Recommendation method based on social network user behavior information

Abstract

The invention relates to a recommendation method based on social network user behavior information, and the method comprises the steps: obtaining user behavior data, and obtaining activity holding position preference and holding time preference of a user; constructing a preference similar network by using the obtained user preferences, and fusing a user online social relation network to obtain a plurality of non-overlapping community structures reflecting the social relation and preference similarity between the users; calculating the value of each implicit point representing the project decision of the user by using a hybrid pairwise Markov random field model; all node states of each item are calculated, and then a recommendation list is formed through sorting selection. The method aims to recommend the optimal activity holding place and holding time for the activity organizer so as to maximize the number of participants of the activity and meet the decision planning requirement of the activity organizer.

Description

Technical Field

[0001] The present invention relates to the technical field of social network recommendation systems and data mining, and in particular to a recommendation method based on social network user behavior information. Background Art

[0002] Online and offline interactive activity social networks: refers to the activity social networks where event organizers publish social activities such as academic conferences and outdoor activities online, and publish details of the activities, including the theme, category, scale, organizer, offline time and location of the activities. Social network users select the activities they are interested in through browsing and searching online, and go offline to participate in the activities at the time of the activities. This combination of online and offline methods breaks the functional limitation of traditional social networks that only have online social functions. Users can communicate face-to-face with users with common interests and make friends when participating in offline activities.

[0003] In an event social network, the number of participants in an event plays a vital role in the success of the event. Therefore, for event organizers, maximizing the number of participants within the scope of the event is an important goal. The theme, location, and time of the event are important factors that affect whether users participate in an event. When an event organizer organizes an event, the biggest challenge is how to choose the location and time of the event to maximize the number of participants.

[0004] Currently, researchers focus on the different needs of users in activity social networks and have proposed a variety of methods, such as activity recommendation methods, activity participant prediction methods, etc., or planning a series of activities to be held by an event organizer. Given a set of activities, a set of activity time, and a set of users, consider the conflicts between activities and how to allocate these activities to different times to maximize the number of participants. Although these studies provide decision support for users and activity social network platforms, helping users quickly obtain activities of interest and improve their experience on the platform, few studies have recommended the best event venues and times for organizers from the perspective of event organizers to help them make decisions and maximize the number of participants in the event. To solve this problem, the present invention proposes a joint recommendation method for event venues and times for event organizers to provide decision support for event organizers. Summary of the invention

[0005] Technical issues solved

[0006] In view of the above-mentioned shortcomings of the prior art, the present invention aims to provide a joint recommendation method for event location and time from the perspective of event organizers based on the behavioral information of social network users, so as to provide decision support for event organizers.

[0007] Technical Solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] The present invention provides a recommendation method based on social network user behavior information, the method is used to recommend n activity area-time period pairs for an activity to be held, wherein the area-time period pair is defined as an item i, and the recommendation method comprises:

[0010] S10, obtaining data information including: user behavior data, a set of multiple event locations, a set of multiple event time periods, and event themes from the social network platform;

[0011] S20, constructing a user-region matrix and a user-time matrix according to the acquired data information, and integrating matrix decomposition and network embedding methods to model the user's regional preference and the user's time preference, respectively, to obtain the user's preference for the event location and the event time;

[0012] S30, constructing a preference similarity network using the acquired user preferences, and then integrating the user online social relationship network to obtain multiple non-overlapping community structures that reflect the social connections and preference similarities between users;

[0013] S40, using a mixed pairwise Markov random field model, calculate the value of each implicit point representing the user's decision on item i, when the implicit point = 1, it means that the user will participate in the activities held in this area-time period;

[0014] S50, calculate all node states of each project i, and count the number of values ​​1, that is, the number of users participating in the activity, and then sort the number of participating users of each project i from high to low, and select the first n projects i to form a recommendation list. The recommendation list is used for activity decision makers to make high-quality and efficient choices.

[0015] Furthermore, in step S20, the steps of constructing the user's regional preference and the user's time preference model are as follows:

[0016] S21: constructing a user-region matrix MR and a user-time matrix MT respectively according to the acquired data information;

[0017] S22: performing matrix decomposition on the user-region matrix MR and the user-time matrix MT respectively;

[0018] S23: construct a user-region bipartite graph and a user-time bipartite graph based on the user's historical behavior data, and generate a user-region embedding representation and a user-time embedding representation through random walk and Node2Vec method;

[0019] S24: The latent factors and embedded representations obtained by matrix decomposition are integrated to construct a preference model, and the preference model is used to calculate the user's final preferences in terms of region and time.

[0020] Furthermore, in step S24, an optimization objective function is set, wherein the optimization objective function combines the loss of matrix decomposition and the loss of network embedding, wherein the optimization objective function for user region preference is specifically:

[0021]

[0022] in,

[0023] In the above formula, L 1 is the objective function, λ, α, β are regularization parameters, M ur and The true value and the predicted value, is the user’s embedding representation in the region, is the user’s potential preference for a region, and the region’s potential feature m r , the embedding representation of the region is e r .

[0024] Furthermore, when calculating the final preference in step S24, the latent factor is integrated With embedding representation Get the final preference of user u in the region Fusion potential factor With embedding representation Get the final preference of user u in the region

[0025] Therefore, the user's comprehensive preference in terms of region and time is obtained as

[0026] Further integration of latent factor m r With the embedding representation e r , we get the potential feature θ of region r r , fusion potential factor m t With the embedding representation e t , we get the potential feature θ at time t t ;

[0027] Finally, the latent feature θ of the item i is obtained i =(θr ,θ t ).

[0028] Furthermore, obtaining the community structure in step S30 includes the following steps:

[0029] S31: Based on the acquired user preference information, calculate the preference similarity between user u and user v. The calculation formula is as follows:

[0030]

[0031] Set a similarity threshold ∈. If the preference similarity sim(u,v)>∈ between user u and user v, an edge is established between user u and user v. Otherwise, there is no edge between user u and user v. By calculating the preference similarity between any users, a social network based on user preferences is obtained.

[0032] S32: converting the user online social relationship network into an undirected graph, and then fusing the new user online social relationship network with the preference similarity network to obtain a single-layer network;

[0033] S33: Using the modularity-based Louvain algorithm, a single-layer network is divided into multiple non-overlapping community structures.

[0034] Furthermore, the mixed pairwise Markov random field model includes two types of nodes, namely, observation nodes and hidden nodes, wherein edges are constructed between the hidden nodes, and the edges represent dependencies between the nodes. The dependencies include two types, one is the relationship between the same user, and the other is the relationship between the same project. For the same project, if two users are located in the same community structure, there is an edge between the two users.

[0035] Further, in step S50, when calculating the value of the implicit point of each item i,

[0036] First, for the hidden node d ui and its edge potential function definition, where d ui Represents user u’s decision on item i;

[0037] The joint probability distribution of the mixed pairwise Markov random fields is expressed as a Gibbs distribution;

[0038] For each node d ui , use the belief propagation algorithm to infer its posterior probability Pd ui │Θ);

[0039] Calculate each node d in the mixed pairwise Markov random field model ui After the posterior probability of , its state is determined by the following calculation,

[0040]

[0041] Where S∈{0,1}

[0042] When ui =1 means user u will participate in the event held in region r and time period t, d ui =0 means that user u will not participate in the activities held in area r and time period t.

[0043] Beneficial Effects

[0044] Compared with the known public technology, the technical solution provided by the present invention has the following beneficial effects:

[0045] This solution innovatively obtains users' location preferences and time preferences through preference models based on users' historical behavior information. At the same time, it reflects multiple non-overlapping community structures that reflect the social connections and preference similarities between users. It can generate a recommendation table of event area-time periods for event planners. For an upcoming event on a social network, it recommends the optimal event location and time for event organizers to maximize the number of participants in the event, meet the decision-making and planning needs of event organizers, reduce the decision-making difficulty of event planners, and effectively help improve the quality of event hosting. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 A flowchart of the recommended method steps of the present invention;

[0048] Figure 2 A flow chart of the steps for modeling preferences of the present invention;

[0049] Figure 3 A flowchart of the steps of obtaining a community structure of the present invention;

[0050] Figure 4 It is a schematic diagram of user-region matrix decomposition of the present invention;

[0051] Figure 5 This is a schematic diagram of the application of the hybrid pairwise Markov random field model of this embodiment; DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] The present invention will be further described below in conjunction with the embodiments.

[0054] Example:

[0055] The present invention provides a recommendation method based on user behavior information of a social network. The object of the recommendation method is an event planner. The recommendation problem is as follows: for an activity social network, given an upcoming social activity e, a candidate region set R and a candidate time period set T and historical behavior information of users, the recommendation problem based on the user behavior information of the social network is defined, with the goal of maximizing the number of participants in the activity, for the event organizer, k pairs of activity hosting regions r (r∈R) and hosting time periods t (t∈T) are recommended for an upcoming event e, that is, k region-time period pairs c=(r,t). In particular, for convenience, a region-time period pair (r,t) is called a project i. For convenience of explanation, a region-time period pair is defined as a project i.

[0056] Reference Figure 1 , the recommended methods for the activity holding area r and time period t of this scheme include:

[0057] S10, obtaining data information including: user behavior data, a set R including multiple event locations, a set T including multiple event time periods, and an event theme X from a social network platform;

[0058] S20, constructing a user-region matrix and a user-time matrix according to the acquired data information, and integrating matrix decomposition and network embedding methods to respectively model the user's region preference and the user's time preference, and respectively obtaining the user's preference for the location and time of the event;

[0059] S30, constructing a preference similarity network using the acquired user preferences, and then integrating the user online social relationship network to obtain multiple non-overlapping community structures that reflect the social connections and preference similarities between users;

[0060] S40, using a mixed pairwise Markov random field model, calculate the value of each implicit point representing the user's decision on item i, when the implicit point = 1, it means that the user will participate in the activities held in this area-time period;

[0061] S50, calculate all node states of each project i, and count the number of values ​​1, that is, the number of users participating in the activity, and then sort the number of participating users of each project i from high to low, and select the first n projects i to form a recommendation list. The recommendation list is used for activity decision makers to make high-quality and efficient choices.

[0062] This solution innovatively obtains the user's location preference and time preference through a preference model based on the user's historical behavior information. At the same time, it obtains multiple non-overlapping community structures that reflect the social connections and preference similarities between users, and establishes a mixed pairwise Markov random field model. It can generate a recommendation table of event area-time period pairs for event planners, and recommend the optimal event location and time for an upcoming social network event to the event organizer, so as to maximize the number of participants in the event, reduce the decision-making difficulty of event planners, and effectively help improve the quality of event hosting.

[0063] In this embodiment, it is specifically defined that based on the locations of all activities held in the social network platform, all geographical locations are divided into a candidate area set R using the k-means clustering algorithm, that is, R = {r 1 ,r 2 ,…,r P}, where r 2 Represents a certain area. At the same time, the time of all activities is divided into a candidate time period set T. In particular, 21 time periods are defined, consisting of 7 time periods separated by days (each time period represents a day of the week) and 3 time periods separated by time (morning, afternoon, and evening), that is, T = Monday morning, Monday afternoon...Sunday afternoon, Sunday evening. Based on the above method, each event location l is quantified into a region r∈R, and each event time d is quantified into a time period t∈T. Finally, each activity in the activity social network corresponds to an activity theme x, an activity region r, and an activity time period t.

[0064] User preference is a key factor that affects whether a user participates in a social activity. Therefore, modeling user preferences based on user behavior information in the active social network is the key to solving the social network recommendation problem. In step S20, referring to Figure 2 ,The steps to model the user's regional preference and the user's time preference are as follows:

[0065] S21: Construct the user-region matrix MR∈R based on the acquired data information M×|R| and the user-time matrix MT∈R M ×|T| , each element in the user-region matrix MR ur∈{0,1} indicates whether the user has participated in an event held in region r in the past.

[0066] Each element MT in the user-time matrix ut ∈{0,1} indicates whether the user has participated in an event held at time t in the past.

[0067] S22: Perform matrix factorization on the user-region matrix MR and the user-time matrix MT respectively;

[0068] S23: Construct a user-region bipartite graph and a user-time bipartite graph based on the user's historical behavior data, and generate user-region embedding representations and user-time embedding representations through random walk and Node2Vec methods;

[0069] S24: Fuse the latent factors obtained from matrix factorization and the embedding representations to construct a preference model, and use the preference model to calculate the user's final preferences in terms of region and time.

[0070] In steps S22 - S24, in this embodiment, the construction of the user's region preference modeling is taken as an example for illustration. In addition, the algorithm principles of the user's time period preference modeling are the same.

[0071] Specifically, based on the constructed user-region matrix, a method that fuses matrix factorization and network embedding is designed to more accurately obtain the user's region preferences. The specific steps are as follows:

[0072] 1. Based on the user-region matrix MR, use the matrix factorization method to decompose it into a user region preference matrix and a region attribute matrix Refer to Figure 4 , where represents the latent region preference of user u, and m r □RR represents the latent attribute of region r. The optimization objective function of matrix factorization is:

[0073]

[0074] 2. Next is the network embedding representation learning based on the user-region matrix. First, based on the user-region matrix MR constructed from the user's historical behavior information, construct a user-region bipartite graph. In this graph, the user's behavior data consists of the binary tuple <u, r>. If user u has participated in an event held in region r, then an edge is established between user u and region r.

[0075] Secondly, based on the constructed user-region bipartite graph, random walks are used to walk any node in the bipartite graph. The walk will be repeated until the predetermined step length is reached. In this way, a random walk can obtain a path that captures the network structure and deep semantics. The walk probability of a random walk is:

[0076]

[0077] Among them, N(v i ) represents the node v in the bipartite graph i The set of neighbors.

[0078] Then, based on the random walk path, we get the node sequence of different starting nodes. Then, based on the co-occurrence of the node sequence and the fixed-length window, we get the node v i Construct a neighborhood set Φ(v i ), and use the Node2Vec method to generate the embedded representation of the node using the Skip-Gram method. The optimization objective function is:

[0079]

[0080] Among them, Φ'(v i ) represents node v i Negative samples.

[0081] Again, based on the above optimization objective function, stochastic gradient descent is used to optimize this objective and learn node v i The embedding representation f(v i ). Finally, based on the above steps, the embedding representation of user u in the region is obtained as The embedding of region r is denoted as e r

[0082] 3. In order to obtain more accurate user regional preferences, a preference calculation method integrating matrix decomposition and network embedding is designed. In step S24, an optimization objective function is set. The optimization objective function combines the loss of matrix decomposition and the loss of network embedding. The optimization objective function for user regional preferences is specifically:

[0083]

[0084] in,

[0085] In the above formula, L 1 is the objective function, λ, α, β are regularization parameters, M ur and The true value and the predicted value, is the user’s embedding representation in the region, is the user’s potential preference for a region, and the region’s potential feature m r , the embedding representation of the region is e r Finally, the fusion potential factor With embedding representation Get the final preference of user u in the region

[0086] The comprehensive preference calculation of the fusion area and time is used to calculate the final preference in step S24, and the fusion potential factor With embedding representation Get the final preference of user u in the region Fusion potential factor With embedding representation Get the final preference of user u in the region Therefore, the user's comprehensive preference in terms of region and time is obtained as Further integration of latent factor m r With the embedding representation e r , we get the potential feature θ of region r r , fusion potential factor m t With the embedding representation e t , we get the potential feature θ at time t t ; Finally, the potential feature θ of item i is obtained i =(θ r ,θ t ).

[0087] Further, in step S30, the community structure is obtained, referring to Figure 3 , including the following steps:

[0088] S31: Based on the acquired user preferences, calculate the preference similarity between user u and user v. The calculation formula is as follows:

[0089]

[0090] Set a similarity threshold ∈. If the preference similarity between user u and user v is sim(u,v)>∈, an edge is established between user u and user v. Otherwise, there is no edge between user u and user v. By calculating the preference similarity between any users, a social network based on user preferences is obtained, namely, a preference similarity network.

[0091] S32: converting the user online social relationship network into an undirected graph, and then fusing the new user online social relationship network with the preference similarity network to obtain a single-layer network;

[0092] S33: Using the modularity-based Louvain algorithm, a single-layer network is divided into multiple non-overlapping community structures.

[0093] By utilizing the probabilistic reasoning ability of Markov random fields, the hybrid pairwise Markov random field model can more effectively improve the accuracy of recommendations;

[0094] Reference Figure 5 In the mixed pairwise Markov random field model, there are two types of nodes: observation nodes and hidden nodes. The circular nodes are observation nodes, and the hexagonal nodes are hidden nodes. For example, as an observation node, θ u represents the preference of user u, θ i Represents the attribute characteristics of item i, as a hidden node d ui Represents the decision of user u on item i, with a value of 0 or 1. Each hidden node d ui The generation connects two observation nodes θ u and θ i ,θ u With θ i The value of is calculated by the user preference modeling module. Each hidden node d ui The state of the observation node to which it is connected (i.e., the user preference θ u 、Item characteristics θ i ) and the hidden nodes connected to it (i.e., nodes d uj , node d vi With node d wi ) is determined by

[0095] Among them, edges are constructed between implicit nodes. Edges represent the dependency relationship between nodes. Dependency relationships include two types. One is the relationship between the same user. For example, node d in the figure ui With node d uj The other is the relationship between the same project. For the same project, if two users are in the same community structure, there is an edge between them. For example, node d ui With node d vi The edge between nodes d ui With node d wi The edge between.

[0096] In step S50, when calculating the value of the implicit point of each item i,

[0097] First, for the hidden node d ui and its edge potential function definition, where d ui represents the decision of user u on item i; specifically, for node d ui , ψ ui (d ui |θ u ,θ i ) represents its node potential function, ψ ui,uj (dui ,d uj ) and ψ ui,vi (d ui ,d vi ) represents the potential function of its edge. ui (d ui |θ u ,θ i ) is node d ui Given user u's preference θ u and θ for item i i The prior probability distribution after is given by θ u With θ i The Cosine similarity is calculated,

[0098]

[0099] For node d ui With d uj The edge between them has a potential function ψ ui,uj (d ui ,d uj ) reflects the similarity between items i and j, and is calculated as follows:

[0100]

[0101] For node d ui With d vi The edge between them has a potential function ψ ui,vi (d ui ,d vi ) reflects the similarity between users u and v in the same community, and is calculated as follows:

[0102]

[0103] The joint probability distribution of the mixed pairwise Markov random fields is expressed as a Gibbs distribution;

[0104] The joint probability distribution of the mixed pairwise Markov random field G = (V, E) is expressed as,

[0105]

[0106] Among them, Z is the normalization factor, V and E are the user set and edge set of the mixed pairwise Markov random field.

[0107] For each node d ui , use the belief propagation algorithm to infer its posterior probability Pd ui │Θ);

[0108] Calculate each node d in the mixed pairwise Markov random field modelui After the posterior probability of , its state is determined by the following calculation,

[0109]

[0110] Where S∈{0,1}

[0111] When ui =1 means user u will participate in the event held in region r and time period t, d ui = 0 means that user u will not participate in the event held in region r and time period t. Then, in order to achieve the goal of maximizing the number of participants in a certain event, for each item i = (r, t), calculate the state of each node and calculate the number of nodes with a state of 1, that is, the number of users participating in the event ∑ u d ui Finally, the number of participating users ∑ for each i is sorted from high to low u d ui , select the first few projects i to form a recommendation list and push it to the project planner.

[0112] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above embodiments can be implemented by software, or by means of software plus necessary general hardware platforms. Based on such understanding, the technical solutions of the above embodiments can be embodied in the form of software products, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the recommended methods described in various embodiments of the present invention.

[0113] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A recommendation method based on social network user behavior information, the recommendation method is used to recommend n activity area-time pairs for an upcoming activity, wherein: A region-time period pair is defined as an item i, wherein the recommendation method comprises: S10, obtaining data information including: user behavior data, a set of multiple event locations, a set of multiple event time periods, and event themes from the social network platform; S20, constructing a user-region matrix and a user-time matrix according to the acquired data information, and integrating matrix decomposition and network embedding methods to respectively model the user's region preference and the user's time preference, and respectively obtaining the user's preference for the location and time of the event; S30, constructing a preference similarity network using the acquired user preferences, and then integrating the user online social relationship network to obtain multiple non-overlapping community structures that reflect the social connections and preference similarities between users; S40, using a mixed pairwise Markov random field model, calculate the value of each implicit point representing the user's decision on item i, when the implicit point = 1, it means that the user will participate in the activities held in this area-time period; S50, calculate all node states of each project i, and count the number of values ​​1, that is, the number of users participating in the activity, and then sort the number of participating users of each project i from high to low, and select the first n projects i to form a recommendation list.

2. The recommendation method based on social network user behavior information according to claim 1, characterized in that: In step S20, the steps of constructing the user's regional preference and the user's time preference model are as follows: S21: constructing a user-region matrix MR and a user-time matrix MT respectively according to the acquired data information; S22: performing matrix decomposition on the user-region matrix MR and the user-time matrix MT respectively; S23: construct a user-region bipartite graph and a user-time bipartite graph based on the user's historical behavior data, and generate a user-region embedding representation and a user-time embedding representation through random walk and Node2Vec method; S24: The latent factors and embedded representations obtained by matrix decomposition are integrated to construct a preference model, and the preference model is used to calculate the user's final preferences in terms of region and time.

3. The recommendation method based on social network user behavior information according to claim 2, characterized in that: In step S24, an optimization objective function is set, which combines the loss of matrix decomposition and the loss of network embedding. The optimization objective function for user region preference is specifically: in, In the above formula, L1 is the objective function, λ, α, β are regularization parameters, and M ur and The true value and the predicted value, is the user’s embedding representation in the region, is the user’s potential preference for a region, and the region’s potential feature m r , the embedding representation of the region is e r .

4. The recommendation method based on social network user behavior information according to claim 3, characterized in that: When calculating the final preference in step S24, the latent factor is integrated With embedding representation Get the final preference of user u in the region Fusion potential factor With embedding representation Get the final preference of user u in the region Therefore, the user's comprehensive preference in terms of region and time is obtained as Further integration of latent factor m r With the embedding representation e r , we get the potential feature θ of region r r , fusion potential factor m t With the embedding representation e t , we get the potential feature θ at time t t ; Finally, the latent feature θ of the item i is obtained i =(θ r ,θ t ).

5. The recommendation method based on social network user behavior information according to claim 1, characterized in that: The community structure is obtained in step S30, including the following steps: S31: Based on the acquired user preference information, the preference similarity between user u and user v is calculated. The calculation formula is as follows: Set a similarity threshold ∈. If the preference similarity between user u and user v is sim(u,v)>∈, an edge is established between user u and user v. Otherwise, there is no edge between user u and user v. By calculating the preference similarity between any users, a social network based on user preferences is obtained, namely, a preference similarity network. S32: converting the user online social relationship network into an undirected graph, and then fusing the new user online social relationship network with the obtained preference similarity network to obtain a single-layer network; S33: Use the modularity-based Louvain algorithm to divide a single-layer network into multiple non-overlapping community structures.

6. The recommendation method based on social network user behavior information according to claim 1, characterized in that: The mixed pairwise Markov random field model includes two types of nodes, namely, observation nodes and hidden nodes. Edges are constructed between the hidden nodes, and the edges represent dependencies between the nodes. The dependencies include two types, one is the relationship between the same user, and the other is the relationship between the same project. For the same project, if two users are located in the same community structure, there is an edge between the two users.

7. The recommendation method based on social network user behavior information according to claim 6, characterized in that: In step S50, when calculating the value of the implicit point of each item i, First, for the hidden node d ui and its edge potential function definition, where d ui Represents user u’s decision on item i; The joint probability distribution of the mixed pairwise Markov random fields is expressed as a Gibbs distribution; For each node d ui , use the belief propagation algorithm to infer its posterior probability Pd ui │Θ); Calculate each node d in the mixed pairwise Markov random field model ui After the posterior probability of , its state is determined by the following calculation, Where S∈{0,1} When ui =1 means user u will participate in the event held in region r and time period t, d ui =0 means that user u will not participate in the activities held in area r and time period t.

Images