Object classification and model training method and device, electronic equipment and storage medium

Abstract

The invention provides an object classification and model training method and device, electronic equipment and a storage medium, and relates to the field of artificial intelligence, in particular to the technical fields of NLP, cloud computing, big data, deep learning and the like. According to the specific implementation scheme, a historical behavior sequence of a target object is obtained; wherein the historical behavior sequence is generated according to historical behavior data of the target object in a plurality of business fields associated with the e-commerce platform; performing feature extraction on the historical behavior sequence to obtain target behavior features; classifying the target behavior characteristics by adopting an object classification model to obtain classification labels; wherein the classification label is used for indicating the life cycle stage of the target object on the e-commerce platform.

Description

Technical Field

[0001] The present disclosure relates to the field of AI (Artificial Intelligence), specifically to technical fields such as NLP (Natural Language Processing), cloud computing, big data, and deep learning, and especially to object classification and model training methods, devices, electronic devices, and storage media. Background Art

[0002] Managing user life cycle stages (or life cycle tags) is a key strategy for e-commerce platforms to optimize user experience and enhance user retention. This process comprehensively covers all important stages of users, from their first contact with the platform, through registration and initial use, to their gradual integration into daily use, and even to the eventual possible loss. By meticulously analyzing the specific life cycle stages of users, including but not limited to the new user exploration period, the second-new user adaptation period, the growth period, the maturity period, the low-activity observation period, the silent warning period, the severe silent period, and even the potential loss risk period, e-commerce platforms can more accurately understand user needs, thereby providing users with more personalized services and improving user satisfaction and loyalty. Summary of the invention

[0003] The present disclosure provides a method, device, electronic device and storage medium for object classification and model training.

[0004] According to one aspect of the present disclosure, there is provided an object classification method, comprising:

[0005] Acquire a historical behavior sequence of the target object; wherein the historical behavior sequence is generated based on historical behavior data of the target object in multiple business fields associated with the e-commerce platform;

[0006] Extracting features from the historical behavior sequence to obtain target behavior features;

[0007] An object classification model is used to classify the target behavior characteristics to obtain a classification label; wherein the classification label is used to indicate the life cycle stage of the target object on the e-commerce platform.

[0008] According to another aspect of the present disclosure, a method for training an object classification model is provided, comprising:

[0009] Acquire a historical behavior sequence of a sample object; wherein the historical behavior sequence is generated based on historical behavior data of the sample object in multiple business fields associated with the e-commerce platform;

[0010] Extracting features from the historical behavior sequence to obtain target behavior features;

[0011] The target behavior features are classified using an object classification model to obtain a predicted classification label; wherein the predicted label is used to indicate the life cycle stage of the sample object on the e-commerce platform;

[0012] The object classification model is trained according to the difference between the annotated classification label of the sample object and the predicted classification label.

[0013] According to another aspect of the present disclosure, there is provided an object classification device, comprising:

[0014] An acquisition module, used to acquire a historical behavior sequence of a target object; wherein the historical behavior sequence is generated based on historical behavior data of the target object in multiple business fields associated with an e-commerce platform;

[0015] An extraction module, used for extracting features from the historical behavior sequence to obtain target behavior features;

[0016] A classification module is used to classify the target behavior characteristics using an object classification model to obtain a classification label; wherein the classification label is used to indicate the life cycle stage of the target object on the e-commerce platform.

[0017] According to another aspect of the present disclosure, there is provided a training device for an object classification model, comprising:

[0018] An acquisition module, used to acquire a historical behavior sequence of a sample object; wherein the historical behavior sequence is generated based on historical behavior data of the sample object in multiple business fields associated with the e-commerce platform;

[0019] An extraction module, used for extracting features from the historical behavior sequence to obtain target behavior features;

[0020] A classification module, used to classify the target behavior features using an object classification model to obtain a predicted classification label; wherein the predicted label is used to indicate the life cycle stage of the sample object on the e-commerce platform;

[0021] The training module is used to train the object classification model according to the difference between the annotated classification label of the sample object and the predicted classification label.

[0022] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0023] at least one processor; and

[0024] a memory communicatively connected to the at least one processor; wherein,

[0025] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the object classification method proposed in the above-mentioned first aspect of the present disclosure, or execute the training method of the object classification model proposed in the above-mentioned second aspect of the present disclosure.

[0026] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium of computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the object classification method proposed in the above-mentioned first aspect of the present disclosure, or to execute the object classification model training method proposed in the above-mentioned other aspect of the present disclosure.

[0027] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the object classification method proposed in the above-mentioned first aspect of the present disclosure, or, when executed, implements the object classification model training method proposed in the above-mentioned other aspect of the present disclosure.

[0028] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0030] Figure 1 A flowchart of an object classification method provided in the first embodiment of the present disclosure;

[0031] Figure 2 A flowchart of an object classification method provided in Embodiment 2 of the present disclosure;

[0032] Figure 3 A flowchart of an object classification method provided in Embodiment 3 of the present disclosure;

[0033] Figure 4 A flowchart of an object classification method provided in Embodiment 4 of the present disclosure;

[0034] Figure 5 A flowchart of a method for training an object classification model provided in Embodiment 5 of the present disclosure;

[0035] Figure 6 A flowchart of a method for training an object classification model provided in Embodiment 6 of the present disclosure;

[0036] Figure 7 It is a schematic diagram of the implementation principle of each embodiment of the present disclosure;

[0037] Figure 8 This is a schematic diagram of the structure of an object classification device provided in Embodiment 7 of the present disclosure;

[0038] Fig. 9 A schematic diagram of the structure of a training device for an object classification model provided in Embodiment 8 of the present disclosure;

[0039] Fig.10 A schematic block diagram of an example electronic device that can be used to implement embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0040] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0041] At present, many companies do not fully realize the importance of user life cycle stage management, and user life cycle stage management is often neglected. For example, some companies focus too much on acquiring new users and ignore the maintenance and retention of existing users, while other companies are aware of user life cycle stage management, but in actual operation, due to the lack of effective methods and tools, they are unable to accurately manage the user life cycle stage.

[0042] Therefore, establishing and improving the management of the user life cycle stages can help companies better understand users, improve user experience, and increase user retention.

[0043] In related technologies, the management of user life cycle stages mainly relies on single-dimensional user behavior data, such as order data. By analyzing or predicting these behavior data, users can be classified and labeled to obtain the user's life cycle stage.

[0044] However, the above management methods for user life cycle stages have at least the following problems:

[0045] (1) High data dependence: If user behavior data is not collected completely or of low quality, the user life cycle stage obtained will be inaccurate and may even mislead the company's marketing decisions;

[0046] (2) Lack of depth: Relying solely on a single dimension of user behavior data (such as order time and order cycle) makes it difficult to deeply understand users' personalized needs and preferences, which, to a certain extent, limits the improvement of user experience;

[0047] (3) Untimely updates: User needs and behavior patterns are constantly changing, but existing user lifecycle management systems are often not updated in a timely manner, making it difficult to capture the latest changes in users.

[0048] (4) Lack of dynamic perspective: The user life cycle is a dynamic process, but existing tag management methods often deal with dynamic problems in a static statistical way. That is, they simply use multiple fixed time stages as the basis for life cycle division and lack differentiated management for user life cycle stages.

[0049] Therefore, in response to at least one of the above-mentioned problems, the present disclosure proposes an object classification and model training method, device, electronic device and storage medium.

[0050] The object classification and model training method, device, electronic device and storage medium of the embodiments of the present disclosure are described below with reference to the accompanying drawings.

[0051] Figure 1 A flowchart of the object classification method provided in the first embodiment of the present disclosure.

[0052] The disclosed embodiment takes the object classification method configured in an object classification device as an example. The object classification device can be applied to any electronic device so that the electronic device can perform an object classification function.

[0053] Among them, the electronic device can be any device with computing capabilities, such as a server, a personal computer, a mobile terminal, etc. The mobile terminal can be, for example, a mobile phone, a tablet computer, a personal digital assistant, a wearable device, and other hardware devices with various operating systems, touch screens and / or display screens.

[0054] like Figure 1 As shown, the object classification method may include the following steps S101 to S103:

[0055] Step S101, obtaining a historical behavior sequence of a target object; wherein the historical behavior sequence is generated based on historical behavior data of the target object in multiple business fields associated with an e-commerce platform.

[0056] The target object can be any user object.

[0057] Among them, multiple business fields include but are not limited to: live e-commerce, video e-commerce, search e-commerce, search, feed (information flow) and other fields.

[0058] In an embodiment of the present disclosure, a long-term behavior sequence of the target object can be generated based on the historical behavior data of the target object in multiple business fields associated with the e-commerce platform, which is recorded as a historical behavior sequence in the present disclosure.

[0059] Step S102: extract features from the historical behavior sequence to obtain target behavior features.

[0060] In the disclosed embodiment, the historical behavior sequence of the target object may be converted into behavior features for user classification, that is, a feature extraction algorithm may be used to extract features from the historical behavior sequence of the target object to obtain target behavior features.

[0061] Exemplarily, the hidden features of the historical behavior sequence can be extracted by counting the behavior frequency and behavior interval of the target object, or using a more complex feature extraction method (such as a sequence model), which are referred to as target behavior features in this disclosure.

[0062] Among them, sequence models include but are not limited to RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory), Transformer and other models.

[0063] Step S103, classify the target behavior features using an object classification model to obtain a classification label; wherein the classification label is used to indicate the life cycle stage of the target object in the e-commerce platform.

[0064] Among them, the life cycle stages include but are not limited to: new user exploration period, second-new user adaptation period, growth period, maturity period, low activity observation period, silent warning period, serious silent period, potential loss risk period, and others.

[0065] In an embodiment of the present disclosure, a trained object classification model may be used to classify target behavior features of a target object to obtain a classification label of the target object; wherein the classification label is used to indicate the life cycle stage of the target object on the e-commerce platform.

[0066] As an example, a trained object classification model can be used to classify the target behavior features of the target object, and the classification probabilities of multiple known classification labels output by the object classification model can be obtained. Thus, in the present disclosure, the classification label to which the target object belongs can be determined from multiple known classification labels according to the classification probabilities of multiple known classification labels. For example, the known classification label with the largest classification probability can be used as the classification label to which the target object belongs.

[0067] The object classification method of the disclosed embodiment integrates the historical behavior data of the target object in multiple business fields associated with the e-commerce platform to generate a complete or long-term historical behavior sequence, and then predicts the life cycle stage of the target object on the e-commerce platform based on the rich, complete or long-term historical behavior sequence. The accuracy and reliability of the prediction results can be improved, so that the e-commerce platform can provide personalized services and recommendations for the target object based on the precise life cycle stage, improve user experience, and increase user retention rate.

[0068] It should be noted that in the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information are all carried out with the user's consent, and are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0069] In order to clearly explain how to obtain the historical behavior sequence of the target object in any embodiment of the present disclosure, the present disclosure also proposes an object classification method.

[0070] Figure 2 A flowchart of the object classification method provided in the second embodiment of the present disclosure.

[0071] like Figure 2 As shown, the object classification method may include the following steps S201 to S205:

[0072] Step S201, obtaining historical behavior data of a target object in at least one first business field of an e-commerce platform; wherein the historical behavior data carries timestamp information.

[0073] The target object can be any user object.

[0074] Among them, the first business field includes but is not limited to: live e-commerce, video e-commerce, search e-commerce and other fields.

[0075] In the embodiment of the present disclosure, historical behavior data of the target object in each first business field of the e-commerce platform may be acquired or collected, wherein the historical behavior data of each first business field carries timestamp information.

[0076] In any embodiment of the present disclosure, the historical behavior data of the target object in each first business field can be obtained from the data application layer of the data warehouse associated with the e-commerce platform. The data warehouse may include the following layers:

[0077] 1. The data access layer is used to extract and store the original behavior data of the target object in each first business field from multiple data sources.

[0078] Exemplarily, the data access layer can extract the original behavior data of the target object in each first business field from data sources such as the log middle platform and the clue middle platform, and store the original behavior data of each first business field.

[0079] It should be noted that, in the related art, different service nodes in a distributed cluster are responsible for different task types. If there are a large number of tasks under the same task type, it may cause the task to be in a queue for a long time and not be executed in time. To address this problem, in any one of the embodiments of the present disclosure, the service nodes in the distributed cluster can be dynamically scheduled to collect the historical behavior data of the target object in each first business field. Exemplarily, the method for obtaining the original behavior data of the target object in each first business field is, for example, based on the load information of each service node in the distributed cluster, determining the target node with the lowest load from each service node, and scheduling the target node to perform the data collection task, so as to collect the original behavior data of each first business field associated with the target object from multiple data sources.

[0080] This can not only improve the resource utilization efficiency in the distributed cluster, but also improve the execution efficiency of data collection tasks.

[0081] 2. The data processing layer is used to process the original behavior data of each first business field to obtain intermediate behavior data, and based on the business theme of each first business field, process the intermediate behavior data of each first business field to obtain a theme portrait of each first business field.

[0082] Among them, business topics include but are not limited to: product portraits, merchant portraits, store portraits, etc.

[0083] Among them, the theme portrait can be expressed in the form of a large wide table.

[0084] Taking product portrait as an example, the basic attributes of the product (such as category, price, etc.), user behavior characteristics (such as average rating, purchase frequency, click-through rate) and sentiment characteristics (such as the proportion of positive reviews) can be extracted from the intermediate behavior data. Machine learning algorithms or deep learning algorithms (such as decision trees, random forests, neural networks, etc.) are used to analyze the relationship between these characteristics, find out which factors most affect the user's purchasing decision, and create product portraits based on the above characteristics and the relationship between the characteristics.

[0085] 3. A data application layer, which is used to generate historical behavior data of each first business field based on the intermediate behavior data and subject portrait of each first business field, and store the historical behavior data of each first business field.

[0086] In summary, processing the extracted raw behavior data and generating subject portraits can not only improve the quality and richness of the historical behavior data finally obtained, but also enable user behavior data to be stored, processed and applied in an orderly and efficient manner.

[0087] In any embodiment of the present disclosure, the data processing layer may include the following three DW layers:

[0088] 1) DWD (Data Warehouse Detail / Discover, basic data layer), used to clean the original behavior data of each first business field, obtain the clean behavior data of each first business field, and store it.

[0089] Exemplarily, outliers, dirty data, etc. in the original behavior data of each first business field may be cleaned to obtain clean behavior data of each first business field.

[0090] Exemplarily, the original behavior data of each first business field may be converted into a unified format, and outliers, dirty data, etc. in the original behavior data after the format conversion may be cleaned to obtain clean behavior data of each first business field.

[0091] 2) DWM (Data WareHouse Middle), which is used to extract data features and aggregate data for the clean behavior data of each first business field that has not triggered the risk control strategy, obtain the intermediate behavior data of each first business field, and store it.

[0092] Exemplarily, a risk control model may be used to identify whether the clean behavior data of each first business field triggers a risk control strategy. If so, subsequent steps are not performed, that is, the life cycle stage of the target object is not identified. If not, a model algorithm such as a cleaning rule is used to extract data features of the clean behavior data of each first business field, and the extracted data features are aggregated according to different dimensions to obtain intermediate behavior data of each first business field.

[0093] 3) DWS (Data Warehouse System / Services, data service layer), which is used to process the intermediate behavior data of each first business field based on the business subject of each first business field, obtain the subject portrait of each first business field, and store it.

[0094] Among them, business topics include but are not limited to: product portraits, merchant portraits, store portraits, etc.

[0095] Taking product portrait as an example, the basic attributes of the product (such as category, price, etc.), user behavior characteristics (such as average rating, purchase frequency, click-through rate) and sentiment characteristics (such as the proportion of positive reviews) can be extracted from the intermediate behavior data. Machine learning algorithms or deep learning algorithms (such as decision trees, random forests, neural networks, etc.) are used to analyze the relationship between these characteristics, find out which factors most affect the user's purchasing decision, and create product portraits based on the above characteristics and the relationship between the characteristics.

[0096] In summary, cleaning the raw behavior data, extracting data features, aggregating data, generating subject portraits, etc. can not only improve the quality of the historical behavior data finally obtained, but also enable user behavior data to be stored, processed and applied in an orderly and efficient manner.

[0097] In any one of the embodiments of the present disclosure, the data warehouse may further include: a data dimension layer, wherein the data dimension layer is used to store basic dimension data; wherein the basic dimension data is used to guide other layers in the data warehouse except the data dimension layer to perform data processing.

[0098] Among them, basic dimension data includes but is not limited to: city level table, vehicle model library information table, etc.

[0099] In the present disclosure, basic dimensional data can provide context for data analysis and is usually associated with other fact tables to support multidimensional analysis. In addition, the basic dimensional data contains metadata that is more related to the business field (such as products, customers, time, region, etc.), which can be used for data classification and induction, making data access and analysis more convenient and quick.

[0100] Step S202, obtaining historical behavior data of the target object in at least one second business field of the information service platform; wherein the historical behavior data carries timestamp information.

[0101] The second business domain includes, but is not limited to, search, feed (information flow) and other domains.

[0102] Among them, the information service platform is not only used to provide information services, but also to provide traffic entrances for e-commerce platforms, promoting the conversion of users from information browsing to shopping behavior.

[0103] In the embodiment of the present disclosure, historical behavior data of the target object in each second business field of the information service platform may be acquired or collected, wherein the historical behavior data of each second business field carries timestamp information.

[0104] It should be noted that the method for acquiring the historical behavior data of the target object in each second business context is similar to the method for acquiring the historical behavior data of the target object in each first business context, and will not be elaborated herein.

[0105] Step S203 , sorting and merging the historical behavior data of each first business field and the historical behavior data of each second business field according to the timestamp information to obtain a historical behavior sequence.

[0106] In the disclosed embodiment, the historical behavior data of the target object in each first business field and the historical behavior data of the target object in each second business field may be sorted and merged according to timestamp information to obtain a historical behavior sequence.

[0107] Exemplarily, the historical behavior data of different business scenarios covers specific behaviors such as user exposure, clicks, views, favorites, and comments. In the present disclosure, a fusion algorithm can be used to combine the historical behavior data of different business scenarios in chronological order to ensure the sequential nature of the basic behavior data of daily users.

[0108] For example, assuming that the target object (userA) watches short video data from 10:00 to 10:05, during which time he watches a 5-minute short video (videoA), clicks on the short video (videoA) 10 times, enters the live broadcast room through the live broadcast avatar in the short video (videoA) at 10:06, and watches for 10 minutes in the live broadcast room, then a historical behavior sequence of the following form can be formed: userA{videoA{duration:5;clicks:10;timestamp:"10:00-10:05"},live{enter_timestamp:"10:06";watch_duration:10;timestamp:"10:06-10:16"},…} or userA{videoA{duration:5;clicks:10},live{watch_duration:10},…}.

[0109] It should be noted that the behavior data of each user may change dynamically. In order to improve the accuracy and reliability of the prediction of the life cycle stage of each user, in the present disclosure, the historical behavior sequence of the target object can also be updated in a timely manner based on the new behavior data of the target object in the first business field and / or the second business field.

[0110] That is, in the present disclosure, when new behavior data of the target object in the first business domain is collected, the new behavior data can be fused with the historical behavior sequence based on the timestamp information in the new behavior data to obtain an updated historical behavior sequence.

[0111] Alternatively, when new behavior data of the target object in the second business field is collected, the new behavior data may be merged with the historical behavior sequence based on timestamp information in the new behavior data to obtain an updated historical behavior sequence.

[0112] Alternatively, when new behavior data of the target object in the first business context and the second business context is collected, the new behavior data may be merged with the historical behavior sequence based on timestamp information in the new behavior data to obtain an updated historical behavior sequence.

[0113] It should be noted that the historical behavior sequence (or long-term behavior sequence) is generated based on the daily historical behavior data of the target object starting from the registration of the target object, and the calculation overhead is very huge if a large amount of historical behavior data is used at one time to generate the historical behavior sequence. In response to this problem, in the present disclosure, the data warehouse can regularly (such as daily) calculate the new behavior data of the target object in the first business field and / or the second business field, and based on the timestamp information in the new behavior data, form a single-day behavior sequence of the target object, and merge the single-day behavior sequence with the historical behavior sequence updated the previous day to obtain the latest historical behavior sequence. In this way, not only the generation efficiency of the historical behavior sequence can be improved, but also the richness and comprehensiveness of the historical behavior sequence can be improved, thereby improving the accuracy and reliability of the subsequent life cycle stage prediction, that is, the historical behavior sequence can reflect the daily behavior changes of the target object, and can deeply understand the personalized needs and preferences of users, so as to make predictions of the life cycle stage based on the dynamically maintained historical behavior sequence, which can improve the rationality and reliability of the prediction results.

[0114] Step S204: extract features from the historical behavior sequence to obtain target behavior features.

[0115] Exemplarily, a feature extraction algorithm may be used to extract features from the latest historical behavior sequence of the target object to obtain target behavior features.

[0116] Step S205, classify the target behavior features using an object classification model to obtain a classification label; wherein the classification label is used to indicate the life cycle stage of the target object in the e-commerce platform.

[0117] For explanations of steps S204 to S205, reference may be made to the relevant descriptions in any embodiment of the present disclosure, and will not be repeated here.

[0118] The object classification method of the disclosed embodiment integrates the historical behavior data of the target object in various business fields of the e-commerce platform and the historical behavior data of the target object in various business fields of the information service platform to generate a historical behavior sequence of the target object, which can improve the richness, comprehensiveness and completeness of the historical behavior sequence, and thus improve the accuracy and reliability of subsequent life cycle stage predictions.

[0119] In order to clearly explain how to extract features from historical behavior sequences in any embodiment of the present disclosure to obtain target behavior features, the present disclosure also proposes an object classification method.

[0120] Figure 3 This is a flowchart of the object classification method provided in the third embodiment of the present disclosure.

[0121] like Figure 3 As shown, the object classification method may include the following steps S301 to S305:

[0122] Step S301, obtaining a historical behavior sequence of the target object; wherein the historical behavior sequence is generated based on historical behavior data of the target object in multiple business fields associated with the e-commerce platform.

[0123] For the explanation of step S301, reference may be made to the relevant description in any embodiment of the present disclosure, and no further details will be given here.

[0124] Step S302: extract features from the historical behavior sequence to obtain a plurality of initial behavior features.

[0125] In the disclosed embodiment, the historical behavior sequence of the target object may be converted into behavior features for user classification, that is, a feature extraction algorithm may be used to extract features from the historical behavior sequence of the target object to obtain a plurality of initial behavior features.

[0126] For example, the initial behavior features of the historical behavior sequence may be extracted by counting the behavior frequency and behavior interval of the target object, or by using a more complex feature extraction method (such as a sequence model).

[0127] Among them, sequence models include but are not limited to RNN, LSTM, Transformer and other models.

[0128] Step S303: Obtain the importance of multiple initial behavior features.

[0129] It should be noted that after feature extraction, a large number of initial behavior features may be obtained. In the present disclosure, the features that are most helpful for user classification (referred to as target behavior features in the present disclosure) can be screened out from the large number of initial behavior features.

[0130] Exemplarily, a feature selection method may be used to calculate the importance of a plurality of initial behavior features, so as to screen the plurality of initial behavior features based on the importance of the plurality of initial behavior features.

[0131] As a possible implementation method, the following steps A to B may be used to calculate the importance of the initial behavior feature:

[0132] Step A: For any initial behavior feature, calculate the statistic between the initial behavior feature and the target variable.

[0133] Among them, the target variable, also known as the dependent variable or response variable, is the test result of the machine learning prediction algorithm and the object that the model needs to predict or explain. In the context of object classification models, the target variable usually represents a certain state, behavior or attribute of the user, which the object classification model hopes to predict or explain based on behavioral characteristics.

[0134] The value of the target object is determined based on multiple known classification labels associated with the object classification model.

[0135] Among them, known classification labels are used to indicate known life cycle stages, including but not limited to: new user exploration period, second-new user adaptation period, growth period, maturity period, low activity observation period, silent warning period, serious silent period, potential loss risk period, and others.

[0136] The statistics include at least one of the following: chi-square value, correlation coefficient, information gain, and mutual information.

[0137] Among them, the chi-square value is used to test the independence between two variables; the correlation coefficient is used to measure the strength and direction of the linear relationship between two variables, where the value of the correlation coefficient is between -1 and 1, close to 1 or -1 indicates a strong linear relationship, and close to 0 indicates a weak linear relationship or no linear relationship; information gain is used to measure how much information the initial behavior feature provides for the target variable, or to measure the reduction of the uncertainty of the initial behavior feature on the target variable; mutual information is used to measure the dependency between two variables.

[0138] Exemplarily, the chi-square test method in the statistical method can be used to calculate the chi-square value between the initial behavioral feature and the target variable; the correlation coefficient method in the statistical method can be used to calculate the correlation coefficient between the initial behavioral feature and the target variable; the information gain method can be used to calculate the information gain between the initial behavioral feature and the target variable; the mutual information method can be used to calculate the mutual information between the initial behavioral feature and the target variable.

[0139] Step B: Determine the importance of the initial behavior feature based on the statistics.

[0140] Among them, the importance is positively correlated with the statistic, that is, the larger the statistic, the higher the importance.

[0141] Therefore, using statistics such as chi-square value, correlation coefficient, information gain, mutual information, etc. to measure the importance of initial behavioral features can improve the objectivity, accuracy, efficiency and interpretability of feature evaluation and screening.

[0142] As another possible implementation, the following steps A' to B' may be used to calculate the importance of the initial behavior feature:

[0143] Step A': For any initial behavior feature, a random forest model is used to predict the importance score of the initial behavior feature based on the target variable; wherein the target variable is determined based on multiple known classification labels associated with the object classification model; the importance score is used to indicate the impurity reduced in the splitting process of the decision tree corresponding to the random forest model for the initial behavior feature.

[0144] In the present disclosure, a model-based method (such as feature importance of random forest) can be used to predict the importance score of each initial behavior feature. In the random forest model, each decision tree is split based on the initial behavior feature to maximize the reduction of sample impurity (such as Gini coefficient or information gain). In this process, each initial behavior feature is assigned an importance score based on its ability to reduce impurity, and the importance score reflects the contribution of the initial behavior feature in the model decision process.

[0145] Step B': Determine the importance of the initial behavior feature according to the importance score.

[0146] Among them, the importance is positively correlated with the importance score, that is, the larger the importance score, the higher the importance.

[0147] Therefore, the random forest model can significantly improve the accuracy and stability of the overall prediction by constructing multiple decision trees and combining their prediction results. In the feature selection process, the random forest can objectively evaluate the degree of influence of each initial behavior feature on the target variable, avoiding the interference of subjective judgment. At the same time, because the random forest uses multiple decision trees for voting or averaging, its prediction results are usually more accurate than a single decision tree, so that the importance of the initial behavior feature can be more accurately evaluated. Therefore, the random forest model is used to predict the importance score of the initial behavior feature based on the target variable, and the importance of each initial behavior feature is determined accordingly, which can improve the objectivity, accuracy, stability, efficiency and scalability of feature evaluation and screening (the construction process of the random forest can be parallelized).

[0148] It should be noted that the above-mentioned feature selection method is only an exemplary description, but the present disclosure is not limited to this. In actual application, other feature selection methods can also be used to calculate the importance of multiple initial behavior features. For example, a learning-based method (such as recursive feature elimination) can also be used to calculate the importance of multiple initial behavior features. The embodiments of the present disclosure do not limit this.

[0149] Step S304: based on the importance of the multiple initial behavior features, the multiple initial behavior features are screened to obtain the target behavior feature.

[0150] Exemplarily, an initial behavior feature whose importance is higher than a set degree may be used as a target behavior feature.

[0151] Exemplarily, multiple initial behavior features may be sorted from high to low according to their importance to obtain a sorting sequence, and a set number or a set proportion of initial behavior features arranged at the front in the sorting sequence are used as target behavior features.

[0152] Step S305, classify the target behavior features using an object classification model to obtain a classification label; wherein the classification label is used to indicate the life cycle stage of the target object in the e-commerce platform.

[0153] For the explanation of step S305, reference may be made to the relevant description in any embodiment of the present disclosure, and no further details will be given here.

[0154] The object classification method of the disclosed embodiment screens initial behavior features based on the importance of a large number of initial behavior features, which can improve the quality of target behavior features, and then predict the life cycle stage of the target object based on the high-quality target behavior features, which can improve the accuracy and reliability of the prediction results.

[0155] In order to clearly illustrate any embodiment of the present disclosure, the present disclosure also proposes an object classification method.

[0156] Figure 4 A flowchart of the object classification method provided in Embodiment 4 of the present disclosure.

[0157] like Figure 4 As shown, the object classification method may include the following steps S401 to S406:

[0158] Step S401, obtaining a historical behavior sequence of the target object; wherein the historical behavior sequence is generated based on historical behavior data of the target object in multiple business fields associated with the e-commerce platform.

[0159] Step S402: extract features from the historical behavior sequence to obtain target behavior features.

[0160] Step S403, classify the target behavior features using an object classification model to obtain a classification label; wherein the classification label is used to indicate the life cycle stage of the target object in the e-commerce platform.

[0161] For explanations of steps S401 to S403, reference may be made to the relevant descriptions in any embodiment of the present disclosure, and will not be repeated here.

[0162] Step S404: acquiring, from the plurality of resource recommendation strategies, a target recommendation strategy adapted to the life cycle stage indicated by the classification tag.

[0163] Among them, multiple resource recommendation strategies are operational strategies formulated for different life cycle stages.

[0164] In an embodiment of the present disclosure, a correspondence or adaptation relationship between different resource recommendation strategies and life cycle stages can be established in advance. Thus, in the present disclosure, the above correspondence or adaptation relationship can be queried according to the life cycle stage indicated by the classification label to which the target object belongs, so as to obtain a resource recommendation strategy adapted to the life cycle stage indicated by the classification label, which is recorded as the target recommendation strategy in the present disclosure.

[0165] Step S405, based on the target recommendation strategy and the portrait label system of the target object, determine the target resource that is adapted to the target object from multiple candidate resources.

[0166] Among them, the portrait label system is a set of label information used to describe the user characteristics and behavior patterns of the target object.

[0167] Among them, candidate resources include but are not limited to: short videos, live broadcast rooms, coupons, product tutorials, etc.

[0168] In the disclosed embodiment, the initial label system of the target object can be obtained based on the basic information of the target object (such as demographic characteristics, social attributes, etc.), and the initial label system can be updated according to the historical behavior sequence to obtain the portrait label system of the target object.

[0169] Exemplarily, historical behavior sequences can be processed based on statistical analysis methods and deep learning methods to obtain portrait labels of the target object in multiple dimensions, and the portrait labels of the target object in multiple dimensions can be used to update the initial label system to obtain a portrait label system of the target object in multiple dimensions.

[0170] Among them, multiple dimensions include but are not limited to the following dimensions: basic attribute dimension, statistical feature dimension, user grouping dimension, predictive indicator dimension, etc.

[0171] In the embodiment of the present disclosure, the recommendation probability of multiple candidate resources can be predicted based on the target recommendation strategy, the image label system of the target object and the resource characteristics of multiple candidate resources, wherein the recommendation probability is used to indicate the preference of the target object for the candidate resources. Thus, in the present disclosure, the target resource adapted to the target object can be determined from the multiple candidate resources based on the recommendation probability of the multiple candidate resources.

[0172] Exemplarily, a candidate resource whose recommendation probability is higher than a set probability threshold may be used as a target resource.

[0173] Exemplarily, multiple candidate resources may be sorted from large to small according to the recommendation probability to obtain a sorting sequence, and thus in the present disclosure, a set number or a set proportion of candidate resources ranked first in the sorting sequence may be used as target resources.

[0174] Step S406: recommending target resources to the target object.

[0175] Exemplarily, each target resource may be pushed to the target object's account in the e-commerce platform.

[0176] Exemplarily, each target resource may be pushed to the target object via email or notification information.

[0177] In any embodiment of the present disclosure, in the process of recommending resources based on the target recommendation strategy, it is also possible to monitor the indicator values ​​of business indicators in multiple business scenarios based on the standard data caliber and standard indicator caliber of business indicators in multiple business scenarios, and adjust the target recommendation strategy according to the indicator values ​​of business indicators in multiple business scenarios.

[0178] Among them, business indicators include but are not limited to: CTCVR (Click-Through & Conversion Rate, which is the product of CTR (Click-Through Rate) and CVR (Conversion Rate), CTR, CVR, user retention rate, user activity, user satisfaction and other business indicators.

[0179] Among them, standard data caliber refers to the standard definition and specification of data items during the data collection stage, including but not limited to data source, data type, time range, geographical area, etc. It ensures that all relevant parties have a consistent understanding of the same data item.

[0180] Among them, the standard indicator caliber refers to the statistics or metrics defined for specific business needs, including the calculation formula, parameter settings, data processing details, units, upper and lower limits, etc. of the first business indicator.

[0181] For example, the real-time changing trends of business indicators can be analyzed, and the key factors and abnormal fluctuation points that affect business performance can be identified based on the real-time changing trends of business indicators. The future changing trends of business indicators can be predicted based on the real-time changing trends of business indicators. According to the future changing trends, key factors and abnormal fluctuation points of business indicators, specific resource recommendation strategy adjustment plans can be formulated: for example, content optimization: adjust the types and sorting of recommended content to better match user interests; time window optimization: optimize push time and frequency based on the user's active time period; resource allocation optimization: reasonably allocate marketing budgets and promotion resources, focusing on high-potential user groups, etc.

[0182] In summary, the use of a unified standard data caliber and indicator caliber can ensure the comparability and consistency of data between different business fields, avoid misjudgments caused by inconsistent calibers, and based on the indicator values ​​of business indicators, more accurately identify which strategies are effective and which strategies need to be optimized. This can not only improve the scientificity and accuracy of decision-making, but also enhance the flexibility and response speed of strategies to adapt to the ever-changing market environment, and can also optimize resource allocation and maximize resource utilization.

[0183] The object classification method of the disclosed embodiment can achieve more effective allocation of marketing resources and ensure that resources are invested in the user groups most likely to generate conversions. It can not only optimize resource allocation, but also ensure that the recommended content is highly consistent with the actual preferences of the target objects, reduce interference from irrelevant information, and enhance the target objects' trust and stickiness to the e-commerce platform.

[0184] The above are various embodiments corresponding to the application method of the object classification model (ie, the object classification method). The present disclosure also proposes a training method for the object classification model.

[0185] Figure 5 A flowchart of the object classification model training method provided in Embodiment 5 of the present disclosure.

[0186] like Figure 5 As shown, the training method of the object classification model may include the following steps S501 to S504:

[0187] Step S501, obtaining a historical behavior sequence of a sample object; wherein the historical behavior sequence is generated based on historical behavior data of the sample object in multiple business fields associated with the e-commerce platform.

[0188] The sample object can be any user object.

[0189] It should be noted that the implementation principle of step S501 is similar to that of step S101, and will not be elaborated here.

[0190] Step S502: extract features from the historical behavior sequence to obtain target behavior features.

[0191] It should be noted that the implementation principle of step S502 is similar to that of step S102, and will not be elaborated here.

[0192] Step S503, classify the target behavior features using the object classification model to obtain a predicted classification label; wherein the predicted label is used to indicate the life cycle stage of the sample object on the e-commerce platform.

[0193] It should be noted that the implementation principle of step S503 is similar to that of step S103, and will not be elaborated here.

[0194] Step S504: training the object classification model according to the difference between the annotated classification label and the predicted classification label of the sample object.

[0195] In an embodiment of the present disclosure, the value of the loss function can be determined based on the difference between the labeled classification label and the predicted classification label of the sample object, which is recorded as the loss value in the present disclosure, so that the above-mentioned object classification model can be trained based on the loss value.

[0196] Among them, the loss value is positively correlated with the difference, that is, the larger the difference, the larger the loss value, and conversely, the smaller the difference, the smaller the loss value.

[0197] Exemplarily, a set classification algorithm (such as decision tree, support vector machine, logistic regression, neural network, etc.) can be used to train the object classification model according to the above loss value.

[0198] In any embodiment of the present disclosure, when the number of sample objects is multiple, the target behavior features of the multiple sample objects can also be grouped to obtain k subsets, and based on the k subsets, the trained object classification model is evaluated using a k-fold cross-validation algorithm to determine the positive evaluation index of the trained object classification model, and to determine whether the positive evaluation index is higher than the index threshold. If so, the training process of the object classification model is terminated, and if not, the object classification model continues to be trained.

[0199] Wherein, k is a positive integer.

[0200] Among them, positive evaluation indicators include but are not limited to accuracy, recall, F1 value (which can be regarded as a harmonic average of precision and recall), etc.

[0201] Among them, the k-fold cross-validation algorithm refers to dividing the data set into K parts in equal proportion, using one of the K parts as the test data and the other K-1 parts as the training data. Then, the K parts of data are used to train and test the model to complete an experiment. It should be noted that the K-fold cross-validation algorithm is considered to have completed a complete experiment only after K experiments. In other words, cross-validation actually repeats the experiment K times, and each experiment selects a different part of the data from the K parts as the test data (to ensure that the data of the K parts have been tested separately), and the remaining K-1 parts are used as training data. Finally, the K experimental results are divided equally.

[0202] It should be understood that in actual application, other methods can also be used to evaluate the trained object classification model to determine the positive evaluation index of the trained object classification model, and the embodiments of the present disclosure are not limited to this. For example, the Leave-One-Out Cross-Validation (LOOCV) method can also be used to evaluate the trained object classification model, where the Leave-One-Out Cross-Validation method is a special case of k-fold cross validation, where k is equal to the size of the data set. Only one sample is used as the test set each time, and the remaining samples are used as the training set.

[0203] Therefore, the k-fold cross-validation algorithm is used to evaluate the trained object classification model, and the training process is terminated when the positive evaluation index is higher than the index threshold. This can not only reduce the randomness that may be introduced by a single evaluation and improve the robustness and accuracy of the evaluation results, but also effectively deal with data sparsity and category imbalance problems, optimize the training process and improve training efficiency.

[0204] The training method of the object classification model of the embodiment of the present disclosure integrates the historical behavior data of the sample objects in multiple business fields associated with the e-commerce platform to generate a complete or long-term historical behavior sequence, and then predicts the life cycle stage of the sample objects on the e-commerce platform based on the rich, complete or long-term historical behavior sequence. The accuracy and reliability of the prediction results can be improved, so that the e-commerce platform can provide personalized services and recommendations for the target objects based on the precise life cycle stage, improve user experience, and increase user retention rate.

[0205] In order to clearly illustrate the above embodiments, the present disclosure also proposes a training method for an object classification model.

[0206] Figure 6 A flowchart of the object classification model training method provided in Embodiment 6 of the present disclosure.

[0207] like Figure 6 As shown, the training method of the object classification model may include the following steps S601 to S607:

[0208] Step S601, obtaining historical behavior sequences of multiple sample objects; wherein the historical behavior sequences are generated based on historical behavior data of the sample objects in multiple business fields associated with the e-commerce platform.

[0209] For the explanation of step S601, reference may be made to the relevant description in any embodiment of the present disclosure, and no further details will be given here.

[0210] Step S602: clustering the multiple sample objects based on their historical behavior sequences to obtain multiple clusters.

[0211] In the embodiment of the present disclosure, a clustering algorithm may be used to cluster multiple sample objects based on their historical behavior sequences to obtain multiple clusters, wherein the historical behavior sequences of each sample object in the same cluster are similar.

[0212] Step S603: based on the historical behavior sequence of each sample object in each cluster, obtain the annotation classification label of each cluster.

[0213] As an example, manual labeling can be used to manually label each cluster based on the historical behavior sequence of each sample object in each cluster to obtain a labeled classification label for each cluster; wherein the labeled classification label for each cluster is used to indicate the life cycle stage of each sample object in the cluster on the e-commerce platform.

[0214] As another example, a machine labeling method can be used with a high-precision labeling model to automatically label each cluster based on the historical behavior sequence of each sample object in each cluster to obtain a label classification label for each cluster.

[0215] Step S604: Use the labeled classification label of each cluster as the labeled classification label of each sample object in each cluster.

[0216] In the embodiment of the present disclosure, for any cluster, the annotated classification label of the cluster can be used as the annotated classification label of all sample objects in the cluster.

[0217] Step S605 , extracting features from the historical behavior sequence of each sample object to obtain target behavior features of each sample object.

[0218] Step S606: classify the target behavior feature of each sample object using the object classification model to obtain a predicted classification label for each sample object.

[0219] Among them, the predicted label is used to indicate the life cycle stage of the sample object on the e-commerce platform.

[0220] Step S607: training the object classification model according to the difference between the annotated classification label and the predicted classification label of each sample object.

[0221] For explanations of steps S605 to S607, reference may be made to the relevant descriptions in any embodiment of the present disclosure, and will not be repeated here.

[0222] The object classification model training method of the disclosed embodiment uses a clustering method to cluster multiple sample objects, which can achieve batch labeling of sample objects and improve labeling efficiency.

[0223] In any embodiment of the present disclosure, based on the experience in the field of e-commerce vertical business, combined with data warehouse, data analysis, data mining, machine learning and other technologies, a system can be constructed as follows: Figure 7 The user life cycle prediction solution shown mainly includes the following parts: data infrastructure construction, user long-term behavior sequence construction, user life cycle prediction and application.

[0224] Part 1: Data infrastructure construction: Relying on users' historical behavior data, build a six-layer data warehouse system.

[0225] 1. Basic behavior data includes e-commerce business data.

[0226] E-commerce business includes user behavior data in live broadcast e-commerce, video e-commerce, search e-commerce and other fields, such as user viewing time, product card clicks, likes, follows, comments, and user purchase-related order data. In addition, the user's activity in other fields of the information service platform, such as search, feed (information flow), etc., is also used as the basic behavior data of e-commerce business, forming a conversion funnel from the wide field of information service platform to the e-commerce field.

[0227] 2. Data warehouse.

[0228] It can clean, count, aggregate, build a data warehouse for various behavioral data, and process and store behavioral data. The data warehouse mainly includes the following six layers:

[0229] 1) DIM (Dimension Data): This layer stores basic dimensional data, such as city level table, vehicle model library information table, etc.

[0230] 2) ODS (Operational Data Store, data access layer): This layer extracts and stores the original behavior data from the log middle platform, clue middle platform, and vehicle model library data.

[0231] 3) DWD (Data Warehouse Detail / Discover, basic data layer): This layer performs simple ETL (Extract-Transform-Load, which is used to describe the process of extracting, transforming, and loading data from the source to the destination) processing on the original behavior data of the ODS layer. It does not involve complex logic or model algorithms. Its purpose is to clean up usable clean behavior data.

[0232] 4) DWM (Data Warehouse Middle): This layer performs complex processing on the clean behavior data of the DWD layer. For example, a risk control model can be used to identify whether the clean behavior data triggers a risk control strategy. If so, the subsequent steps are not executed, that is, the user's life cycle stage is not identified. If not, model algorithms such as cleaning rules are used to extract data features of the clean behavior data, and the extracted data features are aggregated according to different dimensions to obtain the intermediate behavior data of each first business field.

[0233] 5) DWS (Data Warehouse System / Services): This layer is mostly based on business themes and builds relevant subject portraits, which are mostly expressed in the form of large wide tables.

[0234] 6) ADS (Application Data Service / Store): This layer extracts, performs logical operations and processes data in the three DW layers (i.e., DWD, DWM, and DWS) according to specific business needs, and generates historical behavior data or behavior data analysis reports directly for the business end.

[0235] Based on this, on the one hand, we strengthen the management of metadata to make the development and use of behavioral data more convenient. On the other hand, we strengthen tool construction, add resource load, task monitoring and other functions to ensure the timeliness and stability of behavioral data output.

[0236] Part 2: Construction of long-term user behavior sequence: To predict the user life cycle, it is necessary to track and characterize the user's behavior over a long period of time. Therefore, it is particularly important to construct a comprehensive and accurate long-term user behavior sequence. To this end, the long-term user behavior sequence can be constructed based on the behavior data in the e-commerce data warehouse.

[0237] 1. Multi-domain data fusion.

[0238] Based on the historical behavior data in the e-commerce data warehouse, it mainly includes four fields: search, feed, video, and live broadcast, covering specific behaviors such as user exposure, clicks, views, favorites, and comments. Based on the aggregated data of the e-commerce data warehouse, the historical behavior data of different business fields are combined in chronological order using a fusion algorithm, and the order of daily user basic behavior data is ensured to obtain a complete user long-term behavior sequence (referred to as a historical behavior sequence in this disclosure).

[0239] 2. Behavior sequence storage.

[0240] The long-term behavior sequence of users is counted from the time the user registers, and the accumulation period is more than 2 years, so it takes up a large storage space. The daily data warehouse will calculate the user's new behavior data to form the user's single-day behavior sequence, and then merge the user's single-day behavior sequence with the historical behavior sequence and store it using JSON (JavaScript Object Notation, an open standard file format and data exchange format). When it is determined that there is new user behavior, the latest behavior data will be appended and merged with the historical behavior sequence to form the user's complete historical behavior sequence.

[0241] Part III: User life cycle prediction and application.

[0242] 1. Sequence feature extraction: Convert the user's historical behavior sequence into features that can be used for classification. This can be done by counting the behavior frequency and behavior interval of each user, or using more complex methods such as sequence models (such as RNN, LSTM, Transformer, etc.) to extract hidden features of the historical behavior sequence.

[0243] 2. Feature selection: After feature extraction, a large number of behavioral features may be obtained. Various feature selection methods can be used, such as statistical-based methods (such as chi-square test), model-based methods (such as feature importance of random forest), and learning-based methods (such as recursive feature elimination) to select those features that are most helpful for user classification.

[0244] 3. Model training: Use the selected features and appropriate classification algorithms (such as decision trees, support vector machines, logistic regression, neural networks, etc.) to train the object classification model.

[0245] 4. Model evaluation: Use methods such as cross-validation or holdout to evaluate the performance of the object classification model, such as accuracy, recall, F1 value, etc.

[0246] 5. User classification: Use the trained object classification model to classify users and obtain the classification label of each user. The classification labels include but are not limited to the following life cycle labels: new user, second-new user, growth period, mature period, low activity period, silent period, severe silent period, churn period, and others.

[0247] 6. Refined operation strategy based on user life cycle.

[0248] a. User behavior analysis: There are obvious differences in user behavior and needs at each stage. For example, new users may be learning about the product, mature users focus on the user experience in various business fields, and lost users may lose interest in the product.

[0249] b. Strategy formulation: formulate corresponding operation strategies based on the user behavior characteristics and needs at each stage. For example, product tutorials and coupons can be provided to new users, personalized recommendations can be provided to growth-stage users, the needs and confusions of low-activity users can be understood through surveys, and lapsed users can be attracted to use the product again through promotional activities.

[0250] c. Strategy implementation: Concrete the formulated operational strategy into operational activities and implement them. For example, sending emails, pushing notifications, holding online events, etc.

[0251] d. Effect monitoring: After implementing the operation activities, it is necessary to monitor the effects of the operation activities, such as user activity, user satisfaction, user retention rate, etc.

[0252] e. Strategy adjustment: Adjust the operation strategy based on the monitoring results to improve the operation effect.

[0253] In summary, using user behavior data from multiple fields to construct a long-term user behavior sequence and using clustering algorithms to predict the user's actual life cycle stage can better reflect the user's true status. At the same time, implementing operational activities based on accurate life cycle stages can drive changes in user behavior at each stage and improve the retention rate of e-commerce platforms and the growth of GMV (Gross Merchandise Volume).

[0254] With the above Figures 1 to 4 Corresponding to the object classification method provided in the embodiment, the present disclosure also provides an object classification device. Figures 1 to 4 The object classification method provided in the embodiment corresponds to the object classification method, so the implementation of the object classification method is also applicable to the object classification device provided in the embodiment of the present disclosure, and will not be described in detail in the embodiment of the present disclosure.

[0255] Figure 8 This is a schematic diagram of the structure of the object classification device provided in Embodiment 7 of the present disclosure.

[0256] like Figure 8 As shown, the object classification device 800 may include: an acquisition module 810 , an extraction module 820 and a classification module 830 .

[0257] The acquisition module 810 is used to acquire the historical behavior sequence of the target object; wherein the historical behavior sequence is generated based on the historical behavior data of the target object in multiple business fields associated with the e-commerce platform;

[0258] An extraction module 820 is used to extract features from the historical behavior sequence to obtain target behavior features;

[0259] The classification module 830 is used to classify the target behavior characteristics using an object classification model to obtain a classification label; wherein the classification label is used to indicate the life cycle stage of the target object in the e-commerce platform.

[0260] In a possible implementation manner of the embodiment of the present disclosure, the multiple business fields include at least one first business field of the e-commerce platform and at least one second business field of the information service platform, and the information service platform is used to provide a traffic entrance for the e-commerce platform;

[0261] The acquisition module 810 is used to: acquire the historical behavior data of the target object in each first business field; wherein the historical behavior data carries timestamp information; acquire the historical behavior data of the target object in each second business field; wherein the historical behavior data carries timestamp information; sort and merge the historical behavior data of each first business field and the historical behavior data of each second business field according to the timestamp information to obtain a historical behavior sequence.

[0262] In a possible implementation of the embodiment of the present disclosure, the acquisition module 810 is used to: acquire historical behavior data of the target object in each first business field from the data application layer of the data warehouse associated with the e-commerce platform; wherein the data warehouse includes:

[0263] The data access layer is used to extract the original behavior data of the target object in each first business field from multiple data sources and store it;

[0264] The data processing layer is used to process the original behavior data of each first business field to obtain intermediate behavior data, and based on the business theme of each first business field, process the intermediate behavior data of each first business field to obtain a theme portrait of each first business field;

[0265] The data application layer is used to generate and store the historical behavior data of each first business field according to the intermediate behavior data and subject portrait of each first business field.

[0266] In a possible implementation of the embodiment of the present disclosure, the data warehouse further includes:

[0267] The data dimension layer is used to store basic dimension data; the basic dimension data is used to guide other layers in the data warehouse except the data dimension layer to perform data processing.

[0268] In a possible implementation of the embodiment of the present disclosure, the data processing layer includes:

[0269] The basic data layer is used to clean the original behavior data of each first business field, and obtain and store the clean behavior data of each first business field;

[0270] The intermediate data layer is used to extract data features and aggregate data for the clean behavior data of each first business field that has not triggered the risk control strategy, and obtain and store the intermediate behavior data of each first business field;

[0271] The data service layer is used to process the intermediate behavior data of each first business field based on the business subject of each first business field, and obtain and store the subject portrait of each first business field;

[0272] The data application layer is used to generate and store the historical behavior data of each first business field based on the clean behavior data, intermediate behavior data and subject portrait of each first business field.

[0273] In a possible implementation of the embodiment of the present disclosure, the acquisition module 810 is used to: determine the target node from each service node based on the load information of each service node in the distributed cluster; schedule the target node to perform a data collection task to collect the original behavior data of each first business field associated with the target object from multiple data sources.

[0274] In a possible implementation of the embodiment of the present disclosure, the object classification device 800 further includes:

[0275] The fusion module is used to, in response to collecting new behavior data of the target object in the first business field and / or the second business field, fuse the new behavior data with the historical behavior sequence based on timestamp information in the new behavior data to obtain an updated historical behavior sequence.

[0276] In a possible implementation of the embodiment of the present disclosure, the extraction module 820 is used to: extract features from the historical behavior sequence to obtain multiple initial behavior features; obtain the importance of the multiple initial behavior features; and screen the multiple initial behavior features based on the importance of the multiple initial behavior features to obtain the target behavior features.

[0277] In a possible implementation of the embodiments of the present disclosure, the extraction module 820 is used to: obtain statistics between any initial behavioral feature and a target variable; wherein the target variable is determined based on multiple known classification labels associated with an object classification model; determine the importance of any initial behavioral feature based on the statistics; wherein the statistics include at least one of the following: chi-square value, correlation coefficient, information gain, mutual information.

[0278] In a possible implementation of the embodiment of the present disclosure, the extraction module 820 is used to: use a random forest model to predict the importance score of any initial behavior feature based on the target variable; wherein the target variable is determined based on multiple known classification labels associated with the object classification model; the importance score is used to indicate the reduced impurity of any initial behavior feature during the splitting process of the decision tree corresponding to the random forest model; and determine the importance of any initial behavior feature based on the importance score.

[0279] In a possible implementation of the embodiment of the present disclosure, the object classification device 800 further includes:

[0280] A first determination module is used to determine, from a plurality of resource recommendation strategies, a target recommendation strategy adapted to the life cycle stage indicated by the classification label;

[0281] The second determination module is used to determine a target resource that is suitable for the target object from multiple candidate resources based on the target recommendation strategy and the portrait label system of the target object;

[0282] The recommendation module is used to recommend target resources to target objects.

[0283] In a possible implementation of the embodiment of the present disclosure, the object classification device 800 further includes:

[0284] A monitoring module, used to monitor the indicator values ​​of business indicators in multiple business scenarios based on the standard data caliber and standard indicator caliber of business indicators in multiple business scenarios in the process of recommending resources based on the target recommendation strategy;

[0285] The adjustment module is used to adjust the target recommendation strategy according to the indicator values ​​of the business indicators under multiple business scenarios.

[0286] The object classification device of the disclosed embodiment integrates the historical behavior data of the target object in multiple business fields associated with the e-commerce platform to generate a complete or long-term historical behavior sequence, and then predicts the life cycle stage of the target object on the e-commerce platform based on the rich, complete or long-term historical behavior sequence, which can improve the accuracy and reliability of the prediction results, so that the e-commerce platform can provide personalized services and recommendations for the target object based on the precise life cycle stage, improve user experience, and increase user retention rate.

[0287] With the above Figures 5 and 6 Corresponding to the training method of the object classification model provided in the embodiment, the present disclosure also provides a training device for the object classification model. Figures 5 and 6 The training method of the object classification model provided in the embodiment corresponds to the training method of the object classification model, so the implementation method of the training method of the object classification model is also applicable to the training device of the object classification model provided in the embodiment of the present disclosure, and will not be described in detail in the embodiment of the present disclosure.

[0288] Fig. 9 A schematic diagram of the structure of the training device for the object classification model provided in the eighth embodiment of the present disclosure.

[0289] like Fig. 9 As shown, the training device 900 for the object classification model may include: an acquisition module 910 , an extraction module 920 , a classification module 930 and a training module 940 .

[0290] The acquisition module 910 is used to acquire the historical behavior sequence of the sample object; wherein the historical behavior sequence is generated based on the historical behavior data of the sample object in multiple business fields associated with the e-commerce platform;

[0291] An extraction module 920 is used to extract features from the historical behavior sequence to obtain target behavior features;

[0292] The classification module 930 is used to classify the target behavior features using the object classification model to obtain a predicted classification label; wherein the predicted label is used to indicate the life cycle stage of the sample object in the e-commerce platform;

[0293] The training module 940 is used to train the object classification model according to the difference between the annotated classification label and the predicted classification label of the sample object.

[0294] In a possible implementation of the embodiment of the present disclosure, the number of sample objects is multiple, and the training device 900 of the object classification model further includes:

[0295] A grouping module, used for grouping target behavior features of multiple sample objects to obtain k subsets; wherein k is a positive integer;

[0296] An evaluation module, used to evaluate the trained object classification model according to the k subsets by using a k-fold cross-validation algorithm to determine a positive evaluation index of the trained object classification model;

[0297] The end module is used to end the training process of the object classification model when the positive evaluation index is higher than the index threshold.

[0298] In a possible implementation of the embodiment of the present disclosure, there are multiple sample objects, and the acquisition module 910 is further used to: cluster the multiple sample objects based on the historical behavior sequences of the multiple sample objects to obtain multiple clusters; obtain the annotated classification label of each cluster based on the historical behavior sequence of each sample object in each cluster; and use the annotated classification label of each cluster as the annotated classification label of each sample object in each cluster.

[0299] The training device of the object classification model of the embodiment of the present disclosure integrates the historical behavior data of the sample objects in multiple business fields associated with the e-commerce platform to generate a complete or long-period historical behavior sequence, and then predicts the life cycle stage of the sample objects in the e-commerce platform based on the rich, complete or long-period historical behavior sequence, which can improve the accuracy and reliability of the prediction results, so that the e-commerce platform can provide personalized services and recommendations for the target objects based on the precise life cycle stage, improve user experience, and increase user retention rate.

[0300] In order to implement the above embodiments, the present disclosure also provides an electronic device, which may include at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the object classification method or the object classification model training method proposed in any of the above embodiments of the present disclosure.

[0301] In order to implement the above embodiments, the present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the object classification method or object classification model training method proposed in any of the above embodiments of the present disclosure.

[0302] In order to implement the above embodiments, the present disclosure further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the object classification method or the object classification model training method proposed in any of the above embodiments of the present disclosure.

[0303] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0304] Fig.10 A schematic block diagram of an example electronic device that can be used to implement embodiments of the present disclosure is shown.

[0305] Among them, the electronic device may include the server and the client in the above-mentioned embodiment. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0306] like Fig.10As shown, the electronic device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 1002 or a computer program loaded from a storage unit 1007 to a RAM (Random Access Memory) 1003. In the RAM 1003, various programs and data required for the operation of the device 1000 can also be stored. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An I / O (Input / Output) interface 1005 is also connected to the bus 1004.

[0307] A number of components in the device 1000 are connected to the I / O interface 1005, including: an input unit 1006, such as a keyboard, a mouse, etc.; an output unit 1007, such as various types of displays, speakers, etc.; a storage unit 1008, such as a disk, an optical disk, etc.; and a communication unit 1009, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1009 allows the device 1000 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0308] The computing unit 1001 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as the above-mentioned object classification method or the training method of the object classification model. For example, in some embodiments, the above-mentioned object classification method or the training method of the object classification model may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1000 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded into RAM 1003 and executed by computing unit 1001, one or more steps of the object classification method or the object classification model training method described above may be performed. Alternatively, in other embodiments, computing unit 1001 may be configured to perform the object classification method or the object classification model training method described above in any other appropriate manner (e.g., by means of firmware).

[0309] Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and / or interpreted on a programmable system including at least one programmable processor that may be a dedicated or general-purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0310] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0311] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0312] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0313] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

[0314] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services (Virtual Private Server). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0315] It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0316] According to the technical solution of the embodiment of the present disclosure, by integrating the historical behavior data of the target object in multiple business fields associated with the e-commerce platform, a complete or long-period historical behavior sequence is generated, and then based on the rich, complete or long-period historical behavior sequence, the life cycle stage of the target object on the e-commerce platform is predicted. The accuracy and reliability of the prediction results can be improved, so that the e-commerce platform can provide personalized services and recommendations for the target object based on the precise life cycle stage, improve user experience, and increase user retention rate.

[0317] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0318] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for object classification, comprising: Acquire a historical behavior sequence of the target object; wherein the historical behavior sequence is generated based on historical behavior data of the target object in multiple business fields associated with the e-commerce platform; Extracting features from the historical behavior sequence to obtain target behavior features; An object classification model is used to classify the target behavior characteristics to obtain a classification label; wherein the classification label is used to indicate the life cycle stage of the target object on the e-commerce platform.

2. The method according to claim 1, wherein: The multiple business fields include at least one first business field of the e-commerce platform and at least one second business field of the information service platform, and the information service platform is used to provide a traffic entrance for the e-commerce platform; The step of obtaining the historical behavior sequence of the target object includes: Acquire historical behavior data of the target object in each of the first business fields; wherein the historical behavior data carries timestamp information; Acquire historical behavior data of the target object in each of the second business fields; wherein the historical behavior data carries timestamp information; The historical behavior data of each of the first business fields and the historical behavior data of each of the second business fields are sorted and merged according to timestamp information to obtain the historical behavior sequence.

3. The method according to claim 2, wherein: The acquiring of the historical behavior data of the target object in each of the first business fields includes: Acquire historical behavior data of the target object in each of the first business fields from a data application layer of a data warehouse associated with the e-commerce platform; wherein the data warehouse includes: A data access layer, used to extract and store the original behavior data of the target object in each of the first business fields from multiple data sources; A data processing layer, configured to process the original behavior data of each first business field to obtain intermediate behavior data, and based on the business subject of each first business field, process the intermediate behavior data of each first business field to obtain a subject portrait of each first business field; The data application layer is used to generate and store the historical behavior data of each first business field according to the intermediate behavior data and the subject portrait of each first business field.

4. The method according to claim 3, wherein: The data warehouse further includes: The data dimension layer is used to store basic dimension data; wherein the basic dimension data is used to guide other layers in the data warehouse except the data dimension layer to perform data processing.

5. The method according to claim 3, wherein: The data processing layer includes: The basic data layer is used to clean the original behavior data of each first business field, and obtain and store the clean behavior data of each first business field; An intermediate data layer, used to extract data features and aggregate data for the clean behavior data of each first business field that has not triggered the risk control strategy, and obtain and store the intermediate behavior data of each first business field; A data service layer, configured to process the intermediate behavior data of each first business field based on the business subject of each first business field, and obtain and store a subject portrait of each first business field; The data application layer is used to generate and store the historical behavior data of each first business field according to the clean behavior data, intermediate behavior data and subject portrait of each first business field.

6. The method according to claim 3, wherein: The original behavior data of the target object in each of the first business fields is collected by using the following steps: Determine a target node from each of the service nodes based on load information of each service node in the distributed cluster; The target node is scheduled to execute a data collection task to collect original behavior data of each first business field associated with the target object from multiple data sources.

7. The method according to claim 2, wherein: The method further comprises: In response to collecting new behavior data of the target object in the first business context and / or the second business context, the new behavior data is merged with the historical behavior sequence based on timestamp information in the new behavior data to obtain an updated historical behavior sequence.

8. The method according to any one of claims 1 to 7, wherein: The extracting features of the historical behavior sequence to obtain target behavior features includes: Extracting features from the historical behavior sequence to obtain a plurality of initial behavior features; Obtaining the importance of the multiple initial behavior features; Based on the importance of the multiple initial behavior features, the multiple initial behavior features are screened to obtain the target behavior feature.

9. The method according to claim 8, wherein: The obtaining of the importance of the plurality of initial behavior features includes: Obtaining a statistic between any initial behavior feature and a target variable; wherein the target variable is determined based on a plurality of known classification labels associated with the object classification model; Determining the importance of any of the initial behavior features according to the statistics; The statistics include at least one of the following: chi-square value, correlation coefficient, information gain, and mutual information.

10. The method according to claim 8, wherein: The obtaining of the importance of the plurality of initial behavior features includes: A random forest model is used to predict the importance score of any initial behavior feature based on a target variable; wherein the target variable is determined based on a plurality of known classification labels associated with the object classification model; and the importance score is used to indicate the impurity reduced by any initial behavior feature during the splitting process of a decision tree corresponding to the random forest model; The importance of any one of the initial behavior features is determined according to the importance score.

11. The method according to any one of claims 1 to 7, wherein: The method further comprises: Acquire, from a plurality of resource recommendation strategies, a target recommendation strategy adapted to the life cycle stage indicated by the classification label; Based on the target recommendation strategy and the portrait label system of the target object, determine a target resource that is suitable for the target object from multiple candidate resources; The target resource is recommended to the target object.

12. The method according to claim 11, wherein: The method further comprises: In the process of recommending resources based on the target recommendation strategy, based on the standard data caliber and standard indicator caliber of the business indicators under the multiple business scenarios, monitoring the indicator values ​​of the business indicators under the multiple business scenarios; The target recommendation strategy is adjusted according to the indicator values ​​of the business indicators under the multiple business scenarios.

13. A method for training an object classification model, comprising: Acquire a historical behavior sequence of a sample object; wherein the historical behavior sequence is generated based on historical behavior data of the sample object in multiple business fields associated with the e-commerce platform; Extracting features from the historical behavior sequence to obtain target behavior features; The target behavior features are classified using an object classification model to obtain a predicted classification label; wherein the predicted label is used to indicate the life cycle stage of the sample object on the e-commerce platform; The object classification model is trained according to the difference between the annotated classification label of the sample object and the predicted classification label.

14. The method according to claim 13, wherein: The number of the sample objects is multiple, and the method further includes: Grouping the target behavior features of the plurality of sample objects to obtain k subsets; wherein k is a positive integer; According to the k subsets, the trained object classification model is evaluated using a k-fold cross-validation algorithm to determine a positive evaluation index of the trained object classification model; When the positive evaluation index is higher than the index threshold, the training process of the object classification model is terminated.

15. The method according to claim 13, wherein: There are multiple sample objects, and the labeled classification labels of the sample objects are obtained by following the steps below: Based on the historical behavior sequences of the multiple sample objects, clustering the multiple sample objects to obtain multiple clusters; Based on the historical behavior sequence of each sample object in each cluster, obtaining a labeled classification label for each cluster; The labeled classification label of each cluster is used as the labeled classification label of each sample object in each cluster.

16. An object classification device, comprising: An acquisition module, used to acquire a historical behavior sequence of a target object; wherein the historical behavior sequence is generated based on historical behavior data of the target object in multiple business fields associated with an e-commerce platform; An extraction module, used for extracting features from the historical behavior sequence to obtain target behavior features; A classification module is used to classify the target behavior characteristics using an object classification model to obtain a classification label; wherein the classification label is used to indicate the life cycle stage of the target object on the e-commerce platform.

17. The device according to claim 16, wherein: The multiple business fields include at least one first business field of the e-commerce platform and at least one second business field of the information service platform, and the information service platform is used to provide a traffic entrance for the e-commerce platform; The acquisition module is used to: Acquire historical behavior data of the target object in each of the first business fields; wherein the historical behavior data carries timestamp information; Acquire historical behavior data of the target object in each of the second business fields; wherein the historical behavior data carries timestamp information; The historical behavior data of each of the first business fields and the historical behavior data of each of the second business fields are sorted and merged according to timestamp information to obtain the historical behavior sequence.

18. The device according to claim 17, wherein: The acquisition module is used to: Acquire historical behavior data of the target object in each of the first business fields from a data application layer of a data warehouse associated with the e-commerce platform; wherein the data warehouse includes: A data access layer, used to extract and store the original behavior data of the target object in each of the first business fields from multiple data sources; A data processing layer, configured to process the original behavior data of each first business field to obtain intermediate behavior data, and based on the business subject of each first business field, process the intermediate behavior data of each first business field to obtain a subject portrait of each first business field; The data application layer is used to generate and store the historical behavior data of each first business field according to the intermediate behavior data and the subject portrait of each first business field.

19. The device according to claim 18, wherein: The data warehouse further includes: The data dimension layer is used to store basic dimension data; wherein the basic dimension data is used to guide other layers in the data warehouse except the data dimension layer to perform data processing.

20. The device according to claim 18, wherein The data processing layer includes: The basic data layer is used to clean the original behavior data of each first business field, and obtain and store the clean behavior data of each first business field; An intermediate data layer, used to extract data features and aggregate data for the clean behavior data of each first business field that has not triggered the risk control strategy, and obtain and store the intermediate behavior data of each first business field; A data service layer, configured to process the intermediate behavior data of each first business field based on the business subject of each first business field, and obtain and store a subject portrait of each first business field; The data application layer is used to generate and store the historical behavior data of each first business field according to the clean behavior data, intermediate behavior data and subject portrait of each first business field.

21. The device according to claim 18, wherein The acquisition module is used to: Determine a target node from each of the service nodes based on load information of each service node in the distributed cluster; The target node is scheduled to execute a data collection task to collect original behavior data of each first business field associated with the target object from multiple data sources.

22. The device according to claim 17, wherein: The device also includes: A fusion module is used to, in response to collecting new behavior data of the target object in the first business domain and / or the second business domain, fuse the new behavior data with the historical behavior sequence based on timestamp information in the new behavior data to obtain an updated historical behavior sequence.

23. The device according to any one of claims 16 to 22, wherein: The extraction module is used to: Extracting features from the historical behavior sequence to obtain a plurality of initial behavior features; Obtaining the importance of the multiple initial behavior features; Based on the importance of the multiple initial behavior features, the multiple initial behavior features are screened to obtain the target behavior feature.

24. The device according to claim 23, wherein: The extraction module is used to: Obtaining a statistic between any initial behavior feature and a target variable; wherein the target variable is determined based on a plurality of known classification labels associated with the object classification model; Determining the importance of any of the initial behavior features according to the statistics; The statistics include at least one of the following: chi-square value, correlation coefficient, information gain, and mutual information.

25. The device according to claim 23, wherein: The extraction module is used to: A random forest model is used to predict the importance score of any initial behavior feature based on a target variable; wherein the target variable is determined based on a plurality of known classification labels associated with the object classification model; and the importance score is used to indicate the impurity reduced by any initial behavior feature during the splitting process of a decision tree corresponding to the random forest model; The importance of any one of the initial behavior features is determined according to the importance score.

26. The device according to any one of claims 16 to 22, wherein: The device also includes: A first determination module is used to determine, from a plurality of resource recommendation strategies, a target recommendation strategy adapted to the life cycle stage indicated by the classification tag; A second determination module is used to determine a target resource that is suitable for the target object from a plurality of candidate resources based on the target recommendation strategy and the portrait label system of the target object; A recommendation module is used to recommend the target resource to the target object.

27. The device according to claim 26, wherein: The device also includes: A monitoring module, used to monitor the indicator values ​​of the business indicators under the multiple business scenarios based on the standard data caliber and standard indicator caliber of the business indicators under the multiple business scenarios in the process of recommending resources based on the target recommendation strategy; The adjustment module is used to adjust the target recommendation strategy according to the indicator values ​​of the business indicators under the multiple business scenarios.

28. A training device for an object classification model, comprising: An acquisition module, used to acquire a historical behavior sequence of a sample object; wherein the historical behavior sequence is generated based on historical behavior data of the sample object in multiple business fields associated with the e-commerce platform; An extraction module, used for extracting features from the historical behavior sequence to obtain target behavior features; A classification module, used to classify the target behavior features using an object classification model to obtain a predicted classification label; wherein the predicted label is used to indicate the life cycle stage of the sample object on the e-commerce platform; The training module is used to train the object classification model according to the difference between the annotated classification label of the sample object and the predicted classification label.

29. The device according to claim 28, wherein The number of the sample objects is multiple, and the device further includes: A grouping module, used for grouping the target behavior features of the plurality of sample objects to obtain k subsets; wherein k is a positive integer; An evaluation module, configured to evaluate the trained object classification model using a k-fold cross-validation algorithm according to the k subsets to determine a positive evaluation index of the trained object classification model; The end module is used to end the training process of the object classification model when the positive evaluation index is higher than the index threshold.

30. The device according to claim 28, wherein The number of the sample objects is multiple, and the acquisition module is further used to: Based on the historical behavior sequences of the multiple sample objects, clustering the multiple sample objects to obtain multiple clusters; Based on the historical behavior sequence of each sample object in each cluster, obtaining a labeled classification label for each cluster; The labeled classification label of each cluster is used as the labeled classification label of each sample object in each cluster.

31. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the object classification method described in any one of claims 1-12, or execute the training method of the object classification model described in any one of claims 13-15.

32. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to enable the computer to execute the object classification method according to any one of claims 1-12, or to execute the object classification model training method according to any one of claims 13-15.

33. A computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the object classification method according to any one of claims 1 to 12, or implements the steps of the object classification model training method according to any one of claims 13 to 15.

Images