Commodity click rate prediction method for realizing cross-platform sequence fusion based on bridge connection

By collecting and processing user behavior data on e-commerce platforms, using hyperbolic spatial mapping bridge connection method to fusion cross-platform sequences, building product network structure, solving the limitations and complexity of cross-platform recommendations and improving product clicks accuracy and efficiency of rate prediction.

CN118886948APending Publication Date: 2024-11-01CHONGQING UNIV OF POSTS & TELECOMM
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410919903.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-10
Publication Date
2024-11-01

AI Technical Summary

Technical Problem

The existing product recommendation method has limitations in cross-platform user recommendations, which cannot effectively capture the user's global preferences, resulting in unsatisfactory click-through rate prediction; improper weight allocation of personalized recommendations leads to low accuracy; the user matching process is very complex, and the calculation is The large quantity and high real-time requirements make it difficult to meet the needs of e-commerce platforms.

Method used

By collecting user behavior data, generating time series, cutting into time unit sequences, and cross-platform fusion through bridge connection method of hyperbolic space mapping, building a product network structure, and weighting and summing different operations according to the quantity of goods and historical operation time of the product , generate click-through rate prediction value for user personalized recommendations.

Benefits of technology

It solves the problem of hierarchical feature data loss after cross-platform data fusion, reduces the user and product feature dimensions, improves the accuracy and efficiency of recommendations, and generates recommendation sequences that are closer to the real situation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118886948A_ABST
    Figure CN118886948A_ABST
Patent Text Reader

Abstract

The invention belongs to a personalized recommendation technology, and particularly relates to a commodity click rate prediction method for realizing cross-platform sequence fusion based on bridge connection, which comprises the following steps: collecting user behavior data under each platform, preprocessing the user behavior data, and generating a time sequence containing user operation information; cutting the time sequence under each platform into a plurality of time unit sequences, and performing cross-platform fusion on the cut time unit sequences through a hyperbolic space mapping bridge connection mode to obtain a commodity network structure; and performing weighted summation on different operations in the sequence according to the number of commodities in the commodity network structure and historical record operation time, and generating a click rate prediction value of personalized recommendation of potential interested commodities by the user. According to the method, different weights are given to different operations, so that the recommendation result is closer to the real situation, the recommendation sequence with higher accuracy is generated, and meanwhile, a more accurate commodity recommendation sequence can be realized through cross-platform user behavior sequence data fusion.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of personalized recommendation, and in particular relates to a method for predicting the click-through rate of commodities by realizing cross-platform sequence fusion based on bridge connection. Background Art

[0002] With the rapid development of Internet technology and the widespread application of mobile smart devices, various online social media, e-commerce platforms, video entertainment and other network service platforms have emerged rapidly, greatly enriching people's online life, making network services more popular, extensive and diversified. Among the many network services, the e-commerce field is particularly prominent. It not only provides people with a convenient shopping experience, but also helps users quickly find products that meet their interests and needs through personalized recommendation technology. Personalized recommendation not only provides convenience for users to find their favorite products, but also brings more business opportunities and value to e-commerce platforms. Mining user behavior data based on long-term and short-term interests of users to achieve personalized recommendations has received widespread attention from scholars at home and abroad.

[0003] So far, domestic and foreign researchers have conducted a lot of research on product recommendation. Existing product recommendation methods can be generally divided into two categories: recommendation methods based on user attributes and historical behavior data and recommendation methods based on machine learning and deep learning. Methods based on user attributes and historical behavior data use the attribute information disclosed by users on social platforms and the topological structure of social networks to measure the similarity between users; methods based on machine learning and deep learning extract user features through feature engineering, train classification models based on labeled data, and finally use the trained models to predict personalized recommendations.

[0004] In summary, many researchers have made many achievements in personalization research, laying a solid foundation for further research. There are still some challenges in the process of cross-platform user recommendation based on the marketing data of actual e-commerce platforms:

[0005] Problem 1: Due to differences in user preferences for platforms, the platform has limitations in user recommendations, and the user's global preferences are ignored, or even trapped in an information cocoon, resulting in unsatisfactory click-through rate prediction. Therefore, the key to more realistically simulating user behavior is to understand the user's interest preferences. Only by letting the model understand user behavior can we better predict the user's next behavior;

[0006] Problem 2: Weight allocation problem for personalized recommendations. Different operations contribute differently to the final prediction value. For example, users have both long-term static interest preferences and short-term dynamic interest development. If the recommendation decision is made based only on the user's long-term preference or short-term preference, the accuracy of personalized recommendations will inevitably be low.

[0007] Problem 3: High complexity of the user matching process. E-commerce platforms have a large number of products. The amount of computation required to identify and match the massive user-product relationships between large e-commerce platforms is huge, and the real-time requirements are relatively high. In addition, the personalized needs of users are diverse and constantly changing, which increases the complexity and difficulty of the user matching process, resulting in an explosive increase in time consumption. Summary of the invention

[0008] In order to solve the above technical problems, the present invention provides a method for predicting the click-through rate of commodities based on bridge connection to realize cross-platform sequence fusion, comprising:

[0009] S1: Collect user behavior data on each platform, pre-process the user behavior data, and generate a time series containing user operation information;

[0010] S2: Cut the time series under each platform into multiple time unit sequences, and integrate the cut time unit sequences across platforms through the bridge connection method of hyperbolic space mapping to obtain the commodity network structure;

[0011] S3: Perform weighted summation of different operations in the sequence based on the number of products in the product network structure and the historical record operation time to generate a click-through rate prediction value for personalized recommendations of potential products of interest to the user.

[0012] Beneficial effects of the present invention:

[0013] The present invention solves the problem of unsatisfactory recommendation effect caused by loss of hierarchical feature data after cross-platform and cross-domain user behavior sequence fusion, reduces the feature dimensions of users and products, and reduces the complexity of data processing;

[0014] The present invention assigns different weights to different operations to make the recommendation results closer to the actual situation and generate a recommendation sequence with higher accuracy. At the same time, through the fusion of cross-platform user behavior sequence data, a more accurate product recommendation sequence can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The present invention is a flow chart of a method for predicting click-through rate of a product based on bridge connection to achieve cross-platform sequence fusion. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] A method for predicting product click-through rate based on bridge connection to achieve cross-platform sequence fusion, such as Figure 1 As shown, including:

[0018] S1: Collect user behavior data on each platform, pre-process the user behavior data, and generate a time series containing user operation information;

[0019] S2: Cut the time series under each platform into multiple time unit sequences, and integrate the cut time unit sequences across platforms through the bridge connection method of hyperbolic space mapping to obtain the commodity network structure;

[0020] S3: Perform weighted summation of different operations in the sequence based on the number of products in the product network structure and the historical record operation time to generate a click-through rate prediction value for personalized recommendations of potential products of interest to the user.

[0021] In this embodiment, user behavior data on the platform is collected, and the user behavior data is preprocessed to generate a time series containing user operation information, including:

[0022] S11: Using the user account information of the e-commerce platform, collect the user's historical behavior information, including browsing, clicking, collecting and purchasing operation information of goods, and count the operation frequency of similar goods. Each operation corresponds to a field, and the field value k is used to mark whether there is a corresponding operation;

[0023] S12: Generate a user behavior attribute statistical matrix U(m×n) based on the statistical historical behavior information, where m represents the number of products that the user is interested in, and n represents the number of product attributes in the sequence;

[0024] Where m can be expanded from the user's relationship network without enumerating all users. The user attribute matrix is ​​shown in Table 1;

[0025] Table 1 User behavior attribute statistical matrix U

[0026] commodity … Browse Click collect Buy <![CDATA[C1]]> … 1 1 1 1 <![CDATA[C2]]> … 2 0 1 0 … … 1 0 0 0 <![CDATA[C m ]]> … 3 1 0 1

[0027] Among them, C i Indicates the user's potential interest in the product. 0 means no operation, and a non-zero value indicates the number of operations.

[0028] S13: Repeat steps S11 to S12 to set the e-commerce platform P i The user behavior information under the i ,The sequence contains the relationship representation among user historical behavior information, user attribute information and product information.

[0029] In this embodiment, the time series under each platform is cut into multiple time unit sequences, and the cut time unit sequences are fused across platforms through a bridge connection method of hyperbolic space mapping, including:

[0030] By analyzing the user's historical operation behavior in the time series of each platform, the user's operations are divided according to the operation type and operation time. The time units are based on the user's recent operations. The time units are not necessarily consistent. If a user performs multiple closely related operations within an hour, the operations within this hour are divided into one time unit. On the contrary, if the user's operation behaviors in different time periods are separated by a large interval, these operations are divided into different time units.

[0031] The user operations in each divided time period have certain correlations, including: the operation time is close, the operation products are similar, and a product network is constructed by connecting related products. In the product network, nodes represent products and edges represent the correlation between products.

[0032] In the divided time units, the operation time and the length of each divided time unit are not necessarily consistent under the user behavior sequence of different platforms. Whether to add the commodity network constructed during the last fusion to each fusion is determined by the operation time interval. In each fusion, whether to add the commodity network constructed during the last fusion to the current fusion process is determined according to the operation time interval. If the operation time interval between the current time unit and the operation time of the last fusion is short, and there is a strong correlation between the commodities involved, the commodity network constructed during the last fusion is added to the current fusion process to make full use of historical information. On the contrary, if the time interval is long or the commodity correlation is weak, only the information in the current time unit is selected for fusion.

[0033] In a certain period of time, if the historical operation products of users on each platform belong to the same domain, a bidirectional edge is added for each product during the fusion. If they belong to different domains, the products are independent after being integrated into the new behavior sequence. If the two fused time units belong to the same domain in terms of the operation products, it means that the two operations are strongly related. In this way, a bidirectional edge is added to each product while being integrated into the new sequence to build a network structure.

[0034] In this embodiment, a commodity network is constructed by connecting related commodities, including:

[0035] Within a certain period of time, if the products of the historical user operations on each platform belong to the same domain, it means that the two operations are strongly related. In this case, a bidirectional edge is added to each product during the fusion process. If they belong to different domains, the products will be independent after being integrated into the new behavior sequence.

[0036] Determine whether the two operated products belong to the same domain under different e-commerce platform user sequences. If the two merged time units belong to the same domain in the operated products, it means that the two operations are strongly correlated. In this case, a bidirectional edge is added to each product while the two operations are merged into the new sequence. If the two merged time units belong to different domains in the operated products, it means that the two operations are weakly correlated. In this case, the two operations are merged into the new sequence, but no edge is added for the products.

[0037] In this embodiment, determining whether two operations on commodities are in the same domain under different e-commerce platform user sequences includes:

[0038]

[0039] in, Represents products on two different e-commerce platforms Whether they are in the same product domain; X and Y represent two different e-commerce platforms; W ij Represents the association and related attribute matrix between node i and node j.

[0040] In this embodiment, if the two merged time units belong to the same domain in terms of the operating commodity, a bidirectional edge is added to each commodity while being merged into the new sequence, including:

[0041]

[0042] Among them, F(s i ,s2,…,s m ) represents a fusion function of a commodity network structure; s m represents the mth node in the network; C ij represents the correlation between the i-th node and the j-th node; m represents the number of products that the user is interested in; n is the number of product attributes in the sequence; Represents two products Are they the same domain? If they are different domains, C ij =0; Indicates two operations y on the operation time i , The strength of the association between them.

[0043] In this embodiment, a weighted sum of different operations in the sequence is performed according to the number of products in the product network structure and the operation time of the historical records to generate a click rate prediction value of the user's personalized recommendation of potential interest products, including:

[0044]

[0045] Among them, W irepresents the click-through rate prediction value of the user's personalized recommendation of potential interest products, α, β, γ, and δ represent the weighting factors of product quantity, operation time, product network structure connection strength, and historical operation behavior, respectively; N i represents the number of commodities in the ith network block. If it is independent, then N i =1; t and Ti represent the current time and the last operation time of the i-th item respectively; C ij represents the association between the i-th node and the j-th node; H i Represents the historical interaction intensity of the ith network block or product.

[0046] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting product click-through rate based on bridge connection to achieve cross-platform sequence fusion, characterized in that: include: S1: Collect user behavior data on each platform, pre-process the user behavior data, and generate a time series containing user operation information; S2: Cut the time series under each platform into multiple time unit sequences, and integrate the cut time unit sequences across platforms through the bridge connection method of hyperbolic space mapping to obtain the commodity network structure; S3: Perform weighted summation of different operations in the sequence based on the number of products in the product network structure and the historical record operation time to generate a click-through rate prediction value for personalized recommendations of potential products of interest to the user.

2. According to claim 1, a method for predicting product click-through rate based on bridge connection to achieve cross-platform sequence fusion, characterized in that: Collect user behavior data on the platform, pre-process the user behavior data, and generate a time series containing user operation information, including: S11: Using the user account information of the e-commerce platform, collect the user's historical behavior information, including browsing, clicking, collecting and purchasing operation information of goods, and count the operation frequency of similar goods. Each operation corresponds to a field, and the field value k is used to mark whether there is a corresponding operation; S12: Generate a user behavior attribute statistical matrix U(m×n) based on the statistical historical behavior information, where m represents the number of products that the user is interested in, and n represents the number of product attributes in the sequence; S13: Repeat steps S11 to S12 to set the e-commerce platform P i The user behavior information under the i ,The sequence contains the relationship representation among user historical behavior information, user attribute information and product information.

3. According to claim 1, a method for predicting product click-through rate based on bridge connection to achieve cross-platform sequence fusion, characterized in that: The time series of each platform is cut into multiple time unit sequences, and the cut time unit sequences are fused across platforms through the bridge connection method of hyperbolic space mapping, including: By analyzing the user's historical operation behavior in the time series of each platform, the user's operations are divided according to the operation type and operation time. The time units are based on the user's recent operations. The time units are not necessarily consistent. If a user performs multiple closely related operations within an hour, the operations within this hour are divided into one time unit. On the contrary, if the user's operation behaviors in different time periods are separated by a large interval, these operations are divided into different time units. The user operations in each divided time period have certain correlations, including: the operation time is close, the operation products are similar, and a product network is constructed by connecting related products. In the product network, nodes represent products and edges represent the correlation between products. In the divided time units, the operation time and the length of each divided time unit are not necessarily consistent under the user behavior sequence of different platforms. Whether to add the commodity network constructed during the last fusion to each fusion is determined by the operation time interval. In each fusion, whether to add the commodity network constructed during the last fusion to the current fusion process is determined according to the operation time interval. If the operation time interval between the current time unit and the operation time of the last fusion is short, and there is a strong correlation between the commodities involved, the commodity network constructed during the last fusion is added to the current fusion process to make full use of historical information. On the contrary, if the time interval is long or the commodity correlation is weak, only the information in the current time unit is selected for fusion. Within a certain period of time, if the historical user operation products on each platform belong to the same domain, a bidirectional edge is added for each product during the fusion. If they belong to different domains, the products will be independent after being integrated into the new behavior sequence. If the two fused time units belong to the same domain in terms of the operation products, it means that the two operations are strongly correlated. Then, a bidirectional edge is added to each product while being integrated into the new sequence to build a network structure.

4. According to claim 3, a method for predicting click-through rate of commodities based on bridge connection to realize cross-platform sequence fusion is characterized in that: Build a product network by connecting related products, including: Within a certain period of time, if the products of the historical user operations on each platform belong to the same domain, it means that the two operations are strongly related. In this case, a bidirectional edge is added to each product during the fusion process. If they belong to different domains, the products will be independent after being integrated into the new behavior sequence. Determine whether the two operated products belong to the same domain under different e-commerce platform user sequences. If the two merged time units belong to the same domain in the operated products, it means that the two operations are strongly correlated. In this case, a bidirectional edge is added to each product while the two operations are merged into the new sequence. If the two merged time units belong to different domains in the operated products, it means that the two operations are weakly correlated. In this case, the two operations are merged into the new sequence, but no edge is added for the products.

5. According to claim 4, a method for predicting product click-through rate based on bridge connection to achieve cross-platform sequence fusion, characterized in that: Determine whether two operations on products are from the same domain under different e-commerce platform user sequences, including: in, Represents products on two different e-commerce platforms Whether they are in the same product domain; X and Y represent two different e-commerce platforms; W ij Represents the association and related attribute matrix between node i and node j.

6. The method for predicting product click-through rate based on bridge connection to realize cross-platform sequence fusion according to claim 4, characterized in that: If the two time units to be merged belong to the same domain in terms of the operation commodity, a bidirectional edge is added to each commodity while merging into the new sequence, including: Among them, F(s i ,s2,…,s m ) represents a fusion function of a commodity network structure; s m represents the mth node in the network; C ij represents the correlation between the i-th node and the j-th node; m represents the number of products that the user is interested in; n is the number of product attributes in the sequence; Represents two products Are they the same domain? If they are different domains, C ij =0; Indicates two operations y on the operation time i , The strength of the association between them.

7. The method for predicting product click-through rate based on bridge connection to realize cross-platform sequence fusion according to claim 1, characterized in that: According to the number of products in the product network structure and the operation time of the historical records, the weighted sum of different operations in the sequence is performed to generate the click rate prediction value of the user's personalized recommendation of potential interest products, including: Among them, W i represents the click-through rate prediction value of the user's personalized recommendation of potential interest products, α, β, γ, and δ represent the weighting factors of product quantity, operation time, product network structure connection strength, and historical operation behavior, respectively; N i represents the number of commodities in the ith network block. If it is independent, then N i =1; t and Ti represent the current time and the last operation time of the i-th item respectively; C ij Represents the association between the i-th node and the j-th node; H i Represents the historical interaction intensity of the ith network block or product.

Citation Information

Cited By

  • Cross-platform e-commerce user behavior joint prediction and recommendation method and system based on artificial intelligence

    CN122288841A