CN113836444B - Linear time friend recommendation method, system, terminal and storage medium - Google Patents
Linear time friend recommendation method, system, terminal and storage medium Download PDFInfo
- Publication number
- CN113836444B CN113836444B CN202111159485.5A CN202111159485A CN113836444B CN 113836444 B CN113836444 B CN 113836444B CN 202111159485 A CN202111159485 A CN 202111159485A CN 113836444 B CN113836444 B CN 113836444B
- Authority
- CN
- China
- Prior art keywords
- users
- community
- target user
- user
- social
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 84
- 230000003997 social interaction Effects 0.000 claims abstract description 16
- 230000006399 behavior Effects 0.000 claims abstract description 11
- 238000009826 distribution Methods 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 19
- 230000003993 interaction Effects 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 8
- 238000005065 mining Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 12
- 230000009191 jumping Effects 0.000 description 7
- 238000010801 machine learning Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- KJNFMGMNZKFGIE-UHFFFAOYSA-N n-(4-hydroxyphenyl)acetamide;5-(2-methylpropyl)-5-prop-2-enyl-1,3-diazinane-2,4,6-trione;1,3,7-trimethylpurine-2,6-dione Chemical compound CC(=O)NC1=CC=C(O)C=C1.CN1C(=O)N(C)C(=O)C2=C1N=CN2C.CC(C)CC1(CC=C)C(=O)NC(=O)NC1=O KJNFMGMNZKFGIE-UHFFFAOYSA-N 0.000 description 1
- 238000009828 non-uniform distribution Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Primary Health Care (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- Economics (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application relates to a linear time friend recommendation method, a system, a terminal and a storage medium. The method comprises the following steps: estimating the topic interest degree of each user based on text content data issued by the users in the social network, and calculating the interest similarity among the users according to the topic interest degree; calculating social affinities among users according to social interaction behaviors of the users; establishing a weighted heterogeneous network for the target user according to the interest similarity and the social affinity; the community discovery method of bottom-up modularity increment is adopted to mine a community set of a target user in a weighted heterogeneous network, and membership values of communities in the target user affiliated related community set are calculated; and extracting a set number of users from the weighted heterogeneous network according to the community membership value, and generating a friend recommendation set of the target user. According to the method and the device for recommending the friends, the interest similarity and the social affinity among the users are fused, so that the accuracy of friend recommendation is obviously improved, and the efficiency of friend recommendation is obviously improved.
Description
Technical Field
The application belongs to the technical field of social networks, and particularly relates to a linear time friend recommendation method, a system, a terminal and a storage medium.
Background
Friend recommendations are everywhere visible in the social network. The friend recommendation algorithm can help the target user to quickly find friends suitable for the target user from a plurality of other users, so that the target user is assisted in quickly expanding the social circle to improve the use experience.
The existing friend recommending method mainly comprises the following steps:
1. A semantic similarity-based method; according to the method, a user model is built for each user to describe the statistical information or characteristic attributes such as questionnaire investigation answers and the like, and friend recommendation is carried out according to the characteristic attributes. However, the friend recommendation mechanism of the method completely ignores social relations among users, so that high-precision link prediction cannot be performed.
2. Based on social interactivity; the method recommends friends to a target user by utilizing a social network structure (such as the number of common friends) among users. However, this method does not consider rich semantic content in social media, and its recommendation accuracy needs to be further improved.
3. A method based on collaborative filterability; the method provides for analyzing social relationships between users and dependencies between items to estimate new user associations. However, the method only takes the social relationship as one of many factors of machine learning, and fails to realize the important role of social interaction association in friend recommendation, so that the link prediction accuracy is not ideal enough.
Disclosure of Invention
The application provides a linear time friend recommending method, a system, a terminal and a storage medium, which aim to solve one of the technical problems in the prior art at least to a certain extent.
In order to solve the problems, the application provides the following technical scheme:
A linear time friend recommending method comprises the following steps:
Estimating the topic interest degree of each user based on text content data issued by the users in the social network, and calculating the interest similarity among the users according to the topic interest degree;
calculating social affinities among users according to social interaction behaviors of the users in the social network;
Establishing a weighted heterogeneous network for the target user according to the interest similarity and the social affinity; the weighted heterogeneous network comprises potential friend candidates which have common interests or common friends with the target user;
A community discovery method of bottom-up modularity increment is adopted to mine a community set of the target user in a weighted heterogeneous network, and a community membership value of the target user in each community in the community set is calculated;
And extracting a set number of users from the weighted heterogeneous network according to the community membership value, and generating a friend recommendation set of the target user.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the estimating the subject interestingness of each user based on the text content data published by the user in the social network comprises the following steps:
the probability distribution of regarding a text collection posted by a user on a social network as related topics is expressed as The probability distribution of treating each topic as a large number of words is denoted/> And/>Super-parameters/>, each with dirichlet priorsAnd
Estimating distribution of D text sets on a topic using Gibbs samplingAnd distribution of T topics over words/>The topic interest degree estimation result matrix of the user is expressed as:
The matrix of D X T, labeled DT, where D represents the number of users, T represents the number of topics, and DT ij represents the number of words in the text set of user u i that are divided into topics T j;
A matrix of W x T, labeled WT, where W represents the number of unique words in the text, and WT ij represents the number of times unique word W i is divided into topics T j;
Normalizing the matrix DT to DT ' such that each row DT ' i· holds the equation DT ' i·||1 =1; the element DT 'ij in DT' represents the probability that the user u i is interested in the topic t j;
normalizing the matrix WT to WT 'such that each row WT' i· holds the equation W i·||1 = 1; the element WT 'ij in WT' represents the probability that the unique term w i is affiliated with the topic t j.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the estimating the subject interestingness of each user based on the text content data published by the user in the social network further comprises:
calculating the domain influence of each user on the interesting subject by combining the social relationship topological structure; the domain influence r j (i) of the user s i on the topic t j is:
wherein DT' ij represents the interestingness of the user s i on the topic t j; μ is an adjustable parameter, "+1" is a constant.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the calculating the interest similarity among users according to the subject interest degree comprises the following steps:
The similarity of interests S (i, j) between users S i and S j is:
S(i,j)＝Sc(i,j)×Sd(i,j)
Wherein S c (i, j) represents the topic probability similarity of users S i and S j, and S d (i, j) represents the dominant interest similarity of users S i and S j;
The value of the topic probability similarity S c (i, j) between users S i and S j is:
Wherein T represents a set of topics; the interest degree average value of the user s i on all topics is obtained; /(I) Referring to the harmonic mean of two correlation values, it can be calculated as follows:
wherein eta represents a normalization factor for ensuring The value of (1) belongs to the interval [0,1], which can be set equal to the number of subjects; the interest value of the target user on the topic t z is given; /(I) Meaning the occurrence probability of the topic t z, which can be calculated according to the number of times of the unique word attaching topic t z in the matrix WT';
The main interest similarity S d (i, j) between the users S i and S j is calculated as:
Where D i and D j represent the main interest sets of users s i and s j, respectively.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the calculating the social affinity between the users according to the social interaction behaviors of the users in the social network comprises the following steps:
Calculating social interaction strength of the user s i pointing to s j:
If user s i "pays attention to" s j, F (i, j) =1; otherwise F (i, j) =0; n c (i, j) and N f (i, j) represent the number of comments and forwarding operations performed by the user s i for the text content posted by s j, respectively; τ is a regularization parameter.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the building the weighted heterogeneous network for the target user according to the interest similarity and the social affinity specifically comprises the following steps:
According to the interest similarity and the social affinity, potential friend candidates which have common interests or common friends with the target user are found out from a social network; the potential friend candidates comprise a set number of friends with the most common friends and domain high-influence users with the most common interests;
Generating a weighted heterogeneous network G (V, E S,WS,EI,WI) according to the potential friend candidates, wherein V represents a node set and consists of a target user, a friend set of the target user and the potential friend candidate set; e S and E I represent sets of undirected and directed edges, respectively, corresponding to interest similarity and social affinity among users in V; w S and W I indicate the edge weights on E S and E I, respectively.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the method for discovering communities by using bottom-up modularity increment comprises the following steps:
for each node in the weighted heterogeneous network, removing the node from an original community, adding the node into an adjacent community, evaluating the increment of the modularity of the community, and if the increment of the modularity is positive, transferring the node into the adjacent community with the maximum increment of the modularity; otherwise, the node is kept in the original community;
and sequentially repeating the processes for all the nodes until the value of the community modularity is not increased, thereby obtaining a community set where the target user is located.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the calculating the membership value of the target user to the different communities in the community set comprises the following steps:
Assume that Is the ith community associated with node s o,/>Is/>If the node set adjacent to s o in the middle is s o is attachedMembership degree/>The expression is as follows:
wherein, And/>Represents the undirected edge weights between nodes s o and s u, respectively, and the directed edge weights from s o to s u; /(I)Represents the sum of the undirected edge weights connected to s o; /(I)And/>Representing the sum of the outgoing side weights of s o and the sum of the incoming side weights of s u respectively; /(I)And/>
The technical scheme adopted by the embodiment of the application further comprises the following steps: extracting a set number of users from the weighted heterogeneous network according to the community membership value of the target user, and generating a friend recommendation set of the target user comprises:
For each community associated with the target user, carrying out priority ranking on the node set in the communities through a LeaderRank ranking algorithm;
And extracting a proper number of high-influence users from each community according to the community membership value of the target user and the node ordering score value in each community, and generating a friend recommendation set of the target user.
The embodiment of the application adopts another technical scheme that: a linear time friend recommendation system, comprising:
The interest calculation module: the method comprises the steps of estimating the topic interest degree of each user based on text content data issued by the users in a social network, and calculating the interest similarity among the users according to the topic interest degree;
and the interaction calculation module is used for: the method comprises the steps of calculating social affinities among users according to social interaction behaviors of the users in a social network;
And a network construction module: the method comprises the steps of establishing a weighted heterogeneous network for a target user according to the interest similarity and the social affinity; the weighted heterogeneous network comprises potential friend candidates which have common interests or common friends with the target user;
The community mining module: the community discovery method is used for mining a community set of the target user in a weighted heterogeneous network by adopting a bottom-up modularity value-added community discovery method, and calculating a community membership value of the target user in each community in the community set;
friend recommendation module: and the method is used for extracting a set number of users from the weighted heterogeneous network according to the community membership value and generating a friend recommendation set of the target user.
The embodiment of the application adopts the following technical scheme: a terminal comprising a processor, a memory coupled to the processor, wherein:
the memory stores program instructions for implementing the linear time friend recommendation method;
the processor is configured to execute the program instructions stored by the memory to control linear time friend recommendation.
The embodiment of the application adopts the following technical scheme: a storage medium storing program instructions executable by a processor for performing the linear time friend recommendation method.
Compared with the prior art, the embodiment of the application has the beneficial effects that: according to the linear time friend recommendation method, system, terminal and storage medium, the interest similarity and social affinity among users are fused, a weighted heterogeneous network is built for each target user, the network comprises most potential friend candidates with common interests or common friends with the target user, a clustering algorithm is executed for the weighted network of each target user, a relevant community set is mined, and then individuals matched with the interest and social relations of the target user in a proper proportion are selected from different communities to build a friend recommendation set. According to the method and the device, the interest similarity and the social affinity among the users are fused, so that the accuracy of friend recommendation is greatly improved; by constructing a weighted heterogeneous network for each target user, the search range of friend recommendation is narrowed, the efficiency of friend recommendation is improved, and the requirement of a large-scale social network on linear time complexity is met.
Drawings
FIG. 1 is a flow chart of a linear time friend recommendation method according to an embodiment of the present application;
FIG. 2 is a schematic diagram of text analysis based on a latent dirichlet allocation model according to an embodiment of the present application;
FIG. 3 is a schematic diagram of a potential friend candidate acquisition process according to an embodiment of the present application;
FIG. 4 is a schematic diagram of a weighted heterogeneous network construction process according to an embodiment of the present application;
FIG. 5 is a flow chart of a method of community discovery in a multidimensional heterogeneous network in accordance with an embodiment of the present application;
FIG. 6 is a schematic diagram of a friend recommendation set construction process according to an embodiment of the present application;
FIG. 7 is a schematic diagram of a linear time friend recommendation system according to an embodiment of the present application;
fig. 8 is a schematic diagram of a terminal structure according to an embodiment of the present application;
Fig. 9 is a schematic structural diagram of a storage medium according to an embodiment of the present application.
Detailed Description
The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
Fig. 1 is a flowchart of a linear time friend recommendation method according to a first embodiment of the present application. The linear time friend recommending method of the first embodiment of the application comprises the following steps:
s10: acquiring a text content data set published by a user on a social network and a social interaction data set among the users;
In this step, the social network includes, but is not limited to, social software such as microblog, QQ say, weChat, etc. For convenience of explanation, the embodiments of the present application take as an example the acquisition of a Tencentrated microblog dataset (real-world Network of Tencent Weibo, RN-TW) and a twitter dataset (real-World network of Twitter, RN-TT). The method for acquiring the Tencent microblog data set comprises the following steps: starting with an initial person who designates an individual as a dataset, adding all fans of the individual as new members into the dataset; thereafter, all fans of all new members are incorporated in the same way; and so on until the number of people in the dataset meets the experimental requirements. And meanwhile, eliminating users with the number of vermicelli exceeding the set number (1000), and identifying the users as stars or abusing people who pay attention to the relationship. 121 active members are selected in the Tencet microblog data set to serve as target users for friend recommendation, and the target users need to meet the following requirements: at least 5 "attention" relationships are actively established or terminated during an interval of 2 months.
The twitter data set comprises three parts of cleaned Tweet content, user information and link structure, specifically comprises 107660 users and corresponding 4020199 pieces of concerned link information, and isolated users which do not participate in concerned relation and active users which have more than 1000 fans or friends can be removed in advance. For convenience, 45 persons with 200 (accurate numerical values) existing friends are selected from 107660 users as target users for friend recommendation, and the recommendation set adopts cross-validation training set with the size of
S20: performing semantic interest analysis on the text content dataset by adopting an LDA (LATENT DIRICHLET Allocation, potential Dirichlet distribution) model, and estimating the subject interest degree of each user;
In this step, the LDA model is an unsupervised machine learning technique for identifying potential topic information from a text content dataset and estimating the user's interest level in different topics. The LDA model treats each set of text as a "bag of words," and each set of text is treated as a probability distribution of related topics, where each topic is in turn treated as a probability distribution of a large number of words. The topic of interest to the user has a relatively large corresponding probability value. In the LDA model, the text set generation process comprises the following three steps:
(1) Selecting a topic from the associated topic distribution probabilities for each text set;
(2) Extracting a term from the term distribution probabilities associated with the selected topic;
(3) Repeating the steps (1) and (2) until all words in the text set are sampled.
Referring to fig. 2, a schematic diagram of a Bayesian Network (Bayesian Network) based text analysis process according to an embodiment of the present application is shown. Wherein each of the D sets of text is associated with a distribution of multiple items over T topics, which are represented asEach topic is related to a polynomial probability distribution of a large number of words, denoted/> And/>Hyper-parameters/>, each with Dirichlet a prioriAnd/>In fig. 2, double-lined circles and single-lined circle panels represent observations and latent variables, respectively. The directed edges correspond to conditional dependencies between two variables, while the boxes represent the resampling process, the number of which is given by the variables in the lower right corner of the box. z is the polynomial distribution/>, from the polynomials associated with the text setUpsampling a fetched topic, w represents a distribution/>, from a polynomial associated with the topicThe words obtained are upsampled and N d represents the number of words in the text set.
Further, embodiments of the present application employ Gibbs Sampling (Gibbs Sampling) to estimate the distribution of D text sets on a topic from a text content datasetAnd distribution of T topics over words/>The topic estimation results of interest to the user are represented in a matrix form:
(1) The matrix of D x T, labeled DT, where D represents the number of users, T represents the number of topics, and DT ij represents the number of words in the text set of user u i that are partitioned into topics T j.
(2) The |w|×|t| matrix, labeled WT, where |w| represents the number of Unique words (Unique words) and WT ij represents the number of times Unique Word W i is divided into topics T j.
Without loss of generality, the matrix DT is normalized to DT ' such that each row DT ' i· holds the equation DT ' i·||1 =1. Each row of DT 'corresponds to a probability distribution of interest of a user over all topics, i.e. element DT' ij expresses the probability that user u i is interested in topic t j. Similarly, the matrix WT is normalized to WT ' such that each row WT ' i· holds the equation WT ' i·||1 =1; the element WT 'ij in WT' represents the probability that the unique term w i in the text is affiliated with the topic t j.
In the embodiment of the application, the parameter values of the LDA model are as follows: t=100, The setting can be specifically performed according to practical applications.
S30: calculating the domain influence of each user on the interested topics by combining the social relationship topological structure, respectively extracting a certain number of high-influence users on each topic to serve as candidate users for friend recommendation, and sorting all the candidate users on each topic in a descending order according to the domain influence;
In this step, the domain influence of the user depends on the interest probability value of the user on each topic and the social relationship topology structure thereof, and the domain influence calculation of the user mainly includes: the user's interest in the relevant area, the number of fans focused on the user, and the degree to which the user's fans are approved as friends within the area. Thus, the domain influence r j (i) of the user s i on the topic t j can be quantitatively expressed as:
Wherein DT' ij represents the interestingness of the user s i on the topic t j; μ is an adjustable parameter, which is set by the embodiment of the application as the product of the topic number |t| and the logarithmic base; the constant "+1" is to avoid negative values in the logarithmic expression.
After calculating the domain influence of each user on each topic, a certain number of high influence users in each topic (the number is set to be 200 in the embodiment of the application) are arranged in a descending order and stored as a matrix |t|×n r, denoted as TN r, where |t| and N r respectively represent the number of topics of interest of the user and the number of predefined domain high influence users on each topic, and the element (TN r)ij is the jth high influence user on the ith topic).
S40: calculating interest similarity and social affinity among users, finding out potential friend candidates which have common interests or common friends with each target user from a social network according to the interest similarity and the social affinity, and respectively constructing a weighted heterogeneous network in which undirected edges and directed edges coexist for each target user;
In this step, the interest similarity calculation method between users specifically includes: defining the topic with the larger probability value as the main interest of the user, and representing the main interest set of the user s i as D i,Di needs to satisfy the following conditions:
(1) The sum of the probabilities of all topics within the set is greater than a predefined threshold lambda; preferably, in the embodiment of the application, λ=0.5 is set, and the adjustment can be specifically performed according to practical application, and the value of the parameter does not seriously affect the friend recommendation precision.
(2) The probability of any topic in the set is greater than or equal to the probability of any topic outside the set;
(3) The number of topics within the collection is kept to a minimum.
Assuming that the topic probability vector of user s i is denoted as P i＝[p(z1|si),p(z2|si),...,p(zZ|si), rearranging the vector elements in descending probability order may resultWherein/>(If m < n), the number of elements of the user's main interest set can thus be deduced as:
finally, the main interest set of the user is obtained
Further, the similarity of interests between users is not only influenced by the topic probability distribution, but also related to the main interests of each user. The interest similarity S (i, j) between the users S i and S j is calculated as follows:
S(i,j)＝Sc(i,j)×Sd(i,j)(3)
where S c (i, j) represents the topic probability similarity of users S i and S j and S d (i, j) represents the dominant interest similarity of users S i and S j.
Considering topic popularity (i.e., topic popularity) and interest of a target user, embodiments of the present application use a fine-tuned version of the pearson correlation coefficient (Pearson Correlation Coefficient) to calculate the value of topic probability similarity S c (i, j) between users S i and S j as follows:
Wherein T represents a set of topics; the interest degree average value of the user s i on all topics is obtained; /(I) Refers to the Harmonic Mean (Harmonic Mean) of two correlation values,/>The calculation formula of (2) is as follows:
wherein eta represents a normalization factor for ensuring Preferably, in the embodiment of the present application, the value of η is set equal to the number of subjects; /(I)The interest value of the target user s o on the topic t z; /(I)The occurrence probability of the topic t z can be calculated according to the number of times of the unique word attached topic t z in the matrix WT'.
If the primary interests of users s i and s j are different, then the similarity of interests between them is correspondingly smaller. The main interest similarity S d (i, j) between users is calculated as:
D i and D j represent the main interest sets of users s i and s j, respectively. If S d (i, j) is less than the set similarity threshold (the present application sets the threshold to 0.01, which can be specifically adjusted according to the actual application), it is ignored.
Further, the embodiment of the application carries out social affinity calculation according to the interaction behaviors among users and the relationship of 'concern'. Interaction modes between users in a social network include comments, forwarding, etc. Each user interaction mode comprises two important characteristics of interaction frequency and recency. The interaction frequency represents the number of times that a user interacts with friends in a specific mode in a given time period; the recency indicates the time that the interaction pattern has elapsed since the last occurrence. Meanwhile, even though the users do not have any interaction behavior, the relationship of 'concern' plays a decisive role in the social affinity among the users, and the relationship of social relations among the corresponding users is clearly expressed. The importance of the "attention" relationship is manifested by assigning it an appropriate weight.
Based on the above analysis, embodiments of the present application calculate the social affinity strength from user s i to s j according to the following formula based on a series of interactions between users:
If user s i "pays attention to" s j, F (i, j) =1; otherwise F (i, j) =0. N c (i, j) and N f (i, j) represent the number of comments and forwarding operations performed by user s i for the social content of s j, respectively. Regularization parameter τ balances the contribution of "attention" relationships and interaction behavior to social affinity strength. After the social interaction strength between each user and friends of the user is calculated, the social interaction strength value is scaled in a comparable way, so that the maximum value of the social interaction strength value is equal to 1.
Further, referring to fig. 3, a schematic diagram of a potential friend candidate searching process according to an embodiment of the present application is shown. The method specifically comprises the following steps:
(01) Specifying a target user s o and the number k of friends to be recommended;
(02) Potential friend candidate set The expected value of the number of potential friend candidates with social affinity N f (o) =2k; the expected value of the number of potential friend candidates with interest similarity N s (o) =2k;
(03) Potential friend candidate set with social affinity
(04) Selecting N f (o) users from friends of the target user s o to construct a CS f,CSf construction rule that the number of fan-shaped users 'attention' by s o is the largest; if the number of friends of the friends of s o is less than N f (o), allow |CS f|＜Nf (o) to hold;
(05)CS＝CS∪CSf；
(06) Definition of the definition D o represents the main interest set of s o;
(07) Potential friend candidate set with interest similarity
(08) Defining a cycle number r=1;
(09) Definition of the definition
(10) Judging whether N s(o)-|CSs|≥Nsr (o) is met, if so, executing the step (11); if not, executing the step (12);
(11) From the subject matter N sr (o) users with the highest field influence are selected and added into CS s; the new user selection rules here are: the selected new user is not in the CS U CS s; then jumping to the step (14) for execution;
(12) Judging whether 0 is smaller than N s(o)-|CSs|＜Nsr (o) or not, if so, executing the step (13); if not, executing the step (16);
(13) From the subject matter The user with the highest field influence (N s(o)-|CSs I) is selected and added into CS s; the new user selection rules here are: the selected new user does not belong to the set CS U CS s;
(14)r＝r+1；
(15) Defining q o to represent the number of elements of the s o main interest set, judging whether r > q o is met, and executing the step (16) if the r > q o is met; if not, executing the step (09);
(16) A set of potential friend candidates cs=cs-CS f is obtained.
In the embodiment of the application, the potential friend candidates of the target user comprise two parts of users with common friends (namely friends of friends) and high influence of common interests. In a social network, each user may be considered to be his own friends. Thus, no less than 50% of potential friend candidates have common interest. Preferably, the embodiment of the application sets that the number of people with common interests and common friends in the potential friend candidates of the target user is half of each other.
In the embodiment of the application, the weighted heterogeneous network comprises most potential friend candidates which have common interests or common friends with the target user, the undirected edges of the weighted heterogeneous network represent the interest similarity among the users, and the directed edges represent the social affinity among the users. The construction process of the weighted heterogeneous network G is shown in fig. 4, and specifically includes:
(01) Specifying a target user s o, adjusting regularization parameters w of interest similarity and social affinity weight, and giving a potential friend candidate set CS;
(02) Definition V 1 = CS;
(03) Defining V 2＝so U FS, wherein FS represents a friend set of s o;
(04) Traversing unordered pairs (s i∈V1,sj∈V2);
(05) Calculating an interest similarity value S (i, j) between the users S i and S j (formula (3));
(06) Setting threshold=0.01, if S (i, j) > threshold, executing step (07); otherwise, executing the step (08);
(07) Inserting an undirected continuous edge between the nodes S i and S j, wherein the weight value of the undirected continuous edge is w multiplied by S (i, j);
(08) Defining V=CS &. S o &. FS;
(09) Traversing a friend-vermicelli pair (s u,sv) epsilon V;
(10) Calculating social interaction strengths I (u, v) from users s u to s v (formula (7));
(11) Inserting a directed edge from s u to s v, wherein the weight value of the directed edge is I (u, v);
(12) Generating a weighted heterogeneous network G (V, E S,WS,EI,WI), wherein V represents a node set and consists of a target user, a friend set of the target user and all potential friend candidates; e S and E I represent sets of undirected and directed edges, respectively, corresponding to interest similarity and social affinity among users in V; w S and W I indicate the edge weights on E S and E I, respectively.
S50: a community set where the target users are is located is excavated by adopting a community discovery method of bottom-up modularity increment, community division is carried out on weighted heterogeneous networks of each target user, and membership values of the target users belonging to different communities are calculated;
The actual density of the connection lines in the community set is compared with the expected connection line density (the connection lines among the nodes are assumed to have no module structure), so that community modularity in the undirected unowned network is promoted to a weighted heterogeneous network in which undirected edges and directed edges coexist, and a modularity Q value in the weighted heterogeneous network is deduced. The larger the Q value is, the stronger the community structure of the network is. The modularity Q value is derived as follows:
wherein, And/>The undirected edge weights between nodes s i and s j, and the directed edge weights from s i to s j are respectively represented; /(I)Representing the sum of undirected edge weights connected to s i; /(I)And/>Means the sum of the outgoing side weights of s i and the sum of the incoming side weights of s j, respectively; delta (c i,cj) is calculated as a pulse function, delta (c i,cj) =1 if c i＝cj; otherwise, δ (c i,cj) =0; /(I)And/>
In the step, the community discovery method of modular value-added from bottom to top starts by attaching each node to an isolated node community, and for each node, the node is moved out of an original community and added into an adjacent community, and the modular increment is evaluated, if the modular increment is positive, the node is transferred into the adjacent community with the maximum modular value-added; otherwise, the node is retained in its original community. And sequentially repeating the process for all the nodes until no further module degree increment can be obtained, and obtaining the community set where the target user is located. Fig. 5 is a flowchart of a community discovery method in a multidimensional heterogeneous network according to an embodiment of the present application, which specifically includes:
(1) Given a heterogeneous network G (V, E S,WS,EI,WI), a target user s o;
(2) Removing s o from G;
(3) Assigning each node in G to a different isolated community of nodes;
(4) Traversing each node s i in V, wherein the nodes in V are required to be arranged in ascending order of the number of neighbor nodes;
(5) Attempting to transfer s i from the original community to the adjacent community and evaluating the modular increment, and if the modular increment is positive, transferring s i to the adjacent community with the largest modular increment;
(6) Judging whether the value of the modularity is increased, if so, executing the step (4); otherwise, executing the step (7);
(7) Traversing the contiguous community C i of s o in the initial network, adding s o to C i, and then adding C i to the CG;
(8) And generating a community set CG where the target user is.
Further, the target users recommended by friends belong to a plurality of communities, and the membership degree of the target users is in a state of non-uniform distribution. Assume thatIs the ith community associated with node s o, and/>Is/>If the node set adjacent to s o in the middle is s o is attachedMembership degree (Membership Degree)/>The mathematical formula can be expressed as:
wherein, And/>The undirected edge weights between nodes s i and s j, and the directed edge weights from s i to s j are respectively represented; /(I)Representing the sum of undirected edge weights connected to s i; /(I)And/>Means the sum of the outgoing side weights of s i and the sum of the incoming side weights of s j, respectively; /(I)And/>
When it occursWhen the embodiment of the application sets/>Thereby avoiding assigning a negative number to the community membership. Without loss of generality, different community membership of the same target user is scaled equally such that its sum equals 1.
S60: and for the community set of each target user, executing a LeaderRank ordering algorithm to order the priority of the nodes in the same community, extracting a proper number of high-influence users from each community according to the membership value and the node ordering score, and generating a friend recommendation set of the target user.
In the step, for each community associated with the target user, the nodes are subjected to priority ranking in the communities through a LeaderRank ranking algorithm, so that favorable conditions are created for constructing a friend recommendation set. The leader rank ordering algorithm adopts an ordering mechanism based on random walk, and each undirected edge of the original network is temporarily replaced by using two directed edges with different weight halving directions. By introducing a ground node s g, the leader rank adds two directed edges e gu and e ug to each node s u in the original network, assigning it the average weight of all edges in the original network (after replacing the undirected edge). The ranking score q i (t) for the discrete-time t node s i is calculated as follows:
Wherein A j′i represents the sum of directed edge weights from s j to s i, Representing the out-side weight of s j.
Initially, the embodiment of the present application sets the score of the ground node g to r g (0) =0, and the score of each of the other nodes k to r k (0) =1. After steady state is reached, the fraction of ground nodes is equally distributed to all other nodes. Thus, the final ranking score for node j is defined as:
Where r j(t∞) represents the fraction of node j at steady state.
According to the method and the device for selecting the friends, the interest similarity and the social affinity between the target user and the potential friend candidates are considered, and then a preset number of friends are selected from the weighted heterogeneous network corresponding to the target user based on the membership of the target user to the related community, and finally a friend recommendation set of the target user is generated. Specifically, the friend recommendation set construction process is shown in fig. 6, and specifically includes:
(01) Giving a target user s o, an existing friend set FS, a community set CG, a membership set phi, a node priority value set psi in the community and the size k of a friend recommendation set;
(02) Friend recommendation set The sum of membership degrees corresponding to untreated communities rm=1;
(03) Arranging all community membership values of the target user s o in a descending order;
(04) Traversing each element in Φ
(05) Judging whether k-I RS I=0 is met, if yes, jumping to the step (14) for execution; otherwise, executing the step (06);
(06) Judging Whether or not it is true, if so, executing the step (07); otherwise, jumping to the execution of the step (08);
(07) Defining k i =k- |rs| and jumping to the step (09) for execution; k i indicates that the community should be used The number of the selected recommended friends;
(08) Definition of the definition
(09) JudgingIf so, executing the step (10); otherwise, jumping to the execution of the step (11); wherein, |Λ| represents the number of elements in the set Λ;
(10) From the slave Selecting k i nodes with highest priority values, adding the nodes to the RS, and jumping to the step (13) for execution;
(11) Judging If so, executing the step (12); otherwise, jumping to the step (13) to be executed;
(12) Will be All elements in (a) are added to RS;
(13)
(14) Generating a final friend recommendation set RS;
the execution time of the embodiment of the application is in linear relation with the number of network users, and the analysis is as follows:
After LDA model training, the time consumption of Gibbs sampling is in a linear relation with the number of users. Aiming at each LDA theme, the embodiment of the application only saves a small number of users with high influence in the related fields, and avoids the operation of arranging all users in descending order. Thus, the runtime of choosing potential friend candidates with similar interests for all target users is O (n), where n represents the number of users in the social network.
A small heterogeneous network is built for each target user, the interest similarity between the existing friends and the potential friend candidates and the interaction quantity between each pair of friends and vermicelli with 'attention' relation are required to be calculated, and the running time of the interaction quantity and the node quantity in the small heterogeneous network are in linear relation. However, the number of nodes of the small heterogeneous network is almost negligible compared to the number of nodes of the large-scale network. Thus, the time taken to build a respective heterogeneous network for all target users is linear with the number of social network nodes.
The method for discovering communities in the embodiment of the application has the time complexity of O (m ') by transferring nodes between adjacent communities through iterative heuristics to maximize community modularity, wherein m ' represents the number of small network edges comprising n ' nodes and m '. Alpha.n ' 2. Furthermore, the number of nodes in a small network is not high (typically not more than 1000), resulting in a nearly constant time consumption for community discovery for each target user. Thus, the time complexity of community discovery for all target users is O (n).
The most time-consuming part of friend recommendation is solved by the priority ordering problem of the LeaderRank algorithm, the answer can be obtained through iterative multiplication, and the time consumption is O (n' 2). Likewise, the heterogeneous network built for each target user only occupies a small fraction of nodes, which makes the overall ordering process time-complex to O (n).
Based on the above analysis, the embodiments of the present application are essentially linear in execution, i.e., the time complexity is O (n).
Based on the above, the linear time friend recommendation method in the embodiment of the application combines the interest similarity and social affinity between users to construct a weighted heterogeneous network for each target user, wherein the network comprises most potential friend candidates with common interests or common friends with the target user, a clustering algorithm is executed for the weighted network of each target user to mine out related community sets, and then individuals matched with interest and social relations of the target user in proper proportion are selected from different communities to construct friend recommendation sets. According to the method and the device, the interest similarity and the social affinity among the users are fused, so that the accuracy of friend recommendation is greatly improved; by constructing a weighted heterogeneous network for each target user, the search range of friend recommendation is narrowed, the efficiency of friend recommendation is improved, and the requirement of a large-scale social network on linear time complexity is met.
Fig. 7 is a schematic structural diagram of a linear time friend recommendation system according to an embodiment of the application. The linear time friend recommendation system 70 of the embodiment of the application comprises:
interest calculation module 71: the method comprises the steps of estimating the topic interest degree of each user based on text content data issued by the users in a social network, and calculating the interest similarity among the users according to the topic interest degree;
the interaction calculating module 72: the method comprises the steps of calculating social affinities among users according to social interaction behaviors of the users in a social network;
the network construction module 73: the method comprises the steps of establishing a weighted heterogeneous network for a target user according to the interest similarity and the social affinity; the weighted heterogeneous network comprises potential friend candidates which have common interests or common friends with the target user;
community mining module 74: the community discovery method is used for mining a community set of a target user in a weighted heterogeneous network by adopting a bottom-up modularity value-added community discovery method, and calculating community membership values of communities in the target user affiliated community set;
friend recommendation module 75: and the method is used for extracting a set number of users from the weighted heterogeneous network according to the community membership value and generating a friend recommendation set of the target user.
Fig. 8 is a schematic diagram of a terminal structure according to an embodiment of the application. The terminal 80 includes a processor 81, a memory 82 coupled to the processor 81.
The memory 82 stores program instructions for implementing the linear time buddy recommendation method described above.
The processor 81 is configured to execute program instructions stored in the memory 82 to control the linear time friend recommendation.
The processor 81 may also be referred to as a CPU (Central Processing Unit ). The processor 81 may be an integrated circuit chip with signal processing capabilities. Processor 81 may also be a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
Fig. 9 is a schematic structural diagram of a storage medium according to an embodiment of the application. The storage medium according to the embodiment of the present application stores a program file 91 capable of implementing all the methods described above, where the program file 91 may be stored in the storage medium in the form of a software product, and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a Processor (Processor) to execute all or part of the steps of the methods according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (RAM, random Access Memory), a magnetic disk, an optical disk, or other various media capable of storing program codes, or a terminal device such as a computer, a server, a mobile phone, a tablet, or the like.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (11)
1. The linear time friend recommending method is characterized by comprising the following steps of:
Estimating the topic interest degree of each user based on text content data issued by the users in the social network, and calculating the interest similarity among the users according to the topic interest degree;
calculating social affinities among users according to social interaction behaviors of the users in the social network;
Establishing a weighted heterogeneous network for a target user according to the interest similarity and the social affinity, wherein the weighted heterogeneous network comprises potential friend candidates which have common interests or common friends with the target user;
A community discovery method of bottom-up modularity increment is adopted to mine a community set of the target user in a weighted heterogeneous network, and membership values of the target user in different communities are calculated;
Extracting a set number of users from the weighted heterogeneous network according to the membership value, and generating a friend recommendation set of the target user;
the estimating the subject interestingness of each user based on the text content data published by the user in the social network comprises the following steps:
the probability distribution of regarding a text collection posted by a user on a social network as related topics is expressed as The probability distribution of treating each topic as a large number of words is denoted/>And/>Super-parameters/>, each with dirichlet priorsAnd/>
Estimating distribution of D text sets on a topic using Gibbs samplingAnd distribution of T topics over words/>The topic interest degree estimation result matrix of the user is expressed as:
The matrix of D X T, labeled DT, where D represents the number of users, T represents the number of topics, and DT ij represents the number of words in the text set of user u i that are divided into topics T j;
A matrix of W x T, labeled WT, where W represents the number of unique words in the text, and WT ij represents the number of times unique word W i is divided into topics T j;
Normalizing the matrix DT to DT ' such that each row DT ' i· holds the equation DT ' i·||1 =1; the element DT 'ij in DT' represents the probability that the user u i is interested in the topic t j;
Normalizing the matrix WT to WT ' such that each row WT ' i· holds the equation WT ' i·||1 =1; the element WT 'ij in WT' represents the probability that the unique term w i is affiliated with the topic t j.
2. The method for linear time friend recommendation according to claim 1, wherein estimating the topic interest level of each user based on the text content data published by the user in the social network further comprises:
Estimating the domain influence of each user on the interesting subject by combining the social relation topological structure:
the domain influence r j (i) of the user s i on the topic t j is:
wherein DT' ij represents the interestingness of the user s i on the topic t j; μ is an adjustable parameter, "+1" is a constant.
3. The method for linear time friend recommendation according to claim 2, wherein the calculating the interest similarity between users according to the subject interestingness comprises:
The similarity of interests S (i, j) between users S i and S j is:
S(i,j)＝Sc(i,j)×Sd(i,j)
Wherein S c (i, j) represents the topic probability similarity of users S i and S j, and S d (i, j) represents the dominant interest similarity of users S i and S j;
The value of the topic probability similarity S c (i, j) between users S i and S j is:
Wherein T represents a set of topics; the interest degree average value of the user s i on all topics is obtained; /(I) Referring to the harmonic mean of two correlation values, it can be calculated as follows:
wherein eta represents a normalization factor for ensuring The value of (1) belongs to the interval [0,1], which can be set equal to the number of subjects; /(I)The interest value of the target user on the topic t z is given; /(I)Meaning the occurrence probability of the topic t z, which can be calculated according to the number of times of the unique word attaching topic t z in the matrix WT';
The main interest similarity S d (i, j) between the users S i and S j is calculated as:
Where D i and D j represent the main interest sets of users s i and s j, respectively.
4. The method of claim 1, wherein the calculating the social affinities between users according to the social interaction behaviors of the users in the social network comprises:
Calculating social interaction strength of the user s i pointing to s j:
If user s i "pays attention to" s j, F (i, j) =1; otherwise F (i, j) =0; n c (i, j) and N f (i, j) represent the number of comments and forwarding operations performed by the user s i for the text content posted by s j, respectively; τ is a regularization parameter.
5. The method for linear time friend recommendation according to claim 4, wherein the building a weighted heterogeneous network for the target user according to interest similarity and social affinity is specifically as follows:
According to the interest similarity and the social affinity, potential friend candidates which have common interests or common friends with the target user are found out from a social network; the potential friend candidates comprise a set number of friends with the most common friends and domain high-influence users with the most common interests;
Generating a weighted heterogeneous network G (V, E S,WS,EI,WI) according to the potential friend candidates, wherein V represents a node set and consists of a target user, a friend set of the target user and the potential friend candidate set; e S and E I represent sets of undirected and directed edges, respectively, corresponding to interest similarity and social affinity among users in V; w S and W I indicate the edge weights on E S and E I, respectively.
6. The linear-time friend recommendation method of claim 5, wherein the mining the community set of the target user in the weighted heterogeneous network by using the community discovery method with bottom-up modularity added value comprises:
for each node in the weighted heterogeneous network, removing the node from an original community, adding the node into an adjacent community, evaluating the increment of the modularity of the community, and if the increment of the modularity is positive, transferring the node into the adjacent community with the maximum increment of the modularity; otherwise, the node is kept in the original community;
and sequentially repeating the processes for all the nodes until the value of the community modularity is not increased, thereby obtaining a community set where the target user is located.
7. The linear-time friend recommendation method of claim 6, wherein calculating membership values for target users to attach to different communities in the community set comprises:
Assume that Is the ith community associated with node s o,/>Is/>In the node set adjacent to s o, s o is appended/>Membership degree/>The expression is as follows:
wherein, And/>Represents the undirected edge weights between nodes s o and s u, respectively, and the directed edge weights from s o to s u; /(I)Represents the sum of the undirected edge weights connected to s o; /(I)And/>Representing the sum of the outgoing side weights of s o and the sum of the incoming side weights of s u respectively; /(I)And/>
8. The method of claim 7, wherein the extracting a set number of users from the weighted heterogeneous network according to the community membership value of the target user, and generating the friend recommendation set of the target user comprise:
For each community associated with the target user, carrying out priority ranking on the node set in the communities through a LeaderRank ranking algorithm;
And extracting a proper number of high-influence users from each community according to the community membership value of the target user and the node ordering score value in each community, and generating a friend recommendation set of the target user.
9. A linear-time friend recommendation system utilizing the linear-time friend recommendation method of claim 1, comprising:
The interest calculation module: the method comprises the steps of estimating the topic interest degree of each user based on text content data issued by the users in a social network, and calculating the interest similarity among the users according to the topic interest degree;
and the interaction calculation module is used for: the method comprises the steps of calculating social affinities among users according to social interaction behaviors of the users in a social network;
And a network construction module: the method comprises the steps of establishing a weighted heterogeneous network for a target user according to the interest similarity and the social affinity; the weighted heterogeneous network comprises potential friend candidates which have common interests or common friends with the target user;
the community mining module: the community discovery method is used for mining a community set of the target user in a weighted heterogeneous network by adopting a bottom-up modularity value-added community discovery method, and calculating membership values of the target user in each community in the community set;
friend recommendation module: and the friend recommendation set is used for extracting a set number of users from the weighted heterogeneous network according to the membership value and generating the friend recommendation set of the target user.
10. A terminal comprising a processor, a memory coupled to the processor, wherein:
The memory stores program instructions for implementing the linear time friend recommendation method of any one of claims 1 to 8;
the processor is configured to execute the program instructions stored by the memory to control linear time friend recommendation.
11. A storage medium having stored thereon program instructions executable by a processor for performing the linear time buddy recommendation method of any of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111159485.5A CN113836444B (en) | 2021-09-30 | 2021-09-30 | Linear time friend recommendation method, system, terminal and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111159485.5A CN113836444B (en) | 2021-09-30 | 2021-09-30 | Linear time friend recommendation method, system, terminal and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113836444A CN113836444A (en) | 2021-12-24 |
| CN113836444B true CN113836444B (en) | 2024-04-19 |
Family
ID=78967708
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111159485.5A Active CN113836444B (en) | 2021-09-30 | 2021-09-30 | Linear time friend recommendation method, system, terminal and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113836444B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116541617B (en) * | 2023-06-29 | 2023-09-19 | 安徽大学 | Method and system for recommending participating users and places |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105740342A (en) * | 2016-01-22 | 2016-07-06 | 天津中科智能识别产业技术研究院有限公司 | Social relation topic model based social network friend recommendation method |
| CN106506183A (en) * | 2015-09-06 | 2017-03-15 | 国家计算机网络与信息安全管理中心 | The discovery method and device of Web Community |
| CN107526850A (en) * | 2017-10-12 | 2017-12-29 | 燕山大学 | Social networks friend recommendation method based on multiple personality feature mixed architecture |
| CN108460153A (en) * | 2018-03-27 | 2018-08-28 | 广西师范大学 | A kind of social media friend recommendation method of mixing blog article and customer relationship |
| WO2019205795A1 (en) * | 2018-04-26 | 2019-10-31 | 腾讯科技（深圳）有限公司 | Interest recommendation method, computer device, and storage medium |
| CN111143704A (en) * | 2019-12-20 | 2020-05-12 | 北京理工大学 | Online community friend recommendation method and system fusing user influence relationship |
-
2021
- 2021-09-30 CN CN202111159485.5A patent/CN113836444B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106506183A (en) * | 2015-09-06 | 2017-03-15 | 国家计算机网络与信息安全管理中心 | The discovery method and device of Web Community |
| CN105740342A (en) * | 2016-01-22 | 2016-07-06 | 天津中科智能识别产业技术研究院有限公司 | Social relation topic model based social network friend recommendation method |
| CN107526850A (en) * | 2017-10-12 | 2017-12-29 | 燕山大学 | Social networks friend recommendation method based on multiple personality feature mixed architecture |
| CN108460153A (en) * | 2018-03-27 | 2018-08-28 | 广西师范大学 | A kind of social media friend recommendation method of mixing blog article and customer relationship |
| WO2019205795A1 (en) * | 2018-04-26 | 2019-10-31 | 腾讯科技（深圳）有限公司 | Interest recommendation method, computer device, and storage medium |
| CN111143704A (en) * | 2019-12-20 | 2020-05-12 | 北京理工大学 | Online community friend recommendation method and system fusing user influence relationship |
Non-Patent Citations (1)
| Title |
|---|
| "基于社区划分和用户相似度的好友信息服务推荐研究";张继东等;《情报理论与实践》;20190109;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113836444A (en) | 2021-12-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AlNuaimi et al. | Streaming feature selection algorithms for big data: A survey | |
| Huang et al. | Updated review of advances in microRNAs and complex diseases: taxonomy, trends and challenges of computational models | |
| Umer et al. | Sentiment analysis of tweets using a unified convolutional neural network‐long short‐term memory network model | |
| CN111241421B (en) | User forwarding behavior prediction method based on social context information | |
| CN109508385B (en) | Character relation analysis method in webpage news data based on Bayesian network | |
| CN108427756B (en) | Personalized query word completion recommendation method and device based on same-class user model | |
| Lung et al. | Game theory and extremal optimization for community detection in complex dynamic networks | |
| Yang et al. | A multilevel information mining approach for expert recommendation in online scientific communities | |
| Komkhao et al. | Incremental collaborative filtering based on Mahalanobis distance and fuzzy membership for recommender systems | |
| Ni et al. | Comparative convolutional dynamic multi-attention recommendation model | |
| Bezáková et al. | Graph model selection using maximum likelihood | |
| CN113836444B (en) | Linear time friend recommendation method, system, terminal and storage medium | |
| Czibula et al. | Temporal ordering of cancer microarray data through a reinforcement learning based approach | |
| Salam et al. | Optimizing extreme learning machine using GWO algorithm for sentiment analysis | |
| CN116956081A (en) | Heterogeneous social network distribution outward generalization-oriented social label prediction method and system | |
| Wang et al. | Nit: Searching for rumors in social network through neighborhood information transmission | |
| CN114842247A (en) | Graph convolution network semi-supervised node classification method based on feature accumulation | |
| Devi et al. | An innovative word embedded and optimization based hybrid artificial intelligence approach for aspect-based sentiment analysis of app and cellphone reviews | |
| CN114443948A (en) | Ranking model training method, ranking method and device based on multi-scene data | |
| He et al. | A novel graph-based feature interaction model for click-through rate prediction | |
| Mandal et al. | Feature Subset Selection for High-Dimensional, Low Sampling Size Data Classification Using Ensemble Feature Selection With a Wrapper-Based Search | |
| Sridhar et al. | Extending Deep Neural Categorisation Models for Recommendations by Applying Gradient Based Learning | |
| Xiao et al. | A Mobile Bayesian Network Structure Learning Method Using Genetic Incremental K2 Algorithm and Random Attribute Order Technology | |
| Gu et al. | Overlapping community detection in social network based on microblog user model | |
| Cuevas et al. | An improved evolutionary algorithm for reducing the number of function evaluations |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |