Method and device for measuring content homogeneity in Internet media platform

Abstract

The invention provides a method and a device for measuring content homogeneity in an internet media platform. The method comprises the following steps: determining group attribute characteristics of a target group; deploying an artificial intelligence cluster in an internet media platform environment based on the group attribute features of the determined target group; operating an artificial intelligence body to simulate the use behavior of the target group internet media platform; recording content information browsed by each artificial intelligence body on the recommendation page of the Internet media platform in the process of simulating the use behavior of the artificial intelligence body; and based on the content information, calculating and outputting content homogeneity when the target group uses the Internet media platform. According to the method, high-accuracy and high-efficiency measurement of content homogeneity, protection of user privacy and support of cross-platform can be realized.

Description

Technical Field

[0001] This invention relates to the field of internet media technology, and specifically to a method and apparatus for measuring content homogeneity in internet media platforms. Background Technology

[0002] Currently, users on internet media platforms primarily obtain information through platform push notifications. These platforms can personalize content recommendations based on user attributes and characteristics using recommendation algorithms. The content recommended varies significantly across different user groups. Because internet media platforms prioritize attracting user attention, their recommendation algorithms may push too much similar content, trapping users in a highly homogenized "information cocoon." An information cocoon refers to the phenomenon where users, due to factors such as algorithmic recommendations, user preferences, and the homogenization of social networks, become confined to a single viewpoint or area of ​​interest, leading to a closed information environment. The degree of similarity in content provided by internet media platform recommendation algorithms, i.e., the degree of information cocoon formation, can be defined as "content homogeneity." In the field of communication, content homogeneity can serve as an indicator for evaluating internet media platforms or user profiles, and it has significant implications for advertising, marketing, and social governance.

[0003] However, due to the dynamic, random, and black-box nature of platform recommendation algorithms, it is impossible to directly measure the algorithm itself. Therefore, previous research mainly used social surveys to subjectively measure content homogeneity by inquiring about the information accessed by different groups. However, this method has significant limitations in efficiency and accuracy. Another approach is to contact real users or obtain official support from within the platform to calculate content homogeneity by obtaining the browsing history of internet platform accounts. However, this method faces privacy protection issues and cannot be applied to multiple internet media platforms. Therefore, a new method that is accurate, efficient, privacy-free, and supports multiple platforms is urgently needed for calculating content homogeneity. Summary of the Invention

[0004] To address the problems and shortcomings of existing technologies, this invention provides a method and apparatus for measuring content homogeneity in internet media platforms. This method solves the problems of low accuracy, low efficiency, privacy risks, and lack of cross-platform support in existing methods, thereby achieving high accuracy, high efficiency, protection of user privacy, and cross-platform support for content homogeneity measurement.

[0005] The present invention achieves the above objectives through the following technical solutions: A method for measuring content homogeneity in an internet media platform includes: Determine the group attributes of the target group, and combine the activity time distribution Q(t) and activity interval distribution P(t). t) serves as the active time attribute of the target group; among which, , t is the time difference between two user activities, and P is the probability density function. The coefficients are power-law distribution coefficients. Based on the group attribute characteristics of the identified target group, deploy an artificial intelligence agent cluster in the Internet media platform environment; Running an artificial intelligence agent to simulate the internet media platform usage behavior of a target group; During the process of simulating the behavior of AI agents, the content information that each AI agent browses on the recommended pages of Internet media platforms is recorded; Based on content information, the content homogeneity of the target group when using Internet media platforms is calculated and output; among them, the content homogeneity of the AI ​​agent is calculated based on the content information returned by a single AI agent; and the content homogeneity of the target group is calculated based on all AI agents.

[0006] According to the present invention, a method for measuring content homogeneity in an internet media platform is provided, based on a single artificial intelligence agent in an artificial intelligence agent operation module. i The returned content information is used to calculate the artificial intelligence agent. i Content homogeneity includes the following steps: For artificial intelligence entities i For each piece of content returned, a text semantic vector is extracted using a pre-trained language model; Cluster all semantic vectors using the DBSCAN method, dividing all content information into... m One semantic category; Analyze the frequency of content occurrences in each semantic category and calculate the proportion of each category. ; Based on the probability distribution of each semantic category, semantic entropy is calculated as the content homogeneity of the AI ​​agent. The calculation formula is as follows:

[0007] in, For artificial intelligence entities i The returned content information is homogeneous. m This represents the total number of categories after semantic clustering. For the first j The proportion of content items in a semantic category to the total number of content items, that is, the number of content items in that category divided by the total number of content items returned by the AI ​​agent.

[0008] According to the present invention, a method for measuring content homogeneity in an internet media platform is provided, based on all artificial intelligence entities. iCalculating the content homogeneity of the target group includes the following steps: For each AI agent i The returned content information is used to construct its content distribution vector based on the calculated semantic category set: ,in m Number of content categories; Calculate the average vector of content distribution across all AI agents. The calculation formula is:

[0009] based on With each The Jensen-Shannon divergence between the content distribution of all AI agents and the overall average distribution is calculated, and the average value is taken as the content homogeneity index of the target group. The calculation formula is as follows:

[0010] in, This indicates that the target group's content is homogeneous. For the first i Content distribution of individual AI agents The average content distribution of all AI agents. Indicates the first i The Jensen-Shannon divergence between the content distribution of an individual AI agent and the overall average content distribution is calculated using the following formula:

[0011] in, , .

[0012] A method for measuring content homogeneity in an internet media platform according to the present invention further includes: Based on attribute features, partial AI entities can be selected, and the content homogeneity of subgroups within a target group can be calculated: Determine the set of attribute features for the subgroup. This attribute feature set This includes the AI ​​agent's geographic location attribute g and the AI ​​agent's age attribute a; Based on attribute feature set From the deployed AI agent cluster, AI agents that meet the specific attribute characteristics are selected, and the content homogeneity of the segmented group is calculated based on the selected AI agents. ,include: The content homogeneity of a specific group within a geographically defined region; Content homogeneity of a specific age group or age range segment; The calculated content homogeneity results are transmitted to the content homogeneity query module.

[0013] According to the present invention, a method for measuring content homogeneity in an internet media platform is provided, which involves deploying an artificial intelligence agent cluster in the internet media platform environment, including: Based on the group attributes of the target group, the system generates account registration information for the AI ​​agent on internet media platforms. The nickname is generated according to the language habits of the target group; gender and age are set according to the proportion of the target group; the IP address is selected from the address corresponding to the region of the target group; the content preference is determined by data mining of the target group's frequently interacted content; and the initial followers are selected from accounts that the target group generally follows. Design and implement an artificial intelligence entity capable of simulating the internet media platform usage behavior of a target group; Develop an account binding interface to link internet media platform accounts with AI entities; Set the activity parameters of the artificial intelligence agent.

[0014] According to the present invention, a method for measuring content homogeneity in an internet media platform includes, when designing and implementing an artificial intelligence agent capable of simulating the internet media platform usage behavior of a target group, the following steps are taken: For web-based simulation, browser automation control tools or browser plugins are used to enable artificial intelligence to simulate human internet media platform usage behavior. Browser automation control tools and browser plugins include at least Webdriver and Puppeteer. For mobile client simulation, mobile terminal automation control tools are used to enable artificial intelligence agents to simulate human internet media platform usage behavior. Mobile terminal automation control tools include at least Appium and Auto.js. The AI ​​entity has a simulation mode and a recording mode: in simulation mode, it simulates the internet media platform usage behavior of the target group based on preset parameters; in recording mode, it records its own behavior and platform content information. The behavioral logic and operating mode of the AI ​​agent are written into an executable program, placed in the runtime module for execution, and the running status is monitored and managed.

[0015] According to the present invention, a method for measuring content homogeneity in an internet media platform includes, when setting the activity parameters of an artificial intelligence agent, the following: Activity time parameters include: setting the activity time distribution Q(t) and activity interval distribution P(t) based on the activity patterns of the target group. t); Interaction rule parameters include: Based on the frequency of access to different content source pages by the target group, the content source page is selected based on a preset probability P(source); Based on the content themes of interest to the target audience, keywords are set. Based on a preset probability P (click), the AI ​​agent randomly clicks on a piece of content on the current page to enter the details page and perform browsing behavior, depending on whether the content title or tag contains the specified keywords. If no matching information is found, the operation is skipped. On the details page, the AI ​​agent determines whether to like a post based on a preset probability P (like), and whether to follow the content publisher based on a preset probability P (follow). The content source pages include: recommendation pages, content category pages, and search pages; Save the AI ​​agent's activity parameters as a configuration file in JSON or XML text format, and then place the configuration file into the AI ​​agent's running module.

[0016] According to the present invention, a method for measuring content homogeneity in an internet media platform includes, when running an artificial intelligence agent to simulate the internet media platform usage behavior of a target group, the following steps are taken: Activate the simulation mode of the AI ​​agent to enable the AI ​​agent to continuously simulate the target group's internet media platform usage behavior; The behavior of AI agents in simulation mode includes: Log in to the internet media platform account linked to the AI ​​entity; The AI ​​agent enters the recommendation page of an internet media platform and scans the page content according to a preset keyword list. When it finds that the title or summary of a content contains keywords, the AI ​​agent randomly selects a content that meets the criteria and clicks on it to enter the details page of that content. On the content category page, the AI ​​agent selects a corresponding content category from the preset list of content categories; after entering the category page, the AI ​​agent browses the content on the page and randomly clicks on one of the contents to enter the details page; The AI ​​agent enters the search page and inputs preset keywords to search; on the search results page, the AI ​​agent randomly selects a result and clicks on it to enter the details page of that content; the selection of search keywords is also based on the analysis of the target group's search behavior to simulate the target group's process of actively searching for information; Once the AI ​​entity enters the details page, it will perform actions such as liking and following the content publisher based on preset probability values. During operation, the AI ​​agent will receive control commands from the AI ​​agent management module, including: Pause the AI ​​agent's task, suspend the AI ​​agent's task, restart the AI ​​agent, and modify the AI ​​agent's activity parameters.

[0017] According to the present invention, a method for measuring content homogeneity in an internet media platform includes, when recording content information viewed by an AI agent on a recommendation page, the method comprises: Once the preset recording trigger conditions are met, the recording mode of the AI ​​agent is activated; The behavior of AI agents in recording patterns includes: Browsing Recommended Pages: After entering recording mode, the AI ​​entity browses the recommended pages according to its preset browsing rules; Text information storage: During the browsing of the recommendation page, the AI ​​agent extracts and saves the text information contained in the content appearing on the page; Result file generation and return: The saved text information is converted into a JSON text format result file, and the result file is returned to the AI ​​entity management module through the content information upload channel between the AI ​​entity running module and the AI ​​entity management module.

[0018] A device for measuring content homogeneity in an internet media platform, comprising: The AI ​​agent operation module is used to deploy and execute multiple AI agents that can run independently in a distributed manner, and to provide the network access environment required for these AI agents to run. The AI ​​agent management module is used to receive the operation data and behavior logs of the AI ​​agents in the AI ​​agent operation module, and to issue scheduling commands to the AI ​​agent operation module. The content homogeneity calculation module is used to read data in a targeted manner and calculate the homogeneity index of content in Internet media platforms. The content homogeneity query module is used for interactive querying of the calculation results of content homogeneity. Among them, the multiple AI entity operation modules and the multiple AI entity management modules have a content information upload channel and an AI entity control command transmission channel; The content homogeneity calculation module and the artificial intelligence entity management module have a query instruction transmission channel and a data result feedback channel; The content homogeneity query module and the content homogeneity calculation module have a query instruction transmission channel and a content homogeneity result feedback channel.

[0019] Therefore, compared with existing technologies, the method for measuring content homogeneity in internet media platforms proposed in this invention has the following beneficial effects: 1. The content homogeneity measurement method provided by this invention can achieve independent, accurate, and efficient assessment of the content homogeneity of different user groups on Internet media platforms; 2. This invention utilizes an artificial intelligence agent to replace traditional methods of surveying real users, obtaining the recommended content presented by internet media platforms to a corresponding user group during their usage in a simulated manner. This invention simulates the usage behavior of real users on internet media platforms through an artificial intelligence agent, closely approximating the experience of real users on internet media platforms, thereby achieving accurate, efficient, and convenient measurement of content homogeneity. 3. This invention can be applied to any Internet media platform that uses recommendation algorithms to recommend personalized content to users, including but not limited to: short video platforms, news platforms, video platforms, image and text sharing platforms, e-commerce platforms, etc., and has strong cross-platform capabilities.

[0020] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0021] Figure 1 This is a flowchart of an embodiment of a method for measuring content homogeneity in an internet media platform according to the present invention.

[0022] Figure 2 This is a flowchart illustrating the operational logic of an artificial intelligence entity in simulation mode, as described in an embodiment of a method for measuring content homogeneity in an internet media platform according to the present invention.

[0023] Figure 3 This is a flowchart illustrating the operational logic of an artificial intelligence entity in recording mode, as described in an embodiment of a method for measuring content homogeneity in an internet media platform according to the present invention.

[0024] Figure 4 This is a schematic diagram of an embodiment of a device for measuring content homogeneity in an internet media platform according to the present invention.

[0025] The diagram numbers are as follows: 1. AI entity operation module; 2. AI entity management module; 3. Content homogeneity calculation module; 4. Content homogeneity query module; 5. Strategy scheduling unit; 6. Data storage unit. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0028] See Figure 1 This embodiment provides a method for measuring content homogeneity in an internet media platform, including: Step S1: Determine the group attribute characteristics of the target group, and combine the activity time distribution Q(t) and activity interval distribution P(t). t) serves as the active time attribute of the target group; among which, , t is the time difference between two user activities, and P is the probability density function. The coefficients are power-law distribution coefficients. Step S2: Based on the group attribute characteristics of the identified target group, deploy an artificial intelligence agent cluster in the Internet media platform environment; Step S3: Run an AI agent to simulate the target group's internet media platform usage behavior; Step S4: During the process of simulating the usage behavior of the AI ​​agent, record the content information that each AI agent browses on the recommendation page of the Internet media platform; Step S5: Based on content information, calculate and output the content homogeneity of the target group when using the Internet media platform; wherein, the content homogeneity of the AI ​​agent is calculated based on the content information returned by a single AI agent; and the content homogeneity of the target group is calculated based on all AI agents.

[0029] Specifically, this embodiment uses an artificial intelligence agent to replace the survey of real users, and obtains the recommended content presented by the Internet media platform received by the corresponding user group during use in a simulated form, and then uses it to calculate the content homogeneity.

[0030] The target group refers to a group of individual users who are similar in terms of demographic characteristics (geographical location, gender, age, and active time) and content preferences.

[0031] Artificial intelligence refers to automated programs that use programming techniques to enable access to internet media platforms and simulate the internet media platform usage behavior of target user groups in a real environment.

[0032] The AI ​​entity has two functions: in addition to simulating the internet media platform usage behavior of the target user group, it can also record the content information recommended by the internet media platform on the page.

[0033] Content information refers to the information in the content recommended by internet media platforms, including text information, image information, content tag information, and comment information.

[0034] This invention can calculate the content homogeneity of different target groups by binding artificial intelligence entities with different group attribute characteristics.

[0035] In step S1 above, determining the attribute characteristics of the target user group includes: Determine the demographic characteristics of the target group, including: gender, age, geographic location, and active time. Determine the content preference attributes of the target audience, including content category attributes and keyword attributes.

[0036] Furthermore, in determining the demographic characteristics of the target group, this includes: Geographical location attributes can be determined in two ways: geographical region and city line level. When based on geographical region, it will be divided into provincial-level administrative regions, while for overseas geographical locations, it will be based on countries or regions. Cities are classified according to their tier, including: first-tier cities, second-tier cities, third-tier cities, and non-tier cities. Age attributes are determined based on the approximate age range of the target group; Gender attributes include male and female; The active time attribute includes the distribution of active time and the distribution of active intervals; Content preference attributes include two dimensions: content category attributes and keyword attributes. Among them, content category attributes include content categories such as current affairs, sports, entertainment, technology, military, games, and health. The keywords include: keywords related to specific interest areas, keywords related to geographical location, and keywords related to gender.

[0037] In step S2 above, when deploying an AI agent cluster based on the group attribute characteristics of the target group, the following steps are included: Based on the target group's group attributes, the system generates registration information for an AI agent's internet media platform account, including: nickname, gender, age, IP address, content preferences, and initial followers. The nickname is generated based on the target group's language habits; gender and age are set proportionally to the target group; the IP address is selected from addresses corresponding to the target group's geographic location; content preferences are determined through data mining of the target group's frequently interacting content; and initial followers are selected from accounts commonly followed by the target group. Design and implement an artificial intelligence entity capable of simulating the internet media platform usage behavior of a target group; Develop an account binding interface to link internet media platform accounts with AI entities; Set the activity parameters of the artificial intelligence agent.

[0038] Furthermore, in designing and implementing an AI agent capable of simulating the internet media platform usage behavior of a target group, the following are included: For web-based simulation, browser automation control tools or browser plugins are used to enable artificial intelligence to simulate human internet media platform usage behavior. Browser automation control tools and browser plugins include, but are not limited to, Webdriver and Puppeteer. For mobile client simulation, mobile terminal automation control tools are used to enable artificial intelligence agents to simulate human internet media platform usage behavior. Mobile terminal automation control tools include, but are not limited to, Appium and Auto.js. The AI ​​entity has a simulation mode and a recording mode: in simulation mode, it simulates the internet media platform usage behavior of the target group based on preset parameters; in recording mode, it records its own behavior and platform content information. The behavior logic and operating mode of the artificial intelligence agent are written into executable programs using Java, JavaScript, and Python programming languages, placed in the runtime module for execution, and the running status is monitored and managed.

[0039] Furthermore, when setting the activity parameters of the AI ​​agent, the following are included: Activity time parameters include: setting the activity time distribution Q(t) and activity interval distribution P(t) based on the activity patterns of the target group. t); Interaction rule parameters include: Based on the frequency of access to different content source pages by the target group, the content source page is selected based on a preset probability P(source); Based on the content themes of interest to the target audience, keywords are set. Based on a preset probability P (click), the AI ​​agent randomly clicks on a piece of content on the current page to enter the details page and perform browsing behavior, depending on whether the content title or tag contains the specified keywords. If no matching information is found, the operation is skipped. On the details page, the AI ​​agent determines whether to like a post based on a preset probability P (like), and whether to follow the content publisher based on a preset probability P (follow). The content source pages include: recommendation pages, content category pages, and search pages; Save the AI ​​agent's activity parameters as a configuration file in text format such as JSON or XML, and then place the file into the AI ​​agent's running module.

[0040] In step S3 above, when running the AI ​​agent to simulate the target group's internet media platform usage behavior, the following is included: Activate the simulation mode of the AI ​​agent to enable the AI ​​agent to continuously simulate the target group's internet media platform usage behavior; The behavior of AI agents in simulation mode includes: Log in to the internet media platform account linked to the AI ​​entity; The AI ​​agent enters the recommendation page of an internet media platform and scans the page content according to a preset keyword list. When it finds that the title or summary of a content contains keywords, the AI ​​agent randomly selects a content that meets the criteria and clicks on it to enter the details page of that content. On the content category page, the AI ​​agent selects a corresponding content category from the preset list of content categories; after entering the category page, the AI ​​agent browses the content on the page and randomly clicks on one of the contents to enter the details page; The AI ​​agent enters the search page and inputs preset keywords to search; on the search results page, the AI ​​agent randomly selects a result and clicks on it to enter the details page of that content; the selection of search keywords is also based on the analysis of the target group's search behavior to simulate the target group's process of actively searching for information; Once the AI ​​entity enters the details page, it will perform actions such as liking and following the content publisher based on preset probability values. During operation, the AI ​​agent will receive control commands from the AI ​​agent management module, including: Pause the AI ​​agent's task, suspend the AI ​​agent's task, restart the AI ​​agent, and modify the AI ​​agent's activity parameters.

[0041] In step S4 above, recording the content information viewed by the AI ​​agent on the recommendation page includes: Once the preset recording trigger conditions are met, the recording mode of the AI ​​agent is activated; The behavior of AI agents in recording patterns includes: Browsing Recommended Pages: After entering recording mode, the AI ​​entity browses the recommended pages according to its preset browsing rules; Text information storage: During the browsing of the recommendation page, the AI ​​agent extracts and saves the text information contained in the content appearing on the page; Result file generation and return: The saved text information is converted into a result file in text format such as JSON, and the result file is returned to the AI ​​entity management module through the content information upload channel between the AI ​​entity running module and the AI ​​entity management module.

[0042] In step S5 above, When calculating and outputting the content homogeneity of a target group using internet media platforms, the following are included: Based on the content information returned by a single AI agent in the AI ​​agent operation module, calculate the content homogeneity of the AI ​​agents; Based on all AI agents, calculate the content homogeneity of the target group; Optionally, some AI agents can be selected based on attribute features, and the content homogeneity of subgroups within the target group can be calculated; The content homogeneity calculation method based on a single AI agent is as follows: the content information returned by the AI ​​agent is semantically encoded, and the content is divided into multiple semantic categories through a clustering algorithm. Then, the proportion of each category is counted, and the semantic entropy is calculated based on the category distribution as the content homogeneity of the AI ​​agent. The method for calculating the content homogeneity of the target group is as follows: based on the semantic category distribution of each AI agent, calculate the content distribution difference between it and the average distribution of the group, and take the average of the difference results of all AI agents as the content homogeneity of the target group.

[0043] Specifically, based on a single AI agent in the AI ​​agent operation module. i The returned content information is used to calculate the artificial intelligence agent. i Content homogeneity includes the following steps: For artificial intelligence entities i For each piece of content returned, a text semantic vector is extracted using a pre-trained language model; Cluster all semantic vectors using the DBSCAN method, dividing all content information into... m One semantic category; Analyze the frequency of content occurrences in each semantic category and calculate the proportion of each category. ; Based on the probability distribution of each semantic category, semantic entropy is calculated as the content homogeneity of the AI ​​agent. The calculation formula is as follows:

[0044] in, For artificial intelligence entities i The returned content information is homogeneous. m This represents the total number of categories after semantic clustering. For the first j The proportion of content items in a semantic category to the total number of content items, that is, the number of content items in that category divided by the total number of content items returned by the AI ​​agent.

[0045] Specifically, based on all artificial intelligence agents i Calculating the content homogeneity of the target group includes the following steps: For each AI agent i The returned content information is used to construct its content distribution vector based on the calculated semantic category set: ,in m Number of content categories; Calculate the average vector of content distribution across all AI agents. The calculation formula is:

[0046] based on With each The Jensen-Shannon divergence between the content distribution of all AI agents and the overall average distribution is calculated, and the average value is taken as the content homogeneity index of the target group. The calculation formula is as follows:

[0047] in, This indicates that the target group's content is homogeneous. For the first i Content distribution of individual AI agents The average content distribution of all AI agents. Indicates the first i The Jensen-Shannon divergence between the content distribution of an individual AI agent and the overall average content distribution is calculated using the following formula:

[0048] in, , .

[0049] Specifically, by filtering AI agents based on attribute features, the content homogeneity of subgroups within the target group can be calculated: Determine the set of attribute features for the subgroup. The subdivided set of attribute features includes the AI ​​agent's geographic location attribute g and the AI ​​agent's age attribute a; Based on the set of attribute features, AI agents that meet the specific attribute features are selected from the deployed AI agent cluster. The content homogeneity of the selected AI agents is then calculated for the subgroups. ,include: The content homogeneity of a specific group within a geographically defined region; Content homogeneity of a specific age group or age range segment; The calculated content homogeneity results are transmitted to the content homogeneity query module.

[0050] In practical applications, this embodiment will describe the process of measuring the content homogeneity of a target group on the Weibo platform (weibo.com); Determining the attributes and characteristics of the target group includes at least the following steps: Step S100: Determine the group attribute characteristics of the target group, including: Step S110: Determine the user profile of the target group. In this embodiment, the user profile of the target group is determined as: young women living in first-tier cities who are interested in fashion, health, or lifestyle content.

[0051] Step S120: Based on the user profile of the target group, determine the specific demographic characteristics of the target group, including: gender, age, geographic location, and active time.

[0052] In this embodiment, based on the user profile of the target group, the gender attribute is determined to be female; the age attribute is determined to be 14 to 35 years old; the geographical location attribute is determined to be a first-tier city, namely Beijing, Shanghai, Shenzhen or Guangzhou; since the user profile based on the target group does not contain clues about the active time, the active time attribute is determined based on the average active time of Weibo platform users.

[0053] Based on the research results of this embodiment, the peak activity time for Weibo platform users begins between 8:00 AM and 10:00 AM and ends between 9:00 PM and 10:00 PM, with relatively even market activity during this time period. Therefore, the activity time distribution is determined. ,in That is, the probability of a user using the Weibo platform at time point t (in hours) is... .

[0054] Based on the research results of this embodiment, a power-law distribution characteristic exists between two user activities, as shown in Formula 1 below: (1) In the formula, t is the time difference between two user activities, and P is the probability density function. This represents the power-law distribution coefficient. Based on the research results, this embodiment selects... .

[0055] The above activity time distribution Q(t) and activity interval distribution P(t) are used to calculate the activity time distribution Q(t) and activity interval distribution P(t). t) is used as an attribute of the target group's active time.

[0056] Step S130: Based on the user profile of the target group, determine the content preference attribute characteristics of the target group, including: content category attributes and keyword attributes.

[0057] Content preference attributes include two dimensions: content category attributes and keyword attributes. The content category attributes include content categories such as current affairs, sports, entertainment, technology, military, games, and health.

[0058] In this embodiment, based on the user profile of the target group, the content preference attribute is determined to be fashion, health, or lifestyle.

[0059] Keywords include: keywords related to specific interest areas, keywords based on geographic location, and keywords related to gender.

[0060] In this embodiment, a keyword thesaurus containing 28 words was developed based on the user profile of the target group and in combination with relevant content categories.

[0061] Step S140: Based on the initial account information table compiled according to the target group's attribute characteristics, generate the username of each Weibo account according to the target group's language habits.

[0062] Step S200: Deploy an AI agent cluster based on the group attribute characteristics of the target group.

[0063] A key feature of this embodiment is that it uses an artificial intelligence agent to replace the survey of real users in traditional methods, and obtains the recommended content presented by the Internet media platform to the corresponding user group in the process of use in a simulated form.

[0064] The AI ​​agent aims to simulate the usage behavior of real users on internet media platforms, and to closely approximate the experience of real users on internet media platforms.

[0065] In this embodiment, step S200, which involves deploying an AI agent cluster based on the group attribute characteristics of the target group, includes the following steps: Step S210: Set up the registration information for the Internet media platform account, including: nickname, gender, age, IP address, content preferences, and initial followed users.

[0066] In this embodiment, Weibo accounts are registered in batches using mobile phone numbers. During the Weibo account registration process, demographic information such as gender, age, and geographical location, as well as preferred content categories, are set according to the results of step S140. Based on the content category settings, 3 to 5 related accounts are randomly followed.

[0067] Step S220: Design and implement an artificial intelligence agent capable of simulating the internet media platform usage behavior of the target group.

[0068] In this embodiment, the mobile terminal automation control tool Appium is used to enable the artificial intelligence agent to simulate human internet media platform usage behavior. The artificial intelligence agent achieves automated control of the Weibo platform through Appium.

[0069] The designed and implemented artificial intelligence agent has two operating modes: simulation mode and recording mode.

[0070] Use programming languages ​​including but not limited to Java, JavaScript, and Python to save the behavioral logic and operating mode of the AI ​​agent as an executable program, and place it in the AI ​​agent's running module for execution.

[0071] Step S230: Bind the internet media platform account to the AI ​​entity.

[0072] The AI ​​agent is randomly matched with the Weibo account registered in step S210.

[0073] Step S240: Set the activity time parameters of the artificial intelligence agent, including: activity time distribution Q(t) and activity interval distribution P(t). t); Step S250: Set the interaction rule parameters of the AI ​​agent, including the probability P(source) of selecting the source page, the probability P(click) of whether to click the content, the probability P(like) of whether to like, and the probability P(follow) of whether to follow.

[0074] like Figure 2 As shown, the behaviors of the AI ​​in simulation mode include browsing, liking, and following.

[0075] First, select one content source page based on a preset probability P(source). Source pages include: recommendation pages, content category pages, and search pages.

[0076] Second, based on a preset probability P (click), depending on whether the content title or tag contains a specified keyword, the AI ​​agent randomly clicks on a piece of content on the current page to enter the details page and perform browsing behavior; if no matching information is found, the operation is skipped. Third, on the details page, the AI ​​agent determines whether to like the post based on a preset probability P (like), and whether to follow the content publisher based on a preset probability P (follow). In this embodiment, the activity parameters of the aforementioned AI agent are saved as a configuration file in text format such as JSON or XML, and the file is placed into the AI ​​agent's running module.

[0077] Step S300: Run an artificial intelligence agent to simulate the target group's internet media platform usage behavior.

[0078] Step S310: Activate the simulation mode of the AI ​​agent to continuously simulate the target group's Weibo platform usage behavior, including: With human assistance, the registered Weibo account is logged into the matched AI entity.

[0079] Browse the Weibo recommendation page, click on the content with the displayed keywords to enter the details page; On the Weibo content category page, select the corresponding content category, browse the content under the content category, and randomly click on the content to enter the details page; On the Weibo search page, search for keywords and click on the content in the search results to enter the details page; On the Weibo profile page, users can like posts and follow content creators.

[0080] Activity parameters of the agent.

[0081] Step S320: Issue control commands to the artificial intelligence agent.

[0082] For the AI ​​agents in the AI ​​agent operation module, the controller can issue control commands to the AI ​​agents through the AI ​​agent management module, including: pausing the AI ​​agent's tasks; terminating the AI ​​agent's tasks; restarting the AI ​​agent; and modifying the AI ​​agent's activity parameters.

[0083] Step S400: Record the content information viewed by the AI ​​agent on the recommendation page.

[0084] Step S410: Activate the recording mode of the AI ​​agent. The AI ​​agent's behavior in recording mode is as follows: Figure 3 As shown, in this mode, the AI ​​agent browses the Weibo recommendation page and records the text content of the Weibo posts that appear on the page, as well as the publisher, posting time, number of likes, number of comments, and number of reposts.

[0085] Step S420: For Weibo content that enters the details page, simultaneously record the popular comments of that Weibo post, as well as the publisher, posting time, and number of likes of the comments.

[0086] In this embodiment, instead of acquiring all the content information required to calculate the homogeneity of content in a single recording process, the duration and frequency of the recorded content information are determined according to the user's active time attribute, and the information is acquired in multiple sessions.

[0087] Step S430: Convert the content information into a result file in text format such as JSON, and return the result file to the AI ​​agent management module through the content information upload channel between the AI ​​agent running module and the AI ​​agent management module; after receiving the result file, the AI ​​agent management module saves the data.

[0088] Step S440: Repeat the above steps until the AI ​​agent has acquired 500 pieces of content information for content homogeneity calculation.

[0089] Step S500, based on content information, calculate and output the content homogeneity of the target group when using internet media platforms, including the following steps: Step S510, based on the content information returned by a single AI agent i in the AI ​​agent running module, calculate the content homogeneity of AI agent i, including the following steps: Step S511: For each piece of content information returned by AI agent i, extract the text semantic vector using the pre-trained language model "Chinese BERT"; Step S512: Cluster all semantic vectors using the "DBSCAN" method to divide all content information into m semantic categories; Step S513: Count the frequency of content occurrence in each semantic category and calculate the proportion of each category. ; Step S514: Based on the probability distribution of each semantic category, calculate the semantic entropy as the content homogeneity of the AI ​​agent. The calculation formula is as follows: (2) In the formula, Let m represent the content homogeneity of the information returned by AI agent i, and m be the total number of categories after semantic clustering. This represents the proportion of content items in the j-th semantic category to the total number of content items, which is the number of content items in that category divided by the total number of content items returned by the AI ​​agent.

[0090] Step S520, based on all AI agents i, calculate the content homogeneity of the target group, including the following steps: Step S521: For the content information returned by each AI agent i, construct its content distribution vector according to the semantic category set in step S512: ,in m Number of content categories; Step S522: Calculate the average vector of the semantic distribution of all AI agents. The calculation formula is: (3) Step S523, based on With each The Jensen-Shannon divergence between the content distribution of all AI agents and the overall average distribution is calculated, and the average value is taken as the content homogeneity index of the target group. The calculation formula is as follows: (4) In the formula, This indicates that the target group's content is homogeneous. For the first i Content distribution of individual AI agents The average content distribution of all AI agents. Indicates the first i The Jensen-Shannon divergence between the content distribution of an individual AI agent and the overall average content distribution is calculated using the following formula: (5) in, , .

[0091] Optionally, based on attribute features, some AI agents can be selected, and the content homogeneity of subgroups within the target group can be calculated, including the following steps: Step S530: Determine the set of attribute features for the subdivided groups. The subdividable attribute features include: The geographic location attribute (g) of the AI ​​agent, such as one or more cities like Beijing, Shanghai, Shenzhen, Guangzhou, and Hangzhou; The AI ​​agent's age attribute (a), such as one or more specific ages or age ranges, such as 20 years old, 30 years old, 30 to 35 years old.

[0092] Step S540: Select AI agents that meet the subdivided attribute characteristics, and calculate the content homogeneity of the subdivided groups based on some AI agents. ,include: Geographic location represents the homogeneity of content among specific groups within a particular region, for example... This indicates the homogeneity of content recommended by AI-powered content platforms for accounts located in Shanghai within the target audience.

[0093] The content homogeneity of a specific age group or age range segment, for example... This indicates the homogeneity of content recommended by AI-powered content platforms targeting users aged 20 and older.

[0094] The above conditions can be set simultaneously, for example... This indicates the homogeneity of content recommended by AI-powered content platforms for users whose accounts are located in Shenzhen and are 30 years old, within the target audience.

[0095] Step S550: The calculation results in the content homogeneity calculation module are transmitted to the content homogeneity query module.

[0096] Based on the content of this embodiment, the homogeneity of content across multiple internet media platforms for a target group can be measured by extending the implementation process.

[0097] Based on this embodiment (which describes the process of measuring the content homogeneity of a target group on the Weibo platform (weibo.com), the process of applying this invention on multiple Internet media platforms is demonstrated by adding the Douyin short video platform.

[0098] Based on the process of content homogenization on the Weibo platform (weibo.com), the following modifications are made: In step S200 above, an AI agent is added to simulate the operation logic of a real user using the Douyin short video platform, and the account registration process and account binding process of the Douyin short video platform are added. In step S300 above, the AI ​​agent is run on the Douyin short video platform; In step S400 above, a calculation of the homogeneity of content on the Douyin short video platform is added; Content information from the Douyin short video platform can be merged into content information from the Weibo platform to calculate the content homogeneity of the target group across multiple internet media platforms.

[0099] Alternatively, the content homogeneity of the two platforms can be calculated separately using a formula, for example... and .

[0100] By setting weight coefficients for the two platforms and Weighted calculation of content homogeneity across multiple platforms As shown in Formula 6: (6) like Figure 4 As shown, this embodiment also provides a device for measuring content homogeneity in an internet media platform, including: AI Entity Operation Module 1 is used to deploy and execute multiple AI entities that can run independently in a distributed manner, and to provide the network access environment required for these AI entities to run. In this embodiment, the AI ​​agent operation module 1 is implemented by a distributed cloud host, and its features include: Basic hardware: Quad-core CPU, 8GB RAM, 100GB SSD; Operating system: Ubuntu 22.04LTS 64-bit; Runtime environment: OpenJDK 17, Firefox ESR 115 and Java execution framework for AI agents; Network configuration: 2Mbps bandwidth; Independent public IPv4 address.

[0101] In this embodiment, 50 AI agent operation module instances are configured and deployed in a distributed architecture, specifically distributed as follows: 12 units in Shenzhen, 12 units in Hangzhou, 8 units in Chongqing, 10 units in Xi'an, and 8 units in Shenyang. Each AI agent operation module 1 is interconnected via a BGP multi-line network, but the operating environments of each unit are isolated from each other.

[0102] The AI ​​agent management module 2 is used to receive the operation data and behavior logs of the AI ​​agent in the AI ​​agent operation module 1, and to issue scheduling commands to the AI ​​agent operation module 1; wherein, the AI ​​agent management module 2 includes: a data storage unit 6 for storing data and a policy scheduling unit 5 for receiving and issuing scheduling commands.

[0103] In this embodiment, the data storage unit 6 is implemented by a cloud database, and its features include: a 4-core central processing unit, 8GB of RAM, and 800GB of solid-state storage; operating system: Ubuntu 22.04 LTS 64-bit version; database system: MySQL 64-bit version; network configuration: internal transmission bandwidth of 2500Mbps.

[0104] In this embodiment, the policy scheduling unit 5 is implemented by a distributed cloud host, characterized by: a 4-core central processing unit, 8GB of RAM, and 100GB of solid-state storage; system environment: Ubuntu 22.04 LTS 64-bit version; control software: policy scheduling unit software implementation; network configuration: 2Mbps transmission bandwidth.

[0105] Content homogeneity calculation module 3 is used to read data in a targeted manner and calculate the homogeneity index of content in Internet media platforms; In this embodiment, the content homogeneity computing module 3 is implemented through a distributed cloud host, and its features include: basic hardware: 4-core computing unit, 8GB RAM, 100GB solid-state storage; operating system: Ubuntu 22.04 LTS 64-bit version; control software: software implementation of the content homogeneity computing module 3; network configuration: 2Mbps transmission bandwidth.

[0106] Content homogeneity query module 4 is used for interactive querying of the calculation results of content homogeneity.

[0107] In this embodiment, the content homogeneity query module 4 is implemented through a distributed cloud host, and its features include: basic hardware: 4-core computing unit, 8GB RAM, 100GB solid-state storage; operating system: Ubuntu 22.04 LTS 64-bit version; control software: software implementation of the content homogeneity query module 4; network configuration: 20Mbps transmission bandwidth.

[0108] In this embodiment, except for the internal signal path of the AI ​​agent management module 2, the signal paths between the modules are via the public Internet, including: a content information upload channel and an AI agent control command transmission channel between the multiple AI agent operation modules 1 and the multiple AI agent management modules 2; a query command transmission channel and a data result feedback channel between the content homogeneity calculation module 3 and the AI ​​agent management module 2; a query command transmission channel and a content homogeneity result feedback channel between the content homogeneity query module 4 and the content homogeneity calculation module 3; and a bidirectional channel within the AI ​​agent management module 2 via a local area network line.

[0109] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0110] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A method for measuring content homogeneity in an internet media platform, characterized in that, include: Determine the group attributes of the target group, and combine the activity time distribution Q(t) and activity interval distribution P(t). t) serves as the active time attribute of the target group; among which, , t is the time difference between two user activities, and P is the probability density function. The coefficients are power-law distribution coefficients. Based on the group attribute characteristics of the identified target group, deploy an artificial intelligence agent cluster in the Internet media platform environment; Running an artificial intelligence agent to simulate the internet media platform usage behavior of a target group; During the process of simulating the behavior of AI agents, the content information that each AI agent browses on the recommended pages of Internet media platforms is recorded; Based on content information, the content homogeneity of the target group when using Internet media platforms is calculated and output; among them, the content homogeneity of the AI ​​agent is calculated based on the content information returned by a single AI agent; and the content homogeneity of the target group is calculated based on all AI agents.

2. The method according to claim 1, characterized in that, A single AI agent in the AI ​​agent operation module i The returned content information is used to calculate the artificial intelligence agent. i Content homogeneity includes the following steps: For artificial intelligence entities i For each piece of content returned, a text semantic vector is extracted using a pre-trained language model; Cluster all semantic vectors to divide all content information into m One semantic category; Analyze the frequency of content occurrences in each semantic category and calculate the proportion of each category. ; Based on the probability distribution of each semantic category, semantic entropy is calculated as the content homogeneity of the AI ​​agent. The calculation formula is as follows: in, For artificial intelligence entities i The returned content information is homogeneous. m This represents the total number of categories after semantic clustering. For the first j The proportion of content items in a semantic category to the total number of content items, that is, the number of content items in that category divided by the total number of content items returned by the AI ​​agent.

3. The method according to claim 1, characterized in that, Based on all AI agents i Calculating the content homogeneity of the target group includes the following steps: For each AI agent i The returned content information is used to construct its content distribution vector based on the calculated semantic category set: ,in m Number of content categories; Calculate the average vector of content distribution across all AI agents. The calculation formula is: based on With each The Jensen-Shannon divergence between the content distribution of all AI agents and the overall average distribution is calculated, and the average value is taken as the content homogeneity index of the target group. The calculation formula is as follows: in, This indicates that the target group's content is homogeneous. For the first i Content distribution of individual AI agents The average content distribution of all AI agents. Indicates the first i The Jensen-Shannon divergence between the content distribution of an individual AI agent and the overall average content distribution is calculated using the following formula: 。 4. The method according to claim 3, characterized in that, Also includes: Based on attribute features, partial AI entities can be selected, and the content homogeneity of subgroups within a target group can be calculated: Determine the set of attribute features for the subgroup. This attribute feature set This includes the AI ​​agent's geographic location attribute g and the AI ​​agent's age attribute a; Based on attribute feature set From the deployed AI agent cluster, AI agents that meet the specific attribute characteristics are selected, and the content homogeneity of the segmented group is calculated based on the selected AI agents. ,include: The content homogeneity of a specific group within a geographically defined region; Content homogeneity of a specific age group or age range segment; The calculated content homogeneity results are transmitted to the content homogeneity query module.

5. The method according to claim 1, characterized in that, Deploying AI agent clusters in an internet media platform environment includes: Based on the group attributes of the target group, the system generates account registration information for the AI ​​agent on internet media platforms. The nickname is generated according to the language habits of the target group; gender and age are set according to the proportion of the target group; the IP address is selected from the address corresponding to the region of the target group; the content preference is determined by data mining of the target group's frequently interacted content; and the initial followers are selected from accounts that the target group generally follows. Design and implement an artificial intelligence entity capable of simulating the internet media platform usage behavior of a target group; Develop an account binding interface to link internet media platform accounts with AI entities; Set the activity parameters of the artificial intelligence agent.

6. The method according to claim 5, characterized in that, When designing and implementing an AI agent capable of simulating the internet media platform usage behavior of a target group, the following should be included: For web-based simulation, browser automation control tools or browser plugins are used to enable artificial intelligence to simulate human internet media platform usage behavior. Browser automation control tools and browser plugins include at least Webdriver and Puppeteer. For mobile client simulation, mobile terminal automation control tools are used to enable artificial intelligence agents to simulate human internet media platform usage behavior. Mobile terminal automation control tools include at least Appium and Auto.js. The AI ​​entity has a simulation mode and a recording mode: in simulation mode, it simulates the internet media platform usage behavior of the target group based on preset parameters; in recording mode, it records its own behavior and platform content information. The behavioral logic and operating mode of the AI ​​agent are written into an executable program, placed in the runtime module for execution, and the running status is monitored and managed.

7. The method according to claim 5, characterized in that, When setting the activity parameters of an AI agent, the following are included: Activity time parameters include: setting the activity time distribution Q(t) and activity interval distribution P(t) based on the activity patterns of the target group. t); Interaction rule parameters include: Based on the frequency of access to different content source pages by the target group, the content source page is selected based on a preset probability P(source); Based on the content themes of interest to the target audience, keywords are set. Based on a preset probability P (click), the AI ​​agent randomly clicks on a piece of content on the current page to enter the details page and perform browsing behavior, depending on whether the content title or tag contains the specified keywords. If no matching information is found, the operation is skipped. On the details page, the AI ​​agent determines whether to like a post based on a preset probability P (like), and whether to follow the content publisher based on a preset probability P (follow). The content source pages include: recommendation pages, content category pages, and search pages; Save the AI ​​agent's activity parameters as a configuration file in JSON or XML text format, and then place the configuration file into the AI ​​agent's running module.

8. The method according to any one of claims 1 to 7, characterized in that, When running an AI simulation of a target group's internet media platform usage behavior, including: Activate the simulation mode of the AI ​​agent to enable the AI ​​agent to continuously simulate the target group's internet media platform usage behavior; The behavior of AI agents in simulation mode includes: Log in to the internet media platform account linked to the AI ​​entity; The AI ​​agent enters the recommendation page of an internet media platform and scans the page content according to a preset keyword list. When it finds that the title or summary of a content contains keywords, the AI ​​agent randomly selects a content that meets the criteria and clicks on it to enter the details page of that content. On the content category page, the AI ​​agent selects a corresponding content category from the preset list of content categories; after entering the category page, the AI ​​agent browses the content on the page and randomly clicks on one of the contents to enter the details page; The AI ​​agent enters the search page and inputs preset keywords to search; on the search results page, the AI ​​agent randomly selects a result and clicks on it to enter the details page of that content; the selection of search keywords is also based on the analysis of the target group's search behavior to simulate the target group's process of actively searching for information; Once the AI ​​entity enters the details page, it will perform actions such as liking and following the content publisher based on preset probability values. During operation, the AI ​​agent will receive control commands from the AI ​​agent management module, including: Pause the AI ​​agent's task, suspend the AI ​​agent's task, restart the AI ​​agent, and modify the AI ​​agent's activity parameters.

9. The method according to any one of claims 1 to 7, characterized in that, When recording the content information viewed by the AI ​​agent on the recommendation page, this includes: Once the preset recording trigger conditions are met, the recording mode of the AI ​​agent is activated; The behavior of AI agents in recording patterns includes: Browsing Recommended Pages: After entering recording mode, the AI ​​entity browses the recommended pages according to its preset browsing rules; Text information storage: During the browsing of the recommendation page, the AI ​​agent extracts and saves the text information contained in the content appearing on the page; Result file generation and return: The saved text information is converted into a JSON text format result file, and the result file is returned to the AI ​​entity management module through the content information upload channel between the AI ​​entity running module and the AI ​​entity management module.

10. A device for measuring content homogeneity in an internet media platform, characterized in that, include: The AI ​​agent operation module is used to deploy and execute multiple AI agents that can run independently in a distributed manner, and to provide the network access environment required for these AI agents to run. The AI ​​agent management module is used to receive the operation data and behavior logs of the AI ​​agents in the AI ​​agent operation module, and to issue scheduling commands to the AI ​​agent operation module. The content homogeneity calculation module is used to read data in a targeted manner and calculate the homogeneity index of content in Internet media platforms. The content homogeneity query module is used for interactive querying of the calculation results of content homogeneity. Among them, the multiple AI entity operation modules and the multiple AI entity management modules have a content information upload channel and an AI entity control command transmission channel; The content homogeneity calculation module and the artificial intelligence entity management module have a query instruction transmission channel and a data result feedback channel; The content homogeneity query module and the content homogeneity calculation module have a query instruction transmission channel and a content homogeneity result feedback channel.

Images