User portrait generation method and system based on public number private domain traffic

Abstract

The application relates to the technical field of user portrait generation, in particular to a user portrait generation method and system based on private domain traffic of a public account, which comprises the following steps: acquiring recommended information of a public account clicked and followed by a user each time, determining a theme distribution of the recommended information clicked by the user each time by using a probability generation model; composing a probability distribution sequence by using the probability distribution of each theme label of the user in the recommended information clicked all the times; analyzing a change trend and persistence of elements in the probability distribution sequence to determine a situation persistence of each theme label of the user; analyzing a current following degree of each theme label of the user based on the probability distribution sequence, combining the situation persistence to determine an interest retention degree of each theme label of the user; and determining an interest tendency confidence degree of each theme label of the user by using a numerical distribution of elements in the probability distribution sequence and combining the interest retention degree, so as to generate a user portrait. Therefore, the precision of user portrait generation is improved.

Description

A Method and System for Generating User Profiles Based on Private Domain Traffic of WeChat Official Accounts Technical Field

[0001] This application relates to the field of user profile generation technology, specifically to a method and system for generating user profiles based on private domain traffic of WeChat official accounts. Background Technology

[0002] As a social platform-based tool for interaction between businesses and users, WeChat Official Accounts have gradually become a direct communication bridge between merchants and users. Private domain traffic typically refers to a pool of user relationships built through WeChat Official Accounts and other similar platforms. These users have already followed or joined the merchant's platform, and merchants can use WeChat Official Accounts to achieve precise one-to-one or one-to-many marketing. In other words, private domain traffic focuses on precise management and in-depth operation.

[0003] By analyzing user data within private traffic, businesses can generate detailed user profiles, enabling precise personalized marketing, optimizing user experience, and enhancing user stickiness. However, traditional user profiles generated based on keyword frequency statistics primarily focus on surface-level keywords, lacking effective expression of users' deeper needs and complex areas of interest. They often fail to capture subtle changes in user interests in a timely manner, resulting in user profiles that are neither accurate nor effective. Summary of the Invention

[0004] To address the aforementioned technical issues, the purpose of this application is to provide a method and system for generating user profiles based on private domain traffic from WeChat Official Accounts. The specific technical solution adopted is as follows:

[0005] In a first aspect, embodiments of this application provide a method for generating user profiles based on private domain traffic of WeChat official accounts, the method comprising the following steps:

[0006] Obtain the recommendation information of the public accounts that the user clicks to follow each time, and use a probabilistic generative model to determine the topic distribution of the recommendation information for each click;

[0007] The probability distribution of each user's topic tag in all clicks of the recommendation information is used to form a probability distribution sequence; the changing trend of the elements in the probability distribution sequence and the persistence of the change are analyzed to determine the persistence of the user's topic tag status.

[0008] Based on probability distribution sequence analysis, the user's current level of attention to each topic tag is determined, and combined with the persistence of the trend, the user's interest retention rate for each topic tag is determined.

[0009] By combining the numerical distribution of elements in the probability distribution sequence with the interest retention rate, the confidence level of the user's interest tendency for each topic tag is determined, which is used to generate a user profile.

[0010] In one embodiment, determining the topic distribution of the recommended information for each user click using a probability generation model includes:

[0011] The LDA topic model takes the recommended information from each public account that a user clicks to follow as input and outputs the topic tags and their probability values ​​from each recommended information clicked by the user.

[0012] In one embodiment, the order of probability distribution values ​​in the probability distribution sequence is the order in which users clicked on the recommended information.

[0013] In one embodiment, determining the persistence of a user's activity for each topic tag includes:

[0014] The maximum number of consecutive non-zero elements in the probability distribution sequence is calculated, zero elements are removed from the probability distribution sequence, and a straight line is fitted to the probability distribution sequence after removing zero elements.

[0015] The maximum value is calculated as a fusion value of the number of non-zero elements in the probability distribution sequence. The persistence of the user's trend for each topic tag is positively correlated with the fusion value and the slope of the fitted line.

[0016] In one embodiment, the proportion of the fusion value in the total number of elements in the probability distribution sequence is statistically analyzed, and combined with the normalized value of the slope, the persistence of the user's trend for each topic tag is obtained.

[0017] In one embodiment, determining the user's interest retention rate for each topic tag includes:

[0018] Based on the distribution of zero elements in the tail elements of the probability distribution sequence, the short-term shelving degree of each user's topic tag is determined. The negative correlation mapping result of the proportion of the short-term shelving degree in the total number of elements in the probability distribution sequence is combined with the persistence of the user's trend for each topic tag to serve as the user's interest retention degree for each topic tag.

[0019] In one embodiment, determining the short-term shelving level of each user's topic tag includes:

[0020] If the last element in the probability distribution sequence is non-zero, the short-term shelving degree is set to 0; otherwise, the number of elements whose last element in the probability distribution sequence is consecutively zero is counted and used as the short-term shelving degree.

[0021] In one embodiment, determining the confidence level of a user's interest tendency for each topic tag includes:

[0022] Calculate the upper quartile and lower quartile of the probability distribution sequence, and calculate the mean and dispersion of the elements in the probability distribution sequence located between the lower quartile and the upper quartile;

[0023] The product of the mean and the user's interest retention rate for each topic tag is determined. The confidence level of the user's interest tendency for each topic tag is positively correlated with the product and negatively correlated with the degree of dispersion.

[0024] In one embodiment, generating a user profile includes:

[0025] The user's topic tags are used as input to the word cloud function, and the word frequency statistics in the word cloud function are replaced with the confidence level of interest tendency for each topic tag to output a user profile.

[0026] Secondly, embodiments of this application also provide a user profile generation system based on private domain traffic of WeChat official accounts, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0027] This application has at least the following beneficial effects:

[0028] This application obtains recommendation information from the public accounts followed by users each time they click, and uses a probabilistic generation model to determine the topic distribution of the recommendation information for each click. The probabilistic generation model analyzes the recommendation information from the public accounts followed by users each time they click, obtaining a probability distribution sequence. This effectively extracts the potential topic distribution of user interests, contributing to the accuracy of subsequent user profiling. By analyzing the changing trends of elements in the probability distribution sequence, the application determines the persistence of user behavior for each topic tag, identifying long-term changes and trends in user interest in a particular topic tag. This helps predict future changes in user interests, thereby optimizing content delivery, enhancing user stickiness and activity, and improving the persistence of user behavior. This system can accurately assess the stability of a user's current interests, avoiding content loss due to ignoring changing trends in user interests. By combining the level of attention and persistence of each topic tag, it can better understand the concentration and changing trends of user interest in different topics. The confidence level of user interest tendencies provides a more scientific basis for content recommendation, helping to push more personalized and accurate content to users based on user profiles, thereby improving user experience and satisfaction. This application can capture the dynamic changes in user interests in real time, update user profiles in a timely manner, and ensure that recommendations are always optimized based on the latest user behavior data, making the generation of user profiles more in-depth and dynamic. Attached Figure Description

[0029] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 is a flowchart of the steps of a user profile generation method based on private domain traffic of a public account provided in an embodiment of this application;

[0031] Figure 2 is a flowchart for determining the confidence level of topic tags' interest tendencies. Detailed Implementation

[0032] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the user profile generation method and system based on WeChat Official Account private domain traffic proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0034] The following, in conjunction with the accompanying drawings, details the specific solution of the user profile generation method and system based on WeChat Official Account private domain traffic provided in this application.

[0035] Please refer to Figure 1, which shows a flowchart of the steps of a user profile generation method based on WeChat Official Account private domain traffic according to an embodiment of this application. The method includes the following steps:

[0036] S1. Obtain the recommendation information of the public accounts that the user clicks to follow each time, and use a probabilistic generative model to determine the topic distribution of the recommendation information for each click.

[0037] For users in private domain traffic, the system obtains recommendation information from each WeChat Official Account a user clicks and follows. This recommendation information is then used as input to an LDA (Latent Dirichlet Allocation) topic model. The output is the topic distribution for each clicked recommendation, including all topic tags and their probability values, representing the topics the user is interested in. The LDA topic model is a well-known existing technology, and its specific process will not be elaborated upon.

[0038] S2, construct a probability distribution sequence from the probability distribution of each user's topic tag in all clicks of the recommendation information; analyze the changing trend of the elements in the probability distribution sequence and the persistence of the changes to determine the persistence of the user's topic tag.

[0039] Hashtags are identifiers of content that users are interested in, accurately reflecting their preferences. A higher probability value for a hashtag indicates a greater user interest in the content associated with that hashtag. When a public account recommends articles with a specific hashtag to its private domain users, the likelihood of those users clicking and viewing is higher. However, user interests change over time; therefore, the probability changes of hashtags at different times determine these changes in user interest. The faster the probability value of a hashtag decreases, the more likely the user no longer prefers the recommendations associated with that hashtag, and that hashtag is no longer suitable for characterizing the public account's user profile.

[0040] Based on the above analysis, this embodiment constructs a probability distribution sequence using the probability values ​​of each user's topic tag across all clicked recommendations. This sequence represents the changing state of the user's interest tags. If a user clicks on a recommendation that does not contain that topic tag, the corresponding probability value is 0. This ensures that the length of the probability distribution sequence for each user's topic tag is equal, equal to the number of clicked recommendations. The order of the probability values ​​in the probability distribution sequence corresponds to the order in which the user clicked the recommendations.

[0041] In the probability distribution sequence of topic tags, the longer the continuity of non-zero elements, the more long-term the user's interest in the articles corresponding to that topic tag is. This indicates that the topic tag is more representative of the user's interests, and the more accurate the user profile can be created using that topic tag. Therefore, the length of the longest consecutive non-zero element in the probability distribution sequence is denoted as the "long-term value," used to characterize the long-term nature of a user's interests.

[0042] Furthermore, all zero elements are removed from the probability distribution sequence to obtain the user's interest probability sequence, which is used to represent the user's interest preferences. Then, the interest probability sequence is used as input to the linear least squares method, and the output is the fitted line of the interest probability sequence. The slope of the fitted line is obtained to represent the changing trend of the user's interests. The linear least squares method is a well-known existing technique, and its specific process will not be elaborated here.

[0043] Therefore, the persistence of user activity for each topic tag is calculated as follows:

[0044] The fusion value of the long-term degree and the number of non-zero elements in the probability distribution sequence is calculated. The persistence of the user's topic tag is positively correlated with the fusion value and the slope of the fitted line.

[0045] It should be noted that fusion means combining multiple variables, which can be done by addition, multiplication, a combination of addition and multiplication, or by taking the average.

[0046] In this embodiment, the expression for the persistence of a user's topic tag is:

[0047] In the formula, This indicates the persistence of a user's activity on the i-th topic tag. Let represent the slope of the fitted line for the probability sequence of user interest in the i-th topic tag, and sig() is the sigmoid() function. This represents the longevity of the user's i-th topic tag. This represents the number of non-zero elements in the probability distribution sequence of the user's i-th topic tag. This represents the length of the probability distribution sequence, i.e., the number of elements in the probability distribution sequence.

[0048] When a public account recommends information to a user, the user's clicking to browse the information indicates that the user has a certain interest in it. Therefore, for the topic tags present in the recommended information, the greater the continuity of the same topic tags when the user continues to click and browse, the greater the quantification probability of the topic tags in the recommended information. This indicates that the user's interest in the topic tags is constantly increasing, and the long-term nature of this interest is also constantly increasing. Therefore, the greater the sustainability of the topic tag's trend, the more positive and sustainable the development of the topic tag is for the user's interests and hobbies in the long term, and the more accurate the user profile depicted through the topic tag.

[0049] S3. Based on probability distribution sequence analysis, determine the user's current level of attention to each topic tag, and combine this with the trend persistence to determine the user's interest retention rate for each topic tag.

[0050] The importance of hashtags is reflected not only in their long-term persistence but also in their short-term immediacy. Specifically, for the information a user clicks on, the closer the time since the hashtag appeared, the higher the user's short-term preference for that hashtag. As user interests change over time, hashtags that are closer to the present moment are more representative of the user's current interests. Therefore, when a public account's next recommended information includes a hashtag, the user is more likely to click and browse that recommended information.

[0051] Therefore, the short-term abandonment rate of each user's topic tag is obtained to characterize the short-term lack of interest in the user's topic tags. The higher the value, the longer the continuous period that the topic tag has not appeared at the current moment, and the lower the click-through rate of the user's corresponding recommendation information. Specifically:

[0052] For a probability distribution sequence of topic tags, if the last element in the probability distribution sequence is non-zero, the short-term shelving degree of the topic tag is set to 0. Otherwise, the number of elements whose last element in the probability distribution sequence is consecutively zero is counted and used as the short-term shelving degree of the topic tag.

[0053] Furthermore, the interest retention rate for each user's topic tag is calculated. Specifically, the negative correlation mapping result of the proportion of the short-term shelving rate in the total number of elements in the probability distribution sequence is combined with the persistence of the user's activity for each topic tag to obtain the interest retention rate for each topic tag. Here, the negative correlation mapping indicates that the interest retention rate of a topic tag is negatively correlated with the proportion of the short-term shelving rate.

[0054] In this embodiment, the expression for user interest retention for each topic tag is:

[0055] In the formula, This represents the user's interest retention rate for the i-th topic tag. This represents the short-term dwell time of the user's i-th topic tag, and exp() represents an exponential function with the natural constant as the base. This indicates the persistence of a user's activity on the i-th topic tag. This represents the length of the probability distribution sequence.

[0056] Because user interests can change, the longer a topic tag has not appeared in the immediate vicinity of the current moment, the faster the user's interest in that topic tag declines. This decline in user interest leads to a decrease in the frequency and duration of that topic tag's appearance in user-clicked browsing information, resulting in reduced topic tag persistence and increased short-term topic tag abandonment. Consequently, the topic tag's retention rate decreases, indicating a reduced user interest in clicking on the recommended information corresponding to that topic tag, and a decline in the topic tag's importance to the user.

[0057] S4. By combining the numerical distribution of elements in the probability distribution sequence with the interest retention rate, determine the confidence level of the user's interest tendency for each topic tag, and use it to generate a user profile.

[0058] The higher the probability value of different topic tags for recommended information, the more representative the main content of the recommended information is. However, the probability value of recommended information topic tags may vary considerably and cannot truly represent changes in user interests. Conversely, the higher the stability of the probability of recommended information topic tags, the more stable the user's interests corresponding to that topic tag. When a public account recommends information to a user in the next instance, using recommendations with that topic tag increases the likelihood of the user clicking and viewing.

[0059] Regarding the distribution of topic tag probability values, extreme values ​​may occur, affecting the calculation of the stability of the topic tag probability values. Therefore, in this embodiment, elements between the lower quartile and the upper quartile are extracted from the probability distribution sequence to remove the influence of extreme values. The dispersion of all elements extracted from the probability distribution sequence is then calculated to characterize the stability of the topic tag probability values, and the mean of all elements extracted from the probability distribution sequence is calculated to characterize the likelihood of a user's interest in that topic tag. Specifically, the dispersion can be calculated using variance, standard deviation, coefficient of variation, etc.

[0060] In this embodiment, the confidence level of a user's interest in each topic tag is calculated, and the specific expression is as follows:

[0061] In the formula, This represents the confidence level of a user's interest in the i-th topic tag. , Let represent the standard deviation and mean of the elements extracted from the probability distribution sequence of the user's i-th topic tag, respectively. This indicates a preset value greater than 0 to prevent the denominator from being 0. To avoid affecting the calculation of the result, in this implementation... The implementer can set the parameters according to the actual situation; this embodiment does not impose any restrictions. The flowchart for determining the confidence level of topic tags is shown in Figure 2.

[0062] For tags with a relatively stable probability of appearing in recommended information on public accounts, these tags represent relatively stable user interests and preferences. The higher the probability value of a tag, the higher the user's interest, and the more interested the user is in the recommended information corresponding to that tag. As a result, the dispersion of elements extracted from the probability distribution sequence of the tag is smaller, the mean is larger, and the interest retention rate of the tag is higher, which increases the confidence of the tag's interest tendency. This means that the tag is more representative of the user's interests and hobbies, and the user profile depicted through the tag is more accurate.

[0063] Furthermore, all of the user's topic tags are used as input to the WordCloud word cloud function. The word frequency statistics in the WordCloud word cloud function are replaced with the confidence score of the interest tendency for each topic tag, which is used to assess the importance of the topic tags. The higher the confidence score of the interest tendency, the higher the importance. The output is a visualized user profile, which is a word cloud of the user's topic tags. The larger the font of the topic tag in the word cloud and the higher it is displayed, the more representative the topic tag is of the user's interest tendency. The calculation of the WordCloud word cloud function is a well-known technique, and the specific process will not be described in detail here.

[0064] Based on the same inventive concept as the above methods, this application also provides a user profile generation system based on the private domain traffic of a public account, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above methods for generating user profiles based on the private domain traffic of a public account.

[0065] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0066] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0067] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for generating user profiles based on private domain traffic of WeChat official accounts, characterized in that, The method includes the following steps: obtaining recommendation information from the public accounts followed by the user each time; using a probability generation model to determine the topic distribution of the recommendation information clicked by the user each time; forming a probability distribution sequence by composing the probability distribution of each topic tag of the user in all clicks of recommendation information; analyzing the changing trend and persistence of the elements in the probability distribution sequence to determine the persistence of the user's topic tag, including: counting the maximum number of consecutive non-zero elements in the probability distribution sequence; removing zero elements from the probability distribution sequence; fitting a straight line to the probability distribution sequence after removing zero elements; calculating the fusion value between the maximum value and the number of non-zero elements in the probability distribution sequence; the persistence of the user's topic tag is positively correlated with the fusion value and the slope of the fitted straight line. The process involves analyzing the current level of user attention to each topic tag based on a probability distribution sequence, and determining the user's interest retention rate for each topic tag based on the persistence of the trend. Then, by analyzing the numerical distribution of elements in the probability distribution sequence and combining this interest retention rate, the confidence level of the user's interest tendency for each topic tag is determined, which is used to generate a user profile. Determining the confidence level of the user's interest tendency for each topic tag includes: calculating the upper and lower quartiles of the probability distribution sequence; calculating the mean and dispersion of elements in the probability distribution sequence located between the lower and upper quartiles; and determining the product of the mean and the user's interest retention rate for each topic tag. The confidence level of the user's interest tendency for each topic tag is positively correlated with the product and negatively correlated with the dispersion.

2. The user profile generation method based on WeChat Official Account private domain traffic as described in claim 1, characterized in that, The method of using a probabilistic generative model to determine the topic distribution of the recommended information for each user click includes: taking the recommended information of the public account followed by the user each time as the input of the LDA topic model, and outputting the topic tags and their probability values ​​in the recommended information for each user click.

3. The user profile generation method based on WeChat Official Account private domain traffic as described in claim 1, characterized in that, The order of probability distribution values ​​in the probability distribution sequence is the order in which users click on the recommended information.

4. The user profile generation method based on WeChat Official Account private domain traffic as described in claim 1, characterized in that, The percentage of the fusion value in the total number of elements in the probability distribution sequence is statistically analyzed, and combined with the normalized value of the slope, to obtain the persistence of the user's trend for each topic tag.

5. The user profile generation method based on WeChat Official Account private domain traffic as described in claim 1, characterized in that, The determination of user interest retention for each topic tag includes: determining the short-term shelving degree of each user topic tag based on the distribution of 0 elements in the tail elements of the probability distribution sequence; and combining the negative correlation mapping result of the proportion of the short-term shelving degree in the total number of elements in the probability distribution sequence with the persistence of the user's status for each topic tag as the user interest retention degree for each topic tag.

6. The user profile generation method based on WeChat Official Account private domain traffic as described in claim 5, characterized in that, Determining the short-term shelving degree of each user's topic tag includes: if the last element in the probability distribution sequence is non-zero, then the short-term shelving degree is set to 0; otherwise, the number of elements whose last element in the probability distribution sequence is consecutively zero is counted and used as the short-term shelving degree.

7. The user profile generation method based on WeChat Official Account private domain traffic as described in claim 1, characterized in that, The process of generating a user profile includes: taking all of the user's topic tags as input to a word cloud function, replacing the word frequency statistics in the word cloud function with the confidence level of interest tendency for each topic tag, and outputting a user profile.

8. A user profile generation system based on WeChat Official Account private domain traffic, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-7.

Images