Associated user network data mining method based on multi-information fusion

Abstract

The invention discloses an associated user network data mining method based on multi-information fusion. The method comprises the following steps: outputting a user interaction coefficient through a user interaction model; obtaining an article interaction coefficient through the article interaction model; obtaining an article activity coefficient through an article activity model; obtaining a social trust coefficient through the social trust model; constructing a user-article interaction model to output a user-article interaction coefficient; comparing the obtained user-article interaction coefficient of the user with the user-article interaction coefficient of an article which has not been touched, and obtaining an article of which the user-article interaction coefficient is higher than that of the user; and obtaining user-article interaction coefficients of all articles, sorting the user-article interaction coefficients from high to low, and selecting the first N articles as recommendation results to be output. According to the method, a multi-source fusion quantitative result is converted into accurate personalized recommendation based on an interaction coefficient sorting and screening mechanism, and the problem of insufficient recommendation adaptability caused by multi-factor splitting in the prior art is effectively solved.

Description

Technical Field

[0001] This invention belongs to the field of information fusion technology, specifically relating to a method for mining associated user network data based on multi-information fusion. Background Technology

[0002] With the rapid development of internet technology and the dramatic increase in user data, how to accurately mine user interests from massive amounts of online behavior and make personalized recommendations has become a core need in fields such as e-commerce, social networks, and content platforms.

[0003] In the field of personalized recommendation, various solutions have been proposed in existing technologies to alleviate the problem of data sparsity and utilize social information. For example, Chinese patent application No. 202410893430.4 discloses a lightweight graph convolutional collaborative filtering recommendation method that integrates high-order social relationships. First, in the model embedding layer, graph embedding technology is used to map user, item, and friend information to a low-dimensional dense vector space, mitigating the negative impact of data sparsity on the model's recommendation results. Second, in the graph convolutional layer, three stacked graph convolutional layers are used to learn the topological structure of the user's social relationship graph and learn the high-order connection information between users, items, and friends, generating a series of indirect feedbacks from implicit negative feedbacks. Implicit negative feedbacks are indirectly captured by analyzing user behavior and friend intimacy, thereby improving the utilization rate of implicit negative feedbacks. In addition, a graph attention network is integrated to measure the contribution value of neighbors and dynamically assign weights to them adaptively, which can filter noisy neighbors, making the model robust and giving it a certain degree of interpretability.

[0004] This patent discloses a method for making recommendations by constructing a "user-friend social graph" and a "user-item-friend high-order connection graph," and using graph convolutional networks (GCNs) to aggregate multi-order neighborhood information. While this method introduces social information through graph structures, it has the following limitations: First, its core lies in learning the topological structure of social networks, lacking fine-grained dynamic quantification of social behaviors themselves (such as the trust level of forwarding and sharing); second, this scheme mainly integrates user-item interactions and user-user social interactions, failing to systematically consider the important dimension of the dynamic state of items themselves (such as item lifecycle and real-time popularity changes); finally, GNN-based schemes may face challenges such as high training complexity and difficulty in real-time updates when dealing with ultra-large-scale dynamic networks. Summary of the Invention

[0005] The purpose of this invention is to provide a method for mining associated user network data based on multi-information fusion, in order to solve the problems mentioned in the background art. The method for mining associated user network data based on multi-information fusion provided by this invention has the characteristics of achieving adaptive weighting of multiple factors, significantly improving the scenario adaptability and credibility of recommendation results.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for mining associated user network data based on multi-information fusion, comprising the following steps:

[0007] S1. Based on user interaction data of user u to item i, output user interaction coefficients through user interaction model;

[0008] S2. Based on the item interaction data of item i, obtain the item interaction coefficient through the item interaction model;

[0009] S3. Based on the item activity data of item i, obtain the item activity coefficient through the item activity model;

[0010] S4. Based on user social trust data of user u to item i, obtain the social trust coefficient through the social trust model;

[0011] S5. Construct a user-item interaction model based on the user interaction coefficient and item interaction coefficient under the social trust coefficient and item status coefficient, and output the user-item interaction coefficient.

[0012] S6. Compare the user-item interaction coefficient of the obtained user with the user-item interaction coefficient of items that have not been touched, and obtain items with a higher user-item interaction coefficient than the user.

[0013] S7. For user u, obtain the user-item interaction coefficients for all items, sort them from highest to lowest, and select the top N items as the recommendation results for output.

[0014] In this invention, further, in S1, the step of outputting user interaction coefficients through a user interaction model based on user interaction data of user u to item i is as follows:

[0015] S11. Obtain user interaction data, including average user ratings, number of visits to item pages, and duration of visits to item pages;

[0016] S12. Perform maximum-min normalization on the mean user rating, the number of times the item page is accessed, and the duration of the item page access to obtain the user rating index, the number of page accesses index, and the duration of the page access index.

[0017] S13. Import the user rating index, page visit frequency index, and page visit duration index into the user interaction model to obtain the user interaction coefficient.

[0018] In this invention, the user interaction model is further calculated using a weighting function to obtain user interaction coefficients, wherein the user interaction coefficients are positively correlated with the intensity of individual user interaction with items;

[0019] The user interaction model is represented as follows:

[0020] ;

[0021] in, Represents the user interaction coefficient. This represents the user rating index. This represents the page view count index. This represents the page visit duration index. Represents the weighting coefficient, and , The larger the value, the stronger the user's interaction with the item.

[0022] In this invention, further, in S2, the step of obtaining the item interaction coefficient based on the item interaction data of item i through the item interaction model is as follows:

[0023] S21. Obtain item interaction data, including the total number of item ratings, the average item rating, and the total number of item views;

[0024] S22. Sum the total number of item ratings, the average item rating, and the total number of item views with the corresponding average total number of item ratings, average item ratings, and average number of item views to obtain reference values ​​for the number of ratings, the ratings, and the views.

[0025] S23. Ratio the total number of item ratings, the average item rating, and the total number of item views with the corresponding reference values ​​to obtain the rating count index, the average rating index, and the view index.

[0026] S24. Import the rating frequency index, average rating index, and pageview index into the item interaction model to obtain the item interaction coefficient.

[0027] Furthermore, in this invention, the item interaction model calculates the item interaction coefficient through an aggregation function, wherein the item interaction coefficient is positively correlated with the popularity of the item;

[0028] The item interaction model is represented as follows:

[0029] ;

[0030] in, Indicates the interaction coefficient of items. Indicates the number of ratings index. Indicates the average score index. This represents the page view count index. Furthermore, the higher the value, the more popular the item.

[0031] In this invention, further, in S3, the step of obtaining the item activity coefficient based on the item activity data of item i through the item activity model is as follows:

[0032] S31. Obtain item interaction data, including item listing duration, item lifecycle, and item forwarding count;

[0033] S32. Calculate the ratio of the item's listing duration to the item's lifecycle to obtain the item's timeliness index;

[0034] S33. Sum the number of times an item is forwarded with the average number of times all items are forwarded to obtain a forwarding reference value. Ratio the number of times an item is forwarded with the item reference value to obtain the item forwarding index.

[0035] S34. Import the item timeliness index and the item forwarding frequency index into the item activity model to obtain the item activity coefficient.

[0036] In this invention, the item activity model calculates the item activity coefficient through a weighted combination function, wherein the item activity coefficient is positively correlated with item freshness and popularity.

[0037] The item activity model is represented as follows:

[0038] ;

[0039] in, Indicates the activity level of an item. Indicates the timeliness index of the item. This indicates the number of times an item has been forwarded. Represents the balance weight coefficient and , Furthermore, the higher the value, the fresher and more active the item.

[0040] In this invention, further, in S4, the step of obtaining the social trust coefficient based on user u's social trust data regarding item i through a social trust model is as follows:

[0041] S41. Perform maximum-min normalization on the user forwarding time to obtain the user forwarding time index;

[0042] S42. Obtain user social trust data, including the number of times items are shared, the maximum number of mutual friends among the sharing friends, and the user's forwarding time;

[0043] S43. Sum the number of times the item was shared, the maximum number of mutual friends among the sharing friends, and the corresponding average values ​​to obtain reference values ​​for the number of shares and mutual friends.

[0044] S44. Ratio the number of times an item is shared and the maximum number of mutual friends among the sharing friends to the corresponding reference values ​​to obtain the sharing count index and mutual friend index.

[0045] S45. Import the user forwarding time index, sharing frequency index, and mutual friend index into the social trust model to obtain the social trust coefficient.

[0046] Furthermore, in this invention, the social trust coefficient is positively correlated with the sharing frequency index and the mutual friend index in the social trust model, and negatively correlated with the user forwarding time index.

[0047] The social trust model is represented as:

[0048] ;

[0049] in, Indicates the social trust coefficient. Indicates the forwarding time index. This represents the number of times the message was shared. Indicates the mutual friend index. The larger the value, the better for users. For items The greater the social trust, the better.

[0050] In this invention, further, in S5, the step of constructing the user-item interaction model and outputting the user-item interaction coefficient based on the user interaction coefficient and item interaction coefficient under the social trust coefficient and item status coefficient is as follows:

[0051] S51. Obtain the social trust coefficient and the item status coefficient and take their complements. Obtain the social trust coefficient factor and the item status coefficient factor.

[0052] S52. Import the social trust coefficient factor and the item status coefficient factor into the social context enhancement model to obtain the social context enhancement coefficient.

[0053] S53. Construct a user-item interaction model based on the user interaction coefficient and item interaction coefficient under the social context enhancement coefficient, and output the user-item interaction coefficient.

[0054] In this invention, the social context enhancement model is further represented as follows:

[0055] ;

[0056] in, Indicates the social context enhancement coefficient. Indicates the social trust factor. Indicates the activity coefficient factor of an item;

[0057] The user-item interaction model is represented as follows:

[0058] ;

[0059] Wherein, represents the user-item interaction coefficient. Represents the user interaction coefficient. Indicates the interaction coefficient of items. Indicates the social context enhancement coefficient. The larger the value of u, the greater the user's interest in item i.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] 1. This invention effectively addresses the challenges of data sparsity and cold start in recommendation systems through a collaborative processing mechanism of multi-source data. The construction of user interaction coefficients fully utilizes the complementarity of explicit ratings and implicit behavioral data, enabling the system to reliably infer preferences even when user historical behavior is limited. The generation of item interaction coefficients integrates multi-dimensional popularity indicators, allowing newly listed or low-frequency items to obtain reasonable evaluations based on real-time interaction data. The dynamic calculation mechanism of item activity coefficients ensures that recommended content is synchronized with changes in the item lifecycle, avoiding the problem of outdated recommendations due to lack of timeliness. The quantitative introduction of social trust coefficients compensates for the shortcomings of traditional methods that rely solely on direct associations, enabling recommendation results to reflect the implicit influence in social networks.

[0062] 2. The user-item interaction model of the present invention organically integrates social context and item status factors through a nonlinear fusion mechanism, rather than simply superimposing data from various dimensions, thereby achieving adaptive weighting of multiple factors and significantly improving the scenario adaptability and credibility of the recommendation results.

[0063] 3. The ranking and filtering mechanism based on the interaction coefficient of this invention transforms the quantitative results of multi-source fusion into accurate personalized recommendations, effectively solving the problem of insufficient recommendation adaptability caused by multiple factors in the prior art.

[0064] 4. This invention achieves the systematic integration and precise quantification of users' multi-dimensional interactive behaviors, effectively solving the problem of quantification deviation caused by differences in data scale, enabling the recommendation system to more accurately identify the intensity of users' true interest in items, thereby improving the accuracy of recommendation results and user satisfaction.

[0065] 5. The item interaction coefficient of this invention can accurately quantify the true popularity and quality balance of items, effectively avoiding the problem of item weight distortion caused by a single indicator, and making the recommendation results more accurately reflect the user's potential interests and preferences.

[0066] 6. This invention can effectively integrate the timeliness and dynamic behavior data of items, accurately quantify the real-time freshness of items, solve the problem of inaccurate evaluation of newly listed items or items with short life cycles, thereby improving the adaptability and credibility of recommendation results in time-sensitive scenarios.

[0067] 7. This invention enables the calculation of social trust coefficients to comprehensively consider multi-dimensional characteristics of user behavior, including forwarding timeliness, sharing activity, and social closeness, thereby more accurately reflecting the dynamics of trust propagation among users and effectively improving the adaptability and credibility of recommendation results. Attached Figure Description

[0068] Figure 1 This is a block diagram of the method of the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] In the description of this invention, unless otherwise explicitly specified and limited, the terms "connected," "linked," and "fixed" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0071] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0072] In the description of this invention, the terms "upper," "lower," "right," and "left," etc., refer to the orientation or positional relationship shown in the accompanying drawings. They are used only for ease of description and simplification of operation, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention. Furthermore, the terms "first" and "second" are used merely for distinction in description and have no special meaning.

[0073] Please see Figure 1 This embodiment provides the following technical solution: a method for mining associated user network data based on multi-information fusion, comprising the following steps:

[0074] S1. Based on user interaction data of user u to item i, output user interaction coefficients through user interaction model;

[0075] S2. Based on the item interaction data of item i, obtain the item interaction coefficient through the item interaction model;

[0076] S3. Based on the item activity data of item i, obtain the item activity coefficient through the item activity model;

[0077] S4. Based on user social trust data of user u to item i, obtain the social trust coefficient through the social trust model;

[0078] S5. Construct a user-item interaction model based on the user interaction coefficient and item interaction coefficient under the social trust coefficient and item status coefficient, and output the user-item interaction coefficient.

[0079] S6. Compare the user-item interaction coefficient of the obtained user with the user-item interaction coefficient of items that have not been touched, and obtain items with a higher user-item interaction coefficient than the user.

[0080] S7. For user u, obtain the user-item interaction coefficients for all items, sort them from highest to lowest, and select the top N items as the recommendation results for output.

[0081] By adopting the above technical solutions, this invention effectively addresses the challenges of data sparsity and cold start in recommendation systems through a collaborative processing mechanism of multi-source data. The construction of user interaction coefficients fully leverages the complementarity of explicit ratings and implicit behavioral data, enabling the system to reliably infer preferences even with limited historical user behavior. The generation of item interaction coefficients integrates multi-dimensional popularity indicators, allowing newly listed or low-frequency items to receive reasonable evaluations based on real-time interaction data. The dynamic calculation mechanism of item activity coefficients ensures that recommended content is synchronized with changes in the item's lifecycle, avoiding outdated recommendations due to lack of timeliness. The quantitative introduction of social trust coefficients compensates for the shortcomings of traditional methods that rely solely on direct associations, enabling recommendation results to reflect implicit influences within social networks. This invention's user-item interaction model organically integrates social context and item status factors through a non-linear fusion mechanism, rather than simply superimposing data from various dimensions, achieving adaptive weighting of multiple factors and significantly improving the scenario adaptability and credibility of recommendation results. This invention's ranking and filtering mechanism based on interaction coefficients transforms the quantitative results of multi-source fusion into precise personalized recommendations, effectively solving the problem of insufficient recommendation adaptability caused by the fragmentation of multiple factors in existing technologies.

[0082] Specifically, in S1, the steps for outputting user interaction coefficients based on user interaction data of user u to item i through the user interaction model are as follows:

[0083] S11. Obtain user interaction data, including the average user rating (user rating refers to the user's direct score for all items, such as 1-5 stars, the more stars, the higher the rating), the number of times the item page is accessed, and the duration of the access to the item page;

[0084] S12. Perform maximum-min normalization on the mean user rating, the number of times the item page is accessed, and the duration of the item page access to obtain the user rating index, the number of page accesses index, and the duration of the page access index.

[0085] S13. Import the user rating index, page visit frequency index, and page visit duration index into the user interaction model to obtain the user interaction coefficient.

[0086] Specifically, the user interaction model calculates the user interaction coefficient through a weighted function, where the user interaction coefficient is positively correlated with the intensity of individual user interaction with items.

[0087] Specifically, the user interaction model is represented as follows:

[0088] ;

[0089] in, Represents the user interaction coefficient. This represents the user rating index. This represents the page view count index. This represents the page visit duration index. Represents the weighting coefficient, and , The larger the value, the stronger the user's interaction with the item.

[0090] Among them, the mean user rating score refers to the central tendency measure of users' historical rating behavior, which can be achieved using an arithmetic mean or a weighted average method, aiming to quantify the strength of users' explicit preferences; the number of times an item page is visited can be understood as the frequency of user interaction with the item page, which can be obtained through front-end tracking technology or log analysis systems, aiming to reflect the frequency of user attention to items; the duration of an item page visit is specifically a numerical representation of the user's single or cumulative dwell time, which can be achieved through browser session tracking or in-application timers, aiming to capture the depth of interest in users' implicit behavior; max-min normalization is a method of linearly mapping the original data to a unified interval, which can be achieved using dynamic threshold adjustment or segmented normalization strategies, aiming to eliminate the quantitative bias caused by differences in the units and value ranges of data from different dimensions; the user rating score index, page visit index, and page visit duration index can be understood as normalized standardized feature vectors, which provide a comparable basis for multi-source feature fusion by eliminating the scale differences of the original data; the weight coefficients in the user interaction model It refers to the parameter that adjusts the contribution of each behavioral feature. It can be dynamically determined through expert experience assignment, offline training, or online optimization. The purpose is to flexibly allocate the relative importance of explicit scores and implicit behaviors according to the actual scenario.

[0091] Specifically, this invention first acquires three types of heterogeneous data: the average user rating, the number of page visits, and the visit duration. Because these data have different dimensions (e.g., ratings are discrete values ​​from 1 to 5, visit counts are positive integers, and visit duration is a continuous time value), direct fusion would lead to high-dimensional features dominating the calculation results. Therefore, the original data is converted into a dimensionless exponent in the [0,1] interval through max-min normalization, for example, mapping the average user rating to a user rating index. Convert visit counts into page view count index Convert visit duration into page visit duration index Based on this, the normalized three types of indices are imported into the user interaction model, and a weighted linear combination formula is used. Calculate the overall interaction coefficient, where the weighting coefficient is... Constraints Ensure that the output value is strictly limited to the range [0,1]. This process integrates explicit ratings and implicit behavioral features through the system, combining normalization to eliminate scaling bias and a weighted fusion mechanism, so that the user interaction coefficient can simultaneously reflect the intensity of the user's explicit preference for items and the depth of implicit behavior. This provides an accurate quantitative basis for the intensity of interest in subsequent recommendation steps. The weight coefficient is a parameter used to adjust the importance of different factors in decision-making, and can be assigned values ​​through preset strategies or dynamic algorithms.

[0092] As a specific implementation method, the solution of the present invention is implemented as follows: the average user rating score can be calculated based on the user's historical rating records. For example, if a user's ratings for 10 items are 4, 5, 3, 4, and 5 respectively, then the average rating score is 4.2; the number of visits to the item page can be counted by the total number of times the user visits the item details page within 24 hours; the duration of the item page visit can be recorded by averaging the duration of each visit. After the above raw data is processed by max-min normalization, the user rating index is obtained. Page visit index reflects the relative position of the rating among all users. Page visit duration index reflects the activity level of visit frequency. The significance of dwell time is represented. When importing the three indices into the user interaction model, the weighting coefficients... The weighting can be adjusted based on platform characteristics. For example, in content consumption scenarios, the access duration index can be given higher weight, while in e-commerce scenarios, the rating index can be emphasized, ultimately outputting the user interaction coefficient. It can be directly used to compare and rank user-item interaction coefficients.

[0093] Through the above technical solution, the present invention achieves the systematic integration and precise quantification of users' multi-dimensional interactive behaviors, effectively solving the problem of quantification deviation caused by differences in data scale, enabling the recommendation system using the mining method of the present invention to more accurately identify the user's true interest intensity in items, thereby improving the accuracy of recommendation results and user satisfaction.

[0094] Specifically, in S2, the steps to obtain the item interaction coefficients based on the item interaction data of item i through the item interaction model are as follows:

[0095] S21. Obtain item interaction data, including the total number of item ratings and the average item rating (users' direct rating of the item, such as 1-5 stars, the more stars the higher the rating) and the total number of item views.

[0096] S22. Sum the total number of item ratings, the average item rating, and the total number of item views with the corresponding average total number of item ratings, average item ratings, and average number of item views to obtain reference values ​​for the number of ratings, the ratings, and the views.

[0097] S23. Ratio the total number of item ratings, the average item rating, and the total number of item views with the corresponding reference values ​​to obtain the rating count index, the average rating index, and the view index.

[0098] S24. Import the rating frequency index, average rating index, and pageview index into the item interaction model to obtain the item interaction coefficient.

[0099] Specifically, the item interaction model calculates the item interaction coefficient through an aggregation function, where the item interaction coefficient is positively correlated with the popularity of the item.

[0100] Specifically, the item interaction model is represented as follows:

[0101] ;

[0102] in, Indicates the interaction coefficient of items. Indicates the number of ratings index. Indicates the average score index. This represents the page view count index. Furthermore, the higher the value, the more popular the item.

[0103] Among them, the total number of item ratings refers to the cumulative number of user ratings for an item, which can be achieved by counting rating events recorded in the database transaction log, aiming to quantify the breadth of item evaluation; the average item rating refers to the arithmetic mean of user ratings for an item, which can be calculated using a weighted average algorithm based on a 1-5 star rating system, aiming to reflect the depth of item quality; the total number of item page views refers to the cumulative number of times the item page has been accessed, which can be obtained by statistically analyzing request records in the server access log, aiming to measure the coverage of user interests; the rating reference value is the sum of the total number of item ratings and the global average total number of ratings, which can be generated by dynamically calculating the average number of ratings for all items and then adding the current item value, aiming to provide a benchmark for adaptive data distribution; the rating reference value is the sum of the average item rating and the global average rating, which can be achieved through a sliding window statistical analysis of historical rating data, aiming to eliminate... Aside from systemic rating bias; the pageview reference value refers to the sum of the total pageviews of an item and the global average pageviews. It can be constructed based on the benchmark value output by the real-time traffic analysis module, aiming to standardize the scale of user behavior; the rating frequency index is the ratio of the total number of ratings of an item to the rating frequency reference value. It can be generated as a dimensionless indicator through the division operation module, aiming to highlight the relative popularity advantage of the item; the rating mean index is the ratio of the average rating of an item to the rating reference value. It can be converted into a relative quality indicator using the normalization processing unit, aiming to strengthen the competitive representation of the quality dimension; the pageview index is the ratio of the total pageviews of an item to the pageview reference value. It can be output as a relative value of interest breadth using the ratio calculation module, aiming to quantify the expansion of user interest coverage; the item interaction model is a nonlinear calculation framework that integrates multi-dimensional indices. It can be implemented through a software module using a geometric mean function, aiming to balance the contribution of each dimension and constrain the range of coefficients.

[0104] Specifically, this invention first obtains the total number of item ratings, the average item rating, and the total number of item views as basic input data. These data capture the breadth of item popularity, the depth of quality, and the coverage of user interests, respectively. Then, the raw data is summed with the global mean to generate a dynamic reference value. This operation introduces an adaptive benchmark to eliminate scale bias caused by fixed thresholds. Subsequently, the raw data is converted into dimensionless relative indices through ratio processing, so that the rating frequency index, the average rating index, and the view index represent the item's popularity multiple, quality improvement degree, and interest expansion range relative to the overall level, respectively. Finally, a geometric mean model is used to nonlinearly fuse the three indices. This model forces a balanced contribution of each dimension through the square root of the product, avoiding the dominance of a single indicator in the results, while ensuring that the item interaction coefficient is strictly limited to the [0,1] interval, thus providing a reliable input for the user-item interaction model that comprehensively reflects the popularity of items.

[0105] As a preferred embodiment, the specific implementation of the present invention is as follows: In the e-commerce recommendation system, item interaction data is extracted from a distributed database; the total number of item ratings is obtained by aggregating user rating events; the average item rating is calculated based on a 1-5 star rating system; and the total number of item views is statistically analyzed based on Nginx access logs. The rating frequency reference value is generated by summing the average number of ratings for all items using the Spark computing engine; the rating reference value and the view reference value are constructed using the same mechanism. The rating frequency index, the average rating index, and the view index are obtained by performing ratio calculations using the Python data processing module. The item interaction model calculates the geometric mean by calling the sqrt function of the NumPy library, and the output item interaction coefficient is written to the feature storage layer of the recommendation engine in real time for subsequent construction of user-item interaction coefficients.

[0106] Through the above technical solution, the item interaction coefficient of the present invention can accurately quantify the true popularity and quality balance of items, effectively avoid the problem of item weight distortion caused by a single indicator, and make the recommendation results more accurately reflect the user's potential interests and preferences.

[0107] Specifically, in some embodiments of the present invention, a user interaction model is proposed to output user interaction coefficients. However, in its implementation, the multi-dimensional features of user interaction data (such as explicit ratings and page access behavior) are not integrated by the system, and the difference in scale of the original data leads to quantification bias, which makes it impossible to accurately capture the user's true interest intensity in the items and affects the accuracy of the recommendation results.

[0108] Specifically, in S3, the steps to obtain the item activity coefficient based on the item activity data of item i through the item activity model are as follows:

[0109] S31. Obtain item interaction data, including item listing duration, item lifecycle, and item forwarding count;

[0110] S32. Calculate the ratio of the item's listing duration to the item's lifecycle to obtain the item's timeliness index;

[0111] S33. Sum the number of times an item is forwarded with the average number of times all items are forwarded to obtain a forwarding reference value. Ratio the number of times an item is forwarded with the item reference value to obtain the item forwarding index.

[0112] S34. Import the item timeliness index and the item forwarding frequency index into the item activity model to obtain the item activity coefficient.

[0113] Specifically, the item activity model calculates the item activity coefficient through a weighted combination function, where the item activity coefficient is positively correlated with item freshness and popularity.

[0114] Specifically, the item activity model is represented as follows:

[0115] ;

[0116] in, Indicates the activity level of an item. Indicates the timeliness index of the item. This indicates the number of times an item has been forwarded. Represents the balance weight coefficient and , Furthermore, the higher the value, the fresher and more active the item.

[0117] Among them, the item timeliness index refers to the ratio of the item's listing time to its life cycle. It can be calculated using normalization, such as by division to obtain the relative time ratio. Its purpose is to eliminate the influence of differences in the life cycles of different items and objectively reflect the relative freshness of the item within its own life cycle. The item reference value is the sum of the number of times the item is forwarded and the average number of times all items are forwarded. It can be calculated using arithmetic addition, such as by adding the number of forwards to the global average. Its purpose is to provide a benchmark value to handle data sparsity and scaling differences. The item forwarding frequency index is the ratio of the number of times the item is forwarded to the item reference value. It can be calculated using division, such as by dividing the number of forwards by the reference value. Its purpose is to normalize the number of forwards and make the activity assessment comparable among items. The item activity model is a mathematical model that integrates the item timeliness index and the item forwarding frequency index. It can be implemented using an exponential decay weighting function. For example, it can dynamically adjust the contribution ratio of timeliness and activity through configurable balancing weight coefficients. Its purpose is to accurately quantify the overall activity status of the item.

[0118] Specifically, this invention first acquires item interaction data as the basic input source, including item listing duration, item lifecycle, and item forwarding count. This data comprehensively depicts the item's status from both time and user behavior dimensions. Then, by comparing the item listing duration with the item lifecycle, an item timeliness index is calculated. This index effectively normalizes the differences in lifecycles among different items, ensuring that the relative freshness of an item within its lifecycle is objectively quantified, preventing items with longer lifecycles from being misjudged as outdated due to their larger absolute listing duration. Next, the system sums the item forwarding count with the average of all item forwarding counts to obtain an item reference value. The item forwarding count is then compared with this reference value to obtain an item forwarding count index. This solves the evaluation bias problem caused by differences in the absolute value of forwarding counts, making activity assessments comparable across items. Finally, the item timeliness index and item forwarding count index are imported into the item activity model, and weighted by balancing coefficients... Weighted fusion, where the exponential decay term The weighting of new items has been strengthened (the closer an item is to its initial listing period, the lower its expiration index and the greater its contribution to the decay factor), while the weighting coefficient has been balanced. The introduction of this feature enables the system to dynamically adjust the priority between timeliness and activity, thereby ensuring... While standardizing within the range of [0,1], the overall activity status of items is accurately quantified, providing reliable input for subsequent user-item interaction models and balancing weight coefficients. The value can be determined through expert experience or through the analytic hierarchy process.

[0119] As a preferred embodiment, the solution of the present invention is implemented as follows: In the recommendation system of a social e-commerce platform, an item activity model is deployed on the backend server. When processing newly listed fashion items, the system retrieves the listing duration and preset lifecycle of the item from the database and calculates the item timeliness index; simultaneously, it obtains the number of times the item is forwarded and sums it with the average number of times all items are forwarded on the entire site to obtain an item reference value, and then calculates the item forwarding frequency index. Subsequently, these two indices are input into the item activity model and weighted and fused through a configurable balanced weight coefficient to output the item activity coefficient. This coefficient is integrated into the user-item interaction model in real time to generate a personalized recommendation list, ensuring that newly listed items receive reasonable exposure in the recommendation results.

[0120] Through the above solution, the present invention can effectively integrate the timeliness and dynamic behavior data of items, accurately quantify the real-time freshness of items, solve the problem of inaccurate evaluation of newly listed items or items with short life cycles, and thus improve the adaptability and credibility of recommendation results in time-sensitive scenarios.

[0121] Specifically, in S4, the steps to obtain the social trust coefficient based on user u's social trust data regarding item i through the social trust model are as follows:

[0122] S41. Perform maximum-min normalization on the user forwarding time to obtain the user forwarding time index;

[0123] S42. Obtain user social trust data, including the number of times items are shared, the maximum number of mutual friends among the sharing friends, and the user forwarding time (the time from discovering the item to forwarding it).

[0124] S43. Sum the number of times the item was shared, the maximum number of mutual friends among the sharing friends, and the corresponding average values ​​to obtain reference values ​​for the number of shares and mutual friends.

[0125] S44. Ratio the number of times an item is shared and the maximum number of mutual friends among the sharing friends to the corresponding reference values ​​to obtain the sharing count index and mutual friend index.

[0126] S45. Import the user forwarding time index, sharing frequency index, and mutual friend index into the social trust model to obtain the social trust coefficient.

[0127] Specifically, in the social trust model, the social trust coefficient is positively correlated with the sharing frequency index and the mutual friend index, and negatively correlated with the user forwarding time index.

[0128] Specifically, the social trust model is represented as follows:

[0129] ;

[0130] in, Indicates the social trust coefficient. Indicates the forwarding time index. This represents the number of times the message was shared. Indicates the mutual friend index. The larger the value, the better for users. For items The greater the social trust, the better.

[0131] Among them, the user forwarding time index is a quantitative indicator that standardizes the time from when a user discovers an item to when they forward it. It can use a max-min normalization method to map the original time data to a unified interval, aiming to eliminate differences in time scales between different users and ensure the comparability of timeliness factors in the model. User social trust data refers to a set of key indicators reflecting user social behavior, which can include the number of times an item is shared, the maximum number of mutual friends among the sharing friends, and the user's forwarding time, aiming to comprehensively capture social activity, relationship tightness, and timeliness. The sharing frequency reference value is a benchmark value obtained by summing the number of times an item is shared with the global average. It can be based on the statistical mean of the number of times all items are shared, aiming to provide a representative reference benchmark and avoid bias caused by uneven distribution of local data. The mutual friend reference value is the value obtained by summing the maximum number of mutual friends with the global average. The baseline value, obtained by summing the local averages, can be achieved using the statistical characteristics of the overall structure of the social network, aiming to provide an objective reference for quantifying the tightness of social relationships. The sharing frequency index is a relative indicator obtained by comparing the number of times an item is shared with the reference value, which can be expressed as the original number of shares divided by the reference value, aiming to transform sharing activity into a dimensionless standardized value. The mutual friend index is a relative indicator obtained by comparing the maximum number of mutual friends with the reference value of mutual friends, which can be achieved by ratio calculation, aiming to quantify social tightness into a comparable index. The social trust model is a mathematical framework used to integrate multi-dimensional indices to calculate trust coefficients, which can be expressed as a product form combined with an exponential decay term, aiming to ensure that the output coefficients are within a reasonable range and accurately represent the strength of trust by structurally integrating timeliness, activity, and tightness characteristics.

[0132] Specifically, this invention first acquires user social trust data, including the number of shares, the maximum number of mutual friends, and the user's forwarding time. Then, it normalizes the user's forwarding time to generate a forwarding time index, and calculates reference values ​​for the number of shares and mutual friends based on global statistics, subsequently obtaining a standardized index through ratio processing. Finally, these indices are imported into a social trust model, using a product form to fuse multi-dimensional features. The index decay term non-linearly decays the time difference to emphasize timeliness. The overall process ensures the organic unity of data standardization and multi-dimensional comprehensive evaluation. This operating mechanism eliminates time scale differences through normalization, provides a relative benchmark for the calculation of reference values ​​and indices, and the model's product structure effectively integrates timeliness, activity, and density, thereby dynamically reflecting the complex characteristics of trust propagation among users.

[0133] As a preferred embodiment, the specific implementation of the present invention is as follows: In a social e-commerce platform, when a user forwards product information, the system records the time from when the user browses the product to when they forward it, the number of times the product is shared, and the maximum number of mutual friends between the user and their friends; the forwarding time is converted into a forwarding time index through maximum-min normalization; the average number of shares and the average number of mutual friends for all products are calculated as reference values ​​to obtain the sharing frequency index and the mutual friend index; these indices are then substituted into the social trust model. Calculate the social trust coefficient for use in the construction of subsequent user-item interaction models.

[0134] Through the above scheme, the present invention enables the calculation of social trust coefficient to comprehensively consider the multi-dimensional characteristics of user behavior, including forwarding timeliness, sharing activity and social closeness, thereby more accurately reflecting the dynamics of trust transmission among users and effectively improving the adaptability and credibility of recommendation results.

[0135] Specifically, in S5, the steps for constructing the user-item interaction model and outputting the user-item interaction coefficient based on the user interaction coefficient and item status coefficient under the social trust coefficient and item status coefficient are as follows:

[0136] S51. Obtain the social trust coefficient and the item status coefficient and take their complements. Obtain the social trust coefficient factor and the item status coefficient factor.

[0137] S52. Import the social trust coefficient factor and the item status coefficient factor into the social context enhancement model to obtain the social context enhancement coefficient.

[0138] S53. Construct a user-item interaction model based on the user interaction coefficient and item interaction coefficient under the social context enhancement coefficient, and output the user-item interaction coefficient.

[0139] Specifically, the social context enhancement model is represented as:

[0140] ;

[0141] in, Indicates the social context enhancement coefficient. Indicates the social trust factor. Indicates the activity coefficient factor of an item;

[0142] The user-item interaction model is represented as follows:

[0143] ;

[0144] Wherein, represents the user-item interaction coefficient. Represents the user interaction coefficient. Indicates the interaction coefficient of items. Indicates the social context enhancement coefficient. The larger the value of u, the greater the user's interest in item i.

[0145] Specifically, the social trust coefficient factor refers to the complement of the social trust coefficient, which can be achieved by subtracting the normalized value of the original social trust coefficient from 1. Its purpose is to map high trust intensity to low factor values, thus positively responding to positive trust signals in subsequent models. The item state coefficient factor refers to the complement of the item activity coefficient, which can be achieved by subtracting the normalized value of the original item activity coefficient from 1. Its purpose is to reduce the factor value when the item is in a highly active state, providing a basis for the dynamic adjustment of the social context enhancement coefficient. The social context enhancement model refers to the mathematical framework used for nonlinearly fusing the social trust coefficient factor and the item state coefficient factor. It can be implemented using a combination of reciprocal functions and product terms, aiming to accurately quantify the inhibitory or enhancing effect of social context on user interest based on the dynamic changes of the factor product. The user-item interaction model refers to the prediction mechanism integrating user interaction coefficients, item interaction coefficients, and social context enhancement coefficients. It can be implemented using a multiplicative fusion structure, aiming to organically couple multi-source information and output a quantifiable user-item interest intensity index.

[0146] Specifically, this invention first converts the original social trust coefficient and item activity coefficient into complementary factors, so that high trust strength or high activity status corresponds to low factor values, thereby avoiding logical conflicts. Then, it utilizes the nonlinear characteristics of the social context enhancement model. When the social trust coefficient factor or item status coefficient factor increases (indicating weakened trust or reduced activity), the denominator term increases accordingly, leading to a decrease in the enhancement coefficient, accurately simulating the dynamic inhibitory effect of social context on user interest. Finally, through the multiplication fusion mechanism of the user-item interaction model, the enhancement coefficient is multiplied by the user interaction coefficient and the item interaction coefficient. When the enhancement coefficient is high (indicating the coexistence of strong trust and high activity), the interaction coefficient is significantly amplified, realizing the dynamic quantitative integration of multi-source information and forming a complete user interest prediction chain.

[0147] As a specific implementation method, the solution of the present invention is implemented as follows: When the user's social trust coefficient for a certain item is at a high level and the item's activity coefficient is at a high level, the calculated social trust coefficient factor and item state coefficient factor are both low. After being substituted into the social context enhancement model, the denominator term approaches 1, thereby obtaining a high social context enhancement coefficient. After multiplying this enhancement coefficient by the user interaction coefficient and the item interaction coefficient, the user-item interaction coefficient is effectively amplified, accurately reflecting the user's true interest intensity in a strong trust social environment and a highly active item state, providing a reliable basis for the recommendation results.

[0148] Example 1

[0149] Personalized product recommendations on e-commerce platforms: A certain apparel e-commerce platform needs to recommend products that users have not purchased before.

[0150] 1. User interaction coefficient: Based on user A's historical browsing, collection, and purchase records (e.g., browsing dresses 3 times and collecting 2 shirts in the past month), the "Women's Clothing - High Interest" coefficient is calculated through the user interaction model.

[0151] 2. Item Interaction Coefficient: Based on the click and add-to-cart data of dress products (e.g., a floral dress has an add-to-cart rate of 30% in the past 7 days), the "floral dress - high popularity" coefficient is obtained through the item interaction model.

[0152] 3. Item Activity Coefficient: Based on the real-time inventory (sufficient inventory) and the time of new product release (3 days ago) of this floral dress, the "highly recommendable" coefficient is derived through the item activity model.

[0153] 4. Social Trust Coefficient: Based on the fact that User A's friend B recently purchased the same floral dress (friends interact 5 times a week), the "Friend Recommendation - High Trust" coefficient is derived through the social trust model.

[0154] 5. Fusion Calculation: Combine the above coefficients to construct a user-item interaction model and calculate the interaction coefficient of user A with the floral dress (which is 30% higher than the user's historical average).

[0155] 6. Recommended output: Sort the products that user A has not yet touched by according to the interaction coefficient, and select the top 10 (including the floral dress) as the homepage recommendation. In the end, the click-through rate of the product increased by 28%.

[0156] Example 2

[0157] Social content platform video recommendations: A short video platform recommends videos that users have not yet watched.

[0158] 1. User interaction coefficient: Based on user C's historical likes (likes on 8 pet videos in the past 7 days) and comments (comments on 2 "Corgi-related" videos), the "pet content - high preference" coefficient was derived through the user interaction model.

[0159] 2. Item Interaction Coefficient: Based on the completion rate of Corgi-related videos (65% completion rate for a certain video) and the number of shares (2000 times), the "Corgi Video - High Interaction" coefficient is derived through the item interaction model.

[0160] 3. Item Activity Coefficient: Based on the video's release time (2 hours ago) and current playback growth rate (10,000 times per hour), the "high timeliness" coefficient is derived through the item activity model.

[0161] 4. Social Trust Coefficient: Based on the fact that user C's followed blogger D posted the corgi video (user C's interaction rate with D is 80%), the "following blogger - high trust" coefficient was obtained through the social trust model.

[0162] 5. Fusion Calculation: Construct a user-item interaction model and calculate the interaction coefficient of user C for this video (40% higher than the average of similar videos on the platform).

[0163] 6. Recommended output: Sort the videos that user C has not watched by coefficient, select the top 5 (including the Corgi video) and push them to the homepage, which increases user C's video viewing time by 35%.

[0164] Example 3

[0165] Online education platform course recommendations: A certain programming education platform recommends courses that users have not yet taken.

[0166] 1. User interaction coefficient: Based on user D's historical learning records (completed Python basic course, bookmarked machine learning course), the coefficient of "AI direction - medium to high demand" is obtained through user interaction model.

[0167] 2. Item Interaction Coefficient: Based on the student completion rate (70%) and rating (4.8 / 5) of the "Introduction to Deep Learning" course, the "High Quality" coefficient is derived through the item interaction model.

[0168] 3. Item Activity Coefficient: Based on the recent enrollment volume of the course (200 new people per week) and the instructor's live broadcast schedule (Q&A session next week), the "high participation value" coefficient is derived through the item activity model.

[0169] 4. Social Trust Coefficient: Based on the fact that user D's classmate E has enrolled in the course (students interact and check in 3 times a week), the "classmate recommendation - high credibility" coefficient is obtained through the social trust model.

[0170] 5. Fusion Calculation: Construct a user-item interaction model and calculate the interaction coefficient of user D for this course (which is 25% higher than the average of courses at the user's current learning stage).

[0171] 6. Recommended output: Sort the courses that user D has not yet taken by coefficient, select the top 3 (including the deep learning course) and push them to the "To be taken" list, which increases the course enrollment conversion rate by 22%.

[0172] Through the above solution, this application effectively solves the problem of quantitative ambiguity in multi-source information fusion, enabling the recommendation system to accurately capture the dynamic interaction between social trust and item status, and significantly improving the adaptability and credibility of the recommendation results in real-world scenarios.

[0173] In summary, this invention effectively addresses the challenges of data sparsity and cold start in recommendation systems through a collaborative processing mechanism of multi-source data. The construction of user interaction coefficients fully leverages the complementarity of explicit ratings and implicit behavioral data, enabling the system to reliably infer preferences even with limited user historical behavior. The generation of item interaction coefficients integrates multi-dimensional popularity indicators, allowing newly listed or low-frequency items to receive reasonable evaluations based on real-time interaction data. The dynamic calculation mechanism of item activity coefficients ensures that recommended content is synchronized with changes in the item's lifecycle, avoiding outdated recommendations due to lack of timeliness. The quantitative introduction of social trust coefficients compensates for the shortcomings of traditional methods that rely solely on direct associations, allowing recommendation results to reflect implicit influences within social networks. This invention's user-item interaction model organically integrates social context and item status factors through a non-linear fusion mechanism, rather than simply superimposing data from various dimensions, achieving adaptive weighting of multiple factors and significantly improving the scenario adaptability and credibility of recommendation results. Furthermore, this invention's ranking and filtering mechanism based on interaction coefficients transforms the quantitative results of multi-source fusion into precise personalized recommendations, effectively solving the problem of insufficient recommendation adaptability caused by the fragmentation of multiple factors in existing technologies. This invention achieves the systematic integration and precise quantification of multi-dimensional user interaction behavior, effectively solving the quantification bias problem caused by differences in data scale. This enables the recommendation system to more accurately identify the intensity of users' true interest in items, thereby improving the accuracy of recommendation results and user satisfaction. The item interaction coefficient of this invention can accurately quantify the true popularity and quality balance of items, effectively avoiding the distortion of item weights caused by a single indicator, making the recommendation results more accurately reflect users' potential interests and preferences. This invention can effectively integrate item timeliness and dynamic behavior data, accurately quantifying the real-time freshness of items, solving the problem of inaccurate evaluation of newly listed items or items with short lifecycles, thereby improving the adaptability and credibility of recommendation results in time-sensitive scenarios. This invention enables the calculation of social trust coefficients to comprehensively consider multi-dimensional characteristics of user behavior, including forwarding timeliness, sharing activity, and social closeness, thereby more accurately reflecting the dynamics of trust propagation among users and effectively improving the adaptability and credibility of recommendation results.

[0174] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for mining associated user network data based on multi-information fusion, characterized in that, Includes the following steps: S1. Based on user interaction data of user u to item i, output user interaction coefficients through user interaction model; S2. Based on the item interaction data of item i, obtain the item interaction coefficient through the item interaction model; S3. Based on the item activity data of item i, obtain the item activity coefficient through the item activity model; S4. Based on user social trust data of user u to item i, obtain the social trust coefficient through the social trust model; S5. Construct a user-item interaction model based on the user interaction coefficient and item interaction coefficient under the social trust coefficient and item status coefficient, and output the user-item interaction coefficient. S6. Compare the user-item interaction coefficient of the obtained user with the user-item interaction coefficient of items that have not been touched, and obtain items with a higher user-item interaction coefficient than the user. S7. For user u, obtain the user-item interaction coefficients for all items, sort them from highest to lowest, and select the top N items as the recommendation results for output.

2. The method for mining associated user network data based on multi-information fusion according to claim 1, characterized in that: In step S1, the step of outputting user interaction coefficients based on user interaction data of user u to item i through the user interaction model is as follows: S11. Obtain user interaction data, including average user ratings, number of visits to item pages, and duration of visits to item pages; S12. Perform maximum-min normalization on the mean user rating, the number of times the item page is accessed, and the duration of the item page access to obtain the user rating index, the number of page accesses index, and the duration of the page access index. S13. Import the user rating index, page visit frequency index, and page visit duration index into the user interaction model to obtain the user interaction coefficient.

3. The method for mining associated user network data based on multi-information fusion according to claim 2, characterized in that: The user interaction model calculates the user interaction coefficient through a weighting function, wherein the user interaction coefficient is positively correlated with the intensity of individual user interaction with items; The user interaction model is represented as follows: ； in, Represents the user interaction coefficient. This represents the user rating index. This represents the page view count index. This represents the page visit duration index. Represents the weighting coefficient, and , The larger the value, the stronger the user's interaction with the item.

4. The method for mining associated user network data based on multi-information fusion according to claim 1, characterized in that: In step S2, the step of obtaining the item interaction coefficient based on the item interaction data of item i through the item interaction model is as follows: S21. Obtain item interaction data, including the total number of item ratings, the average item rating, and the total number of item views; S22. Sum the total number of item ratings, the average item rating, and the total number of item views with the corresponding average total number of item ratings, average item ratings, and average number of item views to obtain reference values ​​for the number of ratings, the ratings, and the views. S23. Ratio the total number of item ratings, the average item rating, and the total number of item views with the corresponding reference values ​​to obtain the rating count index, the average rating index, and the view index. S24. Import the rating frequency index, average rating index, and pageview index into the item interaction model to obtain the item interaction coefficient.

5. The method for mining associated user network data based on multi-information fusion according to claim 4, characterized in that: The item interaction model calculates the item interaction coefficient using an aggregation function, where the item interaction coefficient is positively correlated with the popularity of the item. The item interaction model is represented as follows: ； in, Indicates the interaction coefficient of items. Indicates the number of ratings index. This represents the average score index. This represents the pageview index. Furthermore, the higher the value, the more popular the item.

6. The method for mining associated user network data based on multi-information fusion according to claim 1, characterized in that: In step S3, the step of obtaining the item activity coefficient based on the item activity data of item i through the item activity model is as follows: S31. Obtain item interaction data, including item listing duration, item lifecycle, and item forwarding count; S32. Calculate the ratio of the item's listing duration to the item's lifecycle to obtain the item's timeliness index; S33. Sum the number of times an item is forwarded with the average number of times all items are forwarded to obtain a forwarding reference value. Ratio the number of times an item is forwarded with the item reference value to obtain the item forwarding index. S34. Import the item timeliness index and the item forwarding frequency index into the item activity model to obtain the item activity coefficient.

7. The method for mining associated user network data based on multi-information fusion according to claim 6, characterized in that: The item activity model calculates the item activity coefficient through a weighted combination function, wherein the item activity coefficient is positively correlated with item freshness and popularity. The item activity model is represented as follows: ； in, Indicates the activity level of an item. Indicates the timeliness index of the item. This indicates the number of times an item has been forwarded. Represents the balance weight coefficient and , Furthermore, the higher the value, the fresher and more active the item.

8. The method for mining associated user network data based on multi-information fusion according to claim 1, characterized in that: In step S4, the step of obtaining the social trust coefficient based on user u's social trust data for item i through the social trust model is as follows: S41. Perform maximum-min normalization on the user forwarding time to obtain the user forwarding time index; S42. Obtain user social trust data, including the number of times items are shared, the maximum number of mutual friends among the sharing friends, and the user's forwarding time; S43. Sum the number of times the item was shared, the maximum number of mutual friends among the sharing friends, and the corresponding average values ​​to obtain reference values ​​for the number of shares and mutual friends. S44. Ratio the number of times an item is shared and the maximum number of mutual friends among the sharing friends to the corresponding reference values ​​to obtain the sharing count index and mutual friend index. S45. Import the user forwarding time index, sharing frequency index, and mutual friend index into the social trust model to obtain the social trust coefficient.

9. A method for mining associated user network data based on multi-information fusion according to claim 8, characterized in that: In the social trust model, the social trust coefficient is positively correlated with the sharing frequency index and the mutual friend index, and negatively correlated with the user forwarding time index. The social trust model is represented as: ； in, Indicates the social trust coefficient. Indicates the forwarding time index. This represents the number of times the message was shared. Indicates the mutual friend index. The larger the value, the better for users. For items The greater the social trust, the better.

10. The method for mining associated user network data based on multi-information fusion according to claim 1, characterized in that: In step S5, the steps for constructing the user-item interaction model and outputting the user-item interaction coefficient based on the user interaction coefficient and item interaction coefficient under the social trust coefficient and item status coefficient are as follows: S51. Obtain the social trust coefficient and the item status coefficient and take their complements. Obtain the social trust coefficient factor and the item status coefficient factor. S52. Import the social trust coefficient factor and the item status coefficient factor into the social context enhancement model to obtain the social context enhancement coefficient. S53. Construct a user-item interaction model based on user interaction coefficients and item interaction coefficients under the social context enhancement coefficient, and output the user-item interaction coefficients. The social context enhancement model is represented as: ； in, Indicates the social context enhancement coefficient. Indicates the social trust factor. Indicates the activity factor of an item; The user-item interaction model is represented as follows: ； Wherein, represents the user-item interaction coefficient. Represents the user interaction coefficient. Indicates the interaction coefficient of items. Indicates the social context enhancement coefficient. The larger the value of u, the greater the user's interest in item i.

Images