WO2023273299A1 - 应用用户行为数据的处理方法、装置、设备及存储介质 - Google Patents

应用用户行为数据的处理方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2023273299A1
WO2023273299A1 PCT/CN2022/071485 CN2022071485W WO2023273299A1 WO 2023273299 A1 WO2023273299 A1 WO 2023273299A1 CN 2022071485 W CN2022071485 W CN 2022071485W WO 2023273299 A1 WO2023273299 A1 WO 2023273299A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
application
preset
business operation
sequence
Prior art date
Application number
PCT/CN2022/071485
Other languages
English (en)
French (fr)
Inventor
沈嘉良
胡英东
李雪丽
贾素苇
徐宁
陶醉
陈煦
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2023273299A1 publication Critical patent/WO2023273299A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2462—Approximate or statistical queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2465—Query processing support for facilitating data mining operations in structured databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • the present application relates to the field of intelligent decision-making of artificial intelligence, and in particular to a processing method, device, equipment and storage medium for application user behavior data.
  • static data of applications are collected, based on data analysis indicators (such as: application The number of daily active users DAU, the number of monthly active users MAU, the number of new users, the retention rate of users, the jump rate between pages, the loss rate and the transaction rate, etc.), and the user business operation intention for the static data of the application forecast, obtain the forecast result, and send the business operation adjustment strategy corresponding to the forecast result to the client.
  • data analysis indicators such as: application The number of daily active users DAU, the number of monthly active users MAU, the number of new users, the retention rate of users, the jump rate between pages, the loss rate and the transaction rate, etc.
  • the inventor realized that in the above method, since the data analysis indicators adopted are based on mechanized data indicators, they can only reflect the overall usage of the user, and cannot conduct in-depth analysis of the operation behavior of a single user. The prediction accuracy is low, which leads to the low accuracy of the business operation adjustment strategy based on the application user behavior data analysis.
  • the present application provides a processing method, device, device, and storage medium for application user behavior data, which are used to improve the accuracy of business operation adjustment strategies based on application user behavior data analysis.
  • the present application provides a processing method, device, device, and storage medium for application user behavior data, which are used to improve the accuracy of business operation adjustment strategies based on application user behavior data analysis.
  • the first aspect of the present application provides a method for processing application user behavior data, including:
  • the second aspect of the present application provides a processing device for applying user behavior data, including a memory, a processor, and computer-readable instructions stored on the memory and operable on the processor, and the processor executes the When the computer-readable instructions are described, the following steps are implemented:
  • the third aspect of the present application provides a computer-readable storage medium, where computer instructions are stored in the computer-readable storage medium, and when the computer instructions are run on the computer, the computer is made to perform the following steps:
  • the fourth aspect of the present application provides a processing device for application user behavior data, including:
  • a statistics module configured to obtain application user behavior data, and perform preset type classification, sequence data conversion, and classification statistics on the application user behavior data to obtain application operation record sequences and application data to be processed;
  • the prediction module is used to predict the subsequent business operation intention of the application operation record sequence through the preset cyclic neural network model, and obtain business operation intention prediction data;
  • the classification module is used to call a preset dynamic classification model, classify the application operation record sequence based on business operation intentions, obtain dynamic score data, and call a preset static classification model to classify the application data to be processed Perform classification based on business operation intentions to obtain static score data;
  • the first acquisition module is configured to sum the dynamic score data and the static score data to obtain target score data, and obtain the target score data from the predicted value according to the business operation intention prediction data and the target score data Get the business operation adjustment strategy to be pushed in the configuration adjustment strategy sequence.
  • the application user behavior data is obtained, and the preset type classification, sequence data conversion and classification statistics are performed on the application user behavior data to obtain the application operation record sequence and application data to be processed;
  • the network model predicts the intention of subsequent business operations on the application operation record sequence, and obtains the business operation intention prediction data; calls the preset dynamic classification model, classifies the application operation record sequence based on the business operation intention, and obtains the dynamic score data.
  • the preset static classification model classify the application data to be processed based on the business operation intention, and obtain the static score data; sum the dynamic score data and the static score data to obtain the target score data, and according to the business
  • the operation intention prediction data and target score data are used to obtain the business operation adjustment strategy to be pushed from the preset adjustment strategy sequence.
  • the application user behavior data is analyzed from the perspective of machine learning and time series deep learning, which ensures the accuracy and comprehensiveness of the prediction results of business operation intentions, and can analyze the operation behavior of a single user itself. In-depth analysis, and realize the classification of static data and dynamic time series data of application user behavior data, improve the prediction accuracy of business operation intention, and then improve the accuracy of business operation adjustment strategy based on application user behavior data analysis.
  • FIG. 1 is a schematic diagram of an embodiment of a method for processing application user behavior data in an embodiment of the present application
  • FIG. 2 is a schematic diagram of another embodiment of the processing method of application user behavior data in the embodiment of the present application
  • FIG. 3 is a schematic diagram of an embodiment of a processing device for applying user behavior data in an embodiment of the present application
  • FIG. 4 is a schematic diagram of another embodiment of a processing device for applying user behavior data in the embodiment of the present application.
  • Fig. 5 is a schematic diagram of an embodiment of a processing device for applying user behavior data in the embodiment of the present application.
  • Embodiments of the present application provide a processing method, device, device, and storage medium for application user behavior data, which improve the accuracy of business operation adjustment strategies based on application user behavior data analysis.
  • An embodiment of the processing method of application user behavior data in the embodiment of the present application includes:
  • the subject of execution of the present application may be a processing device for application user behavior data, and may also be a terminal or a server, which is not specifically limited here.
  • the embodiment of the present application is described by taking the server as an execution subject as an example.
  • the server After the server is authorized by the user, it calls the preset timer to execute the timing task.
  • the timing task calls the preset crawler or grabbing tool to grab the pending application (application, APP)
  • Application user behavior data the number of applications to be processed includes one or more, that is, the application user behavior data can be the data corresponding to one application program to be processed, and the application user behavior data can also be the data corresponding to more than one application program to be processed
  • the application user behavior data includes the record data generated by the application program to be processed based on user operations, user data based on the application program to be processed, and the status data of the user's business operations on the application program to be processed; the missing value of the application user behavior data Filling, same-type merging, and data desensitization are used to obtain preprocessed application user behavior data, which improves the quality and security of application user behavior data.
  • the server classifies the preprocessed application user behavior data according to the preset type (the preset type includes the application operation data type and the application user data type), and obtains the pre-classified application operation data and the pre-classified application data; Data extraction is performed on the pre-classified application operation data in a preset time period to obtain the extracted application operation data; the extracted application operation data is converted into time series data to obtain a sequence of application operation records, wherein the pre-classified application operation data
  • the data includes user application operation record data and user application purchase data; classify the pre-classified application data by functional area to obtain the classified user application data, and perform statistics on the classified user application data according to the preset statistical indicators.
  • the preset statistical indicators include but not limited to the user's login time, login times, operation times, stay time in each section and preference tags based on the application to be processed.
  • the server pre-trains and optimizes the created recurrent neural network model through the preset historical application operation record sequence, and obtains the final recurrent neural network model (recurrent neural network, RNN).
  • the recurrent neural network model is a deep bidirectional RNN.
  • the neural network model can predict the subsequent optimal operation that should be performed to achieve business operations in each current state.
  • the server invokes the preset cyclic neural network model, through the input layer, multiple hidden layers and output layers in the cyclic neural network model, performs sequence feature extraction and activation function calculation based on the memory state on the application operation record sequence to obtain the target follow-up business Operation information
  • the target follow-up business operation information includes the target follow-up business operation data and the probability value of the target follow-up business operation data based on the business operation results
  • the target follow-up business operation data includes the optimal follow-up operation and the suboptimal follow-up operation
  • through the target Follow-up business operation information classify the application operation record sequence based on the business operation achievement of each current state, and obtain the business operation intention prediction data.
  • the server pre-records sequence samples through application operations, trains and optimizes the preset dynamic prediction model, and obtains a dynamic classification model.
  • the dynamic classification model is used to record a large amount of operation data and is a time series. Subsequent operations have an impact on the previous overall operations. Classify data with certain dependencies, that is, the dynamic classification model is used to classify and predict dynamic time series data (that is, application operation record sequence samples).
  • the dynamic classification model can be a long short-term memory network (long short-term memory, LSTM) Model; the server pre-trains and optimizes the preset static prediction model through user application data samples to obtain a static classification model.
  • the static classification model is used to classify the user's business execution status (business operation intention) for the data collected at the same time , that is, the static classification model is used to classify and predict static data (ie user application data samples), the static classification model can be extreme gradient boosting (extreme gradient boosting, xgboost), wherein the application operation records sequence samples and user application data The execution process of obtaining the sample is similar to the execution process of step 101, and will not be repeated here.
  • the server calls the preset dynamic classification model, extracts multi-level time series features from the application operation record sequence, obtains the time series features of the operation records, and performs probability value calculation and probability value discrimination based on business operation intentions on the time series features of the operation records.
  • Obtain dynamic score data call the preset static classification model, perform multi-level feature extraction on the application data to be processed, obtain user application features, perform probability value calculation and probability value discrimination based on business operation intentions on user application features, and obtain static score Value data, business operation intention such as: purchase intention.
  • both the dynamic classification model and the static classification model can be a combination of feature extraction network and classification network.
  • the feature extraction network is used for multi-level feature extraction.
  • the feature extraction network can be the target detection model ET-YOLOV3, which improves the dynamic classification. Accuracy of value data and static score data.
  • the business operation adjustment strategy includes but is not limited to the application adjustment strategy of the business operation, the same type of content adjustment strategy and the function adjustment strategy, for example: if the business operation is the purchase operation on the application program, in order to achieve the purchase or increase the purchase intention, push For applications frequently used by users or with high ratings (i.e. application adjustment strategies), the push operation path is similar to the current operation in the application user behavior data (i.e. the same type of content adjustment strategy), and the push is similar to the business operation function corresponding to the application user behavior data
  • the function module that is, the function adjustment strategy).
  • the server converts the dynamic score data and the static score data into two-dimensional matrix respectively to obtain the dynamic two-dimensional matrix and the static two-dimensional matrix, calculates the arithmetic mean of the dynamic two-dimensional matrix and the static two-dimensional matrix, and obtains the target score data ; Compare and analyze the target score data with the business operation intention prediction data and the preset threshold to obtain the intention analysis result and the threshold analysis result. If the intention analysis result is the result of the corresponding push, obtain the score from the preset adjustment strategy sequence The policy that ranks the first in the value descending order, so as to obtain the business operation adjustment policy to be pushed, or send the intention analysis results and threshold analysis results to the preset terminal, and the person in charge of the preset terminal will follow up manually. If the analysis result corresponds to a result that does not need to be pushed, it will not be processed.
  • the application user behavior data is analyzed from the perspective of machine learning and time series deep learning, which ensures the accuracy and comprehensiveness of the prediction results of business operation intentions, and can analyze the operation behavior of a single user itself.
  • In-depth analysis and realize the classification of static data and dynamic time series data of application user behavior data, improve the prediction accuracy of business operation intention, and then improve the accuracy of business operation adjustment strategy based on application user behavior data analysis.
  • FIG. 2 another embodiment of the processing method of application user behavior data in the embodiment of the present application includes:
  • the preset types include application operation data types and application user data types.
  • the server regularly collects application user behavior data, classifies application user behavior data according to business operation results, and obtains classified application user behavior data.
  • the classified application user behavior data includes but is not limited to lost new users data, user operation data with successful business operations, and old user return business operation data.
  • Pre-purchase operation record data of new users who purchased i.e. user data with successful business operations
  • post-purchase operation record data of new users who successfully purchased i.e. user operation data with successful business operations
  • daily records of new users who successfully purchased Data i.e. user operation data with successful business operations
  • old user operation record data before loss i.e. old user re-business operation data
  • old user re-purchase operation record data before re-purchase i.e. old user re-business operation data
  • old user's post-repurchase operation record data that is, the old user's re-business operation data
  • Classify the classified application user behavior data according to preset types (the preset types include application operation data types and application user data types), to obtain initial application operation data and initial application data.
  • the application operation data type is used to indicate the operation path data generated by the user's operation on the application program, that is, the application operation data.
  • the application user data type is used to indicate the application data generated by the user's operations on the application, that is, application data.
  • the initial application operation data includes but not limited to user identification number (identity document, ID), operation time and operation path.
  • Initial application data includes, but is not limited to, login data, operational application block data, and tag data.
  • the server obtains the user's historical operation sequence data based on the application program to be processed, target user portrait data and recall configuration information, and the target user portrait data includes the score of the operation data; based on the historical operation sequence data and recall configuration information , index, retrieve and read the target user portrait data to obtain the score of the read operation data; score the historical operation sequence data through the score of the read operation data, and obtain the scored historical operation sequence data; According to the size of the scoring value, sort the scored historical operation sequence data in descending order to obtain the preset adjustment strategy sequence.
  • the server collects the user's operation sequence data based on the historical period of the application to be processed in real time to obtain historical operation sequence data; extracts the pre-created target user portrait data, and obtains the recall configuration corresponding to the target user portrait data information, the target user portrait data is used to indicate the user portrait data corresponding to the application user behavior data, the target user portrait data includes the operation data and the score of the operation data, and the recall configuration information includes the number of recalls and the number of sequential reads; create The index of the historical operation sequence data, through which the target user portrait data is recalled to obtain the recall operation data, and the recall operation data is sequentially read according to the recall configuration information to obtain the read operation data, and the read operation data The score of the data is extracted to obtain the score of the read operation data; based on the read operation data, the preset user portrait data is clustered to obtain similar operation data.
  • the preset user portrait data is in addition to the application user behavior data Corresponding to the portrait data of other applicants outside the country; through the similar operation data and the read operation data, match the historical operation sequence data to obtain the target operation data; calculate the score of the target operation data and the score of the read operation data value, get the comprehensive score; score the historical operation sequence data through the comprehensive score to obtain the scored historical operation sequence data; sort the scored historical operation sequence data in descending order according to the score value, and get the preset Adjust strategy sequence.
  • the data quality of the preset adjustment strategy sequence is guaranteed, thereby improving the accuracy of the business operation adjustment strategy to be pushed.
  • the server invokes a preset linked list function, and stores the initial user application operation data as a linked list according to a preset period of time, so as to obtain a sequence of user application operation records.
  • the application operation record sequence includes a user identification number, an operation time stamp, and operation record data, and the application operation record sequence is used to indicate operation record data that changes dynamically over time, that is, dynamic time series data.
  • the server obtains the operation objects of the initial application data, classifies the operation objects according to the preset dimensions, obtains the classified initial application data, and performs statistics on the classified initial application data according to the preset statistical indicators to obtain the application data to be processed.
  • the classification process of medium granularity is realized.
  • the operation object is used to indicate the functional area corresponding to the click button or section in the application.
  • the application data to be processed is used to indicate the data that is counted at the same time, that is, static data.
  • the server uses the preset cyclic neural network model to predict the follow-up business operation based on the result of the business operation on the application operation record sequence, and obtain the target follow-up business operation information.
  • the target follow-up business operation information includes the target follow-up business operation data and target
  • the follow-up business operation data is based on the probability value of the business operation result; obtain the follow-up operation data of the application user behavior data, match the follow-up operation data of the application user behavior data with the target follow-up business operation data, and obtain the matched target follow-up business operation data;
  • the probability value corresponding to the matched target follow-up business operation data is determined as the business operation intention prediction data.
  • the server invokes the preset cyclic neural network model, through the input layer, multiple hidden layers and output layers in the cyclic neural network model, performs sequence feature extraction and activation function calculation based on the memory state on the application operation record sequence to obtain the target follow-up business Operation information
  • the target follow-up business operation information includes the target follow-up business operation data and the probability value of the target follow-up business operation data based on the business operation results
  • the target follow-up business operation data includes the optimal follow-up operation and the suboptimal follow-up operation
  • through the target Subsequent business operation information classify the application operation record sequence based on the business operation achievement of each current state (subsequent operation) and mark the probability of subsequent operations to obtain the business operation intention prediction data; obtain the subsequent operation data of the application user behavior data, call
  • the preset matching algorithm calculates the similarity between the follow-up operation data and the target follow-up business operation data, sorts the similarity in descending order, and determines the target follow-up business operation data corresponding to the first-ranked
  • the server invokes the first classification model among the preset dynamic classification models, performs attention feature extraction, probability value calculation and probability value discrimination based on business operation intentions on the application operation record sequence, and obtains attention score data; through The second classification model in the dynamic classification model performs feature extraction and classification on the application operation record sequence to obtain the original score data; sums the attention score data and the original score data to obtain the dynamic score data; calls the preset Based on the static classification model set, feature extraction, attention feature fusion, probability value calculation and probability value discrimination based on business operation intentions are performed on the application data to be processed to obtain static score data.
  • the preset dynamic classification model includes a first classification model and a second classification model
  • the first classification model is a dynamic prediction model based on an attention mechanism
  • the second classification model is a dynamic prediction model without an attention mechanism.
  • the execution order of the dynamic classification model and the static classification model is not limited. The operation of the dynamic classification model can be performed first, and then the operation of the static classification model can be performed.
  • the dynamic classification model and the static classification model can also be operated at the same time, or the static classification model can be performed first. The operation, and then the operation of the dynamic classification model.
  • the server invokes the first classification model among the preset dynamic classification models, extracts the features of the application operation record sequence based on the attention mechanism, obtains the attention time series features, and calculates the probability value based on the business operation intention for the attention time series features Discriminate against the probability value to obtain the attention score data.
  • the second classification model perform feature extraction on the application operation record sequence to obtain the initial time series features, and calculate the probability value based on the business operation intention and the probability value of the initial time series features. Discriminate, obtain the original score data, perform matrix conversion and matrix weighted summation on the attention score data and the original score data, and obtain dynamic score data;
  • the server calls the preset static classification model, extracts the features of the application data to be processed, and obtains the initial user application features.
  • the features and attention user application features are matrix spliced to obtain the fused user application features, and the probability value calculation and probability value discrimination based on business operation intentions are performed on the fused user application features to obtain static score data.
  • the server calculates the sum of the dynamic score data and the static score data according to the preset weight to obtain the target score data; compares and analyzes the target score data with the business operation intention prediction data and the preset threshold respectively, and obtains Intention analysis results and threshold analysis results; if the result of the intention analysis is that the business operation intention prediction data is less than the target score data, and/or the threshold analysis result is that the target score data is less than the preset threshold, then the preset adjustment strategy sequence is adjusted first
  • the policy extraction is to obtain the business operation adjustment policy to be pushed, and the preset adjustment policy sequence is a set of adjustment policies sorted in reverse order according to the magnitude of the score value of the operation.
  • the server converts the dynamic score data and the static score data into a two-dimensional matrix to obtain a dynamic two-dimensional matrix and a static two-dimensional matrix, and calculates the weighted sum (or weighted value) of the dynamic two-dimensional matrix and the static two-dimensional matrix according to the preset weight Arithmetic mean) to obtain the target score data; compare and analyze the target score data with the business operation intention prediction data and the preset threshold, and obtain the intention analysis result and the threshold analysis result; if the intention analysis result is the business operation intention prediction data is less than the target score data, and/or the threshold analysis result shows that the target score data is less than the preset threshold, then the first adjustment strategy is obtained from the preset adjustment strategy sequence, so as to obtain the business operation adjustment strategy to be pushed, the preset adjustment strategy
  • the sequence is an adjustment policy set sorted in descending order of the operation score value; if the result of the intention analysis is that the business operation intention prediction data is greater than or equal to the target score data, then the recurrent neural network model is subjected
  • the server performs enhanced learning on the cyclic neural network model.
  • the server invokes the preset policy gradients (policy gradients, PG) deep learning algorithm to adjust the learning rate, gradient update and activation of the recurrent neural network model Function reward assignment for reinforcement learning of recurrent neural network models.
  • policy gradients policy gradients, PG
  • forward reinforcement learning or reverse reinforcement learning is performed on the cyclic neural network model to implement reinforcement learning on the cyclic neural network model.
  • the application user behavior data is analyzed from the perspective of machine learning and time series deep learning, which ensures the accuracy and comprehensiveness of the prediction results of business operation intentions, and achieves finer analysis granularity and
  • the detailed classification of all users can conduct in-depth analysis of the operation behavior of a single user itself, and realize the classification of static data and dynamic time series data of application user behavior data, improve the prediction accuracy of business operation intentions, and further improve the The accuracy of business operation adjustment strategies based on application user behavior data analysis.
  • the processing device of the application user behavior data in the embodiment of the application includes:
  • the statistics module 301 is used to obtain application user behavior data, and perform preset type classification, sequence data conversion and classification statistics on the application user behavior data to obtain application operation record sequence and application data to be processed;
  • the prediction module 302 is configured to use a preset cyclic neural network model to predict subsequent business operation intentions on the application operation record sequence, and obtain business operation intention prediction data;
  • the classification module 303 is used to call the preset dynamic classification model, classify the application operation record sequence based on the business operation intention, obtain dynamic score data, and call the preset static classification model to perform business operations on the application data to be processed Classification of intent to obtain static score data;
  • the first acquisition module 304 is used to sum the dynamic score data and the static score data to obtain the target score data, and obtain the target score data from the preset adjustment strategy sequence according to the business operation intention prediction data and the target score data.
  • the pushed business operation adjustment strategy is used to sum the dynamic score data and the static score data to obtain the target score data, and obtain the target score data from the preset adjustment strategy sequence according to the business operation intention prediction data and the target score data.
  • each module in the device for processing application user behavior data corresponds to the steps in the above embodiment of the method for processing application user behavior data, and its functions and implementation processes will not be repeated here.
  • the application user behavior data is analyzed from the perspective of machine learning and time series deep learning, which ensures the accuracy and comprehensiveness of the prediction results of business operation intentions, and can analyze the operation behavior of a single user itself.
  • In-depth analysis and realize the classification of static data and dynamic time series data of application user behavior data, improve the prediction accuracy of business operation intention, and then improve the accuracy of business operation adjustment strategy based on application user behavior data analysis.
  • FIG. 4 another embodiment of the processing device for applying user behavior data in the embodiment of the present application includes:
  • the statistics module 301 is used to obtain application user behavior data, and perform preset type classification, sequence data conversion and classification statistics on the application user behavior data to obtain application operation record sequence and application data to be processed;
  • the statistics module 301 specifically includes:
  • the classification unit 3011 is configured to obtain application user behavior data, classify the application user behavior data according to business operation results and preset types, and obtain initial application operation data and initial application data.
  • the preset types include application operation data types and application users type of data;
  • the storage unit 3012 is configured to store the initial application operation data in a linked list based on a preset period of time to obtain a sequence of application operation records;
  • a statistics unit 3013 configured to classify the operation objects based on preset dimensions and record statistics on the initial application data, to obtain the application data to be processed;
  • the prediction module 302 is configured to use a preset cyclic neural network model to predict subsequent business operation intentions on the application operation record sequence, and obtain business operation intention prediction data;
  • the classification module 303 is used to call the preset dynamic classification model, classify the application operation record sequence based on the business operation intention, obtain dynamic score data, and call the preset static classification model to perform business operations on the application data to be processed Classification of intent to obtain static score data;
  • the first acquisition module 304 is used to sum the dynamic score data and the static score data to obtain the target score data, and obtain the target score data from the preset adjustment strategy sequence according to the business operation intention prediction data and the target score data.
  • the pushed business operation adjustment strategy is used to sum the dynamic score data and the static score data to obtain the target score data, and obtain the target score data from the preset adjustment strategy sequence according to the business operation intention prediction data and the target score data.
  • the prediction module 302 may also be specifically used for:
  • the target follow-up business operation information includes the target follow-up business operation data and the target follow-up business operation data The probability value based on the achievement of business operation results;
  • the probability value corresponding to the matched target follow-up business operation data is determined as the business operation intention prediction data.
  • the classification module 303 can also be specifically used for:
  • the first acquisition module 304 includes:
  • Calculation unit 3041 configured to calculate the sum of dynamic score data and static score data according to preset weights to obtain target score data
  • An analysis unit 3042 configured to compare and analyze the target score data with the business operation intention prediction data and the preset threshold to obtain the intention analysis result and the threshold analysis result;
  • the extraction unit 3043 is configured to extract the first position adjustment strategy for the preset adjustment strategy sequence if the result of the intention analysis is that the business operation intention prediction data is less than the target score data, and/or the threshold analysis result is that the target score data is less than the preset threshold , to get the business operation adjustment strategy to be pushed, and the preset adjustment strategy sequence is a set of adjustment strategies sorted in reverse order according to the magnitude of the score value of the operation.
  • the first obtaining module 304 also includes:
  • the reinforcement learning unit 3044 is configured to perform reinforcement learning on the cyclic neural network model if the result of the intention analysis is that the business operation intention prediction data is greater than or equal to the target score data.
  • the processing device for application user behavior data also includes:
  • the second obtaining module 305 is used to obtain the user's historical operation sequence data based on the application program to be processed, target user portrait data and recall configuration information, and the target user portrait data includes the score of the operation data;
  • the reading module 306 is used to index, retrieve and read the target user portrait data based on the historical operation sequence data and recall configuration information, and obtain the score of the read operation data;
  • the scoring module 307 is configured to score the historical operation sequence data through the score value of the read operation data, and obtain the scored historical operation sequence data;
  • the sorting module 308 is configured to sort the scored historical operation sequence data in descending order according to the score value to obtain a preset adjustment strategy sequence.
  • each module and each unit in the above-mentioned application user behavior data processing device corresponds to each step in the above-mentioned application user behavior data processing method embodiment, and its functions and implementation processes will not be repeated here.
  • the application user behavior data is analyzed from the perspective of machine learning and time series deep learning, which ensures the accuracy and comprehensiveness of the prediction results of business operation intentions, and achieves finer analysis granularity and
  • the detailed classification of all users can conduct in-depth analysis of the operation behavior of a single user itself, and realize the classification of static data and dynamic time series data of application user behavior data, improve the prediction accuracy of business operation intentions, and further improve the The accuracy of business operation adjustment strategies based on application user behavior data analysis.
  • FIG 3 and Figure 4 above describe in detail the processing device for applying user behavior data in the embodiment of the present application from the perspective of modular functional entities, and the following describes the processing device for applying user behavior data in the embodiment of the present application in detail from the perspective of hardware processing describe.
  • FIG. 5 is a schematic structural diagram of a processing device for applying user behavior data provided by an embodiment of the present application.
  • the processing device 500 for applying user behavior data may have relatively large differences due to different configurations or performances, and may include one or more than one Processor (central processing units, CPU) 510 (for example, one or more processors) and memory 520, one or more storage media 530 for storing application programs 533 or data 532 (for example, one or more mass storage devices).
  • the memory 520 and the storage medium 530 may be temporary storage or persistent storage.
  • the program stored in the storage medium 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the processing device 500 for application user behavior data.
  • the processor 510 may be configured to communicate with the storage medium 530, and execute a series of instruction operations in the storage medium 530 on the processing device 500 applying user behavior data.
  • the processing device 500 for applying user behavior data may also include one or more power sources 540, one or more wired or wireless network interfaces 550, one or more input and output interfaces 560, and/or, one or more operating systems 531, Such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc.
  • operating systems 531 Such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc.
  • the present application also provides a processing device for applying user behavior data, including: a memory and at least one processor, instructions are stored in the memory, and the memory and the at least one processor are interconnected through lines; the at least one The processor invokes the instructions in the memory, so that the device for processing application user behavior data executes the steps in the above method for processing application user behavior data.
  • the present application also provides a computer-readable storage medium, the computer-readable storage medium may be a non-volatile computer-readable storage medium, the computer-readable storage medium may also be a volatile computer-readable storage medium, and the computer-readable storage medium may be Instructions are stored in the readable storage medium, and when the instructions are run on the computer, the computer is made to execute the steps of the processing method for the application user behavior data.
  • the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function, etc.; Use the created data etc.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with each other using cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify its Validity of information (anti-counterfeiting) and generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods in various embodiments of the present application.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (read-only memory, ROM), random access memory (random access memory, RAM), magnetic disk or optical disc and other media that can store program codes. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Databases & Information Systems (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Fuzzy Systems (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种应用用户行为数据的处理方法、装置、设备及存储介质，涉及人工智能领域，用于提高基于应用用户行为数据分析的业务操作调整策略的准确性，所述方法包括：对应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；对应用操作记录序列进行后续业务操作意向的预测得到业务操作意向预测数据；对应用操作记录序列进行分类得到动态分值数据，对待处理应用数据进行分类得到静态分值数据；对动态分值数据和静态分值数据求和得到目标分值数据，根据业务操作意向预测数据和目标分值数据获取待推送的业务操作调整策略；应用用户行为数据可存储于区块链中。

Description

应用用户行为数据的处理方法、装置、设备及存储介质
本申请要求于2021年6月30日提交中国专利局、申请号为202110735749.0、发明名称为“应用用户行为数据的处理方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
技术领域
本申请涉及人工智能的智能决策领域，尤其涉及一种应用用户行为数据的处理方法、装置、设备及存储介质。
背景技术
随着互联网技术和计算机技术的发展，用户逐渐通过各种操作应用实现各种对应业务操作，为了能够及时有效地对用户进行应用推送，采集应用程序的静态数据，基于数据分析指标(如：应用程序的日活跃用户数量DAU、月活跃用户人数MAU、新增用户量、用户留存率，各页面间的跳转率、流失率和成交率等)，对应用程序的静态数据进行用户业务操作意向的预测，得到预测结果，将预测结果对应的业务操作调整策略发送至用户端。
发明人意识到上述方法中，由于所采用的数据分析指标为基于机械化的数据指标，只能反映用户整体的使用情况，无法对单一用户的操作行为本身进行深度分析，因而，导致了业务操作意向的预测准确性低，从而导致了基于应用用户行为数据分析的业务操作调整策略的准确性低。
发明内容
本申请提供一种应用用户行为数据的处理方法、装置、设备及存储介质，用于提高基于应用用户行为数据分析的业务操作调整策略的准确性。
本申请提供一种应用用户行为数据的处理方法、装置、设备及存储介质，用于提高基于应用用户行为数据分析的业务操作调整策略的准确性。
本申请第一方面提供了一种应用用户行为数据的处理方法，包括：
获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
本申请第二方面提供了一种应用用户行为数据的处理设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：
获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
本申请第三方面提供了一种计算机可读存储介质，所述计算机可读存储介质中存储计算机指令，当所述计算机指令在计算机上运行时，使得计算机执行如下步骤：
获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
本申请第四方面提供了一种应用用户行为数据的处理装置，包括：
统计模块，用于获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
预测模块，用于通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
分类模块，用于调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
第一获取模块，用于对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
本申请提供的技术方案中，获取应用用户行为数据，并对应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；通过预置的循环神经网络模型，对应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；调用预置的动态分类模型，对应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；对动态分值数据和静态分值数据进行求和，得到目标分值数据，并根据业务操作意向预测数据和目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。本申请实施例中，结合了机器学习和时间序列深度学习的角度对应用用户行为数据进行了分析，保证了业务操作意向的预测结果的准确性和全面性，能够对单一用户的操作行为本身进行深度分析，并实现了应用用户行为数据的静态数据和动态时间序列数据的分类，提高了业务操作意向的预测准确性，进而提高了基于应用用户行为数据分析的业务操作调整策略的准确性。
附图说明
图1为本申请实施例中应用用户行为数据的处理方法的一个实施例示意图；
图2为本申请实施例中应用用户行为数据的处理方法的另一个实施例示意图；
图3为本申请实施例中应用用户行为数据的处理装置的一个实施例示意图；
图4为本申请实施例中应用用户行为数据的处理装置的另一个实施例示意图；
图5为本申请实施例中应用用户行为数据的处理设备的一个实施例示意图。
具体实施方式
本申请实施例提供了一种应用用户行为数据的处理方法、装置、设备及存储介质，提高了基于应用用户行为数据分析的业务操作调整策略的准确性。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”、“第三”、“第四”等(如果存在)是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的实施例能够以除了在这里图示或描述的内容以外的顺序实施。此外，术语“包括”或“具有”及其任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
为便于理解，下面对本申请实施例的具体流程进行描述，请参阅图1，本申请实施例中应用用户行为数据的处理方法的一个实施例包括：
101、获取应用用户行为数据，并对应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据。
可以理解的是，本申请的执行主体可以为应用用户行为数据的处理装置，还可以是终端或者服务器，具体此处不做限定。本申请实施例以服务器为执行主体为例进行说明。
服务器获得用户授权后，调用预置的定时器执行定时任务，当通过该定时任务计时到预设时刻时，调用预置的爬虫或抓取工具，抓取待处理应用程序(application，APP)的应用用户行为数据，待处理应用程序的数量包括一个或一个以上，即应用用户行为数据可为一个待处理应用程序对应的数据，应用用户行为数据也可为一个以上的待处理应用程序分别对应的数据，该应用用户行为数据包括待处理应用程序基于用户操作生成的记录数据、基于待处理应用程序的用户数据以及用户在待处理应用程序上业务操作的状况数据；对应用用户行为数据进行缺失值填充、同类型合并和数据脱敏，得到预处理后的应用用户行为数据，提高了应用用户行为数据的质量和安全性。
服务器按照预设类型(预设类型包括应用操作数据类型和应用用户数据类型)，对预处理后的应用用户行为数据进行分类，得到预分类后的应用操作数据和预分类后的应用数据；按照预设时段对预分类后的应用操作数据进行数据提取，得到提取后的应用操作数据；将提取后的应用操作数据转换为时间序列数据，得到应用操作记录序列，其中，预分类后的应用操作数据包括用户应用操作记录数据和用户应用购买数据；对预分类后的应用数据进行功能区域分类，得到分类后的用户应用数据，按照预设的统计指标，对分类后的用户应用数据进行统计，得到待处理应用数据，该预设的统计指标包括但不限于用户基于待处理应用程序的登录时长、登录次数、操作次数、各板块停留时间和偏好标签。
102、通过预置的循环神经网络模型，对应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据。
服务器预先通过预置的历史应用操作记录序列，对创建的循环神经网络模型进行训练和优化，得到最终的循环神经网络模型(recurrent neural network，RNN)，该循环神经网络模型为深层双向RNN，循环神经网络模型能够对在当前每个状态下要达成业务操作应该进行的后续最优操作进行预测。服务器调用预置的循环神经网络模型，通过循环神经网络模型中的输入层、多个隐藏层和输出层，对应用操作记录序列进行序列特征提取和基于记忆状态的激活函数运算，得到目标后续业务操作信息，目标后续业务操作信息包括目标后续业务操作数据和目标后续业务操作数据基于业务操作结果达成的概率值，目标后续业务操作数据包括最优的后续操作，以及次优的后续操作；通过目标后续业务操作信息，对 应用操作记录序列进行基于当前每个状态的业务操作达成分类，得到业务操作意向预测数据。
103、调用预置的动态分类模型，对应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对待处理应用数据进行基于业务操作意向的分类，得到静态分值数据。
服务器预先通过应用操作记录序列样本，对预置的动态预测模型进行训练和优化，得到动态分类模型，动态分类模型用于对操作记录数据量大且为时间序列，后续操作对前面的整体操作有一定依赖关系的数据进行分类，即动态分类模型用于对动态时间序列数据(即应用操作记录序列样本)进行分类预测，该动态分类模型可为长短期记忆网络(long short-term memory，LSTM)模型；服务器预先通过用户应用数据样本，对预置的静态预测模型进行训练和优化，得到静态分类模型，静态分类模型用于对同一时刻统计的数据进行用户业务执行状况(业务操作意向)的分类，即静态分类模型用于对静态数据(即用户应用数据样本)进行分类预测，该静态分类模型可为极端梯度提升(extreme gradient boosting，xgboost)，其中，该应用操作记录序列样本和用户应用数据样本的获得的执行过程与步骤101的执行过程类似，在此不再赘述。
服务器调用预置的动态分类模型，对应用操作记录序列进行多层级的时间序列特征提取，得到操作记录时间序列特征，对操作记录时间序列特征进行基于业务操作意向的概率值计算和概率值判别，得到动态分值数据；调用预置的静态分类模型，对待处理应用数据进行多层级特征提取，得到用户应用特征，对用户应用特征进行基于业务操作意向的概率值计算和概率值判别，得到静态分值数据，业务操作意向如：购买意向。其中，动态分类模型和静态分类模型均可为特征提取网络和分类网络的组合，特征提取网络用于进行多层级特征提取，例如：特征提取网络可为目标检测模型ET-YOLOV3，提高了动态分值数据和静态分值数据的准确性。
104、对动态分值数据和静态分值数据进行求和，得到目标分值数据，并根据业务操作意向预测数据和目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
其中，业务操作调整策略包括但不限于业务操作的应用调整策略、相同类型内容调整策略和功能调整策略，例如：以业务操作为应用程序上的购买操作，为了达成购买或者增大购买意向，推送用户常用或评分较高的应用程序(即应用调整策略)，推送与应用用户行为数据中当前操作相似的操作路径(即相同类型内容调整策略)，推送与应用用户行为数据对应的业务操作功能相似的功能模块(即功能调整策略)。
服务器将动态分值数据和和静态分值数据分别进行二维矩阵转换，得到动态二维矩阵和静态二维矩阵，计算动态二维矩阵和静态二维矩阵的算术平均值，得到目标分值数据；将目标分值数据分别与业务操作意向预测数据、预设阈值进行比较分析，得到意向分析结果和阈值分析结果，若意向分析结果为对应推送的结果，则从预置调整策略序列中获取评分值降序排序中排序第一的策略，从而得到待推送的业务操作调整策略，或者，将意向分析结果和阈值分析结果发送至预置终端，通过预置终端的负责人员进行人工跟进，若意向分析结果为对应不需推送的结果，则不进行处理。
本申请实施例中，结合了机器学习和时间序列深度学习的角度对应用用户行为数据进行了分析，保证了业务操作意向的预测结果的准确性和全面性，能够对单一用户的操作行为本身进行深度分析，并实现了应用用户行为数据的静态数据和动态时间序列数据的分类，提高了业务操作意向的预测准确性，进而提高了基于应用用户行为数据分析的业务操作调整策略的准确性。
请参阅图2，本申请实施例中应用用户行为数据的处理方法的另一个实施例包括：
201、获取应用用户行为数据，按照业务操作结果和预设类型，对应用用户行为数据进行分类，得到初始应用操作数据和初始应用数据，预设类型包括应用操作数据类型和应用用户数据类型。
服务器获得用户授权后，定时采集应用用户行为数据，按照业务操作结果，对应用用户行为数据进行分类，得到分类后的应用用户行为数据，该分类后的应用用户行为数据包括但不限于流失新用户数据、业务操作成功的用户操作数据和老用户复业务操作数据，例如：以业务操作为购买为例说明，分类后的应用行为数据包括流失新用户操作记录数据(即流失新用户数据)、成功购买的新用户的购买前操作记录数据(即业务操作成功的用户数据)、成功购买的新用户的购买后操作记录数据(即业务操作成功的用户操作数据)、成功购买的新用户的日常记录数据(即业务操作成功的用户操作数据)、老用户的流失前操作记录数据(即老用户复业务操作数据)、老用户的复购前操作记录数据(即老用户复业务操作数据)和老用户的复购后操作记录数据(即老用户复业务操作数据)；
按照预设类型(预设类型包括应用操作数据类型和应用用户数据类型)，对分类后的应用用户行为数据进行分类，得到初始应用操作数据和初始应用数据。其中，应用操作数据类型用于指示用户在应用程序上进行操作而生成的操作路径数据，即应用操作数据。应用用户数据类型用于指示用户在应用程序上进行操作而生成的应用程序数据，即应用数据。初始应用操作数据包括但不限于用户标识号(identity document，ID)、操作时间和操作路径。初始应用数据包括但不限于登录数据、操作应用程序板块数据和标签数据。
具体地，步骤201之前，服务器获取用户基于待处理应用程序的历史操作序列数据、目标用户画像数据和召回配置信息，目标用户画像数据包括操作数据的分值；基于历史操作序列数据和召回配置信息，对目标用户画像数据进行索引检索和读取，得到已读取操作数据的分值；通过已读取操作数据的分值，对历史操作序列数据进行评分，得到评分后的历史操作序列数据；按照评分值大小，对评分后的历史操作序列数据进行降序排列，得到预置调整策略序列。
服务器通过获得用户授权后，对用户基于待处理应用程序的历史时段的操作序列数据进行实时采集，得到历史操作序列数据；提取预先创建的目标用户画像数据，并获取目标用户画像数据对应的召回配置信息，目标用户画像数据用于指示应用用户行为数据对应的用户的画像数据，目标用户画像数据包括操作数据和操作数据的分值，召回配置信息包括召回的数量和依序读取的数量；创建历史操作序列数据的索引，通过该索引对目标用户画像数据进行召回，得到召回操作数据，根据召回配置信息对召回操作数据进行依序读取，得到已读取操作数据，并对已读取操作数据的分值进行提取，得到已读取操作数据的分值；基于已读取操作数据，对预置用户画像数据进行聚类，得到类似操作数据，预置用户画像数据为除了应用用户行为数据对应的用户外其他申请人的画像数据；通过类似操作数据和已读取操作数据，对历史操作序列数据进行匹配，得到目标操作数据；计算目标操作数据的分值和已读取操作数据的分值的和值，得到综合分值；通过综合分值对历史操作序列数据进行评分，得到评分后的历史操作序列数据；按照评分值大小对评分后的历史操作序列数据进行降序排列，得到预置调整策略序列。保证了预置调整策略序列的数据质量，从而提高了待推送的业务操作调整策略的准确性。
202、对初始应用操作数据进行基于预设时段的链表存储，得到应用操作记录序列。
服务器调用预置的链表函数，按照预设时段，将初始用户应用操作数据存储为链表，从而得到用户应用操作记录序列。其中，应用操作记录序列包括用户标识号、操作时间戳和操作记录数据，应用操作记录序列用于指示随着时间动态变化的操作记录数据，即动态时间序列数据。
203、对初始应用数据进行基于预设维度的操作对象分类和记录统计，得到待处理应用数据。
服务器获取初始应用数据的操作对象，按照预设维度对操作对象进行分类，得到分类后的初始应用数据，并按照预设统计指标，对分类后的初始应用数据进行统计，得到待处理应用数据，通过以固定不变的功能对初始应用数据进行分类统计，实现了中等颗粒度的归类处理。
其中，操作对象用于指示点击应用程序中的按钮或板块对应的功能区域，操作对象可为链接，预设维度包括但不限于应用程序的板块、内容载体和标签等，预设统计指标包括但不限于应用程序的登录时长、登录次数、操作次数、各板块停留时间和偏好标签等，待处理应用数据用于指示同一时刻统计的数据，即静态数据。
实现了较细的分析颗粒度，以及对用户全体的细分类，能够对单一用户的操作行为本身进行深度分析，并实现了应用用户行为数据的静态数据和动态时间序列数据的分类。通过过程性指标对初始应用数据进行统计分析，提高了业务操作意向的预测准确性。
204、通过预置的循环神经网络模型，对应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据。
具体地，服务器通过预置的循环神经网络模型，对应用操作记录序列进行基于业务操作结果达成的后续业务操作预测，得到目标后续业务操作信息，目标后续业务操作信息包括目标后续业务操作数据和目标后续业务操作数据基于业务操作结果达成的概率值；获取应用用户行为数据的后续操作数据，将应用用户行为数据的后续操作数据与目标后续业务操作数据进行匹配，得到匹配的目标后续业务操作数据；将匹配的目标后续业务操作数据对应的概率值确定为业务操作意向预测数据。
服务器调用预置的循环神经网络模型，通过循环神经网络模型中的输入层、多个隐藏层和输出层，对应用操作记录序列进行序列特征提取和基于记忆状态的激活函数运算，得到目标后续业务操作信息，目标后续业务操作信息包括目标后续业务操作数据以及目标后续业务操作数据基于业务操作结果达成的概率值，目标后续业务操作数据包括最优的后续操作，以及次优的后续操作；通过目标后续业务操作信息，对应用操作记录序列进行基于当前每个状态(后续操作)的业务操作达成分类并进行后续操作概率标记，得到业务操作意向预测数据；获取应用用户行为数据的后续操作数据，调用预置的匹配算法，计算后续操作数据与目标后续业务操作数据的相似度，将相似度进行降序排序，并将排序第一的相似度对应的目标后续业务操作数据确定为匹配的目标后续业务操作数据，其中，该后续操作数据用于指示应用用户行为数据中每个当前状态下基于达成业务操作的后续操作，该后续操作例如：当前状态为已进行了订单预支付的操作，要完成购买(即达成业务操作)，则后续操作为付款；将匹配的目标后续业务操作数据对应的概率值确定为业务操作意向预测数据。
205、调用预置的动态分类模型，对应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对待处理应用数据进行基于业务操作意向的分类，得到静态分值数据。
具体地，服务器调用预置的动态分类模型中的第一分类模型，对应用操作记录序列进行注意力特征提取、基于业务操作意向的概率值计算和概率值判别，得到注意力分值数据；通过动态分类模型中的第二分类模型，对应用操作记录序列进行特征提取和分类，得到原始分值数据；将注意力分值数据和原始分值数据进行求和，得到动态分值数据；调用预置的静态分类模型，对待处理应用数据进行特征提取、注意力特征融合、基于业务操作意向的概率值计算和概率值判别，得到静态分值数据。
其中，预置的动态分类模型包括第一分类模型和第二分类模型，第一分类模型为基于注意力机制的动态预测模型，第二分类模型为非注意力机制的动态预测模型。动态分类模型和静态分类模型的执行顺序不限制，可先进行动态分类模型的运算，后进行静态分类模型的运算，也可动态分类模型和静态分类模型同是运算，也可先进行静态分类模型的运算，后进行动态分类模型的运算。
服务器调用预置的动态分类模型中的第一分类模型，基于注意力机制对应用操作记录序列进行特征提取，得到注意力时间序列特征，对注意力时间序列特征进行基于业务操作意向的概率值计算和概率值判别，得到注意力分值数据，通过第二分类模型，对应用操作记录序列进行特征提取，得到初始时间序列特征，对初始时间序列特征进行基于业务操作意向的概率值计算和概率值判别，得到原始分值数据，将注意力分值数据和原始分值数据进行矩阵转换和矩阵加权求和，得到动态分值数据；
服务器调用预置的静态分类模型，对待处理应用数据进行特征提取，得到初始用户应用特征，基于注意力机制，对初始用户应用特征进行注意力矩阵运算，得到注意力用户应用特征，将初始用户应用特征和注意力用户应用特征进行矩阵拼接，得到融合用户应用特征，对融合用户应用特征进行基于业务操作意向的概率值计算和概率值判别，得到静态分值数据。
通过上述操作，既能保留原始特征的原始信息，又能保留注意力特征的偏重信息，利用了有限的注意力资源从大量特征信息中快速筛选出高价值的信息，提高了动态分值数据和静态分值数据的准确性。
206、对动态分值数据和静态分值数据进行求和，得到目标分值数据，并根据业务操作意向预测数据和目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
具体地，服务器按照预设权重，计算动态分值数据和静态分值数据的和值，得到目标分值数据；将目标分值数据分别与业务操作意向预测数据、预设阈值进行比较分析，得到意向分析结果和阈值分析结果；若意向分析结果为业务操作意向预测数据小于目标分值数据，和/或阈值分析结果为目标分值数据小于预设阈值，则对预置调整策略序列进行首位调整策略提取，得到待推送的业务操作调整策略，预置调整策略序列为按照操作的评分值的大小进行倒序排序的调整策略集。
服务器将动态分值数据和静态分值数据进行二维矩阵转换，得到动态二维矩阵和静态二维矩阵，按照预设权重，计算动态二维矩阵和静态二维矩阵的加权和值(或加权算术平均数)，得到目标分值数据；将目标分值数据分别与业务操作意向预测数据、预设阈值进行比较分析，得到意向分析结果和阈值分析结果；若意向分析结果为业务操作意向预测数据小于目标分值数据，和/或阈值分析结果为目标分值数据小于预设阈值，则从预置调整策略序列中获取首位的调整策略，从而得到待推送的业务操作调整策略，预置调整策略序列为按照操作评分值从大到小的顺序进行排序的调整策略集；若意向分析结果为业务操作意向预测数据大于或等于目标分值数据，则对循环神经网络模型进行增强学习处理，若阈值分析结果为目标分值数据大于或等于预设阈值，则不进行处理。
具体地，若意向分析结果为业务操作意向预测数据大于或等于目标分值数据，则服务器对循环神经网络模型进行增强学习。
若意向分析结果为业务操作意向预测数据大于或等于目标分值数据，则服务器调用预置的策略梯度(policy gradients，PG)深度学习算法，对循环神经网络模型进行学习率调整、梯度更新和激活函数奖励分配，实现对循环神经网络模型的增强学习。或者，对循环神经网络模型进行正向增强学习或反向增强学习，实现对对循环神经网络模型的增强学习。通过对循环神经网络模型进行增强学习，进一步识别有较强购买意愿的用户的路径习 惯，从而学习如何更好影响购买意愿较弱的用户，保证了业务操作意向预测的准确性和全面性，从而提高了基于应用用户行为数据分析的业务操作调整策略的准确性。
本申请实施例中，结合了机器学习和时间序列深度学习的角度对应用用户行为数据进行了分析，保证了业务操作意向的预测结果的准确性和全面性，实现了较细的分析颗粒度以及对用户全体的细分类，能够对单一用户的操作行为本身进行深度分析，并实现了应用用户行为数据的静态数据和动态时间序列数据的分类，提高了业务操作意向的预测准确性，进而提高了基于应用用户行为数据分析的业务操作调整策略的准确性。
上面对本申请实施例中应用用户行为数据的处理方法进行了描述，下面对本申请实施例中应用用户行为数据的处理装置进行描述，请参阅图3，本申请实施例中应用用户行为数据的处理装置一个实施例包括：
统计模块301，用于获取应用用户行为数据，并对应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
预测模块302，用于通过预置的循环神经网络模型，对应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
分类模块303，用于调用预置的动态分类模型，对应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
第一获取模块304，用于对动态分值数据和静态分值数据进行求和，得到目标分值数据，并根据业务操作意向预测数据和目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
上述应用用户行为数据的处理装置中各个模块的功能实现与上述应用用户行为数据的处理方法实施例中各步骤相对应，其功能和实现过程在此处不再一一赘述。
本申请实施例中，结合了机器学习和时间序列深度学习的角度对应用用户行为数据进行了分析，保证了业务操作意向的预测结果的准确性和全面性，能够对单一用户的操作行为本身进行深度分析，并实现了应用用户行为数据的静态数据和动态时间序列数据的分类，提高了业务操作意向的预测准确性，进而提高了基于应用用户行为数据分析的业务操作调整策略的准确性。
请参阅图4，本申请实施例中应用用户行为数据的处理装置的另一个实施例包括：
统计模块301，用于获取应用用户行为数据，并对应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
其中，统计模块301具体包括：
分类单元3011，用于获取应用用户行为数据，按照业务操作结果和预设类型，对应用用户行为数据进行分类，得到初始应用操作数据和初始应用数据，预设类型包括应用操作数据类型和应用用户数据类型；
存储单元3012，用于对初始应用操作数据进行基于预设时段的链表存储，得到应用操作记录序列；
统计单元3013，用于对初始应用数据进行基于预设维度的操作对象分类和记录统计，得到待处理应用数据；
预测模块302，用于通过预置的循环神经网络模型，对应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
分类模块303，用于调用预置的动态分类模型，对应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
第一获取模块304，用于对动态分值数据和静态分值数据进行求和，得到目标分值数据，并根据业务操作意向预测数据和目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
可选的，预测模块302还可以具体用于：
通过预置的循环神经网络模型，对应用操作记录序列进行基于业务操作结果达成的后续业务操作预测，得到目标后续业务操作信息，目标后续业务操作信息包括目标后续业务操作数据和目标后续业务操作数据基于业务操作结果达成的概率值；
获取应用用户行为数据的后续操作数据，将应用用户行为数据的后续操作数据与目标后续业务操作数据进行匹配，得到匹配的目标后续业务操作数据；
将匹配的目标后续业务操作数据对应的概率值确定为业务操作意向预测数据。
可选的，分类模块303还可以具体用于：
调用预置的动态分类模型中的第一分类模型，对应用操作记录序列进行注意力特征提取、基于业务操作意向的概率值计算和概率值判别，得到注意力分值数据；
通过动态分类模型中的第二分类模型，对应用操作记录序列进行特征提取和分类，得到原始分值数据；
将注意力分值数据和原始分值数据进行求和，得到动态分值数据；
调用预置的静态分类模型，对待处理应用数据进行特征提取、注意力特征融合、基于业务操作意向的概率值计算和概率值判别，得到静态分值数据。
可选的，第一获取模块304包括：
计算单元3041，用于按照预设权重，计算动态分值数据和静态分值数据的和值，得到目标分值数据；
分析单元3042，用于将目标分值数据分别与业务操作意向预测数据、预设阈值进行比较分析，得到意向分析结果和阈值分析结果；
提取单元3043，用于若意向分析结果为业务操作意向预测数据小于目标分值数据，和/或阈值分析结果为目标分值数据小于预设阈值，则对预置调整策略序列进行首位调整策略提取，得到待推送的业务操作调整策略，预置调整策略序列为按照操作的评分值的大小进行倒序排序的调整策略集。
可选的，第一获取模块304还包括：
增强学习单元3044，用于若意向分析结果为业务操作意向预测数据大于或等于目标分值数据，则对循环神经网络模型进行增强学习。
可选的，应用用户行为数据的处理装置，还包括：
第二获取模块305，用于获取用户基于待处理应用程序的历史操作序列数据、目标用户画像数据和召回配置信息，目标用户画像数据包括操作数据的分值；
读取模块306，用于基于历史操作序列数据和召回配置信息，对目标用户画像数据进行索引检索和读取，得到已读取操作数据的分值；
评分模块307，用于通过已读取操作数据的分值，对历史操作序列数据进行评分，得到评分后的历史操作序列数据；
排序模块308，用于按照评分值大小，对评分后的历史操作序列数据进行降序排列，得到预置调整策略序列。
上述应用用户行为数据的处理装置中各模块和各单元的功能实现与上述应用用户行为数据的处理方法实施例中各步骤相对应，其功能和实现过程在此处不再一一赘述。
本申请实施例中，结合了机器学习和时间序列深度学习的角度对应用用户行为数据进行了分析，保证了业务操作意向的预测结果的准确性和全面性，实现了较细的分析颗粒度 以及对用户全体的细分类，能够对单一用户的操作行为本身进行深度分析，并实现了应用用户行为数据的静态数据和动态时间序列数据的分类，提高了业务操作意向的预测准确性，进而提高了基于应用用户行为数据分析的业务操作调整策略的准确性。
上面图3和图4从模块化功能实体的角度对本申请实施例中的应用用户行为数据的处理装置进行详细描述，下面从硬件处理的角度对本申请实施例中应用用户行为数据的处理设备进行详细描述。
图5是本申请实施例提供的一种应用用户行为数据的处理设备的结构示意图，该应用用户行为数据的处理设备500可因配置或性能不同而产生比较大的差异，可以包括一个或一个以上处理器(central processing units，CPU)510(例如，一个或一个以上处理器)和存储器520，一个或一个以上存储应用程序533或数据532的存储介质530(例如一个或一个以上海量存储设备)。其中，存储器520和存储介质530可以是短暂存储或持久存储。存储在存储介质530的程序可以包括一个或一个以上模块(图示没标出)，每个模块可以包括对应用用户行为数据的处理设备500中的一系列指令操作。更进一步地，处理器510可以设置为与存储介质530通信，在应用用户行为数据的处理设备500上执行存储介质530中的一系列指令操作。
应用用户行为数据的处理设备500还可以包括一个或一个以上电源540，一个或一个以上有线或无线网络接口550，一个或一个以上输入输出接口560，和/或，一个或一个以上操作系统531，例如Windows Serve，Mac OS X，Unix，Linux，FreeBSD等等。本领域技术人员可以理解，图5示出的应用用户行为数据的处理设备结构并不构成对应用用户行为数据的处理设备的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
本申请还提供一种应用用户行为数据的处理设备，包括：存储器和至少一个处理器，所述存储器中存储有指令，所述存储器和所述至少一个处理器通过线路互连；所述至少一个处理器调用所述存储器中的所述指令，以使得所述应用用户行为数据的处理设备执行上述应用用户行为数据的处理方法中的步骤。本申请还提供一种计算机可读存储介质，该计算机可读存储介质可以为非易失性计算机可读存储介质，该计算机可读存储介质也可以为易失性计算机可读存储介质，计算机可读存储介质中存储有指令，当指令在计算机上运行时，使得计算机执行应用用户行为数据的处理方法的步骤。
进一步地，计算机可读存储介质可主要包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需的应用程序等；存储数据区可存储根据区块链节点的使用所创建的数据等。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统，装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例方法的全部或部分步骤。 而前述的存储介质包括：U盘、移动硬盘、只读存储器(read-only memory，ROM)、随机存取存储器(random access memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上，以上实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围。

Claims (20)

  1. 一种应用用户行为数据的处理方法，其中，所述应用用户行为数据的处理方法包括：
    获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
    通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
    调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
    对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
  2. 根据权利要求1所述的应用用户行为数据的处理方法，其中，所述获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据，包括：
    获取应用用户行为数据，按照业务操作结果和预设类型，对所述应用用户行为数据进行分类，得到初始应用操作数据和初始应用数据，所述预设类型包括应用操作数据类型和应用用户数据类型；
    对所述初始应用操作数据进行基于预设时段的链表存储，得到应用操作记录序列；
    对所述初始应用数据进行基于预设维度的操作对象分类和记录统计，得到待处理应用数据。
  3. 根据权利要求1所述的应用用户行为数据的处理方法，其中，所述通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据，包括：
    通过预置的循环神经网络模型，对所述应用操作记录序列进行基于业务操作结果达成的后续业务操作预测，得到目标后续业务操作信息，所述目标后续业务操作信息包括目标后续业务操作数据和所述目标后续业务操作数据基于业务操作结果达成的概率值；
    获取所述应用用户行为数据的后续操作数据，将所述应用用户行为数据的后续操作数据与所述目标后续业务操作数据进行匹配，得到匹配的目标后续业务操作数据；
    将所述匹配的目标后续业务操作数据对应的概率值确定为业务操作意向预测数据。
  4. 根据权利要求1所述的应用用户行为数据的处理方法，其中，所述调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据，包括：
    调用预置的动态分类模型中的第一分类模型，对所述应用操作记录序列进行注意力特征提取、基于业务操作意向的概率值计算和概率值判别，得到注意力分值数据；
    通过所述动态分类模型中的第二分类模型，对所述应用操作记录序列进行特征提取和分类，得到原始分值数据；
    将所述注意力分值数据和所述原始分值数据进行求和，得到动态分值数据；
    调用预置的静态分类模型，对所述待处理应用数据进行特征提取、注意力特征融合、基于业务操作意向的概率值计算和概率值判别，得到静态分值数据。
  5. 根据权利要求1所述的应用用户行为数据的处理方法，其中，所述对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数 据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略，包括：
    按照预设权重，计算所述动态分值数据和所述静态分值数据的和值，得到目标分值数据；
    将所述目标分值数据分别与所述业务操作意向预测数据、预设阈值进行比较分析，得到意向分析结果和阈值分析结果；
    若所述意向分析结果为所述业务操作意向预测数据小于所述目标分值数据，和/或所述阈值分析结果为所述目标分值数据小于预设阈值，则对预置调整策略序列进行首位调整策略提取，得到待推送的业务操作调整策略，所述预置调整策略序列为按照操作的评分值的大小进行倒序排序的调整策略集。
  6. 根据权利要求5所述的应用用户行为数据的处理方法，其中，所述若所述意向分析结果为所述业务操作意向预测数据小于所述目标分值数据，和/或所述阈值分析结果为所述目标分值数据小于预设阈值，则对预置调整策略序列进行首位调整策略提取，得到待推送的业务操作调整策略之后，还包括：
    若所述意向分析结果为所述业务操作意向预测数据大于或等于所述目标分值数据，则对所述循环神经网络模型进行增强学习。
  7. 根据权利要求1-6中任一项所述的应用用户行为数据的处理方法，其中，所述获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据之前，还包括：
    获取用户基于待处理应用程序的历史操作序列数据、目标用户画像数据和召回配置信息，所述目标用户画像数据包括操作数据的分值；
    基于所述历史操作序列数据和所述召回配置信息，对所述目标用户画像数据进行索引检索和读取，得到已读取操作数据的分值；
    通过所述已读取操作数据的分值，对所述历史操作序列数据进行评分，得到评分后的历史操作序列数据；
    按照评分值大小，对所述评分后的历史操作序列数据进行降序排列，得到预置调整策略序列。
  8. 一种应用用户行为数据的处理设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：
    获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
    通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
    调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
    对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
  9. 根据权利要求8所述的应用用户行为数据的处理设备，其中，所述获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据，包括：
    获取应用用户行为数据，按照业务操作结果和预设类型，对所述应用用户行为数据进 行分类，得到初始应用操作数据和初始应用数据，所述预设类型包括应用操作数据类型和应用用户数据类型；
    对所述初始应用操作数据进行基于预设时段的链表存储，得到应用操作记录序列；
    对所述初始应用数据进行基于预设维度的操作对象分类和记录统计，得到待处理应用数据。
  10. 根据权利要求8所述的应用用户行为数据的处理设备，其中，所述通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据，包括：
    通过预置的循环神经网络模型，对所述应用操作记录序列进行基于业务操作结果达成的后续业务操作预测，得到目标后续业务操作信息，所述目标后续业务操作信息包括目标后续业务操作数据和所述目标后续业务操作数据基于业务操作结果达成的概率值；
    获取所述应用用户行为数据的后续操作数据，将所述应用用户行为数据的后续操作数据与所述目标后续业务操作数据进行匹配，得到匹配的目标后续业务操作数据；
    将所述匹配的目标后续业务操作数据对应的概率值确定为业务操作意向预测数据。
  11. 根据权利要求8所述的应用用户行为数据的处理设备，其中，所述调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据，包括：
    调用预置的动态分类模型中的第一分类模型，对所述应用操作记录序列进行注意力特征提取、基于业务操作意向的概率值计算和概率值判别，得到注意力分值数据；
    通过所述动态分类模型中的第二分类模型，对所述应用操作记录序列进行特征提取和分类，得到原始分值数据；
    将所述注意力分值数据和所述原始分值数据进行求和，得到动态分值数据；
    调用预置的静态分类模型，对所述待处理应用数据进行特征提取、注意力特征融合、基于业务操作意向的概率值计算和概率值判别，得到静态分值数据。
  12. 根据权利要求8所述的应用用户行为数据的处理设备，其中，所述对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略，包括：
    按照预设权重，计算所述动态分值数据和所述静态分值数据的和值，得到目标分值数据；
    将所述目标分值数据分别与所述业务操作意向预测数据、预设阈值进行比较分析，得到意向分析结果和阈值分析结果；
    若所述意向分析结果为所述业务操作意向预测数据小于所述目标分值数据，和/或所述阈值分析结果为所述目标分值数据小于预设阈值，则对预置调整策略序列进行首位调整策略提取，得到待推送的业务操作调整策略，所述预置调整策略序列为按照操作的评分值的大小进行倒序排序的调整策略集。
  13. 根据权利要求12所述的应用用户行为数据的处理设备，其中，所述若所述意向分析结果为所述业务操作意向预测数据小于所述目标分值数据，和/或所述阈值分析结果为所述目标分值数据小于预设阈值，则对预置调整策略序列进行首位调整策略提取，得到待推送的业务操作调整策略之后，还包括：
    若所述意向分析结果为所述业务操作意向预测数据大于或等于所述目标分值数据，则对所述循环神经网络模型进行增强学习。
  14. 根据权利要求8-13中任一项所述的应用用户行为数据的处理设备，其中，所述获 取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据之前，还包括：
    获取用户基于待处理应用程序的历史操作序列数据、目标用户画像数据和召回配置信息，所述目标用户画像数据包括操作数据的分值；
    基于所述历史操作序列数据和所述召回配置信息，对所述目标用户画像数据进行索引检索和读取，得到已读取操作数据的分值；
    通过所述已读取操作数据的分值，对所述历史操作序列数据进行评分，得到评分后的历史操作序列数据；
    按照评分值大小，对所述评分后的历史操作序列数据进行降序排列，得到预置调整策略序列。
  15. 一种计算机可读存储介质，所述计算机可读存储介质中存储计算机指令，当所述计算机指令在计算机上运行时，使得计算机执行如下步骤：
    获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
    通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
    调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
    对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
  16. 根据权利要求15所述的计算机可读存储介质，其中，所述获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据，包括：
    获取应用用户行为数据，按照业务操作结果和预设类型，对所述应用用户行为数据进行分类，得到初始应用操作数据和初始应用数据，所述预设类型包括应用操作数据类型和应用用户数据类型；
    对所述初始应用操作数据进行基于预设时段的链表存储，得到应用操作记录序列；
    对所述初始应用数据进行基于预设维度的操作对象分类和记录统计，得到待处理应用数据。
  17. 根据权利要求15所述的计算机可读存储介质，其中，所述通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据，包括：
    通过预置的循环神经网络模型，对所述应用操作记录序列进行基于业务操作结果达成的后续业务操作预测，得到目标后续业务操作信息，所述目标后续业务操作信息包括目标后续业务操作数据和所述目标后续业务操作数据基于业务操作结果达成的概率值；
    获取所述应用用户行为数据的后续操作数据，将所述应用用户行为数据的后续操作数据与所述目标后续业务操作数据进行匹配，得到匹配的目标后续业务操作数据；
    将所述匹配的目标后续业务操作数据对应的概率值确定为业务操作意向预测数据。
  18. 根据权利要求15所述的计算机可读存储介质，其中，所述调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分 值数据，包括：
    调用预置的动态分类模型中的第一分类模型，对所述应用操作记录序列进行注意力特征提取、基于业务操作意向的概率值计算和概率值判别，得到注意力分值数据；
    通过所述动态分类模型中的第二分类模型，对所述应用操作记录序列进行特征提取和分类，得到原始分值数据；
    将所述注意力分值数据和所述原始分值数据进行求和，得到动态分值数据；
    调用预置的静态分类模型，对所述待处理应用数据进行特征提取、注意力特征融合、基于业务操作意向的概率值计算和概率值判别，得到静态分值数据。
  19. 根据权利要求15所述的计算机可读存储介质，其中，所述对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略，包括：
    按照预设权重，计算所述动态分值数据和所述静态分值数据的和值，得到目标分值数据；
    将所述目标分值数据分别与所述业务操作意向预测数据、预设阈值进行比较分析，得到意向分析结果和阈值分析结果；
    若所述意向分析结果为所述业务操作意向预测数据小于所述目标分值数据，和/或所述阈值分析结果为所述目标分值数据小于预设阈值，则对预置调整策略序列进行首位调整策略提取，得到待推送的业务操作调整策略，所述预置调整策略序列为按照操作的评分值的大小进行倒序排序的调整策略集。
  20. 一种应用用户行为数据的处理装置，其中，所述应用用户行为数据的处理装置包括：
    统计模块，用于获取应用用户行为数据，并对所述应用用户行为数据进行预设类型分类、序列数据转换和分类统计，得到应用操作记录序列和待处理应用数据；
    预测模块，用于通过预置的循环神经网络模型，对所述应用操作记录序列进行后续业务操作意向的预测，得到业务操作意向预测数据；
    分类模块，用于调用预置的动态分类模型，对所述应用操作记录序列进行基于业务操作意向的分类，得到动态分值数据，并调用预置的静态分类模型，对所述待处理应用数据进行基于业务操作意向的分类，得到静态分值数据；
    第一获取模块，用于对所述动态分值数据和所述静态分值数据进行求和，得到目标分值数据，并根据所述业务操作意向预测数据和所述目标分值数据，从预置调整策略序列中获取待推送的业务操作调整策略。
PCT/CN2022/071485 2021-06-30 2022-01-12 应用用户行为数据的处理方法、装置、设备及存储介质 WO2023273299A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110735749.0 2021-06-30
CN202110735749.0A CN113190599B (zh) 2021-06-30 2021-06-30 应用用户行为数据的处理方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2023273299A1 true WO2023273299A1 (zh) 2023-01-05

Family

ID=76976728

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/071485 WO2023273299A1 (zh) 2021-06-30 2022-01-12 应用用户行为数据的处理方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN113190599B (zh)
WO (1) WO2023273299A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116843237A (zh) * 2023-09-04 2023-10-03 贵州惠智电子技术有限责任公司 一种办公平台应用考核统计分析系统
CN117235062A (zh) * 2023-11-10 2023-12-15 广州汇通国信科技有限公司 基于数据中台的业务系统数据建模方法
CN117290608A (zh) * 2023-11-23 2023-12-26 深圳数拓科技有限公司 一种营销方案智能推送方法、系统及存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113190599B (zh) * 2021-06-30 2021-09-28 平安科技（深圳）有限公司 应用用户行为数据的处理方法、装置、设备及存储介质
CN117238037B (zh) * 2023-11-13 2024-03-29 中国科学技术大学 动态动作识别方法、装置、设备和存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109741112A (zh) * 2019-01-10 2019-05-10 博拉网络股份有限公司 一种基于移动大数据的用户购买意向预测方法
CN110389970A (zh) * 2019-06-11 2019-10-29 中国平安财产保险股份有限公司 用户意向预测方法、装置、计算机设备及存储介质
US20200090056A1 (en) * 2018-09-19 2020-03-19 Tata Consultancy Services Limited Systems and methods for real time configurable recommendation using user data
CN112464094A (zh) * 2020-11-30 2021-03-09 泰康保险集团股份有限公司 信息推荐方法、装置、电子设备及存储介质
CN113190599A (zh) * 2021-06-30 2021-07-30 平安科技（深圳）有限公司 应用用户行为数据的处理方法、装置、设备及存储介质

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9858317B1 (en) * 2012-12-03 2018-01-02 Google Inc. Ranking communities based on members
CN103412882B (zh) * 2013-07-18 2016-11-23 百度在线网络技术（北京）有限公司 一种识别消费意图的方法及装置
CN107463853B (zh) * 2017-08-04 2020-01-21 晶赞广告（上海）有限公司 受众标签分析的方法及系统
CN110033294A (zh) * 2018-01-12 2019-07-19 腾讯科技（深圳）有限公司 一种业务分值的确定方法、业务分值确定装置及介质
CN109684543B (zh) * 2018-12-14 2021-04-02 北京百度网讯科技有限公司 用户行为预测和信息投放方法、装置、服务器和存储介质
CN109818839B (zh) * 2019-02-03 2022-02-25 三星电子（中国）研发中心 应用于智能家居的个性化行为预测方法、装置和系统
CN112905876B (zh) * 2020-03-16 2024-03-15 腾讯科技（深圳）有限公司 基于深度学习的信息推送方法、装置和计算机设备
CN111681051B (zh) * 2020-06-08 2023-09-26 上海汽车集团股份有限公司 购买意向度预测方法、装置、存储介质及终端
CN112835882B (zh) * 2021-01-29 2024-02-02 北京百家科技集团有限公司 信息预测方法、装置、电子设备和存储介质
CN112884505A (zh) * 2021-02-03 2021-06-01 北京百家科技集团有限公司 用户行为预测方法、装置、计算机设备以及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200090056A1 (en) * 2018-09-19 2020-03-19 Tata Consultancy Services Limited Systems and methods for real time configurable recommendation using user data
CN109741112A (zh) * 2019-01-10 2019-05-10 博拉网络股份有限公司 一种基于移动大数据的用户购买意向预测方法
CN110389970A (zh) * 2019-06-11 2019-10-29 中国平安财产保险股份有限公司 用户意向预测方法、装置、计算机设备及存储介质
CN112464094A (zh) * 2020-11-30 2021-03-09 泰康保险集团股份有限公司 信息推荐方法、装置、电子设备及存储介质
CN113190599A (zh) * 2021-06-30 2021-07-30 平安科技（深圳）有限公司 应用用户行为数据的处理方法、装置、设备及存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116843237A (zh) * 2023-09-04 2023-10-03 贵州惠智电子技术有限责任公司 一种办公平台应用考核统计分析系统
CN116843237B (zh) * 2023-09-04 2023-11-21 贵州惠智电子技术有限责任公司 一种办公平台应用考核统计分析系统
CN117235062A (zh) * 2023-11-10 2023-12-15 广州汇通国信科技有限公司 基于数据中台的业务系统数据建模方法
CN117290608A (zh) * 2023-11-23 2023-12-26 深圳数拓科技有限公司 一种营销方案智能推送方法、系统及存储介质

Also Published As

Publication number Publication date
CN113190599B (zh) 2021-09-28
CN113190599A (zh) 2021-07-30

Similar Documents

Publication Publication Date Title
WO2023273299A1 (zh) 应用用户行为数据的处理方法、装置、设备及存储介质
CN110163647B (zh) 一种数据处理方法及装置
Bamakan et al. A new intrusion detection approach using PSO based multiple criteria linear programming
Sculley et al. Detecting adversarial advertisements in the wild
US20060074621A1 (en) Apparatus and method for prioritized grouping of data representing events
KR101563406B1 (ko) 하둡 기반의 대용량 불균형 데이터 분류분석 시스템 및 방법
CN106709370B (zh) 一种基于文本内容的长词识别方法及系统
CN111507470A (zh) 一种异常账户的识别方法及装置
WO2022267456A1 (zh) 车辆定损方法、装置、设备及存储介质
CN111145009A (zh) 用户贷后风险评估方法、装置及电子设备
CN111210335A (zh) 用户风险识别方法、装置及电子设备
CN110347840A (zh) 投诉文本类别的预测方法、系统、设备和存储介质
CN110909195A (zh) 基于区块链的图片标注方法、装置及存储介质、服务器
CN110796171A (zh) 机器学习模型的未分类样本处理方法、装置及电子设备
Niu et al. Detecting malware on X86-based IoT devices in autonomous driving
CN111210336A (zh) 用户风险模型生成方法、装置及电子设备
More et al. An experimental assessment of random Forest classification performance improvisation with sampling and stage wise success rate calculation
CN117113403A (zh) 一种数据脱敏的方法、装置、电子设备及存储介质
CN110855474A (zh) Kqi数据的网络特征提取方法、装置、设备及存储介质
US11921676B2 (en) Analyzing deduplicated data blocks associated with unstructured documents
Vogler et al. Tackling the information overload: Using automated content analysis for crisis communication research
CN111615178B (zh) 识别无线网络类型及模型训练的方法、装置及电子设备
CN111008349B (zh) 一种大数据信息推送处理方法及系统
CN115640325A (zh) 内容推荐方法、装置、可读介质及电子设备
Ferdosi et al. Tracking Digital Device Utilization from Screenshot Analysis Using Deep Learning

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22831135

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE