Computer device and method for acquiring user activeness of deep learning platform

Abstract

The invention discloses a computer device and method for acquiring user activeness of a deep learning platform. According to the method, the random forest algorithm is adopted to construct the user activity prediction model, so that the prediction performance and robustness of the model are enhanced. The random forest can capture a complex nonlinear relation through combined learning of multiple decision trees, and effectively deals with high-dimensional data and noise interference. Meanwhile, through parameter adjustment and optimization, the model has good generalization ability while keeping high prediction accuracy. The strong prediction capability enables the platform to recognize the activity change trend in advance and take intervention measures in time, such as personalized recommendation and motivation activities, so that the user stickiness is effectively improved, and the sustainable and healthy development of the platform is promoted. In conclusion, the method has remarkable beneficial effects in the aspects of improving the user activeness evaluation quality, optimizing the user management strategy and promoting the sustainable development of the platform.

Description

Technical Field

[0001] The present invention belongs to the technical field of deep learning platform users, and specifically relates to a computer device and method for obtaining the activity of deep learning platform users. Background Art

[0002] A deep learning platform is a software environment that integrates multiple deep learning algorithms and tools, designed to provide researchers, developers, and businesses with convenient model development, training, and deployment services. These platforms typically feature powerful computing capabilities, rich dataset support, and user-friendly interfaces, accelerating the application of deep learning technology in areas such as image recognition, natural language processing, and autonomous driving. User activity on a deep learning platform reflects its popularity and practical application value within the industry. Highly active platforms often boast a large number of loyal users who actively communicate, share experiences, and contribute code, driving innovation and development in deep learning technology while also accumulating valuable knowledge resources and community support. Therefore, evaluating the user activity of a deep learning platform can facilitate platform improvements.

[0003] However, traditional user activity evaluation methods are often limited to simple interaction frequency indicators and cannot fully reflect the actual activity of users. Summary of the Invention

[0004] The purpose of the present invention is to provide a computer device and method for obtaining user activity on a deep learning platform in order to solve the above-mentioned problems.

[0005] The technical solution adopted by the present invention is as follows: a computer method for obtaining user activity on a deep learning platform, the method comprising the following steps:

[0006] S1: Use API interfaces or web crawler technology to automatically collect user behavior data on the deep learning platform, including login frequency, number of posts, number of replies, number of code submissions, and number of model training times;

[0007] S2: Clean the collected data to remove noise and outliers to ensure data quality and consistency; normalize or standardize the data to facilitate subsequent analysis and modeling;

[0008] S3: Introducing the "User Interaction Quality Index" feature, which not only considers the frequency of user interaction but also the depth and breadth of interaction. Depth is measured by the technical depth score of users' posts in the discussion area, and breadth is measured by the number of discussions on different topics in which users participate. Natural language processing technology is used to analyze user posts to extract technical keywords and sentiment trends as auxiliary indicators for evaluating interaction quality.

[0009] S4: Select the random forest machine learning algorithm to build a user activity prediction model; use the training data to train the model, adjust the model parameters, and optimize the model performance;

[0010] S5: Use cross-validation to evaluate the model to ensure its stability and generalization ability, calculate the accuracy, recall, and F1 value of the model, and comprehensively evaluate the model performance;

[0011] S6: Use the trained model to predict the activity of users on the deep learning platform; based on the prediction results, divide the users into different activity levels, including high activity, medium activity, and low activity;

[0012] S7: Visualize the user activity calculation results in the form of charts to facilitate intuitive understanding and analysis. The visualization tool uses the Matplotlib library;

[0013] S8: Regularly collect new user behavior data, continuously train and update the model, and continuously optimize feature engineering and model algorithms based on platform development and changes in user behavior to improve the accuracy and effectiveness of user activity calculations.

[0014] In a preferred embodiment, in step S1, the user's login log, posting record, and reply record structured data are regularly obtained through the API interface; at the same time, web crawler technology is used to supplement the collection of users' unstructured data on the platform, including code submission history and model training logs; these data together constitute the data basis for analyzing user activity.

[0015] In a preferred embodiment, in step S2, for the login frequency data, it will be found that the number of logins of some users is abnormally high. This is due to automated scripts or other abnormal behaviors, so these outliers need to be eliminated; then, the data is normalized and the data of different indicators are converted to the same dimension for subsequent analysis.

[0016] In a preferred embodiment, in step S3, the user interaction quality index is composed of three parts: interaction frequency, interaction depth, and interaction breadth; natural language processing (NLP) technology is used to analyze user postings to extract technical keywords and emotional tendencies as auxiliary indicators for evaluating interaction quality;

[0017] The formula for calculating the interaction frequency is:

[0018] F = (number of posts + number of replies + number of likes) / time window; where number of posts refers to the number of threads published by a user within the time window; number of replies refers to the number of replies published by a user within the time window; number of likes refers to the number of likes a user gave to other people's posts within the time window; and the time window refers to the time period over which user behavior is counted, including a week or a month.

[0019] The formula for calculating engagement depth is:

[0020] D = Σ(TF-IDF keyword weight) / number of posts; TF-IDF keyword weight represents the weight of technical keywords in user posts calculated using the TF-IDF algorithm; number of posts represents the number of topic posts published by the user within the time window;

[0021] The formula for calculating the breadth of interaction is: G = number of different topic tags;

[0022] The number of different topic tags indicates the number of different technical topic tags involved in the user's posts within the time window;

[0023] The calculation formula for the user interaction quality index is:

[0024] IQI=α*F+β*D+γ*G+δ*emotion score

[0025] Among them, α, β, γ, and δ are weight parameters, which are adjusted according to actual data to meet the evaluation requirements of user activity on different platforms;

[0026] F is the interaction frequency;

[0027] D is the depth of interaction;

[0028] G is the breadth of interaction;

[0029] Sentiment score: The sentiment tendency score of user posts obtained through sentiment analysis.

[0030] In a preferred embodiment, in step S4, a random forest algorithm is used to construct multiple decision trees for prediction, and the prediction results of these decision trees are aggregated to obtain a final prediction result;

[0031] The specific steps include:

[0032] S4.1: Data preparation: Divide the feature-engineered dataset into a training set and a test set; the training set is used for model training, and the test set is used for model evaluation;

[0033] S4.2: Model initialization, select the random forest algorithm, and initialize the model parameters; the parameters include:

[0034] n_estimators: the number of decision trees;

[0035] max_depth: the maximum depth of the decision tree;

[0036] min_samples_split: the minimum number of samples required to split an internal node;

[0037] min_samples_leaf: the minimum number of samples required for a leaf node;

[0038] max_features: the number of features considered when finding the best split;

[0039] S4.3: Model training: Use the training data to train the random forest model. During the training process, the model combines and analyzes features through multiple decision trees to learn the relationship between user activity and features.

[0040] S4.4: Parameter adjustment and optimization: Adjust model parameters and optimize model performance through cross-validation techniques; S4.5: Model evaluation: Use the test set to evaluate the trained model and calculate evaluation indicators;

[0041] The prediction result of random forest is usually the average or majority vote of the prediction results of multiple decision trees. For user activity prediction, assuming it is a regression problem, the formula is as follows:

[0042]

[0043] in:

[0044] y^ represents the prediction result of random forest (user activity);

[0045] N represents the number of decision trees;

[0046] y^i represents the prediction result of the i-th decision tree;

[0047] When splitting nodes, decision trees usually use information gain, gain rate or Gini impurity criteria; taking information gain as an example, the calculation formula is as follows:

[0048]

[0049] in:

[0050] D represents the sample set corresponding to the node;

[0051] D_j represents the subsample set after splitting according to the features;

[0052] v represents the number of values ​​of the feature;

[0053] |D| and |D_j| represent the number of samples in sample sets D and D_j respectively:

[0054] The formula for calculating entropy is:

[0055] Where K represents the number of categories, and p_k represents the proportion of samples of the kth category in the sample set D.

[0056] In a preferred embodiment, in step S5, the "posting frequency" and "average number of replies" features are extracted from the user's posting records; the "code submission activity" and "last submission time" features are extracted from the code submission history; and based on the user's interactive relationship on the platform, the user's social network graph is constructed, and the "network centrality" and "clustering coefficient" features are extracted.

[0057] In a preferred embodiment, in step S6, the data set is divided into a training set and a test set to ensure that the training and evaluation of the model are performed on different data sets; then the random forest model is trained using the training set data, and the complex relationship between user behavior characteristics and activity is captured through combined learning of multiple decision trees.

[0058] In a preferred embodiment, in step S7, the trained random forest model is first evaluated using the test set data; the mean square error index of the model is calculated to quantify the prediction accuracy of the model; then the prediction results of the model are displayed in the form of a chart using a visualization tool; in addition, a feature importance graph is drawn to show the contribution of different features to the prediction results, providing a basis for subsequent feature selection and model optimization.

[0059] In a preferred embodiment, in step S8, new data is collected once a quarter and the random forest model is retrained to adapt to the new data distribution and user behavior patterns; at the same time, feature engineering and model algorithms are continuously optimized according to platform development and changes in user behavior; through these continuous updates and optimization measures, the accuracy and effectiveness of user activity calculations are ensured to always remain at a high level.

[0060] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0061] 1. This invention significantly improves the comprehensiveness and accuracy of user activity assessments by introducing a "User Interaction Quality Index" feature. This method combines the depth and breadth of interactions, leveraging natural language processing to deeply analyze user posts, extract technical keywords and sentiment, and form a multi-dimensional assessment system. This comprehensive assessment not only more accurately reflects a user's actual activity status but also reveals their potential value and influence, providing platforms with more accurate user profiles and more effective user management strategies.

[0062] 2. In the present invention, a random forest algorithm is used to construct a user activity prediction model, which enhances the model's predictive performance and robustness. Random forest can capture complex nonlinear relationships through combined learning of multiple decision trees, and effectively deal with high-dimensional data and noise interference. At the same time, through parameter adjustment and optimization, the model maintains high prediction accuracy while also having good generalization capabilities. This powerful predictive capability enables the platform to identify activity change trends in advance and take timely intervention measures, including personalized recommendations, incentive activities, etc., thereby effectively improving user stickiness and promoting the sustainable and healthy development of the platform. In summary, this method has significant beneficial effects in improving the quality of user activity assessment, optimizing user management strategies, and promoting the sustainable development of the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a schematic diagram of the process principle of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0065] Reference Figure 1 ,

[0066] A computer method for obtaining user activity on a deep learning platform comprises the following steps:

[0067] S1: Use API interfaces or web crawler technology to automatically collect user behavior data on the deep learning platform, including login frequency, number of posts, number of replies, number of code submissions, and number of model training times;

[0068] S2: Clean the collected data to remove noise and outliers to ensure data quality and consistency. Normalize or standardize the data to facilitate subsequent analysis and modeling.

[0069] S3: Introducing the "User Interaction Quality Index" feature, which not only considers the frequency of user interaction but also the depth and breadth of interaction. Depth is measured by the technical depth score of users' posts in the discussion area, and breadth is measured by the number of discussions on different topics in which users participate. Natural language processing technology is used to analyze user posts to extract technical keywords and sentiment trends as auxiliary indicators for evaluating interaction quality.

[0070] S4: Select the random forest machine learning algorithm to build a user activity prediction model. Use the training data to train the model, adjust the model parameters, and optimize the model performance.

[0071] S5: Use cross-validation and other methods to evaluate the model to ensure its stability and generalization ability, calculate the model's accuracy, recall rate, F1 value and other indicators, and comprehensively evaluate the model performance;

[0072] S6: Use the trained model to predict the activity of users on the deep learning platform. Based on the prediction results, users are divided into different activity levels, such as high activity, medium activity, and low activity;

[0073] S7: Visualize the user activity calculation results in the form of charts to facilitate intuitive understanding and analysis. The visualization tool can use Python libraries such as Matplotlib and Seaborn;

[0074] S8: Regularly collect new user behavior data, continuously train and update the model, and continuously optimize feature engineering and model algorithms based on platform development and changes in user behavior to improve the accuracy and effectiveness of user activity calculations.

[0075] In step S1, structured data such as user login logs, posting records, and reply records are regularly collected through the API interface. Web crawler technology is also used to supplement the collection of unstructured data on the platform, such as code submission history and model training logs. Together, these data form the foundation for analyzing user activity. For example, for a particular user, specific data collected over the past month might include 20 logins, 10 posts, 30 replies, 50 code submissions, and 20 model training sessions.

[0076] In step S2, the login frequency data may reveal unusually high login times for some users. This could be due to automated scripts or other unusual behavior, so these outliers need to be removed. Next, the data is normalized to convert different metrics to the same scale for easier analysis. Missing data is also interpolated or deleted to ensure data consistency and integrity. These preprocessing steps yield a clean, consistent, and analytically viable user behavior dataset.

[0077] In step S3, the user interaction quality index is composed of three components: interaction frequency, interaction depth, and interaction breadth. Natural language processing (NLP) technology is used to analyze user posts and extract technical keywords and sentiment trends as auxiliary indicators for evaluating interaction quality.

[0078] The formula for calculating the interaction frequency is:

[0079] F = (posts + replies + likes) / time window; where posts represents the number of threads a user posts within the time window. Replies represents the number of replies a user posts within the time window. Likes represents the number of likes a user gives to others' posts within the time window. The time window represents the period of time over which user behavior is counted, including a week or a month.

[0080] The formula for calculating engagement depth is:

[0081] D = Σ(TF-IDF keyword weight) / number of posts; where TF-IDF keyword weight represents the weight of technical keywords in user posts calculated using the TF-IDF algorithm. Number of posts represents the number of topic posts published by the user within the time window;

[0082] The formula for calculating the breadth of interaction is: G = number of different topic tags;

[0083] The number of different topic tags indicates the number of different technical topic tags involved in the user's posts within the time window;

[0084] The calculation formula for the user interaction quality index is:

[0085] IQI=α*F+β*D+γ*G+δ*emotion score

[0086] Among them, α, β, γ, and δ are weight parameters, which are adjusted according to actual data to meet the evaluation requirements of user activity on different platforms.

[0087] F is the interaction frequency.

[0088] D is the depth of interaction.

[0089] G is the breadth of interaction.

[0090] Sentiment score: The sentiment tendency score of user posts obtained through sentiment analysis.

[0091] In step S4, a random forest algorithm is used to construct multiple decision trees for prediction, and the prediction results of these decision trees are aggregated to obtain the final prediction result.

[0092] The specific steps include:

[0093] S4.1: Data preparation: Divide the feature-engineered dataset into a training set and a test set. The training set is used for model training, and the test set is used for model evaluation.

[0094] S4.2: Model initialization, select the random forest algorithm, and initialize the model parameters. The parameters include:

[0095] n_estimators: The number of decision trees.

[0096] max_depth: The maximum depth of the decision tree.

[0097] min_samples_split: The minimum number of samples required to split an internal node.

[0098] min_samples_leaf: The minimum number of samples required for a leaf node.

[0099] max_features: The number of features to consider when finding the best split.

[0100] S4.3: Model Training: Use the training data to train the random forest model. During the training process, the model combines and analyzes features through multiple decision trees to learn the relationship between user activity and features.

[0101] S4.4: Parameter adjustment and optimization: Adjust model parameters and optimize model performance through cross-validation and other techniques.

[0102] S4.5: Model evaluation: Use the test set to evaluate the trained model and calculate evaluation metrics such as accuracy, recall, and F1 score;

[0103] The prediction result of random forest is usually the average of the prediction results of multiple decision trees (regression problem) or the majority voting result (classification problem). For user activity prediction, assuming it is a regression problem, the formula is as follows:

[0104]

[0105] in:

[0106] y^ represents the prediction result of random forest (user activity).

[0107] N represents the number of decision trees.

[0108] y^i represents the prediction result of the i-th decision tree.

[0109] When splitting nodes in a decision tree, criteria such as information gain, gain ratio, or Gini impurity are usually used. Taking information gain as an example, the calculation formula is as follows:

[0110]

[0111] in:

[0112] D represents the sample set corresponding to the node.

[0113] D_j represents the subsample set after splitting according to the features.

[0114] v represents the number of feature values.

[0115] |D| and |D_j| represent the number of samples in sample sets D and D_j respectively:

[0116] The formula for calculating entropy is:

[0117] Where K represents the number of categories, and p_k represents the proportion of samples of the kth category in the sample set D.

[0118] In step S5, features such as "posting frequency" and "average number of replies" are extracted from the user's posting history; features such as "code submission activity" and "last submission time" are extracted from the code submission history. Furthermore, "social network features" are creatively introduced. Based on the user's interactions on the platform, a social network graph is constructed, and features such as "network centrality" and "clustering coefficient" are extracted. These features can more comprehensively reflect the user's activity and social influence, providing rich information for subsequent model training.

[0119] In step S6, the dataset is divided into a training set and a test set, ensuring that model training and evaluation are performed on different datasets. Next, the random forest model is trained using the training set data. Through ensemble learning of multiple decision trees, the model captures the complex relationship between user behavioral characteristics and activity. During training, model parameters, such as the number of trees and tree depth, are adjusted to optimize model performance. Through cross-validation and performance evaluation, a trained random forest model is ultimately obtained, which is capable of predicting user activity based on their behavioral characteristics.

[0120] In step S7, the trained random forest model is first evaluated using the test set data. Indicators such as the mean square error (MSE) of the model are calculated to quantify the prediction accuracy of the model. For example, an MSE value of 0.05 may be obtained, indicating that the prediction error of the model is small and the performance is good. Then, visualization tools such as Python libraries such as Matplotlib and Seaborn are used to display the prediction results of the model in the form of graphs. For example, a comparison graph of the true value and the predicted value may be drawn to intuitively show the prediction effect of the model. In addition, a feature importance graph may be drawn to show the contribution of different features to the prediction results, providing a basis for subsequent feature selection and model optimization.

[0121] In step S8, new data is collected quarterly and the random forest model is retrained to adapt to the new data distribution and user behavior patterns. Simultaneously, feature engineering and model algorithms are continuously optimized based on platform development and changes in user behavior. For example, a new user behavior metric may be found to be closely correlated with activity, so it can be incorporated into feature engineering. Alternatively, other machine learning algorithms, such as gradient boosting trees and neural networks, may be tried to explore more optimal predictive models. Through these continuous updates and optimization measures, the accuracy and effectiveness of user activity calculations remain at a consistently high level.

[0122] A computer device for obtaining user activity on a deep learning platform. The computer device is internally provided with an application program that runs a computer method for obtaining user activity on a deep learning platform as described in the above embodiment.

[0123] From the above we can know:

[0124] This invention significantly improves the comprehensiveness and accuracy of user activity assessments by introducing a "User Interaction Quality Index" feature. This method combines the depth and breadth of interactions, leveraging natural language processing to deeply analyze user posts, extract technical keywords and sentiment, and form a multi-dimensional assessment system. This comprehensive assessment not only more accurately reflects a user's actual activity status but also reveals their potential value and influence, providing platforms with more accurate user profiles and more effective user management strategies.

[0125] In the present invention, a random forest algorithm is used to construct a user activity prediction model, which enhances the model's predictive performance and robustness. Random forest can capture complex nonlinear relationships through combined learning of multiple decision trees, and effectively deal with high-dimensional data and noise interference. At the same time, through parameter adjustment and optimization, the model maintains high prediction accuracy while also having good generalization capabilities. This powerful predictive ability enables the platform to identify activity change trends in advance and take timely intervention measures, such as personalized recommendations and incentive activities, thereby effectively improving user stickiness and promoting the sustainable and healthy development of the platform. In summary, this method has significant beneficial effects in improving the quality of user activity assessment, optimizing user management strategies, and promoting the sustainable development of the platform.

[0126] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further limitations, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0127] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A computer method for obtaining user activity on a deep learning platform, characterized by: The method comprises the following steps: S1: Use API interfaces or web crawler technology to automatically collect user behavior data on the deep learning platform, including login frequency, number of posts, number of replies, number of code submissions, and number of model training times; S2: Clean the collected data to remove noise and outliers to ensure data quality and consistency; normalize or standardize the data to facilitate subsequent analysis and modeling; S3: Introduces a "User Interaction Quality Index" feature, which not only considers the frequency of user interactions but also the depth and breadth of interactions. Depth is measured by the technical depth rating of users' posts in the discussion forum, while breadth is measured by the number of discussions on different topics in which users participate. Natural language processing technology is used to analyze user posts, extracting technical keywords and sentiment as auxiliary indicators for evaluating interaction quality. S4: Select the random forest machine learning algorithm to build a user activity prediction model; use the training data to train the model, adjust the model parameters, and optimize the model performance; S5: Use cross-validation to evaluate the model to ensure its stability and generalization ability. Calculate the accuracy, recall, and F1 value of the model to comprehensively evaluate the model performance. S6: Use the trained model to predict the activity of users on the deep learning platform; based on the prediction results, classify users into different activity levels, including high activity, medium activity, and low activity; S7: Visualize the user activity calculation results in the form of charts to facilitate intuitive understanding and analysis. The visualization tool uses the Matplotlib library; S8: Regularly collect new user behavior data, continuously train and update the model, and continuously optimize feature engineering and model algorithms based on platform development and changes in user behavior to improve the accuracy and effectiveness of user activity calculations.

2. The computer method for obtaining user activity on a deep learning platform according to claim 1, characterized in that: In step S1, the user's login log, posting record, and reply record structured data are regularly obtained through the API interface; at the same time, web crawler technology is used to supplement the collection of the user's unstructured data on the platform, including code submission history and model training log.

3. The computer method for obtaining user activity on a deep learning platform according to claim 1, characterized in that: In step S2, for the login frequency data, it is found that the number of logins of some users is abnormally high. This is caused by automated scripts or other abnormal behaviors, so these outliers need to be eliminated; then, the data is normalized and the data of different indicators are converted to the same dimension to facilitate subsequent analysis.

4. The computer method for obtaining user activity on a deep learning platform according to claim 1, characterized in that: In step S3, the user interaction quality index is composed of three parts: interaction frequency, interaction depth, and interaction breadth; natural language processing (NLP) technology is used to analyze the user post content to extract technical keywords and emotional tendencies as auxiliary indicators for evaluating interaction quality; The formula for calculating the interaction frequency is: F = (number of posts + number of replies + number of likes) / time window; where number of posts refers to the number of threads published by a user within the time window; number of replies refers to the number of replies published by a user within the time window; number of likes refers to the number of likes a user gave to other people's posts within the time window; and the time window refers to the time period over which user behavior is counted, including a week or a month. The formula for calculating engagement depth is: D = Σ(TF-IDF keyword weight) / number of posts; TF-IDF keyword weight represents the weight of technical keywords in user posts calculated using the TF-IDF algorithm; number of posts represents the number of topic posts published by the user within the time window; The formula for calculating the breadth of interaction is: G = number of different topic tags; The number of different topic tags indicates the number of different technical topic tags involved in the user's posts within the time window; The calculation formula for the user interaction quality index is: IQI=α*F+β*D+γ*G+δ*emotion score Among them, α, β, γ, and δ are weight parameters, which are adjusted according to actual data to meet the evaluation requirements of user activity on different platforms; F is the interaction frequency; D is the depth of interaction; G is the breadth of interaction; Sentiment score: The sentiment tendency score of user posts obtained through sentiment analysis.

5. The computer method for obtaining user activity on a deep learning platform according to claim 1, characterized in that: In step S4, a random forest algorithm is used to construct multiple decision trees for prediction, and the prediction results of these decision trees are aggregated to obtain a final prediction result; The specific steps include: S4.1: Data preparation: Divide the feature-engineered dataset into a training set and a test set; the training set is used for model training, and the test set is used for model evaluation; S4.2: Model initialization, select the random forest algorithm, and initialize the model parameters; the parameters include: n_estimators: the number of decision trees; max_depth: the maximum depth of the decision tree; min_samples_split: the minimum number of samples required to split an internal node; min_samples_leaf: the minimum number of samples required for a leaf node; max_features: the number of features considered when finding the best split; S4.3: Model training: Use the training data to train the random forest model. During the training process, the model combines and analyzes features through multiple decision trees to learn the relationship between user activity and features. S4.4: Parameter adjustment and optimization: Adjust model parameters and optimize model performance through cross-validation techniques; S4.5: Model evaluation: Evaluate the trained model using the test set and calculate the evaluation metrics; The prediction result of random forest is usually the average or majority vote of the prediction results of multiple decision trees. For user activity prediction, assuming it is a regression problem, the formula is as follows: in: y^ represents the prediction result of random forest (user activity); N represents the number of decision trees; y^i represents the prediction result of the i-th decision tree; When splitting nodes, decision trees usually use information gain, gain rate or Gini impurity criteria; taking information gain as an example, the calculation formula is as follows: in: D represents the sample set corresponding to the node; D_j represents the subsample set after splitting according to the features; v represents the number of values ​​of the feature; |D| and |D_j| represent the number of samples in sample sets D and D_j respectively: The formula for calculating entropy is: Where K represents the number of categories, and p_k represents the proportion of samples of the kth category in the sample set D.

6. The computer method for obtaining user activity on a deep learning platform according to claim 1, characterized in that: In step S5, the "posting frequency" and "average number of replies" features are extracted from the user's posting records; "Code submission activity" and "last submission time" features are extracted from the code submission history; at the same time, based on the user's interactive relationship on the platform, the user's social network graph is constructed, and the "network centrality" and "clustering coefficient" features are extracted.

7. The computer method for obtaining user activity on a deep learning platform according to claim 1, characterized in that: In step S6, the dataset is divided into a training set and a test set to ensure that the model training and evaluation are performed on different datasets; the random forest model is then trained using the training set data to capture the complex relationship between user behavior characteristics and activity through combined learning of multiple decision trees.

8. The computer method for obtaining user activity on a deep learning platform according to claim 1, characterized in that: In step S7, the trained random forest model is first evaluated using the test set data; the mean square error index of the model is calculated to quantify the prediction accuracy of the model; then the prediction results of the model are displayed in the form of a chart using a visualization tool; in addition, a feature importance graph is drawn to show the contribution of different features to the prediction results, providing a basis for subsequent feature selection and model optimization.

9. The computer method for obtaining user activity on a deep learning platform according to claim 1, characterized in that: In step S8, new data is collected once a quarter and the random forest model is retrained to adapt to the new data distribution and user behavior patterns. At the same time, feature engineering and model algorithms are continuously optimized based on platform development and changes in user behavior. Through these continuous updates and optimization measures, the accuracy and effectiveness of user activity calculations are ensured to always remain at a high level.

10. A computer device for obtaining user activity on a deep learning platform, characterized by: An application is provided inside the computer device, and the application runs the computer method for obtaining user activity of a deep learning platform according to any one of claims 1 to 9.

Images