CN104008141A - Product grading method by user based on on-line social network - Google Patents

Abstract

The present invention provides a method based on an online social network to help users rate products. The method first collects historical rating information of users and friends; then processes the collected information; and finally gives users accurate recommendation scores, in which users Collecting historical score information of friends is obtained through online social networks, processing information is mainly calculated according to Bayesian theorem, and users can update the probability distribution of the entire network. The present invention can continuously carry out dynamic learning, make good use of the online social network to help users recommend products, and ensure the balance between recommendation quality and recommendation quantity.

Description

Method for helping users to rate products based on online social network

technical field

The invention relates to an interaction method of an online social network, establishes a novel recommendation system, uses the principle of Bayesian reasoning for efficient recommendation, and belongs to the application fields of software engineering, human-computer interaction, and Internet crossover technology.

Background technique

In recent years, online social networking has become increasingly popular, attracting tens of thousands of users, and has become one of the main platforms for building friend relationships and sharing information today. In online social networks, friend recommendation is a critical task that plays an increasingly important role in our lives. Accurate suggestions can enable users to quickly find ideal items without being overwhelmed by irrelevant information. Suppliers are happy to recommend their satisfactory products to their potential customers and hope to turn them into real buyers.

In real life, people often turn to their friends on their social networks and ask for their opinions before buying goods or consuming services. Research findings in the fields of sociology and psychology have shown that humans tend to gravitate towards people of the same type, which is known as homophily. People are more willing to share their personal opinions with their friends due to stable and long-lasting social relationships, and they trust advice from friends more than strangers and suppliers. Online social networks are not only convenient for users to share their opinions, but also can be used as a platform to automate the implementation of recommendation algorithms based on online social networks in real life.

The Bayesian method is a pattern recognition method that calculates the posterior probability when the prior probability and conditional probability are known. The principle of its classification is to use the Bayesian formula to calculate the posterior probability based on the prior probability of an instance, that is, the probability that the instance belongs to a certain class, and select the class with the largest posterior probability as the class to which the instance belongs. This method uses probability to represent all forms of uncertainty, and uses probability rules to realize learning or other forms of reasoning. It has the advantages of strong robustness and easy implementation. It is a powerful tool for dealing with uncertainty problems. It has been used in human Learning mechanism exploration and Web mining have been widely used, and become one of the research hotspots in the fields of artificial intelligence, machine learning and data mining.

Contents of the invention

Technical problem: The purpose of the present invention is to provide a method based on online social network to help users rate products. The method uses online social network as a platform to measure the similarity of ratings between online social network associated users with a set of conditional probabilities. and infer user ratings based on Bayesian inference, solving the problems of flexible trade-offs between recommendation quality and recommendation volume and sparse ratings.

Technical solution: the method for helping users to rate products based on online social networks described in the present invention, users connect to their own devices through running agents in the online social network, publish their own ratings to friends, and users query to get the information in the online social network. Based on the ratings of friends, the Bayesian inference method is used to infer new recommendation scores based on the obtained historical similarity scores.

The method for helping users rate a product based on an online social network includes the following steps:

Step 1) Obtain the name of the currently purchased product entered by the user in the online social network; the online social network is a platform that connects with other people on the Internet, and the user evaluates the product and shares it with the user's friends , to query the ratings of the user's friends;

Step 2) Query and obtain the rating given by the user's friends in the online social network to the currently purchased product;

Step 3) Classify friends according to the ratings of the currently purchased products by the querying user friends, and friends with the same ratings are classified into one category;

Step 4) Obtain other products purchased by the user and corresponding ratings;

Step 5) Obtain one of the types of friends and the corresponding ratings sequentially, and the ratings are recorded as t _i , where i is the serial number of the friend category;

Step 51) Acquiring other purchased products and corresponding ratings of a friend k among such friends in turn, where k represents the serial number of the friend;

Step 511) Compare other products purchased by the user with other products purchased by this friend, and obtain the same product and corresponding ratings that both the user and the friend have rated; in this same product, calculate when the user rating is At t _i , the friend’s score is also the proportion of t _i , recorded as P(R _k =t _i |R _s =t _i ), where i represents the serial number of the friend class, and t _i represents the score of friend class i , R _s represents the rating of user s, R _k represents the rating of a certain friend k, P(R _k =t _i |R _s =t _i ) means that when the user’s rating is t _i , the rating of friend k is also t The proportion of _i ; when the user and friends do not have the same rating t _i , the value of P(R _k =t _i |R _s =t _i ) is set to 10%;

Step 512) Calculate in turn when the user's rating is t _i , the proportion of other friends who are rated as t _i ;

Step 52) Calculate the average of all the obtained ratios of this type of friends, and obtain the joint relationship between the user and this type of user rated t _i , denoted as P(Q _t =t _i |R _s =t _i )=mean (P(R _k =t _i |R _s =t _i )), said t _i represents the score of friend category i; R _s represents the score of user s; R _k represents the score of a certain friend k; Q _t represents the score The ratings of the same type of friends; mean(P(R _k =t _i |R _s =t _i )) indicates the average of the proportion of friends rated as t _i in a type of users when the user is rated as t _i , k is the serial number of this type of user in turn; P(Q _t = t _i | Rs = t _i ) indicates the proportion of a type of user whose score is t _i when the user is rated as t _i ;

Step 53) sequentially calculate the joint relationship between the user and all kinds of friends with other ratings;

Step 6) Calculate the proportion of each rating in the ratings of other products purchased by the user, which is recorded as P(R _s =t _i ), where t _i represents the rating of friend category i, and R _s represents the user s Rating, P(R _s =t _i ) indicates the proportion of each rating in the ratings of other products purchased by the user; when the user has never rated a certain score, its value is set to 1%;

Step 7) Calculate the proportion of each rating in all ratings according to the Bayesian formula, by

$P(R_{\mathrm{the\; s}}=t_i|Q_t)=\frac{P(Q_t=t_i|R_{\mathrm{the\; s}}=t_i)*P(R_{\mathrm{the\; s}}=t_i)}{{\mathrm{\Σ}}_{r=1}^{N}P(Q_t=r|R_{\mathrm{the\; s}}=r)*P(R_{\mathrm{the\; s}}=r)}$ Calculated; said r represents the rating level, the range is [1, N], N is the highest rating level preset, t _i represents the rating of friend class i, R _s represents the rating of user s, Q _t represents the same rating Ratings of a class of friends, P(R _s =t _i ) indicates the proportion of each rating in the ratings of other products purchased by the user, P(Q _t =t _i |R _s =t _i ) indicates the user’s rating When is t _i , the proportion of a class of users whose rating is t _i , Indicates the sum of all products of P(Q _t =r|R _s =r) and P(R _s = _r ) when r ranges from _{1 to} N. In the case of , the probability that the user scores t _i ;

Step 8) Select the score when the probability value is the maximum value as the user's recommendation score, by calculated, the Indicates the recommendation score, P(R _s =t _i |Q _t ) indicates the probability of the user scoring t _i in the case of excessive ratings by friends, and argmax indicates that P(R _s =t _i |Q _t ) is selected to be the maximum value When the value of t _i .

Beneficial effect:

1) The present invention provides a method based on an online social network to help users rate products. The whole process is clear and complete, and has strong readability.

2) The recommendation process described in the present invention provides a set of calculation formulas, which can convert the relevant data in the actual network into a mathematical model form, so as to obtain the final result.

3) The recommendation method described in the present invention has a dynamic learning part, and the user can continuously update the probability distribution of the entire network, thereby having better recommendation accuracy.

Description of drawings

Fig. 1 is a flow chart of a method for helping users rate products based on an online social network;

Figure 2 Scoring information table.

Detailed ways

In the online social network, the present invention provides accurate recommendations to users in combination with product-related data that has been rated. Below, the present invention will be described in more detail according to FIG. 1 and the embodiment, and the rating information of products that have been rated is shown in FIG. 2 .

1. Obtain the movie F of the movie name input by the user in the online social network;

2. Query the ratings given by the user's friends on the online social network. The ratings are divided into three levels from 1 to 3. The rating of friend 1 is 3, the rating of friend 2 is 3, the rating of friend 3 is 2, and the rating of friend 4 has a score of 1;

3. Classify friend 1 and friend 2 into one category, friend 3 into one category, and friend 4 into one category;

4. Obtain the user's rated movie titles and corresponding ratings;

5. Obtain the score 3 of friend 1 and friend 2, and record the score as _t1 ;

6. Obtain the name of the rated movie and the corresponding rating of friend 1.

7. Compare the name of the movie that the user has rated with the name of the movie that has been rated by friend 1, and obtain the same movie name and the corresponding rating that both the user and the friend have rated; among the movies that have been rated by both, Calculate when the user's rating is 3, the proportion of the friend's rating is also 3, recorded as P(R ₁ =3|R _s =3)=50%;

8. Calculate when the user score is 3, the proportion of friend 2P (R ₂ =3|R _s =3) is 50%;

9. Multiply all the obtained ratios of this type of friends to obtain the joint relationship between the user and the user with a score of 3, which is recorded as P(Q _t =3|R _s =3)=mean(P(R _k =3|R _s =3), the calculated value of P(Q _t =3|R _s =3) is 50%;

10. Calculate the joint relationship between the user and other rated friends in turn, and get the value of P(Q _t =2|R _s =2) to be 33.33%, and the value of P(Q _t =1|R _s =1) to be Default value 10%;

11. Calculate the proportion of each rating in the user's historical rating, the value of P(R _s =3) is 40%, the value of P(R _s =2) is 60%, and P(R _s =1) The value of is set to 1%;

12. Calculate the proportion of each rating in all ratings according to the Bayesian formula, by

$P(R_{\mathrm{the\; s}}=t_i|Q_t)=\frac{P(Q_t=t_i|R_{\mathrm{the\; s}}=t_i)*P(R_{\mathrm{the\; s}}=t_i)}{{\mathrm{\Σ}}_{r=1}^{N}P(Q_t=r|R_{\mathrm{the\; s}}=r)*P(R_{\mathrm{the\; s}}=r)}$ Calculated, the value of P(R _s =3|Q _t ) is 49.9%, the value of P(R _s =2|Q _t ) is 49.85%, and the value of P(R _s =1|Q _t ) is 0.25% ;

13. Select the score when the probability value is the maximum value as the user's recommendation score, which is determined by $\hat{S}=\underset{1\≤t_i\≤3}{\arg \max }P(R_{\mathrm{the\; s}}=t_i|Q_t)$ The calculated recommendation score for the user is 3.

Claims (1)

1. the help user based on online social networks, to a method for product scoring, is characterized in that the method comprises the following steps:

Step 1) obtain the title of current the bought product that user inputs in online social networks; Described online social networks is the platform being associated with other people on the internet, and user marks to product, is shared with this user's friend simultaneously, inquires about this user's friend's scoring;

Step 2) inquiry obtains the scoring that the friend of user in online social networks has provided to current bought product;

Step 3) according to the user friend who inquires, the scoring size of current bought product is classified to friend, the identical friend that marks is classified as a class;

Step 4) obtain user and bought other products and corresponding scoring;

Step 5) obtain successively wherein a class friend and corresponding scoring, scoring is designated as t _i, i is the sequence number of friend's class;

Step 51) obtain successively a friend k in this class friend buy other products and corresponding scoring, described k represents friend's sequence number;

Step 511) other products of having bought of other products that user has been bought and this friend compare, obtain user and all comment undue like products and corresponding scoring with this friend; In this identical product, calculating when user marks is t _itime, this friend's scoring is also t _ishared ratio, is designated as P (R _k=t _i| R _s=t _i), described i represents the sequence number of friend's class, t _irepresent the scoring of friend's class i, R _srepresent the scoring of user s, R _krepresent the wherein scoring of a certain friend k, P (R _k=t _i| R _s=t _i) represent to mark as t as user _itime, the scoring of friend k is also t _ishared ratio; When user and friend do not have identical scoring t _i, P (R _k=t _i| R _s=t _i) value be set as 10%;

Step 512) calculate successively when user scoring be t _itime, all the other friend's scorings are t _ishared ratio;

Step 52) this class friend's who acquires all proportions is averaged, obtaining user is t with scoring _ithis class user's the relation of combining, be designated as P (Q _t=t _i| R _s=t _i)=mean (P (R _k=t _i| R _s=t _i)), described t _irepresent the scoring of friend's class i; R _srepresent the scoring of user s; R _krepresent the scoring of a certain friend k; Q _trepresent the identical class friend's of scoring scoring; Mean (P (R _k=t _i| R _s=t _i)) represent that user's scoring is for t _itime, in a class user, scoring is t _ithe mean value of friend's proportion, k is followed successively by this class user's sequence number; P (Q _t=t _i| Rs=t _i) represent that user's scoring is for t _itime, a class user scoring is t _ishared ratio;

Step 53) calculate successively user and all kinds of friends' of all the other scorings the relation of combining;

Step 6) calculate user buying in the scoring of other products, the ratio that each scoring is shared, is designated as P (R _s=t _i), described t _irepresent the scoring of friend's class i, R _srepresent the scoring of user s, P (R _s=t _i) represent that user is buying in the scoring of other products, the ratio that each scoring is shared; When user never commented certain point, its value is set as 1%;

Step 7) calculate every kind of scoring shared ratio in all scorings according to Bayesian formula, by

$P(R_s=t_i|Q_t)=\frac{P(Q_t=t_i|R_s=t_i)*P(R_s=t_i)}{{\mathrm{\Σ}}_{r=1}^{N}P(Q_t=r|R_s=r)*P(R_s=r)}$ Calculate; Described r represents grading system, and scope is [1, N], and N is the highest grading system setting in advance, t _irepresent the scoring of friend's class i, R _srepresent the scoring of user s, Q _trepresent the identical class friend's of scoring scoring, P (R _s=t _i) represent that user is buying in the scoring of other products, the ratio that each scoring is shared, P (Q _t=t _i| R _s=t _i) represent that user's scoring is for t _itime, a class user scoring is t _iratio, while representing that r gets 1 to N, P (Q _t=r|R _s=r) with P (R _s=all products r) and, P (R _s=t _i| Q _t) be illustrated in friend and comment in undue situation, user's scoring is t _iprobability;

Step 8) scoring when choosing probable value and being maximal value is as user's recommendation score, by calculate, described in represent recommendation score, P (R _s=t _i| Q _t) be illustrated in friend and comment in undue situation, user's scoring is t _iprobability, argmax represents to choose makes P (R _s=t _i| Q _t) t while being maximal value _ivalue.