Activity tag generation method and equipment

Abstract

The embodiment of the invention provides an activity label generation method and equipment, and aims to improve the accuracy of activity labels. The method comprises the following steps: collecting behavior data of a target user, wherein the behavior data comprises time sequence data related to user behaviors; performing semantic analysis on the behavior data through a natural language processing technology to obtain a semantic analysis result corresponding to the behavior data; performing behavior pattern recognition on the behavior data through a pre-trained machine learning model to obtain a pattern recognition result corresponding to the behavior data; according to the semantic analysis result and the mode recognition result, an activity label of the target user is generated, and the activity label is used for representing the interest and / or activity participation degree of the user.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and device for generating an activity tag. Background Art

[0002] Activity tags are mainly used to reflect users' interests and / or activity participation, and are widely used in areas such as personalized recommendations, advertising targeting, and activity effectiveness evaluation to enhance user experience.

[0003] Traditional activity tag generation methods are mainly based on simple rules or statistical methods, such as setting fixed thresholds and conditions to generate activity tags, or generating activity tags by counting the frequency of user behaviors.

[0004] However, the above-mentioned activity tag generation methods based on simple rules or statistical methods lack flexibility and are difficult to capture complex behavior patterns, resulting in low accuracy of activity tags and difficulty in accurately reflecting users' real interests and / or activity participation. Summary of the invention

[0005] In view of the above problems, embodiments of the present application provide a method and device for generating an activity tag, so as to overcome the above problems or at least partially solve the above problems.

[0006] In a first aspect of an embodiment of the present application, a method for generating an activity tag is provided, the method comprising:

[0007] Collecting behavior data of target users, wherein the behavior data includes: time series data related to user behavior;

[0008] Performing semantic analysis on the behavior data through natural language processing technology to obtain semantic analysis results corresponding to the behavior data;

[0009] Performing behavior pattern recognition on the behavior data through a pre-trained machine learning model to obtain a pattern recognition result corresponding to the behavior data;

[0010] An activity tag of the target user is generated according to the semantic analysis result and the pattern recognition result, and the activity tag is used to characterize the user's interest and / or activity participation.

[0011] As a possible implementation, the semantic analysis of the behavior data is performed by natural language processing technology to obtain the semantic analysis result corresponding to the behavior data, including:

[0012] In a case where the behavior data includes time series data related to user feedback behavior, a first semantic analysis is performed on the time series data related to the user feedback behavior by using a natural language processing technology to obtain a first semantic analysis result, wherein the first semantic analysis includes: at least one of sentiment analysis, feedback intention analysis, and topic distribution analysis;

[0013] In the case where the behavior data includes time series data related to user search behavior, a second semantic analysis is performed on the time series data related to the user search behavior through natural language processing technology to obtain a second semantic analysis result, and the second semantic analysis includes: at least one of: search intent analysis and topic distribution analysis.

[0014] As a possible implementation manner, generating the activity tag of the target user according to the semantic analysis result and the pattern recognition result includes:

[0015] In the case where the semantic analysis result includes a sentiment analysis result, determining the feedback object targeted by the positive sentiment in the sentiment analysis result as the object of interest of the target user;

[0016] generating the activity tag according to the object of interest of the target user, other analysis results in the semantic analysis result, and the pattern recognition result;

[0017] The method further comprises:

[0018] A first label for assisting in improving a product is generated according to feedback information associated with negative sentiment in the sentiment analysis result.

[0019] As a possible implementation, the behavior data is subjected to behavior pattern recognition by using a pre-trained machine learning model to obtain a pattern recognition result corresponding to the behavior data, including:

[0020] Extracting a time series feature vector from the behavior data using a pre-trained deep learning model;

[0021] Extracting text feature vectors from the behavior data using a pre-trained word vector model;

[0022] The behavior data, the time series feature vector, and the text feature vector are used as input data sets, and pattern recognition is performed by using a plurality of pre-trained pattern recognition models, wherein the plurality of pattern recognition models are at least one of an integrated learning model and a deep learning model;

[0023] The pattern recognition result corresponding to the behavior data is determined according to the output results of the multiple pattern recognition models.

[0024] As a possible implementation manner, after collecting the behavior data of the target user each time, the method further includes:

[0025] Detecting abnormal behavior patterns on the behavior data using a pre-trained isolation forest model;

[0026] In the case where an abnormal behavior pattern is detected in the behavior data, the behavior data is discarded, and the activity tag of the target user is updated according to the detected abnormal behavior pattern to mark the user behavior that may be risky.

[0027] As a possible implementation, the method further includes:

[0028] Determine a similar user group to which the target user belongs based on characteristics of multiple dimensions of the target user, wherein the characteristics of the multiple dimensions include at least: demographic characteristics, behavioral characteristics, interest characteristics, and social relationship characteristics;

[0029] The activity tag of the target user is updated according to the group-level activity tag corresponding to the similar user group.

[0030] As a possible implementation, the method further includes:

[0031] Monitoring the behavior changes of the target users;

[0032] According to whether the behavior of the target user changes, it is determined whether to re-collect the behavior data of the target user to update the activity tag of the target user.

[0033] As a possible implementation, the method further includes:

[0034] According to the behavior duration of the target user, the activity tag of the target user is divided into a second tag associated with the long-term behavior habit and a third tag associated with the short-term behavior habit;

[0035] The second label and the third label distribution are distributed and stored long-term and temporarily.

[0036] As a possible implementation method, the collecting of target user behavior data includes:

[0037] Based on the preset data interface standards, the behavior data of the target users are collected from multiple platforms through application programming interfaces (APIs), logs and monitoring tools, and the collected behavior data are stored through multi-dimensional data tables.

[0038] According to a second aspect of the embodiments of the present application, a device for generating an activity tag is provided, the device comprising:

[0039] A data collection module is used to collect the behavior data of the target user, wherein the behavior data includes: time series data related to the user behavior;

[0040] A semantic analysis module, used to perform semantic analysis on the behavior data through natural language processing technology to obtain semantic analysis results corresponding to the behavior data;

[0041] A pattern recognition module, used to perform behavior pattern recognition on the behavior data through a pre-trained machine learning model to obtain a pattern recognition result corresponding to the behavior data;

[0042] The tag generation module is used to generate an activity tag of the target user according to the semantic analysis result and the pattern recognition result, wherein the activity tag is used to characterize the user's interest and / or activity participation.

[0043] According to a third aspect of an embodiment of the present application, an electronic device is provided, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the activity tag generation method described in the first aspect of the embodiment of the present application when executing the program stored in the memory.

[0044] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the activity tag generation method as described in the first aspect disclosed in the embodiment of the present application is implemented.

[0045] According to a fifth aspect of the embodiments of the present application, a computer program product is provided, including a computer program / instruction, which, when executed by a processor, implements the activity tag generating method as described in the first aspect disclosed in the embodiments of the present application.

[0046] The embodiments of the present application include the following advantages: natural language processing technology and machine learning models are combined and applied to the generation process of activity tags, and semantic analysis and pattern recognition of behavioral data are performed to achieve a more flexible and accurate understanding and analysis of user behavior from multiple dimensions, thereby improving the accuracy of the generated activity tags, so that they can more accurately reflect the user's real interests and / or activity participation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0048] Figure 1 This is a flowchart of a method for generating an activity tag in an embodiment of the present application;

[0049] Figure 2 This is a schematic diagram of a system flow for generating an activity tag in an embodiment of the present application;

[0050] Figure 3 It is a structural schematic diagram of an activity tag generating device according to an embodiment of the present application;

[0051] Figure 4 It is a schematic diagram of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0053] Activity tags are mainly used to reflect users' interests and / or activity participation, and are widely used in personalized recommendations, advertising targeting, and activity effect evaluation. For example, accurate activity tags can improve the personalization of recommendation systems, or enable accurate advertising to increase advertising conversion rates, or accurately analyze activity effects to guide future activity planning and execution, thereby improving user experience.

[0054] In the related art, traditional activity tag generation methods are mainly based on simple rules or statistical methods, which lack flexibility and have difficulty in capturing complex behavior patterns, resulting in low accuracy of activity tags and difficulty in accurately reflecting users' true interests and / or activity participation.

[0055] For example, the activity label generation method based on simple rules mainly generates activity labels by setting fixed thresholds and conditions, which lacks flexibility; the activity label generation method based on statistical methods mainly generates activity labels by counting the frequency of user behaviors. For example, if a user clicks on a link of a product more than a certain number of times, an activity label of a high-frequency clicker related to the product is generated for the user to indicate that the user is interested in the product, or has a high degree of participation in the marketing activities for the product. However, this frequency counting method is difficult to capture complex behavior patterns.

[0056] In response to the problem of low accuracy of activity labels existing in related technologies, an embodiment of the present application provides an activity label generation solution, which combines natural language processing technology and machine learning models in the activity label generation process, and performs semantic analysis and pattern recognition on behavioral data to achieve a more flexible and accurate understanding and analysis of user behavior from multiple dimensions, thereby improving the accuracy of the generated activity labels.

[0057] Reference Figure 1 FIG. 1 is a flowchart showing a method for generating an activity tag in an embodiment of the present application, and the method includes the following steps:

[0058] Step S11: Collecting behavior data of target users, wherein the behavior data includes: time series data related to user behavior.

[0059] In specific implementation, the target user's behavior data can be collected through the application programming interface (API) or related data monitoring tools configured on the target platform, or by analyzing the log data provided by the target platform. The behavior data includes but is not limited to the time series data of the user's click, browse, purchase, search and comment behaviors.

[0060] Optionally, refer to Figure 2 The system flow chart of activity tag generation shown in the figure performs data cleaning (such as denoising, filtering special characters, filling missing values), format unification, word segmentation and other data preprocessing on the collected behavior data in order to improve data quality and help speed up subsequent processing efficiency.

[0061] Step S12: Performing semantic analysis on the behavior data through natural language processing technology to obtain semantic analysis results corresponding to the behavior data.

[0062] In specific implementations, natural language processing (NLP) technology can be used to perform keyword extraction and synonym conversion on the behavioral data, and then perform semantic analysis on the processed data to identify whether there are semantic descriptions related to the user's interests or activity participation. For example, relevant semantic descriptions of the user's desire to purchase a certain product or to frequently participate in a certain activity can be identified, thereby obtaining the semantic analysis results.

[0063] Step S13: Performing behavior pattern recognition on the behavior data using a pre-trained machine learning model to obtain a pattern recognition result corresponding to the behavior data.

[0064] In specific implementation, behavioral patterns can be manually annotated on historical behavioral data samples to construct training samples, and the machine learning model can be trained by using the training samples, thereby making use of the learning and generalization capabilities of the machine learning model to achieve recognition of complex behavioral patterns.

[0065] Step S14: generating an activity tag of the target user according to the semantic analysis result and the pattern recognition result, wherein the activity tag is used to characterize the user's interest and / or activity participation.

[0066] In a specific implementation, the semantic analysis results and the pattern recognition results can be input into a pre-trained classification task model to obtain activity labels of corresponding categories; alternatively, the semantic analysis results and the pattern recognition results can be used as user features to find similar groups to which the target user belongs through cluster analysis, and then the activity labels of the similar groups can be used as the activity labels of the target user; the embodiment of the present application does not impose any specific restrictions on the implementation method of generating activity labels based on the semantic analysis results and the pattern recognition results.

[0067] It can be understood that with the help of the above-mentioned semantic analysis results reflecting the semantic description of user interests and / or activity participation, and the pattern recognition results reflecting the user's complex behavioral habits, it is possible to achieve a more flexible and accurate understanding and analysis of user behavior from multiple dimensions, thereby overcoming the shortcomings of related technologies such as lack of flexibility and difficulty in capturing complex behavioral patterns, and achieving improved accuracy in activity label generation.

[0068] By adopting the technical solution of the embodiment of the present application, natural language processing technology and machine learning models are combined and applied to the process of generating activity labels. By performing semantic analysis and pattern recognition on behavioral data, a more flexible and accurate understanding and analysis of user behavior from multiple dimensions can be achieved. This can improve the accuracy of the generated activity labels, so that they can more accurately reflect the user's true interests and / or activity participation.

[0069] As a possible implementation, in step S11, collecting the behavior data of the target user includes:

[0070] Step S111: Based on the preset data interface standard, the behavior data of the target user is collected from multiple platforms through the application programming interface API, logs and monitoring tools, and the collected behavior data is stored through a multi-dimensional data table.

[0071] In the specific implementation, by setting a unified data interface standard and aggregating data from different platforms according to the standard, it is conducive to the rapid integration of multi-platform data, thereby solving the "data island" problem and improving the richness of data sources; and using multi-dimensional data tables to store the collected behavioral data is conducive to the subsequent rapid differentiation of data of different dimensions in the behavioral data (such as data of different behaviors), thereby improving processing efficiency.

[0072] As a possible implementation, in step S12, semantic analysis is performed on the behavior data by natural language processing technology to obtain semantic analysis results corresponding to the behavior data, including:

[0073] Step S121: When the behavior data includes time series data related to user feedback behavior, a first semantic analysis is performed on the time series data related to the user feedback behavior through natural language processing technology to obtain a first semantic analysis result, wherein the first semantic analysis includes: at least one of: sentiment analysis, feedback intention analysis and topic distribution analysis.

[0074] In the specific implementation, the first semantic analysis is performed on the time series data related to the user's comments, evaluations (such as ratings) and other feedback behaviors through natural language processing technology, wherein:

[0075] When performing sentiment analysis on the time series data, a pre-trained convolutional neural network with sentiment classification capability can be used, or a sentiment dictionary such as SentiWordNet can be used to perform sentiment polarity (such as positive, negative, and neutral) analysis or sentiment score calculation, thereby obtaining the sentiment analysis result in the first semantic analysis result. It is understandable that with the help of sentiment analysis, the sentiment reflected in the user feedback behavior can be accurately understood, for example, the user's positive emotional comment behavior on a product can be identified, which helps to confirm the user's interest in the product, thereby improving the accuracy of the relevant activity tags.

[0076] When performing feedback intention analysis on the time series data, classification algorithms such as neural networks and support vector machines (SVM) can be used to identify the user's feedback intention, thereby obtaining the feedback intention analysis result in the first semantic analysis result. It can be understood that the user's intention to make comments or evaluations and other feedback can be accurately understood by means of feedback intention analysis, such as identifying whether the user expects to buy a certain product or expects to frequently participate in a certain activity, thereby helping to confirm the user's interest in the product or participation in the activity, thereby improving the accuracy of the relevant activity tags.

[0077] When performing topic distribution analysis on the time series data, a Latent Dirichlet Allocation (LDA) model can be used to extract topic distribution, thereby obtaining the topic distribution analysis result in the first semantic analysis result. It can be understood that the core objects that users pay attention to in feedback behavior can be accurately identified with the help of topic distribution analysis, thereby generating more targeted activity tags.

[0078] Step S122: When the behavior data includes time series data related to user search behavior, a second semantic analysis is performed on the time series data related to the user search behavior through natural language processing technology to obtain a second semantic analysis result, wherein the second semantic analysis includes: at least one of search intent analysis and topic distribution analysis.

[0079] In specific implementation, natural language processing technology is used to perform a second semantic analysis on the time series data related to the user's search behavior, where:

[0080] When performing search intent analysis on the time series data, a classification algorithm such as a neural network or SVM can be used to identify the user's search intent, thereby obtaining a search intent analysis result in the second semantic analysis result. It is understandable that the user's actual search intent can be accurately understood with the help of search intent analysis, such as identifying whether the user expects to purchase a certain product or participate in a certain activity, thereby helping to confirm the user's interest in the product or participation in the activity, thereby improving the accuracy of the relevant activity tags.

[0081] When performing topic distribution analysis on the time series data, the LDA model can be used to extract topic distribution, thereby obtaining the topic distribution analysis result in the second semantic analysis result. It can be understood that the core objects that users pay attention to in search behavior can be accurately identified with the help of topic distribution analysis, thereby generating more targeted activity tags.

[0082] Optionally, in step S14, generating an activity tag of the target user according to the semantic analysis result and the pattern recognition result includes:

[0083] Step S141: when the semantic analysis result includes a sentiment analysis result, the feedback object directed by the positive sentiment in the sentiment analysis result is determined as the object of interest of the target user.

[0084] In specific implementation, based on the sentiment analysis results, positive feedback from users can be screened out, and the objects to which these positive feedbacks are directed can be determined. For example, if a user has recently given positive feedback on multiple products under a certain brand, it means that the user may be more interested in the brand. Therefore, the brand can be used as the object of interest of the target user.

[0085] Step S142: Generate the activity tag according to the object of interest of the target user, other analysis results in the semantic analysis result, and the pattern recognition result.

[0086] In specific implementation, an interest-related activity tag can be directly generated based on the target user's object of interest, such as generating an activity tag for interest in a certain brand; or, when the other analysis results or the pattern recognition results also reflect the interest level or participation level of the object of interest, these reflected situations can be combined to generate relevant activity tags to further improve accuracy.

[0087] The method further comprises:

[0088] Step S143: generating a first label for assisting product improvement based on feedback information associated with negative emotions in the sentiment analysis result.

[0089] In a specific implementation, based on the sentiment analysis results, negative feedback from users can be screened out, and the specific information of these negative feedbacks can be identified. For example, if it is identified that users have provided feedback about poor sound quality on a product in their comments, a first label indicating that the sound quality needs to be improved can be generated accordingly, so that subsequent relevant personnel can make targeted improvements to the product based on the first label.

[0090] As a possible implementation, in step S13, the behavior data is subjected to behavior pattern recognition by using a pre-trained machine learning model to obtain a pattern recognition result corresponding to the behavior data, including:

[0091] Step S131: extracting a time series feature vector from the behavior data using a pre-trained deep learning model.

[0092] In specific implementation, a pre-trained long short-term memory (LSTM) model can be used to extract time series feature vectors from the behavior data. The time series feature vectors may include: regular characteristics of user behavior changes over time, seasons or periodic events, and random change characteristics of users, etc.

[0093] Step S132: extracting text feature vectors from the behavior data using a pre-trained word vector model.

[0094] In specific implementation, word vector models such as Word2Vec, TF-IDF and BERT can be used to convert the behavior data into word vectors to obtain text feature vectors.

[0095] Step S133: Use the behavior data, the time series feature vector and the text feature vector as input data sets, and perform pattern recognition through multiple pre-trained pattern recognition models, where the multiple pattern recognition models belong to at least one of an integrated learning model and a deep learning model.

[0096] In a specific implementation, one or more models related to ensemble learning models, such as random forests and gradient boosting trees, and / or one or more models related to deep learning models, such as Transformer, can be used to perform pattern recognition on the behavioral data.

[0097] It can be understood that due to the differences in the input data types of different models, for each pattern recognition model, in actual application, one or more items can be selected from the behavioral data, the time series feature vector and the text feature vector as the input data of the pattern recognition model based on the actual input data type of the pattern recognition model.

[0098] Optionally, the input data set may also include statistical features of the behavior data, such as the execution frequency or duration of a certain behavior.

[0099] Step S134: Determine the pattern recognition result corresponding to the behavior data according to the output results of the multiple pattern recognition models.

[0100] In specific implementation, for the same input data set, multiple pattern recognition models may produce different pattern recognition results, and these pattern recognition results can be combined through a fusion strategy to determine the final pattern recognition result.

[0101] For example, based on the voting method, the pattern recognition results of multiple pattern recognition models can be used to determine the final pattern recognition result by majority voting.

[0102] For another example, based on the weighted average method, weights can be assigned to different models according to the performance of the models (such as accuracy), and then the final pattern recognition result can be determined by calculating the weighted average.

[0103] For another example, based on the stacking method, the outputs of multiple pattern recognition models can be used as input, and a higher-level model (such as a logistic regression model) can be used for final classification.

[0104] In this embodiment, the behavior pattern recognition is performed by multi-dimensional feature extraction and multi-model recognition, which can identify more complex and hidden behavior patterns, thereby improving the recognition accuracy of the behavior pattern, thereby making the generated activity labels more accurate.

[0105] As a possible implementation manner, after collecting the behavior data of the target user each time, the method further includes:

[0106] Step S15: Detect abnormal behavior patterns of the behavior data using a pre-trained isolation forest model.

[0107] In specific implementation, the time series data of user behavior, the time interval and frequency of user behavior, and the aggregation of user behavior in a specific time period can be used as input data, and the isolation forest model can be used to identify abnormal behavior patterns that deviate from the user's regular behavior patterns.

[0108] Step S16: When an abnormal behavior pattern is detected in the behavior data, the behavior data is discarded, and the activity tag of the target user is updated according to the detected abnormal behavior pattern to mark the user behavior that may be risky.

[0109] In the specific implementation, considering that the detection of abnormal behavior patterns may be caused by the following erroneous values ​​contained in the behavior data, the behavior data can be directly discarded, that is, the behavior data will not be further processed such as semantic analysis or behavior pattern recognition to improve the quality of the data used to generate activity tags; considering that the detection of abnormal behavior patterns may also be due to some malicious or abnormal behaviors of users, such as abnormal transactions (such as a large number of purchases in a short period of time, which may be malicious behavior), abnormal browsing (such as a surge in click behavior in a short period of time, which may be robot behavior) or abnormal feedback (such as a large number of negative comments, which may be malicious attack behavior) and other abnormal behavior patterns, the activity tags can be dynamically updated according to the detected abnormal behavior patterns to mark user behaviors that may be risky (such as cheating, fraud, etc.), so as to facilitate the subsequent processing or monitoring of related users.

[0110] As a possible implementation, the method further includes:

[0111] Step S21: Determine the similar user group to which the target user belongs based on the characteristics of multiple dimensions of the target user, wherein the characteristics of the multiple dimensions include at least: demographic characteristics, behavioral characteristics, interest characteristics and social relationship characteristics.

[0112] Among them, the demographic characteristics include: age, gender, region, occupation, etc.; the behavioral characteristics include: click records, browsing history, purchase frequency, etc.; the interest characteristics include: interest-related activity tags generated based on historical behavior data, etc.; the social relationship characteristics include: friend relationships, social interactions, etc., such as whether there is frequent interaction with a specific group (such as friends who like the same brand) (which can be associated with activity tags such as brand preference under social influence), or whether there are a large number of social interactions such as sharing, likes, comments and forwarding in activities such as live broadcasts (which can be associated with activity tags such as high social interaction users).

[0113] In specific implementation, clustering analysis algorithms such as K-means and hierarchical clustering may be used to group similar users according to the characteristics of the multiple dimensions, and determine the similar user group to which the target user belongs based on the grouping results.

[0114] Step S22: updating the activity tag of the target user according to the group-level activity tag corresponding to the similar user group.

[0115] In a specific implementation, the same activity tags possessed by all users in a similar user group can be determined as group-level activity tags for the similar user group; or, the frequency of occurrence of each activity tag in a similar user group can be counted, and the activity tags that appear frequently can be determined as group-level activity tags for the similar user group; or, the main features (i.e., common features) shared by all users in a similar user group can be analyzed, and group-level activity tags can be generated based on the common features.

[0116] After determining the group-level activity tag of the similar user group to which the target user belongs, the activity tag of the target user is updated using the group-level activity tag, for example, by adding the group-level activity tag to the activity tag of the target user, or by replacing the existing activity tag of the target user with the group-level activity tag.

[0117] In this embodiment, based on the common interests or engagement reflected by the group-level activity tags, the activity tags of the target users are updated, which can further improve the accuracy of the activity tags.

[0118] As a possible implementation, the method further includes:

[0119] Step S31: monitoring the behavior changes of the target user.

[0120] During specific implementation, the behavior changes can be determined by monitoring the execution frequency and duration of the user behavior.

[0121] Step S32: Determine whether to re-collect the behavior data of the target user according to whether the behavior of the target user has changed, so as to update the activity tag of the target user.

[0122] In specific implementation, if the behavior of the target user changes, for example, the target user has newly participated in activity A, or the frequency of participating in activity B has significantly decreased recently, or the frequency of participating in activity C has significantly increased recently, etc., the behavior data of the target user can be collected again, and the activity label of the target user can be updated by performing semantic analysis and behavior pattern recognition on the behavior data.

[0123] In this embodiment, the accuracy of the activity tags can be further improved by dynamically updating the user's activity tags as the user's behavior changes.

[0124] Optionally, refer to Figure 2 As shown, the method also includes:

[0125] Step S33: According to the behavior duration of the target user, the activity tag of the target user is divided into a second tag associated with the long-term behavior habit and a third tag associated with the short-term behavior habit.

[0126] In the specific implementation, the activity label generated based on long-term behavior will be determined as the second label. This type of label is generated based on the user's long-term behavioral habits and usually will not change; and the activity label generated based on short-term behavior will be determined as the third label. This type of label is generated based on the user's short-term behavioral habits and is more likely to change.

[0127] Step S34: performing long-term storage and temporary storage on the second tag distribution and the third tag distribution.

[0128] In a specific implementation, a persistent storage device can be used to store the second tag for a long time to avoid data loss; and a non-persistent storage device can be used to temporarily store the third tag, thereby reducing equipment costs through classified storage.

[0129] It should be noted that, for the method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the described order of actions, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0130] Figure 3: is a schematic diagram of the structure of an activity tag generating device according to an embodiment of the present application, the device comprising:

[0131] A data collection module is used to collect the behavior data of the target user, wherein the behavior data includes: time series data related to the user behavior;

[0132] A semantic analysis module, used to perform semantic analysis on the behavior data through natural language processing technology to obtain semantic analysis results corresponding to the behavior data;

[0133] A pattern recognition module, used to perform behavior pattern recognition on the behavior data through a pre-trained machine learning model to obtain a pattern recognition result corresponding to the behavior data;

[0134] The tag generation module is used to generate an activity tag of the target user according to the semantic analysis result and the pattern recognition result, wherein the activity tag is used to characterize the user's interest and / or activity participation.

[0135] By adopting the technical solution of the embodiment of the present application, natural language processing technology and machine learning models are combined and applied to the process of generating activity labels. By performing semantic analysis and pattern recognition on behavioral data, a more flexible and accurate understanding and analysis of user behavior from multiple dimensions can be achieved. This can improve the accuracy of the generated activity labels, so that they can more accurately reflect the user's true interests and / or activity participation.

[0136] Optionally, the semantic analysis module includes:

[0137] A first analysis module is used to perform a first semantic analysis on the time series data related to the user feedback behavior by natural language processing technology to obtain a first semantic analysis result when the behavior data includes time series data related to the user feedback behavior, wherein the first semantic analysis includes at least one of sentiment analysis, feedback intention analysis, and topic distribution analysis;

[0138] The second analysis module is used to perform a second semantic analysis on the time series data related to the user search behavior through natural language processing technology when the behavior data includes time series data related to the user search behavior, so as to obtain a second semantic analysis result. The second semantic analysis includes: at least one of search intent analysis and topic distribution analysis.

[0139] Optionally, the label generation module includes:

[0140] A first generating module is used for determining, when the semantic analysis result includes a sentiment analysis result, a feedback object targeted by a positive sentiment in the sentiment analysis result as an object of interest to the target user;

[0141] A second generating module, configured to generate the activity tag according to the object of interest of the target user, other analysis results in the semantic analysis result, and the pattern recognition result;

[0142] The device also includes:

[0143] The third generating module is used to generate a first label for assisting in improving the product according to the feedback information associated with the negative sentiment in the sentiment analysis result.

[0144] Optionally, the pattern recognition module includes:

[0145] A first recognition module, configured to extract a time series feature vector from the behavior data using a pre-trained deep learning model;

[0146] A second recognition module is used to extract a text feature vector from the behavior data using a pre-trained word vector model;

[0147] A third recognition module, configured to use the behavior data, the time series feature vector, and the text feature vector as input data sets, and perform pattern recognition through a plurality of pre-trained pattern recognition models, wherein the plurality of pattern recognition models are at least one of an integrated learning model and a deep learning model;

[0148] The fourth recognition module is used to determine the pattern recognition results corresponding to the behavior data according to the output results of the multiple pattern recognition models.

[0149] Optionally, the device further comprises:

[0150] A first processing module is used to detect abnormal behavior patterns of the behavior data by using a pre-trained isolation forest model after collecting the behavior data of the target user each time;

[0151] The second processing module is used to discard the behavior data when an abnormal behavior pattern is detected in the behavior data, and update the activity tag of the target user according to the detected abnormal behavior pattern to mark the user behavior that may be risky.

[0152] Optionally, the device further comprises:

[0153] A third processing module, configured to determine a similar user group to which the target user belongs based on characteristics of multiple dimensions of the target user, wherein the characteristics of the multiple dimensions at least include: demographic characteristics, behavioral characteristics, interest characteristics, and social relationship characteristics;

[0154] The fourth processing module is used to update the activity tag of the target user according to the group-level activity tag corresponding to the similar user group.

[0155] Optionally, the device further comprises:

[0156] A behavior monitoring module, used to monitor the behavior changes of the target user;

[0157] The dynamic update module is used to determine whether to re-collect the behavior data of the target user according to whether the behavior of the target user changes, so as to update the activity tag of the target user.

[0158] Optionally, the device further comprises:

[0159] A label division module, used for dividing the activity label of the target user into a second label associated with a long-term behavior habit and a third label associated with a short-term behavior habit according to the behavior duration of the target user;

[0160] The tag storage module is used for long-term storage and temporary storage of the second tag and the third tag distribution.

[0161] Optionally, the data collection module includes:

[0162] The collection submodule is used to collect the behavior data of the target user from multiple platforms based on the preset data interface standard through the application programming interface API, logs and monitoring tools, and store the collected behavior data through multi-dimensional data tables.

[0163] It should be noted that the device embodiment is similar to the method embodiment, so the description is relatively simple, and the relevant parts can be referred to the method embodiment.

[0164] The present application also provides an electronic device, such as Figure 4 As shown, it includes a processor 401, a communication interface 402, a memory 403 and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other through the communication bus 404.

[0165] Memory 403, used for storing computer programs;

[0166] The processor 401 is used to execute the program stored in the memory 403, such as the steps of the activity tag generating method disclosed in the embodiment of the present application.

[0167] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0168] The communication interface is used for communication between the above terminal and other devices.

[0169] The memory may include a random access memory (RAM) or a non-volatile memory (non-volatile memory), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0170] The above-mentioned processor can be a general-purpose processor, including a central processing unit, a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0171] The embodiment of the present application also provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the activity tag generating method disclosed in the embodiment of the present application is implemented.

[0172] The embodiment of the present application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the activity tag generation method disclosed in the embodiment of the present application.

[0173] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0174] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including 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 restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0175] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0176] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application are included in the protection scope of the present application.

Claims

1. A method for generating an activity tag, characterized in that: The method comprises: Collecting behavior data of target users, wherein the behavior data includes: time series data related to user behavior; Performing semantic analysis on the behavior data through natural language processing technology to obtain semantic analysis results corresponding to the behavior data; Performing behavior pattern recognition on the behavior data through a pre-trained machine learning model to obtain a pattern recognition result corresponding to the behavior data; An activity tag of the target user is generated according to the semantic analysis result and the pattern recognition result, and the activity tag is used to characterize the user's interest and / or activity participation.

2. The method according to claim 1, characterized in that The semantic analysis of the behavior data is performed by natural language processing technology to obtain the semantic analysis results corresponding to the behavior data, including: In a case where the behavior data includes time series data related to user feedback behavior, a first semantic analysis is performed on the time series data related to the user feedback behavior by using a natural language processing technology to obtain a first semantic analysis result, wherein the first semantic analysis includes: at least one of sentiment analysis, feedback intention analysis, and topic distribution analysis; In the case where the behavior data includes time series data related to user search behavior, a second semantic analysis is performed on the time series data related to the user search behavior through natural language processing technology to obtain a second semantic analysis result, and the second semantic analysis includes: at least one of: search intent analysis and topic distribution analysis.

3. The method according to claim 2, characterized in that The step of generating the activity tag of the target user according to the semantic analysis result and the pattern recognition result includes: In the case where the semantic analysis result includes a sentiment analysis result, determining the feedback object targeted by the positive sentiment in the sentiment analysis result as the object of interest of the target user; generating the activity tag according to the object of interest of the target user, other analysis results in the semantic analysis result, and the pattern recognition result; The method further comprises: A first label for assisting in improving a product is generated according to feedback information associated with negative sentiment in the sentiment analysis result.

4. The method according to claim 1, characterized in that: The method of performing behavior pattern recognition on the behavior data by using a pre-trained machine learning model to obtain a pattern recognition result corresponding to the behavior data includes: Extracting a time series feature vector from the behavior data using a pre-trained deep learning model; Extracting text feature vectors from the behavior data using a pre-trained word vector model; The behavior data, the time series feature vector, and the text feature vector are used as input data sets, and pattern recognition is performed by using a plurality of pre-trained pattern recognition models, wherein the plurality of pattern recognition models are at least one of an integrated learning model and a deep learning model; The pattern recognition result corresponding to the behavior data is determined according to the output results of the multiple pattern recognition models.

5. The method according to claim 1, characterized in that After collecting the behavior data of the target user each time, the method further includes: Detecting abnormal behavior patterns on the behavior data using a pre-trained isolation forest model; In the case where an abnormal behavior pattern is detected in the behavior data, the behavior data is discarded, and the activity tag of the target user is updated according to the detected abnormal behavior pattern to mark the user behavior that may be risky.

6. The method according to claim 1, characterized in that The method further comprises: Determine a similar user group to which the target user belongs based on characteristics of multiple dimensions of the target user, wherein the characteristics of the multiple dimensions include at least: demographic characteristics, behavioral characteristics, interest characteristics, and social relationship characteristics; The activity tag of the target user is updated according to the group-level activity tag corresponding to the similar user group.

7. The method according to claim 1, characterized in that The method further comprises: Monitoring the behavior changes of the target users; According to whether the behavior of the target user changes, it is determined whether to re-collect the behavior data of the target user to update the activity tag of the target user.

8. The method according to claim 7, characterized in that The method further comprises: According to the behavior duration of the target user, the activity tag of the target user is divided into a second tag associated with the long-term behavior habit and a third tag associated with the short-term behavior habit; The second label and the third label distribution are distributed and stored long-term and temporarily.

9. The method according to any one of claims 1 to 8, characterized in that: The collected behavior data of the target user includes: Based on the preset data interface standards, the behavior data of the target users are collected from multiple platforms through application programming interfaces (APIs), logs and monitoring tools, and the collected behavior data are stored through multi-dimensional data tables.

10. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is configured to implement the activity tag generating method according to any one of claims 1 to 9 when executing the program stored in the memory.

Images