Method and device for extracting user interest patterns

By processing the item category access sequence, frequency and residence time of click stream data, calculating user similarity and clustering, the problem of inaccurate user interest mining in the prior art is solved, and the rapid extraction of interest modes in e-commerce is achieved and Sales increase.

CN110827044BActive Publication Date: 2024-11-26BEIJING JINGDONG SHANGKE INFORMATION TECH CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201810891298.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2018-08-07
Publication Date
2024-11-26
Estimated Expiration
2038-08-07

AI Technical Summary

Technical Problem

When measuring user interests in the prior art, item ratings combine price, service and logistics factors, resulting in insufficient new customer ratings and inaccurate user interest mining, and the definition based on purchase behavior ignores important factors such as page residence time and access order.

Method used

By processing the user's click stream data into item category access sequence, access frequency and residence time, the similarity between users is calculated and clustered, the user interest modes of each category are extracted, and these modes are used for scenario-based marketing and item recommendation .

Benefits of technology

It has achieved rapid and effective extraction of interest mode for e-commerce users, and improved sales indicators and user consumption promotion effects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN110827044B_ABST
    Figure CN110827044B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for extracting user interest patterns, and relates to the field of computer technology. A specific implementation of the method includes: based on the classification information of items, processing the user's click stream data into the user's item category access sequence, item category access frequency and item category residence time; calculating the similarity between users according to each user's item category access sequence, item category access frequency and item category residence time; clustering users according to the similarity between users; and extracting user interest patterns for each category of users. This implementation reduces the inaccuracy of mining user interests, and thus quickly and effectively extracts multiple interest patterns of e-commerce users based on click stream data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for extracting user interest patterns. Background Art

[0002] Understanding user interests has become increasingly important for online retailers who want to provide personalized services to their target market. In order to attract more customers, e-commerce companies are constantly diversifying their products and adding corresponding categories. Large e-commerce websites usually have more than one million users logging in every day, and these potential customers visit thousands of items and categories every day. Therefore, these e-commerce companies face a special challenge of how to discover the interests of website users and promote sales by effectively managing an increasing number of categories and items.

[0003] Existing techniques for measuring consumer interest can be categorized into the following:

[0004] 1. Rely on customer ratings, that is, the user's rating of the item indicates whether he is interested in it to some extent, and the rating value represents the degree to which the user likes the target item;

[0005] 2. To measure user interest, several characteristics of user behavior are examined, such as the number of times a user visits each item's web page, item ratings, and purchase history, to simply define user interest.

[0006] In the process of implementing the present invention, the inventors found that there are at least the following problems in the prior art: item rating is the final comprehensive rating of the user's evaluation of price, service and logistics, not just the rating of the item itself. Ratings from new customers are not sufficient for reference, and experienced customers may not be willing to give ratings every time they use the website. These factors make it more difficult to discover the user's true interests based solely on ratings. In addition, simply defining user interests based on user purchasing behavior ignores other important factors, such as the time spent on each page and the order of visits, which leads to inaccurate mining of user interests. Summary of the invention

[0007] In view of this, an embodiment of the present invention provides a method and device for extracting user interest patterns, which can quickly and effectively extract multiple interest patterns of e-commerce users based on click stream data. The extracted interest patterns can be used for scenario-based marketing or item recommendations, thereby promoting user consumption and improving sales indicators.

[0008] According to one aspect of an embodiment of the present invention, a method for extracting user interest patterns is provided.

[0009] A method for extracting user interest patterns comprises: processing user click stream data into the user's item category access sequence, item category access frequency and item category residence time based on item classification information; calculating the similarity between users according to each user's item category access sequence, item category access frequency and item category residence time; clustering users according to the similarity between users; and extracting user interest patterns for each category of users respectively.

[0010] Optionally, processing the user's click stream data into the user's item category access sequence includes: generating an access path of the user during a session, wherein the user's access path is mappable to an item classification topology; and generating the user's item category access sequence based on the user's access path.

[0011] Optionally, processing the user's click stream data into the user's item category visit frequency includes: counting the number of clicks by the user on a certain item category, wherein the number of clicks includes the number of visits to the item category and the number of visits to items belonging to the item category; and obtaining the user's visit frequency to the item category based on the number of clicks, wherein the visit frequency is defined as the ratio of the number of clicks to the length of the user's item category visit sequence.

[0012] Optionally, processing the user's click stream data into the user's item category dwell time includes: obtaining the user's item category dwell time according to the user's dwell time on the item category and the user's dwell time on all items belonging to the item category.

[0013] Optionally, calculating the similarity between users based on each user's item category access sequence, item category access frequency, and item category residence time includes: generating a fixed-dimensional vector representation for each user based on each user's item category access sequence, item category access frequency, and item category residence time; and calculating the similarity between users based on the fixed-dimensional vector representation corresponding to each user.

[0014] Optionally, generating a fixed-dimensional vector representation for each user based on each user's item category access sequence, item category access frequency and item category residence time includes: generating a user-item category association vector based on item category access sequence characteristics based on the user's item category access sequence; generating a user-item category association matrix based on item category access frequency characteristics based on the user's item category access frequency, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix; generating a user-item category association matrix based on item category residence time characteristics based on the user's item category residence time, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix.

[0015] Optionally, calculating the similarity between users includes: calculating the access sequence similarity of users p and q, wherein the access sequence similarity is defined as the longest common subsequence length in the item category access sequence of users p and q divided by the maximum sequence length, in which the two users visit the same item category in the common subsequence; calculating the access frequency similarity of users p and q, wherein the access frequency similarity is defined as the similarity between two row vectors representing users p and q respectively from a user-item category association matrix based on item category access frequency characteristics; calculating the residence time similarity of users p and q, wherein the residence time similarity is defined as the similarity between two row vectors representing users p and q respectively from a user-item category association matrix based on item category residence time characteristics; based on the above three similarities, calculating the overall similarity between users p and q, wherein the above three similarities are weighted respectively with three weights whose sum is 1, and the weighted three similarities are summed to obtain the overall similarity between users p and q.

[0016] Optionally, the visit frequency similarity and the residence time similarity are both cosine similarities.

[0017] Optionally, users are clustered using a streaming clustering method with a specified similarity threshold.

[0018] Optionally, the streaming clustering method includes the following steps: 1) initialize the current category set G to empty, and record the similarity threshold as Th; 2) scan each element in the user data set in turn, wherein the user data set is composed of the overall similarity between users; 3) obtain the element; 4) if all the elements in the data set have been scanned, the algorithm ends; otherwise, execute 5); 5) calculate the similarity between the current element and each category cluster center in the current category set G, record the largest similarity as d, and the corresponding category as g; 6) if d>Th and G is not empty, execute 7); otherwise, execute 8); 7) classify the current element into category g, and update the cluster center of category g, and return to 3); 8) create a new category h, classify the current element into category h, and update the cluster center of category h, and classify the category into G; return to 3).

[0019] Optionally, extracting the user interest pattern for each category of users includes: calculating the occurrence frequency of the item categories visited by the users of each category, and taking a set of item categories whose occurrence frequency meets a preset threshold as the user interest pattern of the users of the category.

[0020] According to another aspect of an embodiment of the present invention, a device for extracting user interest patterns is provided.

[0021] A device for extracting user interest patterns, comprising: a processing module, used to process user click stream data into the user's item category access sequence, item category access frequency and item category residence time based on item classification information; a similarity calculation module, used to calculate the similarity between users according to each user's item category access sequence, item category access frequency and item category residence time; a clustering module, used to cluster users according to the similarity between the users; and an extraction module, used to extract user interest patterns for each category of users respectively.

[0022] Optionally, the processing module is further used to: generate an access path of a user during a session, wherein the access path of the user can be mapped to an item classification topology; and generate an item category access sequence of the user according to the access path of the user.

[0023] Optionally, the processing module is further used to: count the number of clicks by users on a certain item category, wherein the number of clicks includes the number of visits to the item category and the number of visits to items belonging to the item category; and obtain the user's visit frequency to the item category based on the number of clicks, wherein the visit frequency is defined as the ratio of the number of clicks to the length of the user's item category visit sequence.

[0024] Optionally, the processing module is further used to obtain the user's item category dwell time according to the user's dwell time on the item category and the user's dwell time on all items belonging to the item category.

[0025] Optionally, the similarity calculation module is further used to: generate a fixed-dimensional vector representation for each user according to each user's item category access sequence, item category access frequency and item category residence time; and calculate the similarity between users according to the fixed-dimensional vector representation corresponding to each user.

[0026] Optionally, the similarity calculation module is also used to: generate a user-item category association vector based on item category access sequence characteristics according to the user's item category access sequence; generate a user-item category association matrix based on item category access frequency characteristics according to the user's item category access frequency, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix; generate a user-item category association matrix based on item category residence time characteristics according to the user's item category residence time, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix.

[0027] Optionally, the similarity calculation module is also used to: calculate the access sequence similarity of users p and q, wherein the access sequence similarity is defined as the longest common subsequence length in the item category access sequence of users p and q divided by the maximum sequence length, in which the two users visit the same item category in the common subsequence; calculate the access frequency similarity of users p and q, wherein the access frequency similarity is defined as the similarity between two row vectors representing users p and q respectively from a user-item category association matrix based on item category access frequency characteristics; calculate the residence time similarity of users p and q, wherein the residence time similarity is defined as the similarity between two row vectors representing users p and q respectively from a user-item category association matrix based on item category residence time characteristics; based on the above three similarities, calculate the overall similarity between users p and q, wherein the above three similarities are weighted respectively with three weights whose sum is 1, and the weighted three similarities are summed to obtain the overall similarity between users p and q.

[0028] Optionally, the visit frequency similarity and the residence time similarity are both cosine similarities.

[0029] Optionally, users are clustered using a streaming clustering method with a specified similarity threshold.

[0030] Optionally, the streaming clustering method includes the following steps: 1) initialize the current category set G to empty, and record the similarity threshold as Th; 2) scan each element in the user data set in turn, wherein the user data set is composed of the overall similarity between users; 3) obtain the element; 4) if all the elements in the data set have been scanned, the algorithm ends; otherwise, execute 5); 5) calculate the similarity between the current element and each category cluster center in the current category set G, record the largest similarity as d, and the corresponding category as g; 6) if d>Th and G is not empty, execute 7); otherwise, execute 8); 7) classify the current element into category g, and update the cluster center of category g, and return to 3); 8) create a new category h, classify the current element into category h, and update the cluster center of category h, and classify the category into G; return to 3).

[0031] Optionally, the extraction module is further used to: calculate the occurrence frequency of the item categories visited by users of each category, and take the set of item categories whose occurrence frequency meets a preset threshold as the user interest pattern of users of the category.

[0032] According to still another aspect of the embodiments of the present invention, an electronic device is provided.

[0033] An electronic device for extracting user interest patterns, comprising: one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method for extracting user interest patterns provided in an embodiment of the present invention.

[0034] According to still another aspect of an embodiment of the present invention, a computer readable medium is provided.

[0035] A computer-readable medium stores a computer program, which, when executed by a processor, implements a method for extracting user interest patterns provided by an embodiment of the present invention.

[0036] One embodiment of the above invention has the following advantages or beneficial effects: because the following technical means are adopted, namely, processing the click stream data into the user's access sequence, access frequency and dwell time on different item categories and calculating the similarity between users and clustering the users based on this data, the technical problem of inaccurate mining of user interests is overcome, and multiple interest patterns of e-commerce users can be quickly and effectively extracted based on click stream data. The extracted interest patterns can be used for scenario-based marketing or item recommendations, thereby promoting user consumption and improving sales indicators.

[0037] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0039] Figure 1 is a schematic diagram of the main process of the method for extracting user interest patterns according to an embodiment of the present invention;

[0040] Figure 2 is a schematic diagram of an item classification topology structure according to an embodiment of the present invention;

[0041] Figure 3 It is a schematic diagram of the main process of the streaming clustering method of an embodiment of the present invention;

[0042] Figure 4 is a schematic diagram of main modules of an apparatus for extracting user interest patterns according to an embodiment of the present invention;

[0043] Figure 5 is an exemplary system architecture diagram to which embodiments of the present invention may be applied;

[0044] Figure 6 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0046] Generally speaking, users' browsing behavior on e-commerce websites or mobile apps is a comprehensive reflection of their interests. Users with different interests will visit multiple different categories of items and the corresponding various items. For different types of users, their browsing order and path, page access frequency, and time spent in each category are very different. Based on these considerations, the present invention proposes a new method for discovering consumer interest patterns by mining clickstream data recorded on e-commerce websites, and conducts a systematic study. Clickstream data is an electronic record of user activities on the Internet. Compared with the method of users rating items, this method can more accurately describe users' interests using more detailed information.

[0047] The basic idea of ​​the present invention is to mine user interest patterns based on the clickstream data of e-commerce users. First, based on the classification information of items, the user's clickstream data is processed into the user's access sequence, access frequency and stay time on different item categories, and a fixed-dimensional vector representation is generated for each user based on this data; then the similarity between users can be calculated, and the streaming clustering method with a specified similarity threshold is used to cluster the users, and it is considered that there are possible interest patterns in the categories containing a large number of users; finally, the frequency of occurrence of item categories in each larger user category is calculated, and the set of item categories with high frequency of occurrence is regarded as a user interest pattern. This method is used to extract user interest patterns.

[0048] First, regarding the measurement of user interest, the following assumptions should be met:

[0049] Each user has a preference when visiting a website. Users usually only visit products that they are interested in. Users with similar interests have similar visit behaviors.

[0050] The frequency of visiting an item page is positively correlated with the user’s interest in the item;

[0051] The duration that a user stays on an item's webpage is positively correlated with the user's interest in the item;

[0052] The order in which items are visited is positively correlated with users’ interests, and users will select the products that they are most interested in first.

[0053] Based on the above assumptions, user interest can be defined as a set of item categories that the user has visited. In order to unify the measurement unit and simplify the calculation, the present invention only evaluates the user's interest in item categories. The visit order, visit frequency and dwell time on the item page are used as the three main indicators to measure user interest.

[0054] Figure 1 FIG. 1 is a schematic diagram of the main process of the method for extracting user interest patterns according to an embodiment of the present invention. Figure 1 As shown, the method for extracting user interest patterns according to an embodiment of the present invention mainly includes steps S101 to S104.

[0055] Step S101: Based on the classification information of the items, the user's click stream data is processed into the user's item category access sequence, item category access frequency and item category dwell time.

[0056] According to one embodiment of the present invention, processing the user's click stream data into the user's item category access sequence includes:

[0057] Generate an access path of a user during a session, wherein the access path of the user can be mapped to an item classification topology;

[0058] Generate the user's item category access sequence based on the user's access path.

[0059] The access path Pi {url1, url2, url3, ..., urln} is a sequence of web pages visited by user i during a session, where i = 1, 2, ..., n, and n is a natural number. The user's access path can be mapped to an item classification topology, which is exemplified in FIG. Figure 2 . Figure 2 1 is a schematic diagram of the item classification topology structure of an embodiment of the present invention. For example, the access path P1{ctg1; item11; ctg1; item12; item21} indicates that user1 first accesses category ctg1, then accesses item11 belonging to ctg1, then returns to ctg1 and accesses item12, and finally accesses item21 belonging to category ctg2.

[0060] Category access sequence: The category access sequence CtgPi{ctg1, ctg2, ctg3, ..., ctgm} is a series of categories accessed by useri during a given session. The length of CtgPi is m, where m<=n, and m and n are natural numbers. For example, based on the user's access path P1{ctg1; item11; ctg1; item12; item21}, the category access path of the user can be generated as CtgP1{ctg1, ctg2}.

[0061] According to one embodiment of the present invention, processing the user's click stream data into the user's item category access frequency includes:

[0062] Count the number of clicks by users on a certain item category, where the number of clicks includes the number of visits to the item category and the number of visits to items belonging to the item category;

[0063] The user's access frequency to the item category is obtained according to the number of clicks, wherein the access frequency is defined as the ratio of the number of clicks to the length of the user's item category access sequence.

[0064] For a website, Indicates the number of times user i clicks on item category j in a session. As shown in the following formula, the number of clicks by user i on an item category consists of two parts: the number of visits to ctgj and the number of visits to items belonging to ctgj:

[0065]

[0066] Where i, j, k, and l are natural numbers. count(user, ctg) represents the number of times the user user visits the item category ctg during the session. belongs to ctgj, and l<=len(Pi), that is, l is less than or equal to the length of Pi. For example, according to user1's access path Can get and

[0067] The access frequency is defined as the ratio of the number of times an item category is accessed to the length of the user's access sequence. To represent the frequency of user i visiting item category ctgj during the session:

[0068]

[0069] For example, according to the access path of user1, It can be deduced that the access frequencies of these two categories are and

[0070] According to another embodiment of the present invention, processing the user's clickstream data into the user's item category dwell time includes: obtaining the user's item category dwell time according to the user's dwell time on the item category and the user's dwell time on all items belonging to the item category. The user may repeatedly visit a page of a certain item category and pages of items belonging to the item category. Represents the total time that the user visits the item category ctgj in the session. The time of visiting each item item will be accumulated to its parent node (the corresponding category). A user can visit a category or item page repeatedly. Therefore, the time of each visit to the same category will be added together (see the formula below).

[0071]

[0072] Among them, time(user, ctg, s) represents the time that the user user stays on the item category ctg in the session. Belongs to ctgj and l<=len(Pi), and time(user, item, t) represents the time that the user user stays on a certain item in the session.

[0073] Step S102: Calculate the similarity between users based on the item category access sequence, item category access frequency and item category residence time of each user.

[0074] According to an embodiment of the present invention, step S102 may specifically include:

[0075] Step S1021: Generate a fixed-dimensional vector representation for each user according to the item category access sequence, item category access frequency and item category residence time of each user;

[0076] Step S1022: Calculate the similarity between users based on the fixed-dimensional vector representation corresponding to each user.

[0077] Wherein, step S1021 may specifically include:

[0078] Generate a user-item category association vector based on item category access sequence features according to the user's item category access sequence;

[0079] Generate a user-item category association matrix based on item category access frequency features according to the item category access frequency of the user, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix;

[0080] A user-item category association matrix based on the item category residence time feature is generated according to the user's item category residence time, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix.

[0081] According to the assumptions mentioned above, users with similar interests show similar access behaviors. Given the indicators of user access behaviors, namely category access paths, access frequency, and dwell time, the similarity of user behaviors can be quantitatively evaluated. Users can be further clustered based on the similarity between users.

[0082] In the embodiment of the present invention, the generated vector representation is high-dimensional, with three different indicators: item category access sequence, item category access frequency, and item category residence time. When calculating the similarity between users, first, a sequence similarity is defined to evaluate the similarity of the access sequences of two users; then, the cosine function is used to evaluate the similarity between the two users from the perspective of access frequency and residence time; finally, these three similarities are integrated into the overall similarity.

[0083] Access sequence similarity: The access sequence similarity sim between two users p and q pq (path) is defined as the length of the longest common subsequence divided by the length of the largest sequence in p and q. In this common sequence, the two users visit the same categories in the same order. If there are multiple common subsequences between two users, the longest common subsequence is used for the calculation of sequence similarity. The calculation formula for access sequence similarity is as follows:

[0084]

[0085] Among them, 0≤sim pq(path)≤1.

[0086] Access frequency similarity: The access frequency similarity between two users p and q is defined as the cosine similarity of two vectors, which are obtained from the matrix The row vector representing users p and q extracted from . The calculation formula for access frequency similarity is as follows:

[0087] sin pq (freq)=COS(R freq [p,·],R freq [q,·]);

[0088] Among them, 0≤sim pq (freq)≤1.

[0089] Dwell time similarity: Similarly, the dwell time similarity between two users p and q is defined as follows:

[0090] sim pq (time) = cos(R time [p,·],R time [q,·]);

[0091] Among them, 0≤sim pq (time)≤1.

[0092] Based on the above three similarities, the overall similarity between two users p and q is defined as follows:

[0093] sim pq =α×sim pq (seq)+β×sim pq (freq)+γ×sim pq (time);

[0094] Among them, α, β, γ are used to adjust the weights of the three similarity indicators of access sequence, access frequency and residence time. In addition, α+β+γ=1, 0≤sim pq ≤1.

[0095] Step S103: clustering users according to the similarities between users.

[0096] According to an embodiment of the present invention, a streaming clustering method with a specified similarity threshold is used to cluster users.

[0097] For a data set, the distance between data in the same category, that is, the intra-class distance, is small (the similarity is large); the distance between data in different categories, that is, the inter-class distance, is large (the similarity is small). Therefore, consider setting a distance threshold Th. If the distance between two data points is less than the threshold, the two data points are classified into one category. The formula is as follows:

[0098]

[0099] That is, for each data point S in category G j , if S i With S j If the distance between them is less than a certain distance threshold Th, then S i Data points classified as class G.

[0100] Similarly, given a similarity threshold Th, the user data set can be clustered so that the elements whose similarity between data points is greater than the similarity threshold are classified into one category. Since the algorithm only needs to scan the data once to complete the clustering, it belongs to streaming clustering.

[0101] Figure 3 FIG. 1 is a schematic diagram of the main process of the streaming clustering method of an embodiment of the present invention. Figure 3 , the basic steps of the algorithm are as follows:

[0102] (1) Initialize the current category set G to be empty and record the similarity threshold as Th;

[0103] (2) Scan each element in the user data set in turn, where the user data set consists of the overall similarity between users;

[0104] (3) Get elements;

[0105] (4) If the elements in the data set have been scanned, the algorithm ends; otherwise, execute (5);

[0106] (5) Calculate the similarity between the current element and each category cluster center in the current category set G, record the largest similarity as d, and record the corresponding category as g;

[0107] (6) If d>Th and G is not empty, execute (7); otherwise execute (8);

[0108] (7) Classify the current element into category g and update the cluster center of category g; return to (3);

[0109] (8) Create a new category h, classify the current element into category h, update the cluster center of category h, and classify the category into G; return to (3).

[0110] Assuming the amount of data in the input data set is n, the time complexity of the algorithm is O(nlogn), where n is the number of records and logn is the number of categories after clustering under normal circumstances. This clustering algorithm does not require the number of clusters to be given in advance, nor does it require the initial cluster center to be manually determined. The clustering results based on the algorithm make large categories contain more users, so that commonalities can be further extracted from them.

[0111] Step S104: extracting user interest patterns for each category of users respectively.

[0112] According to the technical solution of the present invention, the occurrence frequency of the item categories visited by users of each category is calculated, and the set of item categories whose occurrence frequency meets a preset threshold is taken as the user interest pattern of users of this category.

[0113] By using clustering methods, multiple user categories can be generated. Among them, there are possible interest patterns in categories with more users. Then, the frequency of occurrence of item categories in each larger user category is calculated, and the set of item categories with high occurrence frequency is taken as a user interest pattern. For multiple larger user categories, multiple interest patterns can be extracted. When extracting user interest patterns in categories with fewer users, user interest patterns may not be extracted due to various reasons, such as the small number of item categories visited by users or the low frequency of occurrence of item categories.

[0114] In one embodiment, it is assumed that 10,000 users in the test data set are divided into 734 independent user categories. Among these user categories, the top three user categories in terms of the number of users each contain more than 300 members. Each category also includes multiple item categories, and each item category corresponds to the percentage of the group of users accessing the category, and can be sorted in descending order according to the number of visitors.

[0115] For example, suppose a user category with 402 users has the largest user category, and the most visited item categories are dresses, T-shirts, shirts, etc., so these item categories become the user interest patterns of this user category. For another example, suppose a user category with 300 users has the most visited item categories are electronic products and computers, so these item categories become the user interest patterns of this user category.

[0116] In actual use, commercial websites can reorganize categories and items according to user interest patterns. For example, in addition to the traditional category navigation bar, multiple groups are built for different user groups (such as women, men, and children), and in each group, the most common categories are frequently displayed and updated. By referring to the results in the present invention, user groups can be more accurately segmented according to various user interest patterns (such as "fashionable women", "3C experts", and "housewives' favorites"), helping to improve the page navigation structure and reach target users more accurately.

[0117] In extracting user interest patterns, according to the prior art, the frequency of occurrence of item categories in each larger user category is calculated from multiple user categories generated by clustering, and a set of item categories with high occurrence frequency is taken as a user interest pattern. Multiple interest patterns can be extracted for multiple larger user categories. In short, in the step of extracting user interest patterns, various methods in the prior art can be applied, and are not limited to the content introduced in the embodiments of the present invention.

[0118] According to another aspect of an embodiment of the present invention, a device for extracting user interest patterns is provided.

[0119] Figure 4 is a schematic diagram of the main modules of the apparatus for extracting user interest patterns according to an embodiment of the present invention. Figure 4 As shown, the device 400 for extracting user interest patterns according to an embodiment of the present invention mainly includes: a processing module 401, a similarity calculation module 402, a clustering module 403 and an extraction module 404.

[0120] The processing module 401 is used to process the user's click stream data into the user's item category access sequence, item category access frequency and item category residence time based on the item classification information;

[0121] The similarity calculation module 402 is used to calculate the similarity between users based on the item category access sequence, item category access frequency and item category residence time of each user;

[0122] The clustering module 403 is used to cluster users according to the similarity between users;

[0123] The extraction module 404 is used to extract user interest patterns for each category of users.

[0124] According to one embodiment of the present invention, the processing module 401 may also be used for:

[0125] Generate an access path of a user during a session, wherein the access path of the user can be mapped into an item classification topology;

[0126] An item category access sequence of the user is generated according to the access path of the user.

[0127] According to another embodiment of the present invention, the processing module 401 may also be used for:

[0128] Counting the number of clicks by users on a certain item category, wherein the number of clicks includes the number of visits to the item category and the number of visits to items belonging to the item category;

[0129] The user's access frequency to the item category is obtained according to the number of clicks, wherein the access frequency is defined as the ratio of the number of clicks to the length of the user's item category access sequence.

[0130] According to yet another embodiment of the present invention, the processing module 401 may also be used for:

[0131] The user's item category dwell time is obtained according to the user's dwell time on the item category and the user's dwell time on all items belonging to the item category.

[0132] According to an embodiment of the present invention, the similarity calculation module 402 may also be used for:

[0133] Generate a fixed-dimensional vector representation for each user based on each user's item category access sequence, item category access frequency, and item category residence time;

[0134] The similarity between users is calculated based on the fixed-dimensional vector representation corresponding to each user.

[0135] According to one embodiment of the present invention, the similarity calculation module 402 may also be used for:

[0136] Generate a user-item category association vector based on item category access sequence features according to the user's item category access sequence;

[0137] Generate a user-item category association matrix based on item category access frequency features according to the item category access frequency of the user, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix;

[0138] A user-item category association matrix based on the item category residence time feature is generated according to the user's item category residence time, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix.

[0139] Further, according to an embodiment of the present invention, the similarity calculation module 402 may also be used for:

[0140] Calculate the access sequence similarity of users p and q, where the access sequence similarity is defined as the longest common subsequence length in the item category access sequences of users p and q divided by the maximum sequence length, in which the two users access the same item category in the same order;

[0141] Calculate the access frequency similarity of users p and q, wherein the access frequency similarity is defined as the similarity between two row vectors representing users p and q respectively in a user-item category association matrix based on item category access frequency features;

[0142] Calculate the residence time similarity of users p and q, wherein the residence time similarity is defined as the similarity between two row vectors representing users p and q respectively in a user-item category association matrix based on the item category residence time feature;

[0143] Based on the above three similarities, the overall similarity between users p and q is calculated, wherein the above three similarities are weighted respectively with three weights whose sum is 1, and the weighted three similarities are summed to obtain the overall similarity between users p and q.

[0144] In the embodiment of the present invention, both the access frequency similarity and the residence time similarity are cosine similarities.

[0145] According to an embodiment of the present invention, users are clustered by using a streaming clustering method with a specified similarity threshold.

[0146] Specifically, the streaming clustering method mainly includes the following steps:

[0147] 1) Initialize the current category set G to be empty and record the similarity threshold as Th;

[0148] 2) Scanning each element in the user data set in turn, wherein the user data set consists of the overall similarity between users;

[0149] 3) Get the element;

[0150] 4) If all elements in the data set have been scanned, the algorithm ends; otherwise, execute 5);

[0151] 5) Calculate the similarity between the current element and each category cluster center in the current category set G, record the largest similarity as d, and the corresponding category as g;

[0152] 6) If d>Th and G is not empty, execute 7); otherwise execute 8);

[0153] 7) Classify the current element into category g, update the cluster center of category g, and return to 3);

[0154] 8) Create a new category h, classify the current element into category h, update the cluster center of category h, and classify the category into G; return to 3).

[0155] According to yet another embodiment of the present invention, the extraction module 404 may also be used for:

[0156] The occurrence frequencies of the item categories visited by users of each category are calculated, and the set of item categories whose occurrence frequencies meet a preset threshold is taken as the user interest pattern of users of this category.

[0157] According to the technical solution of the embodiment of the present invention, the following technical means are adopted, namely, processing click stream data into user access sequences, access frequencies and dwell times on different item categories and calculating the similarity between users and clustering users based on this data. Therefore, the technical problem of inaccurate mining of user interests is overcome, and multiple interest patterns of e-commerce users can be quickly and effectively extracted based on click stream data. The extracted interest patterns can be used for scenario-based marketing or item recommendations, thereby promoting user consumption and improving sales indicators.

[0158] Figure 5 An exemplary system architecture 500 is shown to which the method for extracting user interest patterns or the apparatus for extracting user interest patterns according to the embodiments of the present invention can be applied.

[0159] like Figure 5 As shown, system architecture 500 may include terminal devices 501, 502, 503, a network 504 and a server 505. Network 504 is used to provide a medium for communication links between terminal devices 501, 502, 503 and server 505. Network 504 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0160] Users can use terminal devices 501, 502, 503 to interact with server 505 through network 504 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 501, 502, 503, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0161] The terminal devices 501 , 502 , and 503 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0162] The server 505 may be a server that provides various services, such as a backend management server that provides support for shopping websites browsed by users using the terminal devices 501, 502, and 503. The backend management server may analyze and process the received data such as product information query requests, and feed back the processing results such as target push information and product information to the terminal device.

[0163] It should be noted that the method for extracting user interest patterns provided in the embodiment of the present invention is generally executed by the server 505 , and accordingly, the device for extracting user interest patterns is generally set in the server 505 .

[0164] It should be understood that Figure 5 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0165] Reference below Figure 6 , which shows a schematic diagram of the structure of a computer system 600 of a terminal device suitable for implementing an embodiment of the present application. Figure 6 The terminal device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0166] like Figure 6 As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage part 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the system 600 are also stored. The CPU 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0167] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage section 608 as needed.

[0168] In particular, according to the embodiment disclosed in the present invention, the above reference flow chart ( Figure 1 ) The process described in the flowchart can be implemented as a computer software program. For example, the embodiment disclosed in the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the central processing unit (CPU) 601, the above-mentioned functions defined in the system of the present application are executed.

[0169] It should be noted that the computer-readable medium shown in the present invention can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0170] The flow chart or block diagram in the accompanying drawings illustrates the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0171] The modules involved in the embodiments of the present invention may be implemented by software or hardware. The modules described may also be set in a processor, for example, may be described as: a processor includes a processing module, a similarity calculation module, a clustering module and an extraction module. The names of these modules do not constitute limitations on the modules themselves under certain circumstances. For example: the processing module may also be described as "a module for processing the user's click stream data into the user's item category access sequence, item category access frequency and item category residence time based on the item classification information".

[0172] As another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device includes: based on the classification information of the items, processing the user's click stream data into the user's item category access sequence, item category access frequency and item category residence time; calculating the similarity between users according to each user's item category access sequence, item category access frequency and item category residence time; clustering users according to the similarity between users; extracting user interest patterns for each category of users.

[0173] According to the technical solution of the embodiment of the present invention, the following technical means are adopted, namely, processing click stream data into user access sequences, access frequencies and dwell times on different item categories and calculating the similarity between users and clustering users based on this data. Therefore, the technical problem of inaccurate mining of user interests is overcome, and multiple interest patterns of e-commerce users can be quickly and effectively extracted based on click stream data. The extracted interest patterns can be used for scenario-based marketing or item recommendations, thereby promoting user consumption and improving sales indicators.

[0174] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for extracting user interest patterns, characterized in that: include: Based on the classification information of items, the user's click stream data is processed into the user's item category access sequence, item category access frequency and item category dwell time; Calculate the similarity between users based on each user's item category access sequence, item category access frequency, and item category residence time; Clustering users according to similarities between the users; Extract user interest patterns for each category of users respectively; Calculating the similarity between users according to the item category access sequence, item category access frequency and item category residence time of each user includes: generating a fixed-dimensional vector representation for each user according to the item category access sequence, item category access frequency and item category residence time of each user; calculating the similarity between users according to the fixed-dimensional vector representation corresponding to each user; Generating a fixed-dimensional vector representation for each user according to each user's item category access sequence, item category access frequency, and item category residence time includes: generating a user-item category association vector based on item category access sequence features according to the user's item category access sequence; generating a user-item category association matrix based on item category access frequency features according to the user's item category access frequency, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix; generating a user-item category association matrix based on item category residence time features according to the user's item category residence time, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix; generating a user-item category association matrix based on item category residence time features according to the user's item category residence time, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the fixed-dimensional vector generated for any user is one of the row vectors in the matrix; Calculating the similarity between users includes: calculating the access sequence similarity of users p and q, wherein the access sequence similarity is defined as the longest common subsequence length in the item category access sequence of users p and q divided by the maximum sequence length, in which the two users visit the same item category in the common subsequence in the same order; calculating the access frequency similarity of users p and q, wherein the access frequency similarity is defined as the similarity between two row vectors representing users p and q respectively from a user-item category association matrix based on item category access frequency characteristics; calculating the residence time similarity of users p and q, wherein the residence time similarity is defined as the similarity between two row vectors representing users p and q respectively from a user-item category association matrix based on item category residence time characteristics; based on the above three similarities, calculating the overall similarity between users p and q, wherein the above three similarities are weighted respectively with three weights whose sum is 1, and the weighted three similarities are summed to obtain the overall similarity between users p and q.

2. The method according to claim 1, characterized in that Processing the user's click stream data into the user's item category access sequence includes: Generate an access path of a user during a session, wherein the access path of the user can be mapped to an item classification topology; An item category access sequence of the user is generated according to the access path of the user.

3. The method according to claim 1, characterized in that: Processing the user's click stream data into the user's item category visit frequency includes: Counting the number of clicks by users on a certain item category, wherein the number of clicks includes the number of visits to the item category and the number of visits to items belonging to the item category; The user's access frequency to the item category is obtained according to the number of clicks, wherein the access frequency is defined as the ratio of the number of clicks to the length of the user's item category access sequence.

4. The method according to claim 1, characterized in that: Processing the user's click stream data into the user's item category dwell time includes: The user's item category dwell time is obtained according to the user's dwell time on the item category and the user's dwell time on all items belonging to the item category.

5. The method according to claim 1, characterized in that The visit frequency similarity and the residence time similarity are both cosine similarities.

6. The method according to claim 1, characterized in that Users are clustered using a streaming clustering method with a specified similarity threshold.

7. The method according to claim 6, characterized in that The streaming clustering method comprises the following steps: 1) Initialize the current category set G to be empty and record the similarity threshold as Th; 2) Scanning each element in the user data set in turn, wherein the user data set consists of the overall similarity between users; 3) Get the element; 4) If all elements in the data set have been scanned, the algorithm ends; otherwise, execute 5); 5) Calculate the similarity between the current element and each category cluster center in the current category set G, record the largest similarity as d, and the corresponding category as g; 6) If d>Th and G is not empty, execute 7); otherwise execute 8); 7) Classify the current element into category g, update the cluster center of category g, and return to 3); 8) Create a new category h, classify the current element into category h, update the cluster center of category h, and classify the category into G; return to 3).

8. The method according to claim 1, characterized in that: Extracting user interest patterns for each category of users includes: The occurrence frequencies of the item categories visited by users of each category are calculated, and a set of item categories whose occurrence frequencies meet a preset threshold is taken as the user interest pattern of users of the category.

9. A device for extracting user interest patterns, characterized in that: include: A processing module, for processing the user's click stream data into the user's item category access sequence, item category access frequency and item category dwell time based on the item classification information; A similarity calculation module, used to calculate the similarity between users based on each user's item category access sequence, item category access frequency and item category residence time; A clustering module, used for clustering users according to the similarities between the users; An extraction module, used to extract user interest patterns for each category of users; The similarity calculation module is also used to: generate a fixed-dimensional vector representation for each user according to the item category access sequence, item category access frequency and item category residence time of each user; calculate the similarity between users according to the fixed-dimensional vector representation corresponding to each user; The similarity calculation module is also used to: generate a user-item category association vector based on item category access sequence characteristics according to the item category access sequence of the user; generate a user-item category association matrix based on item category access frequency characteristics according to the item category access frequency of the user, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the vector of fixed dimension generated for any user is one of the row vectors in the matrix; generate a user-item category association matrix based on item category residence time characteristics according to the item category residence time of the user, wherein the number of rows of the matrix is ​​the total number of users and the number of columns of the matrix is ​​the total number of item categories, and the vector of fixed dimension generated for any user is one of the row vectors in the matrix; The similarity calculation module is also used to: calculate the access sequence similarity of users p and q, wherein the access sequence similarity is defined as the longest common subsequence length in the item category access sequence of users p and q divided by the maximum sequence length, in which the two users visit the same item category in the common subsequence; calculate the access frequency similarity of users p and q, wherein the access frequency similarity is defined as the similarity between two row vectors representing users p and q respectively from a user-item category association matrix based on item category access frequency characteristics; calculate the residence time similarity of users p and q, wherein the residence time similarity is defined as the similarity between two row vectors representing users p and q respectively from a user-item category association matrix based on item category residence time characteristics; based on the above three similarities, calculate the overall similarity between users p and q, wherein the above three similarities are weighted respectively with three weights whose sum is 1, and the weighted three similarities are summed to obtain the overall similarity between users p and q.

10. The device according to claim 9, characterized in that The processing module is also used for: Generate an access path of a user during a session, wherein the access path of the user can be mapped into an item classification topology; An item category access sequence of the user is generated according to the access path of the user.

11. The device according to claim 9, characterized in that The processing module is also used for: Counting the number of clicks by users on a certain item category, wherein the number of clicks includes the number of visits to the item category and the number of visits to items belonging to the item category; The user's access frequency to the item category is obtained according to the number of clicks, wherein the access frequency is defined as the ratio of the number of clicks to the length of the user's item category access sequence.

12. The device according to claim 9, characterized in that The processing module is also used for: The user's item category dwell time is obtained according to the user's dwell time on the item category and the user's dwell time on all items belonging to the item category.

13. The device according to claim 9, characterized in that The visit frequency similarity and the residence time similarity are both cosine similarities.

14. The device according to claim 9, characterized in that Users are clustered using a streaming clustering method with a specified similarity threshold.

15. The device according to claim 14, characterized in that The streaming clustering method comprises the following steps: 1) Initialize the current category set G to be empty and record the similarity threshold as Th; 2) Scanning each element in the user data set in turn, wherein the user data set consists of the overall similarity between users; 3) Get the element; 4) If all elements in the data set have been scanned, the algorithm ends; otherwise, execute 5); 5) Calculate the similarity between the current element and each category cluster center in the current category set G, record the largest similarity as d, and the corresponding category as g; 6) If d>Th and G is not empty, execute 7); otherwise execute 8); 7) Classify the current element into category g, update the cluster center of category g, and return to 3); 8) Create a new category h, classify the current element into category h, update the cluster center of category h, and classify the category into G; return to 3).

16. The device according to claim 9, characterized in that The extraction module is also used for: The occurrence frequencies of the item categories visited by users of each category are calculated, and a set of item categories whose occurrence frequencies meet a preset threshold is taken as the user interest pattern of users of the category.

17. An electronic device for extracting user interest patterns, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 8.

18. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.