Private domain traffic guiding method and system based on collaborative filtering

Abstract

The invention relates to the technical field of private domain traffic, in particular to a private domain traffic guiding method and system based on collaborative filtering, and the method comprises the following steps: S1, collecting user behavior data: collecting the behavior data of a user in a private domain platform in real time through client equipment; s2, constructing a user portrait: extracting the interest preference of the user, generating a user interest label, and forming the user portrait; s3, interest extension and cross-domain recommendation: constructing a relation graph between the user and the content, identifying association between the user interest node and the cross-domain content, calculating the edge weight between the user and the content node, and extending the interest boundary of the user in the private domain traffic; and S4, recommended content pushing: pushing cross-domain recommended content to the user, and meanwhile, optimizing the display sequence of the content in combination with the situation information of the user. According to the invention, the diversity and personalization of the recommended content can be enhanced, and the user experience and the diversity of the recommendation effect are further improved.

Description

Technical Field

[0001] The present invention relates to the field of private domain traffic technology, and in particular to a private domain traffic guiding method and system based on collaborative filtering. Background Art

[0002] In the current Internet environment, the management and utilization of private domain traffic is of great significance to the user growth and commercial value enhancement of the platform. Through accurate content recommendation, the platform can better meet the personalized needs of users and improve user stickiness and satisfaction. In this process, collaborative filtering is a commonly used recommendation algorithm that can infer users' interest preferences based on their historical behavior and realize personalized recommendations in private domain traffic. However, with the diversification of user behavior and the complexity of content needs, a single collaborative filtering algorithm faces certain limitations in dealing with complex interest expansion and cross-domain recommendation needs.

[0003] Among the existing recommendation technologies, although collaborative filtering can achieve basic recommendation effects through user behavior data, it has shortcomings in the following aspects: first, the dimensions of data collection are limited, and it is difficult to fully capture the user's behavioral characteristics, resulting in inaccurate user portraits and limited recommendation effects. Secondly, the existing technologies perform poorly in cross-domain recommendations and interest expansion, and lack effective methods to dynamically capture users' potential interests in different content areas, resulting in insufficient diversity in recommended content. In addition, recommendation systems usually ignore users' contextual information, making it difficult to provide real-time recommendations that meet users' current needs, reducing user experience and the actual effect of recommendations.

[0004] The purpose of the present invention is to provide a private domain traffic guidance method and system based on collaborative filtering to ensure that the recommended content is highly matched with the user's current needs, thereby significantly improving the user experience and the conversion efficiency of the platform. Summary of the invention

[0005] The present invention provides a private domain traffic guiding method and system based on collaborative filtering.

[0006] The private domain traffic guidance method based on collaborative filtering includes the following steps:

[0007] S1, user behavior data collection: collect user behavior data in the private domain platform in real time through client devices, including browsing history, number of clicks, page dwell time and purchase history;

[0008] S2, user portrait construction: Based on the collected behavior data, extract the user's interest preferences, generate user interest tags, and form a user portrait;

[0009] S3, interest expansion and cross-domain recommendation: Based on the existing interest tags in the user portrait, build a relationship graph between users and content, identify user interest nodes and cross-domain content associations, and calculate the edge weights between users and content nodes, expand the user's interest boundaries in private domain traffic, and implement cross-domain recommendations, including:

[0010] S31, relationship graph construction: Based on the interest tags extracted from the user portrait, the Graph Attention Network (GAT) model is used to represent the user and content nodes as nodes in the graph structure. The user is connected with the content he browses, clicks, and purchases to generate a relationship graph between the user and the content.

[0011] S32, cross-domain interest node identification: In the constructed relationship graph, the edges between the user and the content nodes are analyzed to identify potential cross-domain interest nodes related to the user's current interests, thereby expanding the user's interest range;

[0012] S33, weight calculation and recommendation output: Calculate the edge weights between the user and the content node to form a cross-domain recommendation list, sort the recommended content according to the weight value, prioritize the cross-domain content that best matches the user's interests, and generate the final cross-domain recommendation output;

[0013] S4, recommended content push: based on the results of interest expansion and cross-domain recommendation, cross-domain recommended content is pushed to users. At the same time, the display order of content is optimized based on the user's contextual information (access time, geographic location).

[0014] Optionally, the user behavior data collection in S1 includes:

[0015] S11, Browsing record collection: When a user visits a private domain platform, the pages visited by the user, the number of times, the order of visits, and the length of stay are automatically recorded to generate a browsing track;

[0016] S12, click count collection: monitor all user interactions on the page and record the click count and click location of each page;

[0017] S13, page dwell time collection: calculate the user's dwell time on each page, and trigger the recording of the dwell time through the user's page leaving or switching operations;

[0018] S14, purchase history collection: after the user completes the purchase operation, the information of the goods or services purchased by the user is recorded, including the product category, price, frequency, and purchase time.

[0019] Optionally, the user portrait construction in S2 includes:

[0020] S21, interest preference extraction: Calculate the user's interest weight W for each category based on the user's browsing times, click frequency, page dwell time and purchase history on different content categories. i ;

[0021] S22, user interest tag generation: based on the interest weight W of each category i , by comparing with the preset weight threshold T, when W i >T, the tag of category i is added to the user's interest tag set, indicating that the user has a preference for this type of content;

[0022] S23, user portrait generation: based on the generated user interest tags, a user portrait is formed, including interest weights and interest categories.

[0023] Optionally, the relationship graph construction in S31 includes:

[0024] S311, node representation: users and content are represented as nodes in the graph respectively, user node U i Represents the portrait of user i, which includes the user's interest tags and content node C j Indicates the category and label of content j;

[0025] S312, connection relationship construction: establish user node U according to the user's behavior data on the content i With content node C j The connection relationship between them, the weight of the connection relationship E ij Indicates the interest level of user i in content j;

[0026] S313, graph attention mechanism calculation: Calculate each node U through the graph attention network (GAT) model i Connect node C to it j The attention coefficient A ij ;

[0027] S314, relationship graph generation: attention coefficient A calculated by the graph attention network (GAT) model ij As edge weights, a relationship graph between users and content is generated.

[0028] Optionally, the cross-domain interest node identification in S32 includes:

[0029] S321, interest correlation calculation: In the relationship graph, the interest correlation of each user node U is calculated based on the behavior data in the past 30 days. i With content node C j The correlation R ij ;

[0030] S322, cross-domain node attention weight calculation: Use the graph attention network (GAT) model to calculate the attention weight of the user node U i and cross-domain content node C k Calculate the attention weight A ik ;

[0031] S323, Identification of potential cross-domain interest nodes: Based on attention weight A ik Relevance to interest R ij , filter out cross-domain content nodes related to the user's current interests, when A ik ×R ij When the expansion threshold θ is exceeded, the cross-domain content node C k A node that is identified as a user's potential cross-domain interest node indicates that the node content matches the user's interest and is used to expand the user's interest range.

[0032] Optionally, the weight calculation and recommendation output in S33 include:

[0033] S331, edge weight calculation: In the relationship graph, by combining the user interest correlation R ij And the graph attention weight A ik Calculate user node U i and cross-domain content node C k The edge weight W between ik ;

[0034] S332, recommendation list generation: The calculated edge weight W ik Sort by value from high to low to form a cross-domain recommendation list;

[0035] S333, recommended content output: based on the sorting results, the cross-domain content nodes with the highest weight values ​​are recommended to the user first, and a cross-domain recommendation output is generated.

[0036] Optionally, the recommended content push in S4 includes:

[0037] S41, display of recommended content: according to the generated cross-domain recommendation output, the sorted cross-domain recommended content is selected to generate a user recommended content list, and the content with the highest weight value is displayed first;

[0038] S42, context information optimization: optimizing the display order of recommended content based on the user's context information (access time, geographic location).

[0039] Optionally, the context information optimization in S42 includes:

[0040] S421, context weight calculation: For each recommended content, the context weight S is calculated based on the user's access time and geographical location. ik, to measure the display priority of content;

[0041] S422, display order optimization: In the generated recommended content list, the context weight S ij And the result of edge weight calculation W ik Calculate the comprehensive display weight R ik ;

[0042] S423, final recommendation display: all recommended contents are displayed according to the comprehensive display weight R ik Sort from high to low, and prioritize showing users content with the highest comprehensive weight.

[0043] The private domain traffic guidance system based on collaborative filtering is used to implement the above-mentioned private domain traffic guidance method based on collaborative filtering, and includes the following modules:

[0044] User behavior data collection module: collects user behavior data in the private domain platform in real time through client devices. The behavior data includes user browsing history, number of clicks, page dwell time and purchase history;

[0045] User portrait building module: Based on the collected behavior data, extract the user's interest preferences, generate the user's interest tags, and build a user portrait;

[0046] Interest expansion and cross-domain recommendation module: Based on the interest tags in the user portrait, a relationship graph between users and content is constructed, the user interest nodes and cross-domain content associations are identified, and the edge weights between the user and content nodes are calculated to expand the user's interest boundaries in private domain traffic and generate cross-domain recommended content;

[0047] Recommended content push module: Based on the results of interest expansion and cross-domain recommendations, cross-domain recommended content is pushed to users, and the display order of recommended content is optimized based on the user's contextual information.

[0048] Beneficial effects of the present invention:

[0049] The present invention collects user behavior data in real time and builds user portraits based on these data, so that the recommendation system can deeply understand the user's interest preferences. Through refined behavior data collection and weighted analysis, the system can accurately capture the user's preference intensity in different content categories and generate multi-dimensional user portraits. This not only provides a rich data foundation for the recommendation system, but also ensures the real-time and adaptability of the portrait by dynamically adjusting the generation method of interest tags, thereby significantly improving the accuracy and response speed of the recommendation system.

[0050] The present invention realizes interest expansion and cross-domain recommendation by utilizing the graph attention network model, constructs a relationship graph between users and content, combines user interest correlation and attention weight, generates cross-domain recommended content, and enables the recommendation system to effectively expand the user's interest boundaries, explore the user's potential interests, and recommend more relevant cross-domain content. By preferentially showing users content that highly matches their interests, the system can enhance the diversity and personalization of recommended content, and further improve the user experience and the diversity of recommendation effects.

[0051] The present invention dynamically optimizes the display order of recommended content by combining the user's contextual information, generates a comprehensive display weight by calculating the contextual weight and combining it with the original recommendation weight of the content, and realizes context-related personalized recommendations. This method ensures that the recommended content not only meets the user's interests, but also is highly consistent with the user's current context, making the recommendation more accurate and timely, significantly improving the relevance of the recommendation and user experience, and enhancing the adaptability and practicality of the recommendation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 A schematic diagram of a boot method flow chart of an embodiment of the present invention;

[0054] Figure 2 Schematic diagram of system function modules according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0056] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0057] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0058] like Figure 1 As shown, the private domain traffic guidance method based on collaborative filtering includes the following steps:

[0059] S1, user behavior data collection: collect user behavior data in the private domain platform in real time through client devices, including browsing history, number of clicks, page dwell time and purchase history;

[0060] S2, user portrait construction: Based on the collected behavior data, extract the user's interest preferences, generate user interest tags, and form a user portrait;

[0061] S3, interest expansion and cross-domain recommendation: Based on the existing interest tags in the user portrait, build a relationship graph between users and content, identify user interest nodes and cross-domain content associations, and calculate the edge weights between users and content nodes, expand the user's interest boundaries in private domain traffic, and implement cross-domain recommendations, including:

[0062] S31, relationship graph construction: Based on the interest tags extracted from the user portrait, the Graph Attention Network (GAT) model is used to represent the user and content nodes as nodes in the graph structure. The user is connected with the content he browses, clicks, and purchases to generate a relationship graph between the user and the content.

[0063] S32, cross-domain interest node identification: In the constructed relationship graph, the edges between the user and the content nodes are analyzed to identify potential cross-domain interest nodes related to the user's current interests, thereby expanding the user's interest range;

[0064] S33, weight calculation and recommendation output: Calculate the edge weights between the user and the content node to form a cross-domain recommendation list, sort the recommended content according to the weight value, prioritize the cross-domain content that best matches the user's interests, and generate the final cross-domain recommendation output;

[0065] S4, recommended content push: based on the results of interest expansion and cross-domain recommendation, cross-domain recommended content is pushed to users to ensure that the recommended content is consistent with the user's interests. At the same time, the display order of the content is optimized based on the user's contextual information (access time, geographic location) to enhance the relevance and timeliness of the recommendation;

[0066] Through the above content, based on the collection and portrait construction of user behavior data, the graph attention network is used to realize interest expansion and cross-domain recommendation, and the personalized content push is carried out in combination with user context information, so as to accurately match user interests, expand the diversity and cross-domain relevance of recommended content, and improve the personalization and dynamic adaptability of the recommendation system, provide more suitable recommendation results, and effectively enhance user experience and the utilization efficiency of private domain traffic.

[0067] User behavior data collection in S1 includes:

[0068] S11, Browsing record collection: When a user visits a private domain platform, the pages visited by the user, the number of times, the order of visits, and the length of stay are automatically recorded to generate a browsing track;

[0069] S12, click count collection: monitor all user interactions on the page, record the number of clicks and click positions on each page to reflect the user's attention to different content;

[0070] S13, page dwell time collection: calculate the user's dwell time on each page, and trigger the recording of the dwell time through the user's page leaving or switching operations;

[0071] S14, purchase history collection: after the user completes the purchase operation, the information of the goods or services purchased by the user is recorded, including the category, price, frequency, and purchase time of the goods, so as to improve the user's consumption preference information;

[0072] Through the above content, various types of user behavior data in the private domain platform can be recorded in real time and comprehensively, including browsing habits, content preferences, interaction frequency, consumption history and other information, providing a high-quality data foundation for user portraits and personalized recommendations. By capturing user behavior in detail, the accuracy and response speed of the recommendation system are improved, which helps to achieve content push that is more in line with user interests and needs, thereby improving user experience and platform conversion rate.

[0073] The construction of user portraits in S2 includes:

[0074] S21, interest preference extraction: Calculate the user's interest weight W for each category based on the user's browsing times, click frequency, page dwell time and purchase history on different content categories. i , expressed as:

[0075] W i =a×X+b×Y+c×Z+d×P;

[0076] Among them, W irepresents the user's interest weight in category i, X represents the number of views, Y represents the page dwelling time, Z represents the click frequency, P represents the purchase frequency, and a, b, c, and d are the weight coefficients of each behavioral feature respectively;

[0077] S22, user interest tag generation: based on the interest weight W of each category i , by comparing with the preset weight threshold T, when W i >T, the tag of category i is added to the user's interest tag set, indicating that the user has a preference for this type of content;

[0078] The weight threshold T is determined through statistical analysis of user behavior data, including:

[0079] Calculate the mean and standard deviation of user behavior data: collect the number of views, dwell time, click frequency, and purchase frequency of a large number of users, and calculate the mean μ and standard deviation σ of these data in each interest category;

[0080] Dynamic threshold setting: Based on the mean and standard deviation, define the interest threshold T, expressed as:

[0081] T = μ + k·σ;

[0082] Among them, T is the threshold of interest, k is the adjustment factor;

[0083] S23, user portrait generation: based on the generated user interest tags, a user portrait is formed, including interest weights and interest categories. The interest weights represent the user's preference strength for different tags, and the interest categories aggregate the tags into different content topics.

[0084] Through the above content, we conduct a weighted analysis of users' behaviors in different content categories, calculate interest weights, and generate interest tags based on dynamic threshold settings, so that the portrait can accurately reflect the user's true preferences, accurately screen the user's core interest content, and adapt to changes in user behavior through dynamically adjusted thresholds, ensuring the real-time and accuracy of the portrait, providing efficient and accurate data support for personalized recommendations, thereby improving user experience and recommendation effects.

[0085] The relationship graph construction in S31 includes:

[0086] S311, node representation: users and content are represented as nodes in the graph respectively, user node U i Represents the portrait of user i, which includes the user's interest tags and content node C j Indicates the category and label of content j;

[0087] S312, connection relationship construction: establish user node U according to the user's behavior data on the content iWith content node C j The connection relationship between them, the weight of the connection relationship E ij Indicates the interest of user i in content j, expressed as:

[0088] E ij =α×V ij +β×K ij +γ×P ij ;

[0089] Among them, V ij represents the number of times user i browses content j, K ij represents the click frequency of user i on content j, P ij represents the purchase frequency of content j by user i, α, β, γ are the weight coefficients of the behavior characteristics;

[0090] S313, graph attention mechanism calculation: Calculate each node U through the graph attention network (GAT) model i Connect node C to it j The attention coefficient A ij , expressed as:

[0091]

[0092] in, and They are user nodes U i and content node C j The feature vector of , W is the trainable weight matrix, || represents the connection operation of the feature vector, σ is the LeakyReLU activation function softmax is used for normalization, so that the sum of the attention weights of each node is 1;

[0093] S314, relationship graph generation: attention coefficient A calculated by the graph attention network (GAT) model ij As edge weights, a relationship graph between users and content is generated;

[0094] Through the above content, the user's interest weights in different content can be dynamically calculated, so that the relationship graph can accurately reflect the intensity and changes of user interests, thereby providing a more accurate and personalized user portrait model for the recommendation system, improving the relevance and adaptability of recommended content, and helping to significantly improve user experience and recommendation accuracy.

[0095] The cross-domain interest node identification in S32 includes:

[0096] S321, interest correlation calculation: In the relationship graph, the interest correlation of each user node U is calculated based on the behavior data in the past 30 days. i With content node C j The correlation Rij , expressed as:

[0097] R ij =λ×V ij ′+μ×K ij ′+v×P ij ′;

[0098] Among them, V ij ′ represents the number of times user i browsed content j in the past 30 days, K ij ′ represents the number of clicks of user i on content j in the past 30 days, P ij ′ represents the purchase frequency of content j by user i in the past 30 days, and λ, μ, and v are the corresponding weight coefficients respectively;

[0099] S322, cross-domain node attention weight calculation: Use the graph attention network (GAT) model to calculate the attention weight of the user node U i and cross-domain content node C k Calculate the attention weight A ik , to identify the user's potential interests in other content areas, expressed as:

[0100]

[0101] in, and are the feature vectors of user nodes and cross-domain content nodes respectively, W is a trainable weight matrix, || represents the connection operation of feature vectors, and σ is the LeakyReLU activation function;

[0102] S323, Identification of potential cross-domain interest nodes: Based on attention weight A ik Relevance to interest R ij , filter out cross-domain content nodes related to the user's current interests, when A ik ×R ij When the expansion threshold θ is exceeded, the cross-domain content node C k Identified as a user's potential cross-domain interest node, indicating that the node content matches the user's interests and is used to expand the user's interest range;

[0103] The expansion threshold θ is set based on the user interest correlation, including:

[0104] Calculate the mean and standard deviation of user association: Get the interest association value R of all users and content nodes from the relationship graph ij , and calculate the average of these values and standard deviation σ R ;

[0105] Threshold setting: Set the threshold θ to the average value Add a multiple of the standard deviation, expressed as:

[0106]

[0107] Among them, θ is the threshold of interest relevance, is the average value of interest correlation between all users and content nodes, σ R is the standard deviation of interest relevance, k is the adjustment factor;

[0108] Through the above content, cross-domain content nodes related to the user's current interests can be accurately identified, ensuring that the screening results can flexibly adapt to changes in user behavior and identifying content that is highly matched with the user. This not only effectively expands the user's interest range, but also improves the relevance and diversity of recommended content, providing users with richer cross-domain content recommendations that meet their personalized needs, thereby significantly enhancing the user experience.

[0109] The weight calculation and recommendation output in S33 include:

[0110] S331, edge weight calculation: In the relationship graph, by combining the user interest correlation R ij And the graph attention weight A ik Calculate user node U i and cross-domain content node C k The edge weight W between ik , expressed as:

[0111] W ik =A ik ×R ij ;

[0112] S332, recommendation list generation: The calculated edge weight W ik Sort by value from high to low to form a cross-domain recommendation list. The higher the weight value, the more cross-domain content nodes C k The higher the match with user interests;

[0113] S333, recommended content output: based on the sorting results, the cross-domain content nodes with the highest weight values ​​are preferentially recommended to the user, and a cross-domain recommendation output is generated;

[0114] Through the above content, the matching degree between users and cross-domain content nodes is accurately measured, and the recommendation list is generated by sorting the edge weights. It can give priority to recommending content that is highly consistent with user interests, ensuring the accuracy and relevance of the recommendations. It not only fully taps the user's potential interests to expand the content, but also improves the diversity and personalization level of the recommendation system, thereby significantly enhancing the user experience.

[0115] Recommended content push in S4 includes:

[0116] S41, display of recommended content: according to the generated cross-domain recommendation output, the sorted cross-domain recommended content is selected to generate a user recommended content list, and the content with the highest weight value is displayed first to ensure that the recommended content is highly matched with the user's interests and realize personalized recommendation;

[0117] S42, context information optimization: optimizing the display order of recommended content based on the user's context information (access time, geographic location);

[0118] Through the above content, we ensure that the recommended content is highly matched with the user's interests, enhance the accuracy of personalized recommendations, and at the same time, further optimize the content display order based on the user's contextual information, making the recommendations more timely and context-relevant. This can improve the recommendation effect while providing content display that is more in line with the user's current needs, significantly improving the user experience and the practicality of the recommendations.

[0119] Situational information optimization in S42 includes:

[0120] S421, context weight calculation: For each recommended content, the context weight S is calculated based on the user's access time and geographical location. ik , to measure the display priority of the content, expressed as:

[0121] S ik =w1×T i +w2×L i ;

[0122] Among them, S ik is the priority display weight of recommended content k in the current context of user i, T i represents the access time weight of user i, L i represents the geographical location weight of user i, w1 and w2 are the corresponding adjustment coefficients, w1 is 0.6, and w2 is 0.4;

[0123] S422, display order optimization: In the generated recommended content list, the context weight S ij And the result of edge weight calculation W ik Calculate the comprehensive display weight R ik , expressed as:

[0124] R ik =W ik +S ik ;

[0125] Among them, W ik is the original recommendation weight between user i and recommended content k, reflecting the matching degree between content and user interest, S ik is the situational weight;

[0126] S423, final recommendation display: all recommended contents are displayed according to the comprehensive display weight R ik Sort from high to low, and prioritize showing users the content with the highest comprehensive weight;

[0127] Through the above content, the contextual weight is calculated in combination with the user's contextual information, and it is taken into consideration together with the original weight of the recommended content to dynamically optimize the display order of the recommended content. On the basis of matching the recommended content with the user's interests, it can give priority to displaying content that is highly relevant to the user's current context, ensuring the real-time and personalization of the recommendation. This not only improves the relevance of the recommended content, but also greatly enhances the user experience, making the recommendation system more adaptable and accurate.

[0128] like Figure 2 As shown, the private domain traffic guidance system based on collaborative filtering is used to implement the above-mentioned private domain traffic guidance method based on collaborative filtering, and includes the following modules:

[0129] User behavior data collection module: collects user behavior data in the private domain platform in real time through client devices. The behavior data includes user browsing history, number of clicks, page dwell time and purchase history;

[0130] User portrait building module: Based on the collected behavior data, extract the user's interest preferences, generate the user's interest tags, and build a user portrait;

[0131] Interest expansion and cross-domain recommendation module: Based on the interest tags in the user portrait, a relationship graph between users and content is constructed, the user interest nodes and cross-domain content associations are identified, and the edge weights between the user and content nodes are calculated to expand the user's interest boundaries in private domain traffic and generate cross-domain recommended content;

[0132] Recommended content push module: Based on the results of interest expansion and cross-domain recommendations, cross-domain recommended content is pushed to users, and the display order of recommended content is optimized based on the user's contextual information.

[0133] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0134] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A private domain traffic guidance method based on collaborative filtering, characterized in that: The following steps are involved: S1, user behavior data collection: collect user behavior data in the private domain platform in real time through client devices, including browsing history, number of clicks, page dwell time and purchase history; S2, user portrait construction: Based on the collected behavior data, extract the user's interest preferences, generate user interest tags, and form a user portrait; S3, interest expansion and cross-domain recommendation: Based on the existing interest tags in the user portrait, build a relationship graph between users and content, identify user interest nodes and cross-domain content associations, and calculate the edge weights between users and content nodes, expand the user's interest boundaries in private domain traffic, and implement cross-domain recommendations, including: S31, Relationship graph construction: Based on the interest tags extracted from the user portrait, the graph attention network model is used to represent the user and content nodes as nodes in the graph structure. The user is connected with the content he browses, clicks and purchases to generate a relationship graph between the user and the content. S32, cross-domain interest node identification: In the constructed relationship graph, the edges between the user and the content nodes are analyzed to identify potential cross-domain interest nodes related to the user's current interests, thereby expanding the user's interest range; S33, weight calculation and recommendation output: calculate the edge weights between the user and the content node, form a cross-domain recommendation list, sort the recommended content according to the weight value, prioritize the cross-domain content that best matches the user's interests, and generate the final cross-domain recommendation output; S4, recommended content push: based on the results of interest expansion and cross-domain recommendation, cross-domain recommended content is pushed to users. At the same time, the display order of content is optimized based on the user's contextual information.

2. The method for guiding private domain traffic based on collaborative filtering according to claim 1, characterized in that: The user behavior data collection in S1 includes: S11, Browsing record collection: When a user visits a private domain platform, the pages visited by the user, the number of times, the order of visits, and the length of stay are automatically recorded to generate a browsing track; S12, click count collection: monitor all user interactions on the page and record the click count and click location of each page; S13, page dwell time collection: calculate the user's dwell time on each page, and trigger the recording of the dwell time through the user's page leaving or switching operations; S14, purchase history collection: after the user completes the purchase operation, the information of the goods or services purchased by the user is recorded, including the product category, price, frequency, and purchase time.

3. The method for guiding private domain traffic based on collaborative filtering according to claim 2 is characterized in that: The user portrait construction in S2 includes: S21, interest preference extraction: Calculate the user's interest weight W for each category based on the user's browsing times, click frequency, page dwell time and purchase history on different content categories. i ; S22, user interest tag generation: based on the interest weight W of each category i , by comparing with the preset weight threshold T, when W i >T, the tag of category i is added to the user's interest tag set, indicating that the user has a preference for this type of content; S23, user portrait generation: based on the generated user interest tags, a user portrait is formed, including interest weights and interest categories.

4. The method for guiding private domain traffic based on collaborative filtering according to claim 1, characterized in that: The relationship graph construction in S31 includes: S311, node representation: users and content are represented as nodes in the graph respectively, user node U i Represents the portrait of user i, which includes the user's interest tags and content node C j Indicates the category and label of content j; S312, connection relationship construction: establish user node U according to the user's behavior data on the content i With content node C j The connection relationship between them, the weight of the connection relationship E ij Indicates the interest level of user i in content j; S313, graph attention mechanism calculation: Calculate each node U through the graph attention network model i Connect node C to it j The attention coefficient A ij ; S314, relationship graph generation: attention coefficient A calculated by graph attention network model ij As edge weights, a relationship graph between users and content is generated.

5. The method for guiding private domain traffic based on collaborative filtering according to claim 4 is characterized in that: The cross-domain interest node identification in S32 includes: S321, interest correlation calculation: In the relationship graph, the interest correlation of each user node U is calculated based on the behavior data in the past 30 days. i With content node C j The correlation R ij ; S322, cross-domain node attention weight calculation: Use the graph attention network model to calculate the attention weight of the user node U i and cross-domain content node C k Calculate the attention weight A ik ; S323, Identification of potential cross-domain interest nodes: Based on attention weight A ik Relevance to interest R ij , filter out cross-domain content nodes related to the user's current interests, when A ik ×R ij When the expansion threshold θ is exceeded, the cross-domain content node C k A node that is identified as a user's potential cross-domain interest node indicates that the node content matches the user's interest and is used to expand the user's interest range.

6. The method for guiding private domain traffic based on collaborative filtering according to claim 5, characterized in that: The weight calculation and recommendation output in S33 include: S331, edge weight calculation: In the relationship graph, by combining the user interest correlation R ij And the graph attention weight A ik Calculate user node U i and cross-domain content node C k The edge weight W between ik ; S332, recommendation list generation: The calculated edge weight W ik Sort by value from high to low to form a cross-domain recommendation list; S333, recommended content output: based on the sorting results, the cross-domain content nodes with the highest weight values ​​are recommended to the user first, and a cross-domain recommendation output is generated.

7. The method for guiding private domain traffic based on collaborative filtering according to claim 6, characterized in that: The recommended content push in S4 includes: S41, display of recommended content: according to the generated cross-domain recommendation output, the sorted cross-domain recommended content is selected to generate a user recommended content list, and the content with the highest weight value is displayed first; S42, context information optimization: optimizing the display order of recommended content based on the user's context information.

8. The method for guiding private domain traffic based on collaborative filtering according to claim 7, characterized in that: The context information optimization in S42 includes: S421, context weight calculation: For each recommended content, the context weight S is calculated based on the user's access time and geographical location. ik , to measure the display priority of content; S422, display order optimization: In the generated recommended content list, the context weight S ij And the result of edge weight calculation W ik Calculate the comprehensive display weight R ik ; S423, final recommendation display: all recommended contents are displayed according to the comprehensive display weight R ik Sort from high to low, and prioritize showing users content with the highest comprehensive weight.

9. A private domain traffic guidance system based on collaborative filtering, used to implement the private domain traffic guidance method based on collaborative filtering as described in any one of claims 1 to 9, characterized in that: Includes the following modules: User behavior data collection module: collects user behavior data in the private domain platform in real time through client devices. The behavior data includes user browsing history, number of clicks, page dwell time and purchase history; User portrait building module: Based on the collected behavior data, extract the user's interest preferences, generate the user's interest tags, and build a user portrait; Interest expansion and cross-domain recommendation module: Based on the interest tags in the user portrait, a relationship graph between users and content is constructed, the user interest nodes and cross-domain content associations are identified, and the edge weights between the user and content nodes are calculated to expand the user's interest boundaries in private domain traffic and generate cross-domain recommended content; Recommended content push module: Based on the results of interest expansion and cross-domain recommendations, cross-domain recommended content is pushed to users, and the display order of recommended content is optimized based on the user's contextual information.

Images