Content recommendation strategy detection method and device, equipment, medium and product

Abstract

The invention discloses a content recommendation strategy detection method and device, equipment, a medium and a product, and relates to the field of recommendation system detection.The method comprises the steps that a robot process automation technology is adopted to publish a preset task to a crowdsourcing platform to determine a plurality of participants; after each participant performs account login on the to-be-detected platform, a robot process automation technology is adopted to simulate operation of each participant on the to-be-detected platform to complete a preset task, and multiple pieces of data of each participant on the to-be-detected platform are obtained; performing data vectorization processing on the acquired data to obtain vectors; aggregating the vectors to obtain feature vectors; integrating the user ID data, the recommendation object ID data and the feature vectors; and inputting the integrated feature vector of each participant into the content recommendation strategy detection model to obtain the content recommendation strategy of the to-be-detected platform, so that the method does not depend on manual intervention, a lot of manpower and time are saved, and complex recommendation strategies of a large-scale platform and cross-platform application can be processed.

Description

Technical Field

[0001] This application relates to the field of recommendation system detection, and in particular to a method, apparatus, equipment, medium, and product for detecting content recommendation strategies. Background Technology

[0002] With the continuous development of technology, while providing convenience to people, recommendation systems have also brought about problems such as information cocoons and price discrimination based on big data. These problems pose a huge challenge to the detection of recommendation strategies. Most existing content recommendation strategy detection relies on manual intervention, which consumes a lot of manpower and time. Manual detection is difficult to handle the complex recommendation strategies of large-scale platforms and cross-platform applications. Summary of the Invention

[0003] The purpose of this application is to provide a content recommendation strategy detection method, apparatus, device, medium, and product that does not rely on manual intervention, saving a lot of manpower and time, and can also handle complex recommendation strategies of large-scale platforms and cross-platform applications.

[0004] To achieve the above objectives, this application provides the following solution:

[0005] Firstly, this application provides a method for detecting content recommendation strategies, including:

[0006] Robotic process automation (RPA) technology is used to publish pre-set tasks to a crowdsourcing platform to identify multiple participants.

[0007] After logging into their accounts on the platform to be tested, each participant uses robotic process automation (RPA) technology to simulate their operations on the platform to complete preset tasks. During the simulation, the system obtains user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform to be tested.

[0008] The browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform to be tested are processed into data vectors to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on the platform to be tested.

[0009] For any participant, the feature vector of the participant is obtained by aggregating the browsing record vector, operation record vector, recommended content vector, user basic information vector and personal habit information vector of the participant on the platform to be detected.

[0010] The integrated feature vector of the participant is obtained by integrating the participant's user ID data and recommendation object ID data on the platform to be detected, as well as the participant's feature vector.

[0011] The integrated feature vectors of each participant during the simulation are input into the content recommendation strategy detection model to obtain the content recommendation strategy of the platform to be detected.

[0012] Optionally, the browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform to be tested are processed into data vectors to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on the platform to be tested, specifically including:

[0013] For any participant, perform Log bucketing embedding operation on the participant's browsing record data and operation record data on the platform to be detected to obtain the participant's browsing record vector and operation record vector on the platform to be detected.

[0014] One-Hot embedding operation is performed on the participant's basic user information data and personal habit information data on the platform to be detected to obtain the participant's basic user information vector and personal habit information vector on the platform to be detected.

[0015] Text embedding is performed on the recommended content data of the participant on the platform to be detected to obtain the recommended content vector of the participant on the platform to be detected.

[0016] Optionally, the content recommendation strategy detection model is obtained by training a multilayer perceptron model.

[0017] Optionally, the specific process of training a multilayer perceptron model includes:

[0018] Robotic process automation (RPA) technology is used to publish pre-set sample tasks to a crowdsourcing platform to identify multiple sample participants.

[0019] After logging into their accounts on multiple sample platforms, participants in each sample were simulated using robotic process automation (RPA) technology to complete preset sample tasks. During the simulation, user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on each sample platform were obtained.

[0020] The browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on each sample platform are processed into data vectors to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on each sample platform.

[0021] For any sample participant on any sample platform, the browsing record vector, operation record vector, recommended content vector, user basic information vector, and personal habit information vector of the sample participant on the sample platform are aggregated to obtain the feature vector of the sample participant on the sample platform.

[0022] The integrated feature vector of the sample participants is obtained by integrating the feature vectors, user ID data, and recommendation object ID data of the sample participants on the sample platform.

[0023] A content recommendation strategy detection model is obtained by training a multilayer perceptron model based on the integrated feature vectors of each participant on each sample platform and the content recommendation strategies of each sample platform.

[0024] Optionally, text embedding is performed on the recommended content data of the participants on the platform to be detected, specifically including:

[0025] The BETR model is used to process the recommended content data of the participants on the platform to be detected to obtain text embedding vectors;

[0026] The text embedding vector is processed by multi-layer perceptron processing to obtain the recommended content vector of the participant on the platform to be detected.

[0027] Optionally, the participant's browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector on the platform to be detected are aggregated to obtain the participant's feature vector, specifically including:

[0028] The participant's browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector on the platform to be detected are aggregated to obtain the participant's aggregated feature vector.

[0029] The aggregated feature vectors of the participants are weighted using an attention mechanism to obtain the feature vectors of the participants.

[0030] Secondly, this application provides a content recommendation strategy detection device, comprising:

[0031] The participant identification module is used to identify multiple participants by publishing a pre-set task to a crowdsourcing platform using robotic process automation technology.

[0032] The simulation module is used to simulate the operations of each participant on the platform to be tested after the participants log in to their accounts. It uses robotic process automation technology to simulate the operations of each participant on the platform to be tested to complete preset tasks and obtain user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data and personal habit information data of each participant on the platform to be tested during the simulation operation.

[0033] The data vectorization processing module is used to perform data vectorization processing on the browsing record data, operation record data, recommended content data, user basic information data and personal habit information data of each participant on the platform to be tested, to obtain the browsing record vector, operation record vector, recommended content vector, user basic information vector and personal habit information vector of each participant on the platform to be tested.

[0034] The aggregation module is used to aggregate the browsing record vector, operation record vector, recommended content vector, user basic information vector, and personal habit information vector of any participant on the platform to be detected, so as to obtain the feature vector of the participant.

[0035] The integration module is used to integrate the user ID data and recommendation object ID data of the participant on the platform to be detected, as well as the feature vector of the participant, to obtain the integrated feature vector of the participant;

[0036] The content recommendation strategy detection module is used to input the integrated feature vectors of each participant during the simulation operation into the content recommendation strategy detection model to obtain the content recommendation strategy of the platform to be detected.

[0037] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the content recommendation strategy detection method described in any one of the above.

[0038] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the content recommendation strategy detection method described in any one of the above descriptions.

[0039] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the content recommendation strategy detection method described in any one of the above statements.

[0040] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0041] This application provides a method, apparatus, device, medium, and product for detecting content recommendation strategies. It employs Robotic Process Automation (RPA) technology to publish pre-set tasks to a crowdsourcing platform, identifying multiple participants. Each participant logs in to the platform under test, and RPA technology simulates their own operations on the platform to complete the pre-set tasks. During the simulated operation, it acquires user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data for each participant on the platform. RPA technology can simulate human operation, automatically executing a series of repetitive tasks without human intervention. Furthermore, RPA technology can seamlessly integrate across multiple platforms. Through the crowdsourcing platform, data from different user groups can be collected. This solves the problem that existing content recommendation strategy detection methods largely rely on manual intervention, consuming significant manpower and time. Manual detection struggles to handle complex recommendation strategies on large-scale platforms and cross-platform issues. This method eliminates the need for manual intervention, saving substantial manpower and time, and can also handle complex recommendation strategies on large-scale platforms and cross-platform applications. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments 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.

[0043] Figure 1 A flowchart illustrating a content recommendation strategy detection method provided in an embodiment of this application;

[0044] Figure 2 A framework diagram of a content recommendation strategy detection method provided in this application;

[0045] Figure 3 A flowchart for task assignment and distribution in robotic process automation technology provided for this application;

[0046] Figure 4 The robotic process automation technology for data acquisition provided in this application is illustrated in the flowchart.

[0047] Figure 5 This is a structural diagram of a multilayer perceptron model;

[0048] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0049] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0050] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] Robotic Process Automation (RPA) technology, as an advanced automation tool, has been widely applied. RPA technology can simulate human operation, automatically performing a series of repetitive tasks through software robots, making it particularly suitable for data collection, processing, and analysis scenarios. Using RPA, data collection and processing can be completed automatically without human intervention, greatly improving efficiency and accuracy. Furthermore, RPA technology can be seamlessly integrated across multiple platforms, providing technical support for comprehensive data collection and processing. On the other hand, crowdsourcing, as an effective method, can be used to widely collect user opinions and feedback on recommendation strategies. Through crowdsourcing platforms, opinions and data from different user groups can be collected, thereby increasing sample diversity and improving the efficiency and coverage of data collection. Questionnaire design can delve into user preferences, habits, and satisfaction, obtaining more comprehensive feedback on recommendation algorithms. After collecting this data, Natural Language Processing (NLP) technology is used for in-depth analysis to extract recommendation features from each platform. Based on these features, this application provides a content recommendation strategy detection method to construct a comprehensive evaluation system for Internet information service recommendation algorithms, fully reflecting the application of recommendation algorithms on different platforms.

[0052] In one exemplary embodiment, such as Figure 1 and Figure 2 As shown, a content recommendation strategy detection method is provided, including the following steps:

[0053] Step 201: Use robotic process automation (RPA) technology to publish the pre-set task to the crowdsourcing platform and identify multiple participants.

[0054] Step 202: After logging into their accounts on the platform to be tested, each participant uses robotic process automation (RPA) technology to simulate their operations on the platform to be tested and complete preset tasks. During the simulated operation, the RPA acquires user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform to be tested.

[0055] Step 203: Perform data vectorization on the browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform to be tested, to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on the platform to be tested.

[0056] Step 204: For any participant, aggregate the participant's browsing record vector, operation record vector, recommended content vector, user basic information vector, and personal habit information vector on the platform to be detected to obtain the participant's feature vector.

[0057] Step 205: Integrate the participant's user ID data and recommendation object ID data on the platform to be detected, as well as the participant's feature vector, to obtain the participant's integrated feature vector.

[0058] Step 206: Input the integrated feature vectors of each participant during the simulation into the content recommendation strategy detection model to obtain the content recommendation strategy of the platform to be tested. The content recommendation strategies include: relevance recommendation strategy, environment recommendation strategy, popularity recommendation strategy, and collaborative recommendation strategy (wherein the relevance recommendation strategy is based on the relevance between users and items, recommending suitable objects to users by analyzing the similarity between users' historical behavior (such as clicks, browsing, purchases, etc.) and content; the environment recommendation strategy provides personalized recommendations based on the user's current context (such as time, location, weather, etc.); the environment recommendation strategy recommends based on the popularity of a certain item or content; and the collaborative recommendation strategy recommends content to users by analyzing the similarity between users or between items). Input the integrated feature vectors of each participant during the simulation into the content recommendation strategy detection model to obtain the probabilities of the relevance recommendation strategy, environment recommendation strategy, popularity recommendation strategy, and collaborative recommendation strategy, and determine the recommendation strategy corresponding to the highest probability as the content recommendation strategy of the platform to be tested.

[0059] Implementing steps 201 to 206 above can eliminate the need for manual intervention, saving a significant amount of manpower and time. Furthermore, it can handle complex recommendation strategies for large-scale platforms and cross-platform applications.

[0060] In an exemplary embodiment, the browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform to be tested are processed into data vectors to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on the platform to be tested, specifically including:

[0061] For any participant, the browsing record data and operation record data of the participant on the platform to be detected are subjected to Log bucketing embedding operation to obtain the browsing record vector and operation record vector of the participant on the platform to be detected.

[0062] One-Hot embedding operation is performed on the participant's basic user information data and personal habit information data on the platform to be tested to obtain the participant's basic user information vector and personal habit information vector on the platform to be tested.

[0063] Text embedding is performed on the recommended content data of the participant on the platform to be detected to obtain the recommended content vector of the participant on the platform to be detected.

[0064] In an exemplary embodiment, the text embedding operation on the participant's recommended content data on the platform to be detected specifically includes:

[0065] The BETR model is used to process the recommended content data of the participants on the platform to be detected to obtain text embedding vectors.

[0066] The text embedding vector is processed by multi-layer perceptron processing to obtain the recommended content vector of the participant on the platform to be detected.

[0067] In an exemplary embodiment, the participant's browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector on the platform to be detected are aggregated to obtain the participant's feature vector, specifically including:

[0068] The participant's browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector on the platform to be detected are aggregated to obtain the participant's aggregated feature vector.

[0069] The aggregated feature vectors of the participants are weighted using an attention mechanism to obtain the feature vectors of the participants.

[0070] In one exemplary embodiment, the content recommendation strategy detection model is obtained by training a multilayer perceptron (MLP) model.

[0071] In an exemplary embodiment, the specific process of training a multilayer perceptron model includes:

[0072] Robotic process automation (RPA) technology is used to publish pre-set sample tasks to a crowdsourcing platform to identify multiple sample participants.

[0073] After logging into their accounts on multiple sample platforms, participants in each sample were simulated using robotic process automation (RPA) technology to complete preset sample tasks. During the simulation, user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on each sample platform were obtained.

[0074] The browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on each sample platform are processed into data vectors to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on each sample platform.

[0075] For any sample participant on any sample platform, the browsing record vector, operation record vector, recommended content vector, user basic information vector, and personal habit information vector of the sample participant on the sample platform are aggregated to obtain the feature vector of the sample participant on the sample platform.

[0076] The integrated feature vector of the sample participants is obtained by integrating the feature vectors, user ID data, and recommendation object ID data of the sample participants on the sample platform.

[0077] A content recommendation strategy detection model is obtained by training a multilayer perceptron model based on the integrated feature vectors of each participant on each sample platform and the content recommendation strategies of each sample platform.

[0078] In practical applications, RPA systems automate the entire process of task preparation, deployment, data collection, data processing, and storage. First, RPA automatically generates task templates (setting interests and browsing recommended web pages) and adjusts parameters (for example, for personalized recommendations on platforms like Chao News, RPA can set the number of interests users need to set and the number of recommended web pages they are required to browse, such as requiring users to select at least 5 interest tags and browse no fewer than 10 recommended web pages) to ensure the effectiveness of the task design. Then, the task is deployed to appropriate crowdsourcing platforms and assigned to suitable participants. Participants must log in to the platform and fill in necessary personal information. RPA then simulates the participants' operations on various platforms, collecting data corresponding to various attributes (user ID, recommended object ID, browsing history, operation history, recommended content, basic user information (such as gender, age, interests, etc.), and personal habit information). The collected data is then processed and structured and stored in a database. Finally, RPA generates a feedback report and optimizes the entire process. In one exemplary embodiment, robotic process automation (RPA) technology is used to publish a pre-defined sample task to a crowdsourcing platform to identify multiple sample participants. After logging into their accounts on multiple sample platforms, each participant uses RPA technology to simulate their operations on each platform to complete the pre-defined sample task. During the simulated operation, user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data for each participant on each sample platform are obtained, such as... Figure 4 As shown, the process includes automated task allocation, automated data collection and preprocessing, data verification and quality control, automated feedback and reward mechanisms, dynamic task optimization, and report generation and monitoring. The specific workflow is as follows:

[0079] S1.1: Initialization, i.e., the task preparation phase. First, task design and setup are automated via RPA. RPA can automatically generate task templates, including the task content, details, and data collection requirements. Then, RPA automatically adjusts task parameters, such as the collection scope and frequency, based on historical data or preset rules. Next, the incentive mechanism is designed and managed automatically by RPA. RPA can dynamically set rewards and optimize incentive effects through data analysis to ensure the task's attractiveness and execution efficiency.

[0080] S1.2: Task Issuance and Assignment. (Refer to...) Figure 3As shown, after the task (preset sample task) is prepared, RPA will automate the task publishing and allocation process. First, RPA will automatically break down the task into multiple subtasks, ensuring that these subtasks do not interfere with each other. Next, RPA will automatically publish these tasks on crowdsourcing platforms and search for and filter suitable crowdsourcing participants (sample participants). Through intelligent matching, RPA will assign the task to the most suitable workers (sample participants) and select the best platforms, such as Bilibili, YouTube, Weibo, and Xiaohongshu, so that sample participants can successfully log in remotely and participate in the task.

[0081] S1.3: Automatic User Data Collection. After task allocation, RPA automatically executes the data collection process. RPA simulates user operations on various platforms, executing each task. When simulating a user executing an assigned task, it sequentially executes all subtasks within that task (liking, browsing web pages), automatically logging in remotely and collecting account data according to a preset process. This includes: user ID data of each sample participant on multiple sample platforms, recommendation object ID data (such as the ID of the recommended post author), browsing history data (such as the titles of content in viewed or liked web pages), operation history data (such as likes, comments, reposts, and ratings), recommended content data (such as the number of likes, views, and posting time of recommended posts), basic user information data (such as gender, age, and interests), and personal habit information data (such as the user's personal geographical location information and interests). During the collection process, RPA automatically replaces sensitive information related to user identity with unique identifiers to protect user privacy. Simultaneously, RPA also performs preliminary data processing to ensure consistent formatting.

[0082] S1.4: Data Aggregation and Organization. Once data collection is complete, RPA will automatically collect data from each platform and aggregate it. RPA is responsible for organizing the recommendation information from each platform into a structured data format (the organization process involves labeling each data entry with its corresponding attributes), ensuring that all data can be stored and analyzed in a unified format. This process is fully automated, reducing manual intervention and improving the efficiency and accuracy of data processing.

[0083] S1.5: Data Storage and Management. After data processing is complete, RPA stores this structured data in a database or files. RPA automatically manages the data storage location, ensuring data security, accessibility, and durability. To further ensure data security, RPA also automatically performs data backup tasks and synchronizes data to multiple storage locations when necessary.

[0084] S1.6: Feedback and Optimization. Finally, RPA can automatically generate reports and analyze the effectiveness of the entire data acquisition and processing flow. RPA will automatically generate reports on task completion, data quality, participant performance, etc., and send these reports to system administrators or relevant parties. Through continuous feedback, RPA can also dynamically adjust and optimize task parameters, incentive mechanisms, and data processing flows to improve the overall efficiency and effectiveness of the system.

[0085] Because the data collected and processed by RPA includes browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data from different platforms, preprocessing is required to make this data usable as input for machine learning models. Therefore, in an exemplary embodiment, the browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on each sample platform are vectorized to obtain browsing history vectors, operation history vectors, recommended content vectors, user basic information vectors, and personal habit information vectors for each participant on each sample platform. These vectors are then aggregated to obtain the feature vectors of the sample participants on the sample platforms, specifically:

[0086] S2.1: The following three main preprocessing methods are used to convert this data (the participants' browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data on the sample platform) into input vectors usable by the model:

[0087] Log-based binning embedding for long-tail data: The collected data exhibits statistically significant quantitative characteristics such as the number of tasks completed and the number of page views, with their distribution largely exhibiting long-tail characteristics. Conventional equal-interval binning is insufficient to fully capture the features of long-tail data. Therefore, this application employs a log transformation method to bin these statistical data. Specifically, for long-tail distributed data such as operation record data and page view data, the Log-based binning embedding method reduces the impact of extreme values ​​on model training. The formula is shown below:

[0088] bucket = log(C feature (1)

[0089] emb bucket =onehot(bucket) (2)

[0090] Among them, C featureThis represents browsing or operation record data, which is mapped to a specific bucket through a log transformation. Finally, the bucket features are converted into model input emb using a one-hot encoding method. bucket .

[0091] One-Hot Embedding of Categorical Features: For some categorical features, such as clustering features based on user preferences with a class size of 10, where users with similar preferences are clustered into the same class, this application directly performs One-Hot embedding on these limited categorical features. That is, for categorical features such as user basic information data and personal habit information data, the One-Hot embedding method is used to convert them into sparse vector representations, which are easier for the model to process. The formula is shown below:

[0092] emb category =onehot(Category) (3)

[0093] Where Category represents basic user information or personal habit information, and emb represents... category This is the vector representation of the eigenvalue.

[0094] Text embedding based on Natural Language Processing (NLP) technology: For natural language text describing user information, in order to better understand the semantic information contained therein, this application uses pre-trained models such as BERT to output text embedding vectors. These pre-trained models benefit from a large amount of training data in the real world, and their output text embeddings can better reflect the semantic features between different texts. That is, for text data in recommendation content data, NLP technology is used to perform text embedding operations, converting it into dense vectors to preserve semantic information. The text embedding calculation formula is shown below:

[0095] emb bert =BERT(Text) (4)

[0096] emb text =MLP(emb bert (5)

[0097] Where Text represents the text data in the recommended content data, and emb bert This represents the text embedding vector output by BERT. Since the text embedding vectors output by pre-trained models have high dimensionality, this application utilizes a trainable MLP to... bert Dimensionality reduction is performed to obtain the text feature emb. bert .

[0098] S2.2: After obtaining the vector representations of various types of data (the browsing record vector, operation record vector, recommended content vector, user basic information vector, and personal habit information vector of the sample participants on the sample platform), considering that these features come from different data sources, it is necessary to merge these features into a multi-dimensional feature vector (the aggregated feature vector of the sample participants on the sample platform).

[0099] S2.3: The aggregated feature vectors of the sample participants on the sample platform obtained in S2.2 are weighted by an attention mechanism to obtain the feature vectors of the sample participants on the sample platform.

[0100] In an exemplary embodiment, the feature vectors, user ID data, and recommendation object ID data of the sample participants on the sample platform are integrated to obtain the integrated feature vector of the sample participants; a content recommendation strategy detection model is obtained by training a multilayer perceptron model based on the integrated feature vectors of each participant on each sample platform and the content recommendation strategy of each sample platform, specifically including:

[0101] S3.1: Construct a dataset that includes the integrated feature vectors of all sample participants across all sample platforms, as well as the content recommendation strategies for each sample platform. Integrate the user ID data and recommendation object ID data of the sample participants on each sample platform with the feature vectors obtained in S2.3 to obtain the integrated feature vector of the sample participants on each sample platform. The integrated feature vector should include user ID data, recommendation object ID data, and feature vectors, where the user ID data and recommendation object ID data are mapped to numerical indices, and the feature vectors are fixed-length numerical arrays.

[0102] S3.2: Initialize the MLP model parameters and set an appropriate learning rate.

[0103] S3.3: Training the model. The dataset is divided into small batches and fed into the model batch by batch for training. The input data is then processed by the model to calculate the predicted value. The loss is calculated using the cross-entropy loss function based on the predicted value and the actual value (the content recommendation strategy of the sample platform). The gradient is calculated using the backpropagation algorithm, and the model parameters are updated using the Adam optimizer. The training is iterated until the preset number of training rounds is reached or the loss function converges to obtain the content recommendation strategy detection model.

[0104] MLP takes the integrated feature vector as input, extracts higher-order features layer by layer, performs nonlinear transformations, and then uses multiple output nodes (predicting the probabilities of different classes) to complete the classification. The structure is as follows: Figure 5 As shown, the formula is expressed as:

[0105] H=φ(W o i+bo (6)

[0106] O = Softmax(HW) o +b o (7)

[0107] Where H represents the output of the hidden layer, φ() refers to the activation function (including ReLU, sigmoid, tanh, etc.) used to extract higher-order features, and W o This is the weight matrix of the output layer, where i is the input vector and b is the weight matrix of the output layer. o O is the bias vector of the output layer, O is the output result, i = [i1, i2, i3] is the input feature vector, O = [o1, o2, o3, o4] is the output feature vector, Softmax is the activation function, which converts the model output into a probability distribution, i1, i2, i3, o1, o2, o3, o4 represent statistical features (operation record data and browsing record data), category features (user basic information data and personal habit information data), text features (text data in the recommended content data), and the probability of recommendation strategy based on relevance, recommendation strategy based on environment, recommendation strategy based on popularity, and recommendation strategy based on collaboration, respectively.

[0108] In an exemplary embodiment, after training the multilayer perceptron model to obtain a content recommendation strategy detection model based on the integrated feature vectors of each participant on each sample platform and the content recommendation strategies of each sample platform, the method further includes: evaluating and validating the S3 model in terms of accuracy, coverage, diversity, and user fairness. If the evaluation and validation fail, the multilayer perceptron model is retrained, wherein:

[0109] Accuracy: Root Mean Squared Error (RMSE) is used as the evaluation metric for the model. RMSE is a discriminant function that calculates the root mean square error between the observed and the true values. First, the predicted and actual results are collected. Then, the mean squared error (MSE) and RMSE are calculated according to formulas (8) and (9) respectively. The calculation formulas are as follows:

[0110]

[0111] Where M represents the total number of samples, y m This is the true rating for the m-th sample (a participant's browsing history, operation history, recommended content, basic user information, and personal habit information on a certain sample platform). This rating is a recommendation score between the user and the post (a probability value between 0 and 1). This determines the probability of the model outputting the content recommendation strategy for the m-th sample. If RMSE < 0.5, the model is considered acceptable.

[0112] Coverage: The coverage of a model is evaluated by calculating the proportion of items that can be recommended in a single user interaction with the recommendation system. Multiple recommendation algorithms will be executed multiple times. If the set of items recommended to the user in a specific recommendation session is S... r S a If the set is all available items, then the coverage can be calculated as follows:

[0113]

[0114] Generally, a coverage rate >0.7 indicates good coverage.

[0115] Diversity: This section compares the diversity metrics of content recommendation strategy detection methods across different platforms, evaluating their effectiveness in detecting recommendation strategies on various platforms. The item set c1,...,c n The diversity is defined as the average dissimilarity among all item pairs in a project set:

[0116]

[0117] Among them, diversity(c1,...,c n ) represents the entire project set {c1, c2, c3…c n The diversity index of}, where n represents the number of projects in the project set, similarity(c q ,c j ) represents the q-th item c q and the j-th project c j The similarity between them. A diversity value greater than 0.5 is considered to reflect the richness of the recommended content.

[0118]

[0119] Where similarity(t,c) represents the similarity between item t and item c, sim(t p ,c p ), ω t Let represent the similarity between the p-th feature of item t and the p-th feature of item c, and the weight of the p-th feature, respectively.

[0120] User fairness: Measuring user fairness ensures that the recommendation system maintains consistent recommendation quality across different users. It is determined by calculating the User Fairness Index (UFI).

[0121] UFI = std({RMSE u|u∈U}) (13)

[0122] Where std is the standard deviation function, U represents the set of all users, and RMSE u UFI refers to the root mean square error of user u, which is calculated according to formulas (8) and (9) above. It is used to measure the recommendation accuracy of the recommendation system for user u. The smaller the UFI, the better, to ensure that the recommendation quality among users is relatively balanced.

[0123] To address the "black box" nature of platform content recommendation strategies, which prevents users from understanding the recommendation criteria and may lead to biased content recommendations, this application proposes a content recommendation strategy detection method. Specifically, it's a method based on RPA technology and crowdsourcing collaboration. First, RPA technology and crowdsourcing mechanisms are used for data collection and processing. Then, features are extracted from the data obtained in step S1. Next, the vectors obtained in step S2 are used to train an Multilayer Perceptron (MLP) model. Finally, the model in step S3 is evaluated and validated in terms of accuracy, coverage, diversity, and user fairness. The validated model yields the content recommendation strategy detection model. This application, by combining RPA technology and crowdsourcing data collection methods with Natural Language Processing (NLP) technology and a Multilayer Perceptron (MLP) model, offers the following advantages compared to existing technologies:

[0124] (1) This application combines RPA technology and crowdsourcing, enabling the system to efficiently and automatically collect and process large amounts of user data, ensuring the richness and accuracy of the information. It offers greater flexibility in task allocation and data collection, allowing for real-time adjustment of detection strategies and ensuring more accurate and efficient evaluation of the recommendation algorithm. For ease of understanding of this embodiment, compared to traditional data collection methods, automated RPA technology has significant advantages in versatility, integration capabilities, visual development, error handling, and compliance.

[0125] (2) This application uses natural language processing technology for feature extraction, including Log bucketing embedding of long-tail data, One-Hot embedding of category features, and NLP-based text embedding, etc., to aggregate various features into a unified content space so that subsequent models can perform more accurate and efficient processing.

[0126] (3) This application uses MLP to identify and classify recommendation features. The system can accurately identify the core features of recommendation strategies and provide targeted analysis, providing a reference for the optimization and evaluation of recommendation systems.

[0127] (4) This application uses RPA technology to automatically collect and process multi-dimensional data, which can easily handle complex user behavior information.

[0128] This application combines advanced NLP and MLP models to perform in-depth analysis of large amounts of unstructured data, thereby enabling more accurate evaluation of recommendation strategies.

[0129] This application can be applied to various internet platforms, especially social media, e-commerce, news and video sharing platforms that require precise recommendation strategies, and is expected to achieve the following effects:

[0130] (1) Detecting the Transparency and Fairness of Recommendation Algorithms: By analyzing the platform's recommendation strategy, this application can reveal the transparency and fairness of the recommendation algorithm in content distribution. It can identify whether the recommendation strategy is biased and ensure that content distribution is more equitable, avoiding the impact of algorithmic bias on the user experience on the platform.

[0131] (2) Enhancing User Experience and Satisfaction: By collecting and analyzing user operation records and content interaction data, this application can help the platform gain a deeper understanding of user preferences and habits, as well as user opinions and feedback on recommendation strategies. This ensures the quality of recommendation services while avoiding various risks caused by recommendation algorithms and helps relevant personnel understand the internal mechanisms of the recommendation platform. This insight can guide the platform to optimize recommendation strategies, making recommended content more aligned with user interests, thereby improving the overall user experience and satisfaction.

[0132] (3) Optimizing the diversity and innovation of recommendation strategies: This application can identify the shortcomings of current recommendation systems, such as the problem of content homogenization. The platform can use this information to adjust the recommendation algorithm, increase the diversity of recommended content, promote the dissemination of innovative content, and enhance the user's exploration experience.

[0133] Based on the same inventive concept, this application also provides a content recommendation strategy detection device for implementing the content recommendation strategy detection method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the content recommendation strategy detection device provided below can be found in the limitations of the content recommendation strategy detection method described above, and will not be repeated here.

[0134] In one exemplary embodiment, a content recommendation strategy detection device is provided, comprising:

[0135] The participant identification module is used to identify multiple participants by publishing a pre-set task to a crowdsourcing platform using robotic process automation (RPA) technology.

[0136] The simulation module is used to simulate the operations of each participant on the platform to be tested after the participants log in to their accounts. It uses robotic process automation technology to simulate the operations of each participant on the platform to be tested to complete preset tasks and obtain user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data and personal habit information data of each participant on the platform to be tested during the simulation operation.

[0137] The data vectorization processing module is used to perform data vectorization processing on the browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform to be tested, to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on the platform to be tested.

[0138] The aggregation module is used to aggregate the browsing record vector, operation record vector, recommended content vector, user basic information vector, and personal habit information vector of any participant on the platform to be detected, so as to obtain the feature vector of the participant.

[0139] The integration module is used to integrate the user ID data and recommendation object ID data of the participant on the platform to be detected, as well as the feature vector of the participant, to obtain the integrated feature vector of the participant.

[0140] The content recommendation strategy detection module is used to input the integrated feature vectors of each participant during the simulation operation into the content recommendation strategy detection model to obtain the content recommendation strategy of the platform to be detected.

[0141] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6 As shown, the computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores content recommendation strategy detection data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a content recommendation strategy detection method.

[0142] Those skilled in the art will understand that Figure 6 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method embodiments.

[0143] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the above-described method embodiments.

[0144] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the above-described method embodiments.

[0145] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0146] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0147] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0148] 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.

[0149] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A content recommendation strategy detection method, characterized in that, The content recommendation strategy detection method includes: Robotic process automation (RPA) technology is used to publish pre-set tasks to a crowdsourcing platform to identify multiple participants. After logging into their accounts on the platform to be tested, each participant uses robotic process automation (RPA) technology to simulate their operations on the platform to complete preset tasks. During the simulation, the system obtains user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform to be tested. The browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform to be tested are processed into data vectors to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on the platform to be tested. For any participant, the feature vector of the participant is obtained by aggregating the browsing record vector, operation record vector, recommended content vector, user basic information vector and personal habit information vector of the participant on the platform to be detected. The integrated feature vector of the participant is obtained by integrating the participant's user ID data and recommendation object ID data on the platform to be detected, as well as the participant's feature vector. The integrated feature vectors of each participant during the simulation are input into the content recommendation strategy detection model to obtain the content recommendation strategy of the platform to be detected.

2. The content recommendation strategy detection method according to claim 1, characterized in that, The browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on the platform under test are vectorized to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on the platform under test, specifically including: For any participant, perform Log bucketing embedding operation on the participant's browsing record data and operation record data on the platform to be detected to obtain the participant's browsing record vector and operation record vector on the platform to be detected. One-Hot embedding operation is performed on the participant's basic user information data and personal habit information data on the platform to be detected to obtain the participant's basic user information vector and personal habit information vector on the platform to be detected. Text embedding is performed on the recommended content data of the participant on the platform to be detected to obtain the recommended content vector of the participant on the platform to be detected.

3. The content recommendation strategy detection method according to claim 1, characterized in that, The content recommendation strategy detection model was obtained by training a multilayer perceptron model.

4. The content recommendation strategy detection method according to claim 3, characterized in that, The specific process of training a multilayer perceptron model includes: Robotic process automation (RPA) technology is used to publish pre-set sample tasks to a crowdsourcing platform to identify multiple sample participants. After logging into their accounts on multiple sample platforms, participants in each sample were simulated using robotic process automation (RPA) technology to complete preset sample tasks. During the simulation, user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on each sample platform were obtained. The browsing history data, operation history data, recommended content data, user basic information data, and personal habit information data of each participant on each sample platform are processed into data vectors to obtain the browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector of each participant on each sample platform. For any sample participant on any sample platform, the browsing record vector, operation record vector, recommended content vector, user basic information vector, and personal habit information vector of the sample participant on the sample platform are aggregated to obtain the feature vector of the sample participant on the sample platform. The integrated feature vector of the sample participants is obtained by integrating the feature vectors, user ID data, and recommendation object ID data of the sample participants on the sample platform. A content recommendation strategy detection model is obtained by training a multilayer perceptron model based on the integrated feature vectors of each participant on each sample platform and the content recommendation strategies of each sample platform.

5. The content recommendation strategy detection method according to claim 2, characterized in that, The text embedding operation is performed on the recommended content data of the participants on the platform to be detected, specifically including: The BETR model is used to process the recommended content data of the participants on the platform to be detected to obtain text embedding vectors; The text embedding vector is processed by multi-layer perceptron processing to obtain the recommended content vector of the participant on the platform to be detected.

6. The content recommendation strategy detection method according to claim 1, characterized in that, The feature vector of the participant is obtained by aggregating the browsing record vector, operation record vector, recommended content vector, user basic information vector, and personal habit information vector of the participant on the platform to be detected, specifically including: The participant's browsing history vector, operation history vector, recommended content vector, user basic information vector, and personal habit information vector on the platform to be detected are aggregated to obtain the participant's aggregated feature vector. The aggregated feature vectors of the participants are weighted using an attention mechanism to obtain the feature vectors of the participants.

7. A content recommendation strategy detection device, characterized in that, The content recommendation strategy detection device includes: The participant identification module is used to identify multiple participants by publishing a pre-set task to a crowdsourcing platform using robotic process automation technology. The simulation module is used to simulate the operations of each participant on the platform to be tested after the participants log in to their accounts. It uses robotic process automation technology to simulate the operations of each participant on the platform to be tested to complete preset tasks and obtain user ID data, recommended object ID data, browsing history data, operation history data, recommended content data, user basic information data and personal habit information data of each participant on the platform to be tested during the simulation operation. The data vectorization processing module is used to perform data vectorization processing on the browsing record data, operation record data, recommended content data, user basic information data and personal habit information data of each participant on the platform to be tested, to obtain the browsing record vector, operation record vector, recommended content vector, user basic information vector and personal habit information vector of each participant on the platform to be tested. The aggregation module is used to aggregate the browsing record vector, operation record vector, recommended content vector, user basic information vector, and personal habit information vector of any participant on the platform to be detected, so as to obtain the feature vector of the participant. The integration module is used to integrate the user ID data and recommendation object ID data of the participant on the platform to be detected, as well as the feature vector of the participant, to obtain the integrated feature vector of the participant; The content recommendation strategy detection module is used to input the integrated feature vectors of each participant during the simulation operation into the content recommendation strategy detection model to obtain the content recommendation strategy of the platform to be detected.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the content recommendation strategy detection method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the content recommendation strategy detection method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the content recommendation strategy detection method as described in any one of claims 1-6.

Images