Community recommendation method based on interest induction propagation

Abstract

The invention provides a community recommendation method based on interest induction propagation. According to the method, the accuracy and reliability of group recommendation are improved. The method comprises the following steps: firstly, carrying out preprocessing, interest intensity calculation and interest stability calculation on interest tags of community users, and then carrying out social behavior analysis; thirdly, calculating the self-sensing trust degree and mutual sensing trust degree of the user, obtaining the interest sensing trust degree through fusion calculation, building a trust degree network according to the interest sensing trust degree, calculating the interest influence, simulating the influence of behavior interaction in the community on the interest of the user by utilizing a sorting algorithm, and in the trust degree network, calculating the interest sensing trust degree of the user; and simulating a transmission process of interest influence by adopting a PageRank algorithm, calculating group interestingness in the community, and realizing community information recommendation through the group interestingness. According to the method, individual selection and group influence factors of community user interest decision making are innovatively combined, and the accuracy and reliability of group recommendation are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of community recommendation, specifically a community recommendation method based on interest-based propagation. Background Technology

[0002] In recent years, with the rise of internet technology, social networks, primarily based on platforms like Dianping (a popular Chinese review platform) and group-buying services, have developed rapidly. Through activities such as interest-based friendships and personalized recommendations, online groups have gradually formed several online social groups centered on the transmission of interests. These online social groups, or communities for short, face numerous challenges in information selection due to the rapid expansion of information resources within them. Therefore, identifying user interest characteristics and building online information services based on personalized recommendations is urgently needed. Unlike recommendation methods targeting individual users, group recommendations, which focus on group users, can effectively integrate the differentiated interest needs within a group and establish multi-scenario decision-making based on group preferences. This maximizes group satisfaction while improving recommendation efficiency. In this context, accurately identifying the interaction between individual preferences and group preferences is a key aspect of improving the effectiveness of community recommendations.

[0003] Early research on group recommendation models tended to express group preferences by utilizing the similarity of preferences between individuals. This involved using similarity calculations based on inter-group interests and clustering algorithms to recommend shared resources to the target group. However, with the expansion of research scope and the rapid increase in user group activities, calculations based on inter-individual interest similarity faced practical problems such as feature sparsity, cold start of interests, and dynamic changes in interests. The research focus of group recommendation gradually shifted to using multi-strategy fusion for group recommendation. Multi-strategy fusion recommendation methods can be mainly divided into recommendation fusion, model fusion, and feature fusion. Model fusion refers to using multiple strategies to generate a model of the group's interest preferences. The method involves refining individual recommendations based on feedback to achieve group recommendation. Feature fusion primarily extracts user or interest features from the group to establish a multi-feature matrix oriented towards group interest selection or interest influence, thereby enabling group resource discovery. Furthermore, social influence factors are considered as an important factor in group recommendation, and the weighting of interest dissemination is achieved by quantifying the influence of group members. This approach has yielded good recommendation results in group decision-making. However, the above method requires high computational accuracy for core user identification and lacks correlation analysis between quantifying influence dissemination and interest dissemination. The recommendation effect for non-core users within the group needs improvement.

[0004] In summary, existing group recommendation methods do not organically integrate individual choices that determine group members' interest decisions with group influence factors. Some methods only consider the similarity of individual choices without considering the degree of interest influence between groups, or rely solely on user influence for resource item rating without considering the intrinsic relationships between multiple factors. Furthermore, given the dynamic nature of group interests, it is necessary to introduce feature factors that can quantify the overall interest propagation path to improve the interpretability of information resource services while enhancing the accuracy of knowledge recommendation. Against this backdrop, this invention studies the temporal evolution of group interests in social networks. By integrating group interest choices and interest influence, it introduces interest trust into the influence propagation network, proposing a community recommendation method based on interest-sensory propagation. This method uses individual interest choices as the perceived trust of group interests, while simultaneously using user-related interest similarity and influence factors to treat the interest influence within the group as mutual trust. Finally, by establishing an interest influence propagation network centered on perceived trust, it achieves the calculation of group interest. Summary of the Invention

[0005] To address the above issues, this invention quantifies the degree of interest influence among groups into mutual interest trust by integrating social choice and social influence factors. This provides a new research perspective for in-depth exploration of knowledge flow in the dynamic interest drift process of groups. Based on this, this invention proposes a community recommendation method based on interest-sensing propagation.

[0006] Therefore, the technical solution adopted by the present invention is as follows:

[0007] S1: Extract the interest tag set of community users through tag preprocessing, calculate the interest intensity and interest stability based on the interest tag set, then divide the community users into influence users and selection users, and perform social behavior analysis to extract user activity dataset and user dissemination dataset.

[0008] S2: Analyze the interest trust level of community users. The analysis process of interest trust level is as follows: calculate the self-perceived trust level by analyzing the interest behavior characteristics of selective users; calculate the mutual perception trust level by analyzing the interest behavior characteristics of influential users; calculate the perceived trust level based on the self-perceived trust level and the mutual perception trust level.

[0009] S3: Use the analysis results of interest trust to construct a trust network; calculate the interest influence based on the trust network and the PageRank algorithm; further calculate the group trust based on the interest influence to complete the community interest recommendation.

[0010] Furthermore, the intensity of interest is characterized by the frequency of user access to the tag;

[0011] For the "user-tag" pair, the formula for calculating the intensity of a user's interest is as follows:

[0012]

[0013] in, For users For tags Access frequency, For users Total access frequency for all tags Indicates user For tags j The interest intensity is s, where s is the total number of users and p is the total number of tags.

[0014] Furthermore, the interest stability is expressed by introducing a forgetting function to represent the stability of the interest intensity over time, and the calculation formula is as follows:

[0015]

[0016] Where k is the user's interest half-life, For tags The time of the nth occurrence for Current users For tags Intensity of interest For tags The interval between two consecutive occurrences;

[0017] Constructing a user interest tag matrix , express Current users For tags The intensity of interest, i.e. .

[0018] Furthermore, the user activity dataset includes the number of user forwards. Number of comments Number of likes ;

[0019] The user propagation dataset includes the number of times a user's message has been forwarded. Number of comments Number of likes .

[0020] Furthermore, the perceived level of trust includes relational trust and interactive trust.

[0021] The relationship trust level is established by acquiring the friend relationships between users and building a friend relationship matrix. ,in =1 indicates user and There is a friend relationship between them, and the following conditions are met: = ;

[0022] The interaction trust level is established by calculating the interaction behavior between users and building a behavior relationship matrix. ,in For users right The weighted value of comments and likes on the post. The formula is as follows:

[0023]

[0024] Where N(xy) represents the user right The number of likes, N(x) represents the number of likes by a user. Total number of likes, N p (xy) represents the user In the p-th sampling period The number of comments, N(x) represents the number of users. The total number of comments in the p-th sampling period, where n is the time span of the comments. The time-related impact value of the commenting behavior. , These represent the current moment and the moment the commenting action occurred, respectively.

[0025] The perceived trust level is calculated by weighting relational trust level and interaction trust level, as shown in the following formula.

[0026]

[0027] in, Indicates user right Self-perceived trust level It is represented as the self-perception coefficient and is used to adjust the weight ratio between relationship trust and interaction trust.

[0028] Furthermore, the mutual trust level includes interest similarity factors and user influence factors;

[0029] The interest similarity factor Influenced by the frequency with which users access the same tags, the value is calculated using the user interest tag matrix, as shown in the following formula:

[0030]

[0031] in, Indicates a label; and Representing users respectively and The set of tags accessed. and Representing users respectively and For tags Intensity of interest Indicates the label The set of users who have accessed the site, w p,l Indicates all tags The sum of the intensity of interest of the users who made the visit; For users , The personalization index of the tags that are accessed together.

[0032] Furthermore, the personalization index is related to user interest span and tag popularity. The interest span refers to the total number of tags accessed by a user, and the tag popularity refers to the number of users in the community accessing the same tag. The formula for the personalization index is as follows:

[0033]

[0034] in, For users The total number of tags visited, N(l) is the number of users in the community who have visited tag l.

[0035] Furthermore, the user influence factor is calculated from user activity and user reach. The user activity is used to calculate the quantity of information, and the user reach is used to calculate the quality of information. Influence factors The calculation formula is as follows:

[0036]

[0037]

[0038]

[0039] Where C and N are the information quality value and the information quantity value, respectively. , , , , and To calculate the weights.

[0040] Furthermore, the mutual trust level is obtained by weighting the interest similarity factor and the user influence factor. and The formula for calculating mutual trust is as follows:

[0041]

[0042] in, The similarity threshold;

[0043] The perceived trust level is calculated by fusing self-perceived trust level and mutual-perceived trust level. and Trust level The calculation formula is as follows:

[0044]

[0045] in, The sensing coefficient is used to adjust the weighting relationship between self-perceived trust in interest and perceived trust.

[0046] Furthermore, the nodes of the trust network are users, the edges of the trust network represent the perceived trust levels between users, and the interest influence propagates from a node along the edges to other nodes. The aforementioned interest influence The formula for calculating ) is:

[0047]

[0048] in, User and The level of trust in the sense, ; q is the jump coefficient, User The sensed trust level is higher than the preset trust threshold. The total number of users;

[0049] when hour, ,when hour, ;

[0050] The group trust level The calculation is as follows:

[0051]

[0052] in, Represents the user at time t and tags Group trust level.

[0053] Compared with the prior art, the advantages of the present invention are as follows:

[0054] 1. This invention proposes a method for organically integrating individual choices and group influencing factors in interest decision-making among group members. This method calculates mutual trust between users by extracting their own influence characteristics and the interest similarity between different users. Social relationship weights are introduced into interactive behaviors such as commenting, forwarding, and liking to calculate self-trust. The mutual trust and self-trust are combined to obtain the interest-sensing trust and interest influence among members. This method can significantly improve the quality of information push in virtual communities, objectively reproduce the random walk path of interest vectors, and improve the accuracy and reliability of group recommendations.

[0055] 2. This invention addresses the dynamic changes in group interests by introducing feature factors that can quantify the overall interest propagation path, thereby improving the interpretability of information resource services while enhancing the accuracy of knowledge recommendation. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart of the community interest recommendation framework of the present invention;

[0058] Figure 2 This is a schematic diagram illustrating the propagation of interest influence in the trust network of the present invention;

[0059] Figure 3 This is a schematic diagram illustrating the correlation analysis of the self-inductance coefficient of the present invention;

[0060] Figure 4 This is a schematic diagram comparing the accuracy of interest tag prediction in this invention;

[0061] Figure 5 This is a schematic diagram illustrating the impact of the perceived trust level on the ranking difference according to the present invention;

[0062] Figure 6 This is a schematic diagram illustrating the impact of the number of groups on the recommendation effect in this invention;

[0063] Figure 7 This is a schematic diagram comparing the F-values ​​of different group recommendation algorithms of the present invention; Detailed Implementation

[0064] To achieve the above objectives, the present invention is implemented through the following technical solutions, combined with Figure 1-7 This invention provides a community recommendation method based on interest-based propagation, the method comprising:

[0065] S1. Extract the interest tags of users in the community, and preprocess the interest tags, calculate the interest intensity and interest stability, and then perform social behavior analysis to provide decision-making reference for judging the random walk pattern of interest vectors.

[0066] Within a community, users express their interests and information needs through social tagging behavior, using text tags. These text tags are called interest tags.

[0067] When extracting interest tags, it is necessary to first perform tag preprocessing, interest intensity calculation, and interest stability calculation.

[0068] Tag preprocessing: Analyze user text using syntactic rules and convert it into a set of feature phrases. Syntactic rules refer to the rules of how words, phrases and sentences are combined in a language, and are mainly used to analyze and construct sentence structure.

[0069] Intensity of interest: The tagging behavior of internet users on different tags reflects significant differences in their interest direction and intensity. The tagging frequency can objectively reflect the intensity of users' interest preferences; that is, the higher the tagging frequency, the higher the degree of interest in the content of that tag. Therefore, we first represent the ternary relationship of "user-item-tag" as a "user-tag" binary tuple, and then use the tagging frequency to characterize the tag interest intensity. The tagging behavior mainly refers to collecting, following, and commenting. The calculation formula is as follows:

[0070]

[0071] in, For users For tags Access frequency, For users Total access frequency for all tags.

[0072] Interest stability: Considering the impact of time on interest evolution, a forgetting function is introduced to achieve dynamic updates of interest tags. The intensity of a user's interest typically changes over time, undergoing both forgetting and strengthening. To characterize the dynamic features of user interest tags, a forgetting function is introduced into the tag interest calculation to obtain the stability of interest intensity over time. The calculation formula is as follows:

[0073]

[0074] Where k is the user's interest half-life, For tags The time of the nth occurrence ( , )for Current users For tags Intensity of interest For tags The interval between two consecutive occurrences. By comprehensively considering the intensity and stability of interest, the "user-tag" binary is ultimately transformed into a user-tag matrix. ,in Indicates user For tags The intensity of interest.

[0075] Social Behavior Analysis: From the perspective of the explicit characteristics of user interests, the differences in the intensity and stability of user interests within a community are not only influenced by their own interest choices but also by the deeper impact of user interactions. For example, driven by long-term interests, users join communities they are familiar with, often exhibiting more active social behaviors such as expressing opinions and sharing information. In group recommendations, users tend to label their interests based on their past preferences, meaning social choice dominates, and the probability of autonomous choice is higher than that of group influence. However, under the influence of external stimuli or short-term interests, users will exhibit lower-frequency social behaviors in communities where the discussion content is unfamiliar. Their motivation for participation may be to learn and acquire new information and knowledge. In this case, users' interest decisions are usually influenced by the group, and in group recommendations, users label their interests based on conformity, meaning social influence dominates, and the probability of group influence is higher than that of autonomous choice.

[0076] By analyzing users' social behavior characteristics at different times, we can provide a reliable decision-making reference for accurately judging the random walk pattern of interest vectors. To this end, we use user activity and dissemination as evaluation factors. Activity refers to the number of times a user forwards, comments, and likes within a certain period of time, while dissemination refers to the number of times a user is forwarded, commented on, and liked within a certain period of time. Activity can quantify the real-time changes in a user's interests, that is, users selectively communicate or tag users based on their own interests, which is called interest selection. Dissemination reflects the degree and coverage of the user's interest dissemination within the community, that is, the process by which a user spreads messages outward with their influence, which is called interest influence.

[0077] In addition, in order to acquire users within the community With users Interest and trust levels can be determined by calculating the level of activity among users. Interest received by users To assess the degree of influence, considering that user interactions are dynamic and that more recent social behaviors tend to have a higher level of trust, we analyzed user text tags and their tagging times to statistically determine the impact. Construct a user-to-user reposting dataset based on the number of reposts, comments, and likes from other users. Comment dataset Likes dataset Furthermore, to quantify the influence of core users on the overall dissemination of interests within the community, we can calculate users' community influence to determine the degree and scope of impact of different users' social behaviors on changes in community interests. This can be achieved by compiling a dataset of shared posts, comments, and likes within the community over a specific period. The dataset being commented on Liked dataset .

[0078] S2. Input the interest tags of users in the community into the IIP model, calculate the user's self-perceived trust level and mutual perception trust level respectively, and obtain the interest perception trust level through fusion calculation, that is, the degree to which the user's interests are influenced by other users in the community.

[0079] The interest trends of users within a community are not only related to their own interest choices, but are also influenced by the interests of other users in the community, thus driving the overall interest evolution within the community. The level of trust in the interests of community users can be divided into self-perceived trust and mutual trust.

[0080] Self-perceived trust level:

[0081] By analyzing a user group where social choice plays a dominant role—namely, selective users—who express their interests, opinions, and emotional stances through social behaviors such as interest-based friendships, forwarding, and commenting, we can identify interest communities formed by this user group. These communities typically exhibit high levels of individual autonomy and selectivity, resulting in a relatively high probability of dispersion in the evolution of group interests. Therefore, from the perspective of social choice in interest evolution, we analyze the characteristics of user interaction behaviors to calculate the perceived trust level among users. Specifically, perceived trust level mainly stems from the localization characteristics of individual user choices, primarily including relational trust and interaction trust.

[0082] Relationship trust refers to the homogeneity of a user's interests with those of their close friends; that is, people with close relationships are more likely to establish interest-based interactions. Relationship trust is primarily assessed by acquiring friend relationships among users and building a friend relationship matrix. ,in =1 indicates user u x with u y There is a friend relationship between them, and the following conditions are met: = ,

[0083] Interaction trust refers to the ability of users to tag text labels that interest them through activities such as commenting and liking, thereby spreading their interests within the community. Interaction trust is established by calculating the interactive behaviors between different users and building a behavioral relationship matrix. ,in For user u x For u y The weighted value of comments and likes on a post, considering the difficulty in obtaining the time of a user's like action, only adds an interaction time factor to the comment action. The calculation formula is as follows:

[0084]

[0085] Where N(xy) represents user u x For u y The number of likes, N(x) represents the number of likes for user u. x Total number of likes, N p (xy) represents user u x For u in the p-th sampling period y The number of comments, N(x) represents the number of comments by user u. x The total number of comments in the p-th sampling period, where n is the time span of the comments. t represents the time-dependent impact of the commenting behavior. c t f These represent the current moment and the moment the commenting action occurred, respectively.

[0086] By calculating user u separately x and u y Relationship trust and interaction trust are weighted to obtain user u x For u y Self-confidence ), Calculation formula:

[0087]

[0088] Where 0≤ ≤1 represents the self-perception coefficient, used to adjust the weight ratio of the two levels of trust.

[0089] Mutual trust level:

[0090] Users whose social influence dominates, known as influencer users, are easily influenced by herd mentality and have a significant acceptance of group influence. These users tend to alter their inherent interests during behavioral interactions, resulting in a high probability of convergence in interest evolution. Unlike social selection, when individuals are deeply influenced by the external environment of a community, they are more likely to be influenced by the opinions of a "specific group" when making interest decisions. This "specific group" refers to a user group that shares similar interests and possesses core user characteristics.

[0091] Specifically, considering the characteristics of short-term and long-term interest evolution, users often periodically tag certain labels that express their interests and preferences based on their information needs within a certain time period. This gradually allows them to find several "interest partners" with similar interests. These "interest partners" can establish implicit friendships based on their tagging behavior of the same labels. For example, within a time window, users... and All tags This generates tagging behavior, thereby forming a shared set of interest tags between the two parties. Accordingly, multiple different users tag the tags. The accumulated labeling behavior gradually improved the labeling quality. The popularity of an individual's interests increases, and this rising popularity, along with the increase in the number of users and the amount of tagging activity within the community, in turn influences every user who has tagged the user. This ultimately creates an interest-based influence based on similarity. Furthermore, the intensity and scope of influence from other users within the community vary significantly among individuals. High-influence users, i.e., a "specific group," exhibit significantly enhanced speed, scope, and duration of interest dissemination within the community compared to ordinary users. This means that interest dissemination and influence dissemination are always intertwined. Therefore, when calculating a user's interest influence, it is necessary to consider the user's own community influence on the interest influence among members, thereby quantifying the degree to which an individual's interest is affected by global interest influence.

[0092] Based on the above analysis, the degree to which a user is influenced by the interests of other users can be represented by mutual trust, which is affected by interest similarity factors and user influence factors.

[0093] Interest similarity factor: The interest similarity factor essentially reflects the closeness of the relationship between different users who "make friends through shared interests" by using resource tags, that is, the degree of similarity between users. and The higher the frequency of accessing the same tags within a shared interest set, the higher the similarity of interests between the two users, and the more likely they are to influence each other's random walks of interests. Based on the user tag matrix, a user interest tag matrix is ​​constructed. , express Current users For tags The intensity of interest, i.e. Based on the intensity of user's tag interests, the user u is calculated. x u y Interest similarity factor Calculation formula:

[0094]

[0095] in, Indicates label; L(u) x ), L (u y ) represent user u x u y The set of tags accessed. , They represent user u respectively x u y For tags Intensity of interest Indicates the label The set of users who have accessed the site, w p,l Indicates all tags The sum of the intensity of interest of the users who made the visit; For users , The personalization index of commonly accessed tags; that is, the degree of influence of non-mainstream tags in the tag set on interest similarity, calculated as follows:

[0096]

[0097] Wherein, N(u) x ) for user u x The total number of tags visited, where N(l) is the number of users in the community who have visited tag l.

[0098] The role of the tag personalization index: First, by considering the user The span of interests can quantify the relevance of a user's tag set to their interests. The scope of the effect of jointly labeled tags on their interest similarity is determined to prevent excessive influence on their interest similarity factor due to excessive access frequency of individual tags by both parties. For example, users... Some tags are labeled with strong emphasis, but users... The wide range of interests will offset some of the impact of high-intensity tags on the interest similarity factor, thus reflecting the contribution of each tag to the degree of interest in a relatively balanced way.

[0099] Secondly, hashtag popularity reflects its popularity within a community. The more popular a hashtag is, the lower its ability to represent the similarity of interests among users. For example, highly popular hashtags... and low heat label There are annotations The number of users is only u x and u y That is, N( )=2, and the annotation The number of users is 1000, that is, N( )=1000, the former indicates Belongs to the user User-specific interest tags, therefore for user u x u y , The effect on the interest similarity factor is significantly higher than .

[0100] Furthermore, it should be noted that the user u is calculated based on... x u y The interest similarity factor does not satisfy symmetry, that is... Its rationale lies in, for example, user u x For user u y They pay more attention to certain tags, and thus become more interested in those tags within a certain period of time. As time accumulates, Gradually increasing and stabilizing, however, user u y User u may not yet be aware x The existence of u x Interest selection for u y The labeled resource tags have a relatively small impact. This characteristic, when applied to group recommendations, manifests as user u x Interest migration is easily influenced The influence of similarity increases gradually, and in contrast, user u y The effect of interest similarity factor is relatively low, meaning that interest similarity has a relatively low impact on user u. y The dynamic changes in one's own interests have a relatively small impact.

[0101] User Influence Factor:

[0102] Core users of a community, or a "specific group," typically possess relatively high influence, enabling them to spread information faster and more widely within the community. This, in turn, strengthens their ability to influence the interests of other users. Therefore, this invention incorporates different user social behaviors into the calculation of the p-index, comprehensively considering the content characteristics of user texts to further quantify the degree of influence diffusion within the community. The p-index is an indicator that measures individual excellence.

[0103] Based on the calculation approach of the p-index, combined with the forwarded dataset The dataset being commented on Liked dataset By using user activity and dissemination as evaluation factors, the p-index is improved by first refining the number of user information disseminations into the number of reposts. Number of comments Number of likes The quality of information dissemination is broken down into the number of times a user receives a forwarded message. Number of comments Number of likes This leads to the construction of user u x Influence Factor I (u x The formula is:

[0104]

[0105] Where C represents the information quality value and N represents the information quantity value.

[0106] Information quality value C formula:

[0107]

[0108] Formula for the information quantity value N:

[0109]

[0110] Where C and N are the information quality value and the information quantity value, respectively. , , and , , To calculate the weights.

[0111] Through weighted user u x By using related interest similarity factors and user influence factors, the influence of other users' interests within the community is quantified, ultimately yielding user u. x With user u y Mutual trust between Calculation formula:

[0112]

[0113] Where θ is the similarity threshold, I( ) for users Influence factors For user u x right Interest similarity factor, to determine Once the interest similarity threshold is reached, then the user It will spread and affect u x Interest-based resource tags, and the degree of influence depends on u. x right Interest similarity factor, and One's own influence factors.

[0114] Interest-based trust level:

[0115] By integrating self-perceived trust and mutual trust, the user u is ultimately obtained. x For any user u in the group y Inter-user interest perception trust, i.e., the degree to which a user's interests are influenced by other users within the community, is mainly used to adjust the weighting relationship between self-perceived interests and mutual interest perception. The calculation formula is:

[0116]

[0117] Among them, ITI (u x u y ) for user u y For any user u in the group y Interest and trust level The inductance coefficient,

[0118] Induction coefficient It is mainly used to adjust the weight ratio of self-perceived interest and mutual interest. The more significant the social influence on a group of users, that is, the stronger the mutual interest effect, the better. The smaller the value, the more dominant social choice is, meaning the perceived effect of interest is significant. The larger the value, the more effective the introduction of the inductance coefficient. This involves redistributing the interest sensitivity generated by self-perceived and mutual interest perception, i.e., the probability of the spread of interest influence globally.

[0119] when When =0, interest sensitivity transforms into relying solely on mutual interest perception, i.e., social influence, as the vehicle for the spread of interest influence. When the value is 1, it transforms into relying solely on self-perceived interest, i.e., social choice, as the vehicle for the spread of interest influence.

[0120] S3. Construct a trust network based on the interest-based trust level, calculate the interest influence, and use a ranking algorithm to simulate the influence of user interests on behavioral interactions within the community. In the trust network, the PageRank algorithm is used to simulate the transmission process of interest influence, calculate the group interest level within the community, and realize community information recommendation through the group interest level.

[0121] The interest-based trust level among community users reflects the degree to which the intensity of interests among community members is influenced by social selection and social impact, i.e., the degree of interest association. Using interest trust level as an influencing factor on changes in group interests, a ranking algorithm is employed to simulate the impact of user interests on in-group behavioral interactions, thereby realizing the propagation process of interest influence within the trust network.

[0122] The adjoining matrix of a trust network can be represented by CM. Based on a social network, user u... x and user u y The relationships between nodes are quantified as edges with mutual interest weights. , indicating user u y For u x The perceived trust level is defined as follows: each node in the network represents a user, edges represent interest transfer relationships between users, and edge weights represent the degree of interest influence between nodes. Furthermore, the edges exhibit asymmetry. User u x Interest influence is a key factor in trust networks. x The ability to communicate interests externally reflects the global reach of user u y The assigned interest weights, and accordingly, u x It also gradually accumulates its influence share in the network by establishing interest weight connections with other users, thereby constructing a dynamic interest discovery mechanism oriented towards the spread of interest influence.

[0123] Figure 2 This is a diagram illustrating the propagation of interest influence in a trust network. At time t, User3 is influenced by User1's interest, causing User3's interest-based trust level to increase, i.e., through the connections between them. This enables the transfer of User1's interest influence. In group recommendations, User3 leverages the share of interest influence gained by User1 in a single dissemination, not only realizing its own interest changes towards User1 but also accumulating interest influence and transferring it to User4 through the level of trust and connection. , and pass it to User4.

[0124] Group interest level:

[0125] Inspired by the recursive iterative approach, this study simulates the behavior of randomly accessing user nodes using interest vectors. It utilizes the stationary probability value of each node, i.e., the PageRank value, to quantify the importance of the node, i.e., its interest influence. Interest influence is essentially a propagation mechanism based on the perceived trust between nodes. Globally, it projects the reference associations of text tags onto the interest interaction relationships between users, ultimately establishing a weighted interest network, i.e., a trust network. Users are connected along the weighted edges... x Transmitted to user u y Interest Influence Inf ( ), Calculation formula:

[0126]

[0127] Among them, c xy User u x and u y The level of trust in the sense, c xy =ITI(u x , u y ); q is the jump coefficient, User u y The sensed trust level is higher than the preset trust threshold. The total number of users; when hour ,when hour, ;

[0128] For the target community U={ , ,…, Assume there exists a resource tag set L={ , ,…, }, the interest level of this group at time t (U,L) calculation formula:

[0129]

[0130] User u at time t i and tag l j Group trust level.

[0131] The experimental verification results based on this invention are shown in the image. Figures 3-7 .

[0132] This invention proposes a community recommendation method based on interest-sensing propagation. This method fully integrates the roles of social influence and social choice factors in group interest identification. By comprehensively considering users' own interest choices and the interest influence of social groups, a trust network including self-perceived interest and mutual interest perception is established. On this basis, an interest influence propagation model with interest sensing as the propagation medium is constructed, and group recommendation services are carried out based on this model. Finally, the effectiveness and rationality of the method are verified on a real dataset.

[0133] Group recommendation services can fully explore the common needs of social network users, providing a practical theoretical basis for building group decision-making and intelligent information services for multiple scenarios.

[0134] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for community recommendation based on interest-induced propagation, characterized in that, The method comprises: S1: extracting an interest label set of a community user through label preprocessing, calculating interest intensity and interest stability based on the interest label set, then dividing the community user into an influential user and a selective user, and performing social behavior analysis to extract a user activity data set and a user propagation data set; S2: analyzing interest trust degree of the community user, the analysis process of the interest trust degree being: calculating self-perceived trust degree by analyzing interest behavior characteristics of the selective user; calculating mutual-perceived trust degree by analyzing interest behavior characteristics of the influential user; and calculating induced trust degree based on the self-perceived trust degree and the mutual-perceived trust degree; S3: using the analysis result of the interest trust degree to construct a trust degree network; calculating interest influence based on the trust degree network and combining a PageRank algorithm; further calculating group trust degree based on the interest influence, and completing community interest recommendation. 2.The interest-induced propagation based community recommendation method according to claim 1, characterized in that, The interest intensity is represented by the access frequency of the user to the label; For a "user-label" binary tuple, the interest intensity calculation formula of the user is as follows: wherein, is the user access frequency to the tag , is the user total access frequency to all tags, is the user interest intensity to the tag , j s is the total number of users, and p is the total number of tags. 3.The interest-induced propagation based community recommendation method according to claim 2, characterized in that, The interest stability is used to represent the stability of the interest intensity in the time dimension by introducing a forgetting function, and the calculation formula is as follows: wherein k is the interest half-life of the user, is the label is the time of the n-th occurrence, is the label is the user's interest intensity at the time of the n-th occurrence, is the label is the interest intensity of the user at the time of the n-th occurrence, is the label is the interval between the two consecutive occurrences; Constructing a user interest tag matrix , represents the interest intensity of the user in the tag at the moment, i.e. . 4.The interest-induced propagation based community recommendation method according to claim 1, characterized in that, The user activity dataset includes a number of user retweets , a number of user comments , and a number of user likes ​ The user propagation data set includes the number of times the user is forwarded , the number of times the user is commented on , and the number of times the user is liked . 5.The interest-induced propagation based community recommendation method according to claim 1, characterized in that, The self-perceived trust degree comprises relationship trust degree and interaction trust degree, The relationship trust degree is obtained by acquiring a friend relationship between users, and a friend relationship matrix is established wherein =1 indicates that there is a friend relationship between the user and , and satisfies = ; The interaction trust level is established by calculating the interaction behavior between users and building a behavior relationship matrix. ,in For users right The weighted value of comments and likes on the post. The formula is as follows: wherein N(xy) represents the number of likes of the user to the post , N(x) represents the total number of likes of the user , N p (xy) represents the number of comments of the user to the post in the pth sampling period, N(x) represents the total number of comments of the user in the pth sampling period, n is the time span of the comments, is the time influence value of the comment behavior, , respectively represent the current time and the time when the comment behavior occurs. The self-perceived trust degree is obtained by weighted calculation of the relationship trust degree and the interaction trust degree, and the formula is as follows, in, Indicates user right Self-perceived trust level It is represented as the self-perception coefficient and is used to adjust the weight ratio between relationship trust and interaction trust. 6.The interest-induced propagation based community recommendation method according to claim 3 or 5, characterized in that, The mutual-perceived trust degree comprises an interest similarity factor and a user influence factor; the interest similarity factor affected by the frequency of accessing the same tag between users, calculated by the user interest tag matrix, as follows: in, Indicates a label; and Representing users respectively and The set of tags accessed. and Representing users respectively and For tags Intensity of interest Indicates the label The set of users who have accessed the site, w p,l Indicates all tags The sum of the intensity of interest of the users who made the visit; For users , The personalization index of the tags that are accessed together. 7.The interest-based social recommendation method of claim 6, wherein, The individualization index is related to user interest span and label heat, the interest span refers to the total number of labels accessed by the user, and the label heat refers to the number of users accessing the same label in the community, and the formula of the individualization index is as follows: wherein, the total number of tags visited by the user, N(l) is the total number of users in the community who have visited tag l. 8.The interest-based propagation-induced community recommendation method of claim 4 or 7, wherein, The user influence factor is calculated by user activity and user propagation, the information quantity value is calculated by the user activity, the information quality value is calculated by the user propagation, and the user influence factor The calculation formula is as follows: Wherein, C, N are information quality value and information quantity value respectively, , , , , and is the calculation weight. 9.The interest-based propagation-induced community recommendation method of claim 8, wherein, The mutual-sense trust degree is obtained by weighting the interest similarity factor and the user influence factor. and The mutual-sense trust degree calculation formula of the user A and the user B is as follows: wherein, is a similarity threshold value; The induced trust degree is calculated by fusing the self-induced trust degree and the mutual-induced trust degree and of the user The calculation formula is as follows: wherein, is a coefficient of induction, used to adjust the weight relationship between the self-induction trust and the induction trust of interest. 10.The interest-based propagation-induced community recommendation method according to claim 1 or 8, characterized in that, The nodes of the trust network are users, the edges of the trust network are the induced trust degrees between users, the interest influence propagates from a node to other nodes along edges, and a user The calculation formula of the interest influence of the user ) is as follows: wherein, is the user and induced trust degree, ; q is a jump coefficient, is the user whose induced trust degree is higher than a preset trust threshold total number of users When time, When time, ; The population trustworthiness is calculated as follows: wherein, represents the group trust of the user at time t and the label at time t.

Images