\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; orig</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; syn</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; TermCount</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; TermCount</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; TermCount</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; orig</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; orig</span>}$

Where TermCount(q) represents the length of the original query, TermCount(orig) represents the length of the original word in the original query, and TermCount(pi) represents the length of the i-th synonymous context. Because in this case, there will be words in non-synonymous contexts in the original query, so M is a value between [0, 1]. A matching degree threshold θ is set in advance, and when the value of M is [θ, 1], it indicates that the synonymous context matches the original query, and the synonymous words are replaced with the original words in the original query to obtain the synonymous query, and then search Module 111 searches and obtains the collection of web pages of original query results and the collection of webpages of synonymous query results according to the original query and synonymous query (step 424), and the result synthesis module 113 merges the results of original query and synonymous query according to a predetermined merging strategy (step 424). 425). The result merging strategy will be described in detail later. When the value of M is [0, θ], it indicates that the synonym context does not match the original query, that is, in this semantic environment, it is not suitable to replace the original word with a synonym, and then the search module 111 will only search according to the original query Execute the search and obtain the webpage set of the original query result (step 426), and then the result synthesis module 113 obtains the search result list according to the predetermined correlation weight between each webpage and the original query (step 425). When the original query only contains the original word (ie q=orig) and the matching degree M=1, replace the original query with between synonyms, and then execute steps 424 and 425.

The search engine analyzes the semantic environment of the current user's query needs to determine whether the synonym transformation is applicable to execute the synonymous expansion query, so as to ensure the accuracy of the synonymous expansion query, make the expansion query meet the user's needs as much as possible, and then ensure that the user has a good use experience.

Figure 4 and Figure 5 reveal the second embodiment of the search engine of the present invention. Compared with the first embodiment, the search engine 200 in this embodiment mainly adjusts the position of the synonymous query results in the search result list presented to the user by judging the escape probability of the synonymous query results. As shown in FIG. 4 , the search engine 200 includes a search component 11 , a data repository 12 , an indexer 13 , a crawler 14 , and a user interface 15 . Functional modules such as data storage 12 , indexer 13 , grabber 14 , and user interface 15 are basically the same as those in the above-mentioned embodiment, so the applicant will not repeat them here. In this embodiment, the search component 11 includes a search module 111 , a query analysis module 112 , and a coincidence degree calculation and result combination module 114 .

The synonymous extension query performed by the search engine in this embodiment will be described in detail below in conjunction with FIG. 5 . First, the query analysis module 112 receives a user's original query (step 431). Next, analyze the query formula (step 432), including performing word segmentation on the original query to obtain a set of query words, identify the original words in the original query based on the thesaurus 123 and obtain a synonym pair comprising the original word and its synonyms, and directly Synonyms replace original words to obtain synonymous queries. The search module 111 searches the original query and the synonymous query to obtain a web page set of original query results and a web page set of synonymous query results (step 433 ). Next, the coincidence degree calculation and result merging module 114 calculates the coincidence degree of webpages in the original query result and the synonymous query result (step 434 ). The coincidence degree is mainly used to reflect the number of identical webpages in the original query results and synonymous query results. If the number of identical webpages is large enough, it indicates that the synonymous query results are relatively close to the original query results, and the synonymous query results will appear in turn. On the contrary, it indicates that the probability of escape in synonymous query results is high, and it is necessary to suppress the synonymous query results to prevent the results that do not meet the user's search requirements from appearing in the forefront of the result list.

The coincidence degree can be calculated in many ways, such as only calculating the number of overlapping webpages |U1∩U2| in the original query results and synonymous query results, that is, determining the number of identical URLs; or calculating the top 100 in each of the two result sets The number of overlapping webpages of each result, and then compare and judge with the predetermined threshold. As a preferred mode, the calculation of the coincidence degree also includes determining the lesser Min(|U1|, |U2|) in the number of webpages of the original query result and the number of webpages of the synonymous query result; then the degree of coincidence I(U1, U2 )=|U1∩U2|/Min(|U1|, |U2|). Or in other implementations, the calculation of the coincidence degree also includes calculating the sum |U1∪U2| of the number of webpages of the original query result and the number of webpages of the synonymous query result; then coincidence degree I(U1, U2)=|U1∩U2 |/|U1∪U2|. When the value of the coincidence degree is calculated, it will judge whether the value is within the predetermined coincidence degree range to determine whether to suppress the synonymous query results, and then determine the position of the synonymous query results in the search result list and output the merged results (step 435). Taking the coincidence degree calculation method I(U1, U2)=|U1∩U2|/Min(|U1|, |U2|) as an example, the value I of the coincidence degree is a floating point number between [0, 1]. Pre-set a coincidence degree threshold σ, when I is [σ, 1], it indicates that the coincidence degree between the original query result and the synonymous query result is high. In this case, there is no need to suppress the synonymous query result, just The results of the original and synonymous queries are combined according to a predetermined relevance weight for each web page. When I is in [0, σ], it indicates that the coincidence degree between the original query result and the synonymous query result is low, and the escaping probability of the synonymous query result is high, so it is necessary to suppress the synonymous query result. The method of suppressing can be to lower the weight of the relevance weight of the web pages in the synonymous query results, so that the synonymous query results are in a lower position in the combined search result list; or insert the synonymous query results into the After a specific page in the search result list, such as adjusting the synonymous query result to the second page of the search result list; in addition, you can also adjust the synonymous query result to the back of the original query result, that is, the synonymous query result appears in the search result at the end of the list.

The search engine determines the probability of escaping in synonymous query results by judging the coincidence degree of original query results and synonymous query results, and suppresses synonymous query results when the escaping probability is high, so as not to meet user search needs The results appear in the forefront of the search result list, so as to ensure that users have a good experience. In this embodiment, before replacing the original word with a synonym to perform a synonym expansion query, it is not necessary to use the synonymous context judgment mentioned in Embodiment 1 to determine whether to perform synonym replacement. However, those of ordinary skill in the art It can be easily imagined that if this embodiment is combined with the first embodiment, that is, before the synonyms are replaced, the judgment of the synonymous context is performed first, and then after the synonymous query results come out, the search is combined according to the degree of coincidence between the original and synonymous query results As a result, it is clear that more accurate search results can be obtained in this way, thereby further improving the user experience.

Figure 6 and Figure 7 reveal the third embodiment of the search engine of the present invention. This embodiment is based on the synonymous query results, and further judges the escape probability of the synonymous query results by analyzing the semantic topic distribution of the synonymous query result web pages, and then adjusts the synonymous query results in the search result list that is finally presented to the user. Location. As shown in FIG. 6 , similar to the first embodiment, the search engine 300 includes a search component 11 , a data repository 12 , an indexer 13 , a grabber 14 , a user interface 15 , a user query log database 16 , and a log analyzer 17 . The functional modules such as indexer 13 , grabber 14 , user interface 15 , user query log database 16 , and log analyzer 17 are the same as those in the first embodiment, and the applicant will not repeat them here. In this embodiment, the search component 11 includes a search module 111 , a query analysis module 112 , a result synthesis module 113 , and an escape judgment module 115 . The data repository 12 includes an index repository 121 , a webpage repository 122 , a thesaurus 123 , and a webpage semantic theme repository 124 . The index library 121 , the web page library 122 , and thesaurus 123 are the same as those in the first embodiment, and the applicant will not repeat them here. The search engine 300 also includes a topic analysis module 18. In this embodiment, the topic analysis module 18 includes a probabilistic latent semantic analysis (Probabilitistic Latent Semantic Analysis, hereinafter referred to as PLSA) model.

The PLSA model is a tool for natural language processing, which is mainly used to analyze the latent semantics of documents. A document can be represented as a set of words, but due to the existence of synonyms, words are not the most basic elements of the document, so it can be considered that there is a potential semantic level between words and documents, that is, topics. For example, the query type entered by the user is "Swiss Army Knife Green", since {"Green", "Green"} is a synonym pair, so you can replace "Green" with "Green" to perform a synonymous expansion query, but this The results of the recall may contain a web page titled "System Swiss Army Knife - Perfect Uninstall V2007 Green Edition". This is because the theme corresponding to "Swiss Army Knife Green" is "item", and the theme corresponding to "System Swiss Army Knife - Perfect Uninstall V2007 Green Edition" is "software". Obviously, search engines cannot understand these implicit themes. The PLSA model is a topic model that analyzes latent semantic topics by calculating the distribution of co-occurring words in documents. It introduces a latent semantic layer between documents and words, and the latent semantic layer is composed of n latent semantic topics. Assuming that documents and words are independent of each other, the probability of co-occurrence of documents and words is determined by their probabilistic relationship with topics. Therefore, the relationship between documents or words and latent semantic topics can be calculated through the PLSA model. Based on this, in this embodiment, the semantic topic distribution of the synonymous context and the synonymous query result webpage can be obtained through the PLSA model, and the matching degree of the two can be calculated to determine the escape probability of the synonymous query result. A detailed description will be given next.

As shown in FIG. 6 , the topic analysis module 18 acquires a webpage from the webpage library 122 , removes distracting words such as border advertisements in the webpage, and then extracts a set of keywords that can represent the webpage. Subsequently, the topic analysis module 18 calculates and obtains the webpage-latent semantic topic vector S2={s21, s22,...,s2n} representing the semantic topic distribution of the webpage through the PLSA model, where s2n indicates that the webpage is in the nth semantic topic The probability score on . In this embodiment, the acquisition of the semantic topic distribution of the webpage is obtained offline, that is, the topic analysis module 18 analyzes all captured webpages to obtain the semantic topic distribution, and then stores it in the webpage semantic topic database 124 . Of course, this process can also be obtained in the state of online search, that is, after the synonymous query results are obtained, the topic analysis module 18 only analyzes the webpages in the query results, and then gives the semantic topic distribution of these webpages to the escape judgment module 115 to judge. In this embodiment, the acquisition of the semantic topic distribution of the synonymous context is realized online. After the query analysis module 112 divides the original query to obtain a keyword set, the topic analysis module 18 obtains the keyword set, and obtains a corresponding set of terms contained in the synonymous context from the synonymous context database 1232 . Then, combine the keyword set and the entry set of the synonymous context, submit it to the PLSA model to calculate and obtain the synonymous context-latent semantic topic vector S1={s11 , s12,...s1n}, where s1n refers to the probability value of the synonymous context on the nth semantic topic. After getting the vector S1, the topic analysis module 18 sends it to the escape judging module 115 to judge the similarity between S1 and S2. The steps of judging will be described in detail later.

Next, the detailed steps of the search engine 300 executing the synonymous expansion query in this embodiment will be described in detail with reference to FIG. 7 . First, the query analysis module 112 receives the original query searched by the user (step 441), and then analyzes the original query (step 442). The query analysis module 112 will perform a word segmentation operation on the original query. Like the first embodiment, the word segmentation operation is based on the dictionary constructed by the synonymous context for maximum forward segmentation. After the word segmentation operation, the original keyword set is obtained. On the one hand, the query analysis module 112 sends the original keyword set to the search module 111 to execute the original query (step 449), and obtain the original query result (step 450). On the other hand, the query analysis module 112 identifies the original words included in the original query based on the thesaurus 123 , and obtains corresponding potential synonym pairs and synonymous contexts of the potential synonym pairs. After the analysis and query module 112 obtains the above data, it can directly replace the original words with synonyms to obtain a synonym query, and hand it over to the search module 111 to perform a synonym expansion query (step 443 ). As a preferred embodiment, before performing the synonym replacement operation, it is possible to judge whether it conforms to the synonymous context of the original word, and if so, perform the synonymous replacement operation, which can further improve the accuracy of the synonymous query result . The operation of performing synonym replacement according to the matching degree judgment of the synonymous context has been described in detail in the first embodiment, and the applicant will not repeat it here. In addition, the query analysis module 112 also submits the original keyword set to the topic analysis module 18, which calculates and obtains the semantic topic distribution of the synonymous context through the PLSA model (step 447), and submits the calculated result to the escape judgment module 115 .

After the search module 111 executes the synonymous query to obtain the synonymous query result (step 444), the escape judgment module 115 obtains the semantic topic distribution of the result webpage from the webpage semantic topic database according to the synonymous query result, that is, the vector of webpage-latent semantic topic S2={s21, s22, . . . , s2n} (step 445). On the other hand, the escape decision module 115 obtains the semantic topic distribution of the synonymous context from the topic analysis module, that is, the synonymous context-latent semantic topic vector S1={s11, s12,...s1n}, and then , escape judging module 115 judges the degree of matching of two semantic theme distributions, promptly calculates the similarity degree of two vectors S1, S2 (step 446); Then filter the synonymous query result (step 448) according to degree of matching, promptly determine synonymous The method of suppressing the query results, and combining the results of the original query and synonymous queries accordingly to generate a search result list (step 451). There are many ways to calculate the similarity between two vectors, such as inner product similarity, cosine similarity, etc. The following is an example of a calculation formula for calculating the similarity between vectors S1 and S2 using cosine similarity.

$\mathrm{<span\; class="notranslate">\; sim</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; i</span>}}}{\sqrt{\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}$

If the calculated similarity value is very high, it indicates that the webpage and the synonymous context have a high probability on the nth semantic topic, then it can be judged that the two semantic topic distributions have a high matching degree, that is, the escape probability of the webpage On the contrary, if the calculated similarity value is very low, it indicates that the escape probability of the web page is relatively high, so the result needs to be suppressed. Specifically, the value sim(S1, S2) of the similarity is a floating point number between [0, 1]. A threshold α can be set in advance. When sim(S1, S2) is in [α, 1], it indicates that the matching degree of the two semantic topic distributions is high. In this case, there is no need to suppress the synonymous query results, only The results of the original and synonymous queries need to be combined according to the predetermined relevance weight of the web pages. When sim(S1, S2) is in [0, α], it indicates that the matching degree of the two semantic topic distributions is low, and the escaping probability of the synonymous query results is high, so it is necessary to suppress the synonymous query results. The method of suppressing can be to lower the weight of the relevance weight of the synonymous query result webpage, so that the synonymous query result is in a lower position in the combined search result list; or insert the synonymous query result into the search After a specific page of the result list, such as adjusting the synonymous query result to the second page of the search result list; in addition, the synonymous query result can also be adjusted to the back of the original query result, that is, the synonymous query result appears in the search result list Last.

By comparing the matching degree between the synonymous context and the semantic topic distribution of the synonymous query result webpage, the search engine can judge whether the synonymous query result meets the potential needs of users, so that it can control the synonymous query result in the whole search results accordingly. The sorting in the list avoids escaping results at the top of the search results, thereby ensuring a good user experience. In addition to the PLSA model introduced in the above embodiments, other topic models can also be used to analyze the potential semantic topics of synonymous context and synonymous query result web pages, such as latent semantic analysis (Latent Semantic Analysis, LSA) model, or latent Dirichlet Allocation (Latent Dirichlet Allocation, LDA) model, etc.

8 to 10 disclose the fourth embodiment of the search engine of the present invention. This implementation manner mainly describes the presentation of synonyms in the search results. FIG. 8 shows a block diagram of the working principle of the search engine 400 , which includes a search component 11 , a data repository 12 , an indexer 13 , a crawler 14 , and a user interface 15 . Functional modules such as data storage 12 , indexer 13 , grabber 14 , and user interface 15 are basically the same as those in the above-mentioned embodiment, so the applicant will not repeat them here. In this embodiment, the search component 11 includes a search module 111, a query analysis module 112, a result synthesis module 113, an analysis module 116 for analyzing the similarity level between synonyms and original words, and a labeling module 117 for determining the display mode of synonyms.

The synonymous extension query performed by the search engine in this embodiment will be described in detail below in conjunction with FIG. 9 . First, the query analysis module 112 receives the original query searched by the user (step 461), and then analyzes the original query (step 462). The query analysis module 112 performs a word segmentation operation on the original query to obtain the original keyword set. The query analysis module 112 identifies the original word included in the original query based on the thesaurus 123 , and obtains a synonym pair including the original word and its synonyms. On the one hand, the analysis query module 112 replaces the original words with synonyms to obtain a synonym query, and then the search module 111 executes the original query and the synonym expansion query according to the original query and the synonym query (step 463 ). After the search module 111 acquires the original query result and the synonym query result, the result synthesis module 113 will merge and generate a search result list (step 464 ). The method for merging original and synonymous queries has been described in detail in the above embodiments, and the applicant will not repeat them here. On the other hand, the query analysis module 112 sends the synonym pair to the similarity level analysis module 116 , which judges the similarity level between the synonym and the original word (step 465 ), and sends the judgment result to the labeling module 117 . Next, the tagging module 117 determines the display mode of synonyms according to the judgment result of the similarity level, and finally displays the tagged search result list to the user through the user interface 15 (step 466 ).

The judgment of the similarity level between the synonym and the original word and the corresponding display manner are further illustrated below with reference to FIG. 10 . Similarity level analysis module 116 obtains synonym pair (step 471) from query analysis module 112, at first judges whether the synonym in the synonym pair and original word belong to high similarity level (i.e. the higher first level of similarity level) (step 472 ). In this embodiment, the situation where synonyms and original words belong to high similarity level includes proper noun abbreviation (such as "Peking University" and "Peking University", "Sina" and "sina"), or digital transformation (such as "fifth "Set" and "Fifth Set"), or regional word conversion (such as "Beijing" and "Beijing"), etc. If it belongs to a high similarity level, then the synonym is marked with a specific color (step 473), usually the specific color is a more eye-catching color, such as red in the present embodiment; and whether the original word belongs to the middle similarity level (ie the lower second level of the similarity level) (step 474). In this embodiment, the determination of the similarity level between the original word and the synonym includes the determination of semantic similarity or morphological similarity.

The following is a specific example of the semantic similarity calculation formula:

$\mathrm{<span\; class="notranslate">\; SSim</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; orig</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; syn</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; ClickQueryCount</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; orig</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; syn</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; QueryCount</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; orig</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>$

Among them, ClickQueryCount(orig, syn) indicates that the original word orig is included in the query formula, and the title of the web page that is clicked at the same time does not contain the original word orig but contains the synonym syn; QueryCount(orig) indicates that the original word orig is included in the query formula of historical queries. For example, if the historical query type entered by the user is "Where is Peking University", and then clicks the webpage titled "Where is Peking University" in the search results, then this query will be accumulated to ClickQueryCount(orig, syn) and QueryCount( orig); and if the user just clicks on the webpage titled "Where is Peking University" in the search results for the historical query "Where is Peking University", then this query will only be accumulated to QueryCount(orig). Obviously, the value of the semantic similarity is a floating point number between [0, 1]. A threshold β can be set in advance, when the value of semantic similarity is in [β, 1], it indicates that the original word and the synonym belong to the middle similarity level; and when the value of semantic similarity is in [0, β], then Next, word form similarity will be judged. If it has been determined that the synonym pair belongs to the medium similarity level, the synonym is marked with a specific font (step 475 ), such as bold or italic, which is bold in this embodiment.

The following is a specific example of the word form similarity calculation formula:

$\mathrm{<span\; class="notranslate">\; WSim</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; orig</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; syn</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; CoocAlphaCount</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; orig</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; syn</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; AllAlphaCount</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; orig</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; syn</span>}<span\; class="notranslate">\; )</span>}$

Among them, CoocAlphaCount(orig, syn) indicates how many words are the same in the original word orig and the synonym syn, and AllAlphaCount(orig, syn) indicates the total number of different words contained in the original word orig and the synonym syn. For example: for the synonym pair {"how", "how"}, CoocAlphaCount("how", "how")=2, because the words "how" and "what" appear in the original word and "how" at the same time in the synonym pair Among the synonyms; AllAlphaCount(orig, syn)=3, because there are 3 different words "how", "what" and "like" in the synonyms. For English, count the number of letters, for example: for synonym pairs {"man", "men"}, CoocAlphaCount("man", "men")=2, and AllAlphaCount("man", "men")=4 . Obviously, the value of word form similarity is also a floating point number between [0, 1]. A threshold γ can be preset. When the value of the semantic similarity is [γ, 1], it indicates that the original word and the synonym belong to the medium similarity level, and the labeling module 117 will mark the synonym in bold; and when the value of the semantic similarity When it is [0, γ], it indicates that the synonym and the original word in the synonym pair belong to a low similarity level (that is, the third level of similarity level is lower than the second level), so the synonym is not marked (step 476). Compared with the label of a specific color, the eye-catching degree of a specific font is weaker, but it can still attract users' attention, so it is suitable for synonyms with a medium similarity level, because although its semantics or word form have changed, it is still the same as the original word Relatively close; and synonyms with low similarity levels have a relatively large gap between semantics or word form and the original word, if marked, it will bring abruptness to the user; so the preferred way is not to mark.

By distinguishing the similarity level between synonyms and original words, the search engine can appropriately mark the synonyms in the search results, so that users can quickly locate the information they need while avoiding giving users a sense of abruptness, thereby improving user experience. experience.

Those skilled in the art can easily imagine that the way of judging the similarity level of synonyms, the way of displaying synonyms, and the corresponding relationship between different similarity levels and different ways of displaying are not limited to those described in the above embodiments. For example, the similarity level can also be judged by the edit distance, or a way of highlighting synonyms can be used. In addition, more similarity levels can be set, such as splitting semantic similarity and morphological similarity into two different levels. Of course, the similarity level can also be reduced, that is, all synonyms are only classified into high similarity level or low similarity level. For example, when the synonym and the original word belong to proper noun abbreviation, digital transformation, or regional word transformation; or when the semantic similarity, morphological similarity, or edit distance between the original word and the synonym is greater than or equal to the specified threshold, it can be considered as high similarity degree level, and the rest are low similarity level.

It should be understood that although this description is described according to implementation modes, not each implementation mode only contains an independent technical solution, and this description in the description is only for clarity, and those skilled in the art should take the description as a whole, and each The technical solutions in the embodiments can also be properly combined to form other implementations that can be understood by those skilled in the art.

The series of detailed descriptions listed above are only specific descriptions for feasible implementations of the present invention, and they are not intended to limit the protection scope of the present invention. Any equivalent implementation or implementation that does not depart from the technical spirit of the present invention All changes should be included within the protection scope of the present invention.

Claims (12)

1. an implementation method of a search engine, characterized in that the method comprises the steps: Receive the raw query that the user searches for; analyzing the original query to obtain an original word present in the original query and synonyms of the original word, and a synonymous context of the original word and the synonyms, and based on the original query and the synonyms environment, the synonym is replaced by the original word in the original query to obtain a synonymous query; Searching and obtaining the original query result webpage set and the synonymous query result webpage set according to the original query and the synonymous query; calculating the coincidence degree of web pages in the original query result and the synonymous query result; Merge the result webpage sets of the original query and the synonym query according to a predetermined merging strategy corresponding to the coincidence degree, and generate a search result list; wherein, The synonymous context is mined using the following steps: Acquiring historical user query and click data, said data including historical query and query result web pages returned in response to the query and accessed by clicking; identifying synonym pairs comprising an original term present in the historical query and a corresponding synonym present in the query results web page; At least the historical query is recorded and determined as a synonym context for the synonym pair. 2. The method for implementing a search engine according to claim 1, wherein the calculation of the coincidence degree comprises calculating the number |U1∩U2| of overlapping web pages in the original query result U1 and the synonymous query result U2. 3. the realization method of search engine according to claim 2, is characterized in that, the calculation of described degree of coincidence also comprises determining less one Min(| U1|, |U2|); the coincidence degree I(U1, U2)=|U1∩U2|/Min(|U1|, |U2|). 4. The method for realizing a search engine according to claim 2, wherein the calculation of the coincidence degree also includes calculating the sum |U1∪U2| of the number of webpages of the original query result and the number of webpages of the synonymous query result; The coincidence degree I(U1, U2)=|U1∩U2|/|U1∪U2|. 5. The method for implementing a search engine according to claim 1, wherein the merging strategy comprises: when the value of the degree of coincidence is less than a predetermined threshold in a predetermined degree of coincidence interval, the predetermined merging strategy In order to perform suppression processing when merging synonymous query results, the suppression processing includes: reduce the relevancy weight of web pages in the synonymous query results; or Insert synonym query results after a particular page of the search results list; or Adjust the synonym query results behind the original query results. 6. The method for implementing a search engine according to claim 1, wherein the merging strategy comprises: when the value of the coincidence degree is greater than a predetermined threshold in a predetermined coincidence degree interval, according to the original query result and the synonym The relevancy weight of each web page in the query result combines the original and synonymous query results. 7. A search engine, characterized in that the search engine comprises a search component, and the search component comprises: The query analysis module is used to receive the original query searched by the user; analyze the original query to obtain the original word and the synonym of the original word existing in the original query, and the synonymous context of the original word and the synonym , and according to the original query and the synonymous context, the synonym is replaced by the original word in the original query to obtain a synonymous query; A search module, configured to search and obtain the original query result webpage collection and the synonymous query result webpage collection according to the original query and the synonymous query; A coincidence degree calculation and result merging module, used to calculate the coincidence degree of webpages in the original query results and synonymous query results; and merge the original query and synonymous query result webpages according to a predetermined merging strategy corresponding to the coincidence degree collection, and generate a list of search results; where, The search engine also includes a log analyzer for Acquiring historical user query and click data, said data including historical query and query result web pages returned in response to the query and accessed by clicking; identifying synonym pairs comprising an original term present in the historical query and a corresponding synonym present in the query results web page; and At least the historical query is recorded and determined as a synonym context for the synonym pair. 8. The search engine according to claim 7, wherein the calculation of the coincidence degree comprises calculating the number |U1∩U2| of overlapping webpages in the original query result U1 and the synonymous query result U2. 9. The search engine according to claim 8, wherein the calculation of the coincidence degree also includes determining the lesser Min(|U1|, |U2|); the coincidence degree I(U1, U2)=|U1∩U2|/Min(|U1|, |U2|). 10. The search engine according to claim 8, wherein the calculation of the degree of coincidence further comprises calculating the sum |U1∪U2| of the number of webpages of the original query result and the number of webpages of the synonymous query result; Degree I(U1, U2)=|U1∩U2|/|U1∪U2|. 11. The search engine according to claim 7, wherein the merging strategy comprises: when the value of the degree of coincidence is less than a predetermined threshold in a predetermined degree of coincidence interval, the predetermined merging strategy is to merge Suppression processing is performed for synonymous query results, and the suppression processing includes: reduce the relevancy weight of web pages in the synonymous query results; or Insert synonym query results after a particular page of the search results list; or Adjust the synonym query results behind the original query results. 12. The search engine according to claim 7, wherein the merging strategy comprises: when the value of the coincidence degree is greater than a predetermined threshold in a predetermined coincidence degree interval, according to the original query result and the synonym query result The relevance weight for each web page combines the results of the original and synonymous queries.