Data pushing method and system based on user attribute analysis

Abstract

The invention relates to the field of data pushing, solves the technical problem that recommended data does not meet user requirements due to the fact that dynamic changes of user preferences are not considered in the prior art, and particularly relates to a data pushing method and system based on user attribute analysis. Comprising the following steps: S1, acquiring basic attribute information of a historical user and a target user, and generating attribute information of the historical user and attribute information of the target user; s2, establishing a target model according to the attribute information of the historical users, and classifying the historical users to generate user types of the historical users; s3, analyzing the user type of the target user by taking the attribute information of the target user as the input of the target model; s4, calculating the priority degree of each piece of corresponding recommended data based on the user type of the target user; compared with a common data recommendation method, dynamic changes of user preferences are considered, the priority degree of each piece of recommended data is calculated according to the recent behaviors of the user to be pushed, and the user experience and the transaction rate can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of data push, and in particular to a data push method and system based on user attribute analysis. Background Art

[0002] The common data push method is generally to classify users according to their basic attributes, and then recommend data to users based on the data corresponding to the classification. However, this method does not take into account that users' preferences will change dynamically in daily life. Therefore, when users are classified according to the above method and data is recommended based on the classification, the recommended data will have a low match with the user's current preferences, resulting in a poor user experience. Summary of the invention

[0003] In view of the deficiencies of the prior art, the present invention provides a data push method and system based on user attribute analysis, which solves the technical problem in the prior art that the dynamic changes of user preferences are not taken into account, resulting in the recommended data not meeting user needs.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A data push method and system based on user attribute analysis, comprising the following steps:

[0006] S1. Obtain basic attribute information of historical users and target users, and generate historical user attribute information and target user attribute information;

[0007] S2. Establish a target model based on historical user attribute information, and classify historical users to generate user types of historical users;

[0008] S3, using the target user attribute information as the input of the target model to analyze the user type of the target user;

[0009] S4. Calculate the priority level P(U, i) of each corresponding recommended data based on the user type of the target user;

[0010] S5. Push recommended data to target users based on priority P(U, i).

[0011] Furthermore, in step S1, the basic attribute information includes purchase and consumption records, browsing records, location information and registration information.

[0012] Furthermore, in step S2, the following steps are specifically included:

[0013] S21. Historical users and their behaviors are used as nodes and edges of the graph network respectively. The expression of the node is:

[0014] N(u)=(a1 , …, a 6 )

[0015] Wherein, N(u) represents the node corresponding to the u-th user; a 1 , …, a 6 respectively represent the registered name, age, gender, location, registration time, and total consumption amount of the user;

[0016] S22. Calculate the second weight W purchase (u, v), the third weight W location (u, v), and the fourth weight W browse (u, v) of the edge, and construct an edge weight set W, W = (W 1 , …, W 3 ), where W 1 , …, W 3 respectively represent the second weight W purchase (u, v), the third weight W location (u, v), and the fourth weight W browse (u, v);

[0017] S23. Calculate the first weight W(u, v) of the edge according to the edge weight set W;

[0018] S24. Generate a user classification graph network for classifying historical users according to the first weight W(u, v);

[0019] S25. Train the graph neural network to obtain a target model, and input the user classification graph network into the target model to output the user type of the historical user.

[0020] Furthermore, in step S22, the calculation formula of the second weight W purchase (u, v) is:

[0021] W purchase (u, v) = |{p|p ∈ P u ∩P v}|

[0022] Wherein, P u and P v respectively represent the sets of purchased goods in the purchase consumption records of the u-th and v-th users; p represents the goods jointly purchased by the u-th and v-th users;

[0023] The calculation formula of the third weight W location (u, v) is:

[0024]

[0025] Wherein, loc uand loc v Respectively represent the geographic location coordinates of the uth and vth users; d euclidean (loc u ,loc v ) represents the Euclidean distance function between the geographic location coordinates of the u-th and v-th users;

[0026] Fourth weight W browse The calculation formula for (u, v) is:

[0027]

[0028] In the formula, B u and B v Represent the sets of products and videos browsed by the u-th and v-th users respectively.

[0029] Furthermore, in step S23, the calculation formula of the first weight W(u, v) is:

[0030]

[0031] In the formula, w i and w jk The first influencing parameter and the second influencing parameter with respect to the first weight are represented respectively.

[0032] Further, in step S25, the expression of the node v update of the objective function is:

[0033]

[0034] In the formula, u represents the neighbor node of node v; Represents the feature vector of node v at layer l+1; Represents the feature vector of node v at layer l; represents the feature vector of node u at layer l; N(v) represents the set of all neighbor nodes u of node v; u∈N(v)∪{v} represents the set of node v and all neighbor nodes u of node v; W (l) Represents the learnable weight matrix from layer l to layer l+1; ω uv represents the first weight W(u, v) of the edge between node v and its neighbor node u; B (l) represents the bias term; σ represents the activation function.

[0035] Furthermore, in step S4, the following steps are specifically included:

[0036] S41, obtaining a recommendation data set consisting of multiple groups of recommendation data;

[0037] S42, calculating the priority influencing parameters of each recommended data in the recommended data set, the priority influencing parameters including the time decay factor D(t), the behavior intensity weight W(b), the content relevance R(U, i) and the content freshness bonus item F(i);

[0038] S43. Calculate the priority P(U, i) according to the priority influencing parameter.

[0039] Furthermore, in step S42, the calculation formula of the time attenuation factor D(t) is:

[0040] D(t)=e -λt

[0041] Where t represents the time interval between the last time the target user saw the same type of recommended data; λ represents the decay rate parameter;

[0042] The expression of behavior intensity weight W(b) is:

[0043]

[0044] Among them, b represents the target user’s behavior, which are browsing, searching, and purchasing; w view 、w search and w purchase They respectively represent the behavior intensity weight values ​​corresponding to different behaviors of the target user;

[0045] The calculation formula of content relevance R(U, i) is:

[0046] R(U,i)=V I ×100%

[0047] Where V I Indicates the number of target users browsing data of the same type as the recommended data;

[0048] The calculation formula for the content freshness bonus item F(i) is:

[0049]

[0050] In the formula, view y Indicates the current number of views of the recommended data.

[0051] Furthermore, in step S43, the calculation formula of the priority P(U, i) is:

[0052]

[0053] Where m represents the total number of target users’ behaviors, namely, browsing, searching, and purchasing.

[0054] A data push system based on user attribute analysis, comprising:

[0055] An information acquisition unit, the information acquisition unit is used to acquire basic attribute information of historical users and target users;

[0056] A user classification unit, the user classification unit is used to calculate the user type of the target user;

[0057] The data push unit is used to calculate the priority of the recommended data corresponding to the user type and push the data according to the priority.

[0058] Compared with the prior art, the present invention provides a data push method and system based on user attribute analysis, which has the following beneficial effects:

[0059] 1. Compared with common data recommendation methods, the present invention takes into account the dynamic changes of user preferences, calculates the priority of each recommended data according to the user's recent behavior and pushes it, which can effectively improve user experience and transaction rate.

[0060] 2. Compared with other user classification methods, the present invention adopts a neural network model for classification, which can explore the implicit relationship between different users and basic attribute information to improve the accuracy of user classification. In addition, when constructing a user classification graph network, the mutual influence between each basic attribute information of the user is further explored to further improve the accuracy of user classification.

[0061] 3. When calculating the first weight of an edge, the present invention takes into account the influence of the user's purchase and consumption records, location information and browsing records on the weight of the edge. Compared with other calculation methods, the mutual influence between different basic attribute information on the first weight is also taken into account. Therefore, the present invention is more accurate in calculating the first weight, which is convenient for improving the accuracy of user classification in the later stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0063] Figure 1 A flowchart of a data push method based on user attribute analysis of the present invention;

[0064] Figure 2 The block diagram of a data push system based on user attribute analysis of the present invention. DETAILED DESCRIPTION

[0065] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific implementation methods, so that the implementation process of how the present application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0066] In order to improve user experience, increase user engagement and achieve more effective marketing, it is necessary to analyze user attributes and push data based on the analysis results. Common user attribute analysis methods are generally based on user registration information, location information, purchase history information and browsing behavior, etc., and the analysis based on the above information does not take into account that users are dynamic and their preferences and purchasing tendencies are not static, so it will lead to inaccurate data push and reduce user experience. For this reason, Figure 1 As shown, this embodiment proposes a data push method based on user attribute analysis, which can fully consider the dynamic changes of user preferences, calculate the priority of each recommended data according to the user's recent behavior, and push it, which can effectively improve user experience and transaction rate. The data push method includes the following steps:

[0067] S1. Obtain basic attribute information of historical users and target users, and generate historical user attribute information and target user attribute information; specifically, historical users represent historical users on the platform, target users represent users to be pushed data, historical user attribute information represents basic attribute information of historical users, and target user attribute information represents basic attribute information of target users. The basic attribute information of users can represent the purchasing tendency of users to a certain extent, so before analyzing user attributes, it is necessary to obtain the basic attribute information of users. Therefore, in step S1, the basic attribute information includes purchase and consumption records, browsing records, location information and registration information;

[0068] Purchase records represent the items that users have purchased. Each item is uniquely encoded to facilitate subsequent calculations.

[0069] Browsing history includes information about products that users have browsed;

[0070] Location information indicates the user's location;

[0071] Registration information includes the user's registration name, registration time, age and gender information.

[0072] S2. Establish a target model based on historical user attribute information, and classify historical users to generate user types of historical users; specifically, when analyzing user attributes, if each user can be classified, and then data is pushed based on the purchase records of other users in the corresponding classification, the accuracy of data push can be effectively improved, thereby improving user experience and marketing effects. To this end, in step S2, the following steps are specifically included:

[0073] S21. Historical users and their behaviors are used as nodes and edges of the graph network respectively. The expression of the node is:

[0074] N(u)=(a 1 , …, a 6 )

[0075] Where N(u) represents the node corresponding to the u-th user; a 1 , …, a 6 They represent the user's registered name, age, gender, location, registration time, and total consumption amount respectively. Specifically, when classifying users, users with similar characteristics or behavior patterns need to be grouped together, so that the common interests of these user groups can be found, and then content or products that better meet their needs can be provided. When classifying users, it is necessary to consider the different effects of different historical user attribute information on user classification.

[0076] S22. Calculate the second weight W of the edge based on historical user attribute information purchase (u, v), the third weight W location (u, v) and the fourth weight W browse (u, v), and construct the edge weight set W, W = (W 1 , …, W 3 ), where W 1 , …, W 3 Respectively represent the second weight W purchase (u, v), the third weight W location (u, v) and the fourth weight W browse (u, v); Specifically, when two users have purchased more than 2 items of the same product, the distance between them is no more than 500 kilometers, and they have browsed the same video or product for no less than 10 times, an edge is constructed between the two users; As a further implementation of the present invention, the common method of constructing an edge only considers the influence of each basic attribute information on user classification, but does not consider the mutual influence between each basic attribute information. Therefore, in step S22, the second weight W purchase The calculation formula for (u, v) is:

[0077] W purchase (u, v) = |{p|p∈P u∩P v}|

[0078] Where P u and P v represents the set of goods purchased in the purchase and consumption records of the u-th and v-th users respectively; p represents the goods purchased by the u-th and v-th users together;

[0079] The third weight W location The calculation formula for (u, v) is:

[0080]

[0081] In the formula, loc u and loc v Respectively represent the geographic location coordinates of the uth and vth users; d euclidean (loc u ,loc v ) represents the Euclidean distance function between the geographic location coordinates of the u-th and v-th users;

[0082] Fourth weight W browse The calculation formula for (u, v) is:

[0083]

[0084] In the formula, B u and B v Represent the sets of products and videos browsed by the u-th and v-th users respectively.

[0085] S23. Calculate the first weight W(u, v) of the edge according to the edge weight set W. Specifically, since the common method of calculating edge weights only considers the influence of each basic attribute information on the calculated edge weight, for example, if two users are located close to each other, then the two users may be influenced by the surrounding users, and the information they pay attention to is more inclined to the information paid attention to by users in the same location. And when the browsing history of users is similar, even if they do not directly purchase the same product, this similarity may indicate that there will be more common purchases in the future. The above calculation method does not take into account the mutual influence between these basic attribute information, which ultimately leads to errors in the calculated edge weights and affects the accuracy of user classification in the later stage. Therefore, in step S23, the calculation formula of the first weight W(u, v) is:

[0086] W(u,v)=(∑ i w i ×W i )+(∑ j<k w jk ×(W j ×W k ))

[0087] In the formula, w i and w jk denote the first influencing parameter and the second influencing parameter on the first weight respectively; in the present invention, w i and W jk They are 0.33 and 0.17 respectively.

[0088] In step S23 of the present invention, when the first weight W(u, v) needs to be calculated, in order to take into account the influence of the user's purchase and consumption records, location information and browsing records on the weight of the edge, the second weight W is calculated in sequence. purchase (u, v), the third weight W location (u, v) and the fourth weight W browse (u, v), and then the mutual influence between the basic attribute information is taken into account, which eventually leads to errors in the calculated edge weights. Therefore, when calculating the first weight W(u, v), the mutual influence value between different basic attribute information is introduced to improve the calculation accuracy of the first weight W(u, v); when calculating the first weight of the edge, the influence of the user's purchase consumption record, location information and browsing record on the edge weight is taken into account. Compared with other calculation methods, the mutual influence between different basic attribute information on the first weight is also taken into account. Therefore, the present invention is more accurate in calculating the first weight, which is convenient for improving the accuracy of user classification in the later stage.

[0089] S24, generating a user classification graph network for classifying historical users according to the first weight W(u, v);

[0090] S25. Train the graph neural network to obtain the target model, and input the user classification graph network into the target model to output the user type of the historical user. Specifically, a training set is constructed based on the user classification graph network. Before the graph neural network is put into use, a large amount of training is required to improve its robustness. Therefore, some nodes are selected from the user classification graph network and the edges are reconstructed according to the method of step S21 and the weight of each edge is calculated. These nodes are used as training samples, and the type of each user is manually labeled as a sample label. The above steps are repeated several times to increase the amount of training data.

[0091] In addition, the expression for updating node s in common graph neural network models is:

[0092]

[0093] In the formula, u represents the neighbor node of node v; Represents the feature vector of node v at layer l+1; Represents the feature vector of node v at layer l; represents the feature vector of node u at layer l; N(v) represents the set of all neighbor nodes u of node v; u∈N(v)∪{v} represents the set of node v and all neighbor nodes u of node v; W (l) represents the learnable weight matrix from layer l to layer l+1; σ represents the activation function;

[0094] The calculation formula for the above node update is a common node update expression in graph neural networks. However, the aggregation of node features in common graph neural network models is usually unweighted or only uses a simple adjacency matrix, without considering the impact of the first weight W(u, v) of the edge on node aggregation, resulting in low accuracy in user classification.

[0095] In step S25, the expression of the node v update of the objective function is:

[0096]

[0097] In the formula, u represents the neighbor node of node v; Represents the feature vector of node v at layer l+1; Represents the feature vector of node v at layer l; represents the feature vector of node u at layer l; N(v) represents the set of all neighbor nodes u of node v; u∈N(v)∪{v} represents the set of node v and all neighbor nodes u of node v; W (l) Represents the learnable weight matrix from layer l to layer l+1; ω uv represents the first weight W(u, v) of the edge between node v and its neighbor node u; B (l) represents the bias term; σ represents the activation function.

[0098] In step S25 of the present invention, when updating the node, the first weight W(u, v) of the edge is introduced, which can more finely control the transmission of information from one node to another, and this weighted aggregation method allows the model to more accurately reflect the actual association strength between users, thereby improving the ability to understand user behavior patterns. For the data push system, this means that it can more accurately capture changes in user interests and provide more personalized recommendations.

[0099] In step S2 of the present invention, in order to input the information of historical users into the graph network model for user classification, a user classification graph network is first constructed, and then in order to ensure the robustness of the target model, part of the information is selected from the user classification graph network to construct a training set, and then the user classification graph network is input into the target model for user classification; compared with other user classification methods, the use of a neural network model for classification can explore the implicit relationship between different users and basic attribute information to improve the accuracy of user classification, and when constructing the user classification graph network, the mutual influence between each basic attribute information of the user is further explored to further improve the accuracy of user classification.

[0100] S3. Use the target user attribute information as the input of the target model to analyze the user type of the target user; specifically, when analyzing the user type of the target user, it is only necessary to add the target user attribute information of the target user to the user classification graph network, rebuild the edges, and then input the user classification graph network into the target model to obtain the first user type of the target user.

[0101] S4, calculating the priority level P(U, i) of each corresponding recommended data based on the user type of the target user; specifically, after obtaining the user type of the target user, data of the user type can be recommended to the target user, while common recommendation methods only recommend based on the user type of the target user analyzed in the past, without considering the changes of the target user, which may result in poor user experience. Therefore, in step S4, the following steps are specifically included:

[0102] S41, obtaining a recommendation data set consisting of multiple groups of recommendation data;

[0103] S42, calculate the priority influencing parameters of each recommended data in the recommended data set, the priority influencing parameters include the time decay factor D(t), the behavior intensity weight W(b), the content relevance R(U, i) and the content freshness bonus item F(i); specifically, the time when the target user last saw the corresponding recommended data and the behavior of the target user will affect the priority of the recommended data. Therefore, in step S42, the calculation formula of the time decay factor D(t) is:

[0104] D(t)=e -λt

[0105] Wherein, t represents the time interval since the target user last saw the same type of recommended data; λ represents the decay rate parameter; in the present invention, when the target user has only browsed the same type of data, λ is 0.099; when the target user has only searched for the same type of data, λ is 0.023; when the target user has purchased the same type of data, λ is 0.089;

[0106] The expression of behavior intensity weight W(b) is:

[0107]

[0108] Among them, b represents the target user’s behavior, which are browsing, searching, and purchasing; w view 、w search and w purchase represent the behavior intensity weight values ​​corresponding to different behaviors of the target user; in the present invention, w view 、w search and W purchase They are 2, 6 and 10 respectively;

[0109] The calculation formula of content relevance R(U, i) is:

[0110] R(U,i)=V I ×100%

[0111] Where V I Indicates the number of target users browsing data of the same type as the recommended data;

[0112] The calculation formula for the content freshness bonus item F(i) is:

[0113]

[0114] In the formula, view y Indicates the current number of views of the recommended data.

[0115] S43, calculating the priority P(U, i) according to the priority influence parameter; As a further implementation of the present invention, when recommending data to target users, the influence of the priority of each data type on the data is not considered, resulting in the recommended data not being the data the customer wants, resulting in reduced user experience. Therefore, in step S43, the calculation formula of the priority P(U, i) is:

[0116]

[0117] Where m represents the total number of target users’ behaviors, namely, browsing, searching, and purchasing.

[0118] In step S4 of the present invention, before the recommended data is pushed, the recommended data with the highest priority needs to be pushed to the target user first to improve the transaction rate and user experience. The time interval since the user last viewed this type of data and the user's behavior will affect the priority of the recommended data. Therefore, the time attenuation factor D(t), the behavior intensity weight W(b), the content relevance R(U, i) and the content freshness bonus item F(i) are calculated first, and the priority P(U, i) is calculated based on them. Compared with the common data recommendation method, the dynamic changes of user preferences are taken into account, and the priority of each recommended data is calculated according to the user's recent behavior for pushing, which can effectively improve the user experience and transaction rate.

[0119] S5. Push recommended data to target users according to the priority level P(U, i); specifically, sort the recommended data in descending order according to the priority level P(U, i), and recommend them to target users in this order, so that users can obtain the data they want in the shortest time, thereby improving user experience and transaction rate.

[0120] like Figure 2 As shown, the purpose of this embodiment is to propose a data push system based on user attribute analysis, including:

[0121] An information acquisition unit, the information acquisition unit is used to acquire basic attribute information of historical users and target users;

[0122] A user classification unit, the user classification unit is used to calculate the user type of the target user;

[0123] The data push unit is used to calculate the priority of the recommended data corresponding to the user type and push the data according to the priority.

[0124] Compared with other user classification methods, the present invention adopts a neural network model for classification, which can explore the implicit relationship between different users and basic attribute information to improve the accuracy of user classification. In addition, when constructing a user classification graph network, the mutual influence between each basic attribute information of the user is further explored to further improve the accuracy of user classification.

[0125] Those skilled in the art can understand that all or part of the steps in the following embodiments can be completed by instructing the relevant hardware through a program, so the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0126] The above implementation methods have been described in detail. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A data push method based on user attribute analysis, characterized in that: The following steps are involved: S1. Obtain basic attribute information of historical users and target users, and generate historical user attribute information and target user attribute information; S2. Establish a target model based on historical user attribute information, and classify historical users to generate user types of historical users; S3, using the target user attribute information as the input of the target model to analyze the user type of the target user; S4. Calculate the priority level P(U, i) of each corresponding recommended data based on the user type of the target user; S5. Push recommended data to target users based on priority P(U, i).

2. The data push method according to claim 1, characterized in that: In step S1, basic attribute information includes purchase and consumption records, browsing records, location information and registration information.

3. The data push method according to claim 1, characterized in that: In step S2, the following steps are specifically included: S21. All users and behaviors of historical users are used as nodes and edges of the graph network respectively. The expression of the node is: N(u)=(a1,…,a6) Where N(u) represents the node corresponding to the u-th user; a1,…,a6 represent the user's registered name, age, gender, location, registration time, and total consumption amount, respectively; S22. Calculate the second weight W of the edge based on historical user attribute information purchase (u, v), the third weight W location (u, v) and the fourth weight W browse (u, v), and construct an edge weight set W, W = (W1, ..., W3), where W1, ..., W3 represent the second weight W purchase (u, v), the third weight W location (u, v) and the fourth weight W browse (u, v); S23. Calculate the first weight W(u, v) of the edge according to the edge weight set W; S24, generating a user classification graph network for classifying historical users according to the first weight W(u, v); S25. Train the graph neural network to obtain a target model, input the user classification graph network into the target model, and output the user types of historical users.

4. The data push method according to claim 3, characterized in that: In step S22, the second weight W purchase The calculation formula for (u, v) is: W purchase (u,v)=|{p|p∈P u ∩P v }| Where P u and P v Respectively represent the set of goods purchased in the purchase and consumption records of the u-th and v-th users; p represents the goods purchased by the u-th and v-th users together; The third weight W location The calculation formula for (u, v) is: In the formula, loc u and loc v Represent the geographic location coordinates of the uth and vth users respectively; d euclidean (loc u ,loc v ) represents the Euclidean distance function between the geographic location coordinates of the u-th and v-th users; Fourth weight W browse The calculation formula for (u, v) is: In the formula, B u and B v Represent the sets of products and videos browsed by the u-th and v-th users respectively.

5. The data push method according to claim 3, characterized in that: In step S23, the calculation formula of the first weight W(u, v) is: W(u,v)=(Σ i w i ×W i )+(∑ j<k w jk ×(W j ×W k )) In the formula, w i and w jk The first influencing parameter and the second influencing parameter with respect to the first weight are represented respectively.

6. The data push method according to claim 3, characterized in that: In step S25, the expression of the node v update of the objective function is: In the formula, u represents the neighbor node of node v; Represents the feature vector of node v at layer l+1; Represents the feature vector of node v at layer l; represents the feature vector of node u at layer l; N(v) represents the set of all neighbor nodes u of node v; u∈N(v)∪{v} represents the set of node v and all neighbor nodes u of node v; W (l) Represents the learnable weight matrix from layer l to layer l+1; ω uv represents the first weight W(u, v) of the edge between node v and its neighbor node u; B (l) represents the bias term; σ represents the activation function.

7. The data push method according to claim 1, characterized in that: In step S4, the following steps are specifically included: S41, obtaining a recommendation data set consisting of multiple groups of recommendation data; S42, calculating the priority influencing parameters of each recommended data in the recommended data set, the priority influencing parameters including the time decay factor D(t), the behavior intensity weight W(b), the content relevance R(U, i) and the content freshness bonus item F(i); S43. Calculate the priority P(U, i) according to the priority influencing parameter.

8. The data push method according to claim 7, characterized in that: In step S42, the calculation formula of the time attenuation factor D(t) is: D(t)=e -λt Where t represents the time interval between the last time the target user saw the same type of recommended data; λ represents the decay rate parameter; The expression of behavior intensity weight W(b) is: Among them, b represents the target user’s behavior, which are browsing, searching, and purchasing; w view , W search and w purchase They respectively represent the behavior intensity weight values ​​corresponding to different behaviors of the target user; The calculation formula of content relevance R(U, i) is: R(U,i)=V I ×100% Where V I Indicates the number of target users browsing data of the same type as the recommended data; The calculation formula for the content freshness bonus item F(i) is: In the formula, view y Indicates the current number of views of the recommended data.

9. The data push method according to claim 7, characterized in that: In step S43, the calculation formula of the priority P(U, i) is: Where m represents the total number of target users’ behaviors, namely, browsing, searching, and purchasing.

10. A data push system for implementing the data push method based on user attribute analysis as described in any one of claims 1 to 9, characterized in that: include: An information acquisition unit, the information acquisition unit is used to acquire basic attribute information of historical users and target users; A user classification unit, the user classification unit is used to calculate the user type of the target user; The data push unit is used to calculate the priority of the recommended data corresponding to the user type and push the data according to the priority.

Images