User classification method and related equipment based on multivariate data fusion

Abstract

The present invention relates to artificial intelligence, and provides a user classification method based on multivariate data fusion and related equipment. The method collects user information of a user to be tested; pre-processes the user information to obtain multiple structured features; selects a first feature pair and a second feature pair based on the feature similarity of any two structured features among the multiple structured features; identifies the feature importance of each structured feature based on a pre-trained classification model; fuses the first feature pair according to the feature importance to obtain a fused feature, merges the second feature pair to obtain a merged feature, and generates multiple target features based on the fused feature and the merged feature; identifies and processes the multiple target features based on the classification model to improve the accuracy of the classification result. In addition, the present invention also relates to blockchain technology, and the classification result can be stored in the blockchain.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a user classification method based on multivariate data fusion and related equipment. Background Art

[0002] In the field of financial marketing, there are heterogeneous data of different dimensions. The sources of these data are complex and the structures are varied. It is impossible to fully explore the characteristics of users with only a single dimension and a single data form. However, in the process of identifying potential customers, it is currently impossible to accurately identify whether a user is a potential customer due to the inability to effectively integrate the user's multi-dimensional data. Summary of the invention

[0003] In view of the above, it is necessary to provide a user classification method and related equipment based on multivariate data fusion, which can solve the technical problem of being unable to accurately identify whether a user is a potential customer.

[0004] On the one hand, the present invention proposes a user classification method based on multivariate data fusion, and the user classification method based on multivariate data fusion includes:

[0005] Collect user information of the user to be tested;

[0006] Preprocessing the user information to obtain multiple structured features;

[0007] Selecting a first feature pair and a second feature pair based on feature similarity between any two structured features among the plurality of structured features;

[0008] Identify the feature importance of each structural feature based on the pre-trained classification model;

[0009] According to the feature importance, the first feature pair is fused to obtain a fused feature, the second feature pair is fused to obtain a combined feature, and a plurality of target features are generated according to the fused feature and the combined feature;

[0010] The multiple target features are identified and processed based on the classification model to obtain a classification result of the user to be tested.

[0011] According to a preferred embodiment of the present invention, the preprocessing of the user information to obtain a plurality of structured features includes:

[0012] Identifying structured information in the user information based on a collection source of the user information;

[0013] Building a user portrait wide table based on the structured information;

[0014] Identify abnormal data of field information corresponding to each wide table field in the user portrait wide table;

[0015] Filtering out target users corresponding to the abnormal data from the users to be tested;

[0016] Detecting whether the target user is an abnormal user based on the abnormal data;

[0017] If the target user is not the abnormal user, the abnormal data corresponding to the target user is replaced to obtain the multiple structured features.

[0018] According to a preferred embodiment of the present invention, the preprocessing of the user information to obtain a plurality of structured features further comprises:

[0019] Identifying unstructured information in the user information based on a collection source of the user information;

[0020] Performing data cleaning on the unstructured information based on preset characters to obtain text information;

[0021] Extracting text sequence and time sequence from the text information;

[0022] The text sequence is subjected to text analysis based on preset features, and the time series is subjected to time analysis based on the preset features to obtain the multiple structural features.

[0023] According to a preferred embodiment of the present invention, the selecting the first feature pair and the second feature pair based on the feature similarity of any two structured features among the plurality of structured features comprises:

[0024] Calculating the feature similarity based on the first vector and the second vector generated by constructing any two structured features;

[0025] Determine any two structural features whose feature similarity is greater than or equal to a first preset similarity threshold as the first feature pair;

[0026] Any two structural features whose feature similarity is less than or equal to a second preset similarity threshold are determined as the second feature pair, and the first preset similarity threshold is greater than the second preset similarity threshold.

[0027] According to a preferred embodiment of the present invention, the feature importance of each structural feature identified based on the pre-trained classification model includes:

[0028] Performing an amplification process on the classification model to obtain multiple processing models;

[0029] Calculating a ratio of the number of features of the plurality of structured features to the number of models of the plurality of processing models;

[0030] Selecting features from the plurality of structured features based on the ratio and inputting them into the plurality of processing models for parallel processing respectively, so as to obtain an output category of each structured feature by the classification model;

[0031] Identifying the prediction accuracy of the classification model for each structured feature based on the comparison results of the labeled categories of the plurality of structured features and the output category;

[0032] The prediction accuracy is normalized to obtain the feature importance of each structured feature.

[0033] According to a preferred embodiment of the present invention, the fusing the first feature pair according to the feature importance to obtain a fused feature, merging the second feature pair to obtain a merged feature, and generating multiple target features according to the fused feature and the merged feature includes:

[0034] Calculating a first importance ratio of a first left feature in the first feature pair based on the feature importance, and calculating a second importance ratio of a first right feature in the first feature pair;

[0035] Based on the first importance ratio and the second importance ratio, a weighted sum calculation is performed on the first left feature and the first right feature to obtain the fused feature;

[0036] Performing mutually exclusive merging processing on the second left feature in the second feature pair and the second right feature in the second feature pair to obtain the merged feature;

[0037] The fused features and the combined features are determined as the multiple target features, and the feature similarity of any two target features among the multiple target features is greater than the second preset similarity threshold and less than the first preset similarity threshold.

[0038] According to a preferred embodiment of the present invention, the second left feature in the second feature pair and the second right feature in the second feature pair are subjected to mutually exclusive merging processing to obtain the merged feature, comprising:

[0039] For the second left feature, identifying any feature associated with the second left feature from the second right feature as a primary screening feature, and constructing a feature cluster based on the second left feature and the primary screening feature;

[0040] Identify a feature that is associated with both the second left feature and the primary screening feature from the second right feature as an intermediate feature;

[0041] Counting the number of configuration values ​​that are different between the intermediate feature and the feature values ​​in the feature cluster as the number of contradictions between the intermediate feature and the feature cluster;

[0042] If the number of contradictions is less than a preset contradiction threshold, merging the intermediate features into the feature cluster;

[0043] Each merged feature cluster is fused based on the feature importance to obtain the merged feature.

[0044] On the other hand, the present invention further proposes a user classification device based on multivariate data fusion, the user classification device based on multivariate data fusion comprising:

[0045] A collection unit, used to collect user information of the user to be tested;

[0046] A preprocessing unit, used for preprocessing the user information to obtain a plurality of structured features;

[0047] A selection unit, configured to select a first feature pair and a second feature pair based on feature similarity between any two structured features among the plurality of structured features;

[0048] A recognition unit, used to identify the feature importance of each structural feature based on a pre-trained classification model;

[0049] a generating unit, configured to fuse the first feature pairs according to the feature importance to obtain a fused feature, fuse the second feature pairs to obtain a combined feature, and generate a plurality of target features according to the fused feature and the combined feature;

[0050] The recognition unit is further used to perform recognition processing on the multiple target features based on the classification model to obtain a classification result of the user to be tested.

[0051] On the other hand, the present invention further provides an electronic device, comprising:

[0052] a memory storing computer readable instructions; and

[0053] The processor executes the computer-readable instructions stored in the memory to implement the user classification method based on multivariate data fusion.

[0054] On the other hand, the present invention further proposes a computer-readable storage medium, in which computer-readable instructions are stored. The computer-readable instructions are executed by a processor in an electronic device to implement the user classification method based on multivariate data fusion.

[0055] It can be seen from the above technical solutions that the present invention can not only eliminate the influence of interference data on features by preprocessing the user information, but also convert the unstructured information in the user information into structured features, thereby improving the analysis efficiency. Furthermore, the classification model can accurately identify the feature importance, thereby improving the fusion accuracy of the first feature pair. At the same time, the present invention can effectively realize the fusion of the first feature pair and the second feature pair separately by combining the feature importance with the first feature pair and merging the second feature pair, thereby improving the feature validity of the multiple target features, thereby improving the accuracy of the classification result. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of a preferred embodiment of the user classification method based on multivariate data fusion of the present invention.

[0057] Figure 2 It is a functional module diagram of a preferred embodiment of the user classification device based on multivariate data fusion of the present invention.

[0058] Figure 3 It is a structural schematic diagram of an electronic device of a preferred embodiment of the present invention for implementing a user classification method based on multivariate data fusion. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0060] like Figure 1 FIG. 1 is a flowchart of a preferred embodiment of a user classification method based on multivariate data fusion according to the present invention. According to different requirements, the order of the steps in the flowchart can be changed, and some steps can be omitted.

[0061] The user classification method based on multivariate data fusion can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0062] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0063] The user classification method based on multivariate data fusion is applied to one or more electronic devices, and the electronic device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored computer-readable instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital signal processors (Digital Signal Processor, DSP), embedded devices, etc.

[0064] The electronic device may be any electronic product that can perform human-computer interaction with a user, such as a personal computer, a tablet computer, a smart phone, a personal digital assistant (PDA), a game console, an interactive network television (IPTV), a smart wearable device, etc.

[0065] The electronic device may include a network device and / or a user device. The network device includes, but is not limited to, a single network electronic device, an electronic device group consisting of multiple network electronic devices, or a cloud consisting of a large number of hosts or network electronic devices based on cloud computing.

[0066] The network where the electronic device is located includes, but is not limited to: the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VPN), etc.

[0067] 101. Collect user information of the user to be tested.

[0068] In at least one embodiment of the present invention, the users to be tested may be a plurality of consumer users mined from a big data platform.

[0069] The user information includes data information of the user to be tested in multiple different dimensions. For example, the user information may include, but is not limited to: user basic attributes (such as age, gender, education level, occupation, etc.), user static behavior data (such as user's product ownership), user dynamic behavior data (such as APP click status, etc.) and voice communication data, text communication data and other information in different dimensions.

[0070] In at least one embodiment of the present invention, the electronic device collects relevant information about the user to be tested from multiple collection sources as the user information.

[0071] 102. Preprocess the user information to obtain a plurality of structured features.

[0072] In at least one embodiment of the present invention, the multiple structured features include features obtained after data cleaning of structured information, and the multiple structured features also include features obtained after statistical feature, time feature and trend feature construction analysis of unstructured information.

[0073] In at least one embodiment of the present invention, the electronic device preprocesses the user information to obtain a plurality of structured features including:

[0074] Identifying structured information in the user information based on a collection source of the user information;

[0075] Building a user portrait wide table based on the structured information;

[0076] Identify abnormal data of field information corresponding to each wide table field in the user portrait wide table;

[0077] Filtering out target users corresponding to the abnormal data from the users to be tested;

[0078] Detecting whether the target user is an abnormal user based on the abnormal data;

[0079] If the target user is not the abnormal user, the abnormal data corresponding to the target user is replaced to obtain the multiple structured features.

[0080] The collection sources may include relational databases, non-relational databases and other platforms.

[0081] The structured information generally refers to information extracted from the relational database.

[0082] The user portrait wide table refers to a data table obtained by associating dimension data stored in different data tables with the user to be tested as the primary key. The user portrait wide table stores user information of multiple users to be tested.

[0083] The abnormal data refers to the field information in each wide table field whose deviation from the field information is greater than a preset deviation. The preset deviation can be determined according to actual needs.

[0084] The abnormal user refers to a user containing the abnormal data.

[0085] By constructing the user portrait wide table, the abnormal data can be accurately identified, thereby improving the accuracy of determining the target user. Furthermore, when it is detected that the target user is not the abnormal user, the influence of the abnormal data of the target user on the multiple structured features can be avoided by replacing the abnormal data.

[0086] Specifically, the electronic device performs replacement processing on the abnormal data corresponding to the target user, and obtains the multiple structured features including:

[0087] If the data type of the abnormal data corresponding to the target user is character type, then obtaining the mode information in the wide table field where the abnormal data is located from the user portrait wide table, and replacing the abnormal data based on the mode information to obtain the multiple structured features; or

[0088] If the data type of the abnormal data corresponding to the target user is a numerical type, the abnormal data is replaced based on the configuration value to obtain the multiple structured features.

[0089] The configuration value may be 0, or may be the average value in the wide table field where the abnormal data is located.

[0090] In other embodiments, if the target user is the abnormal user, the target user is removed from the users to be tested, and the identification of the target user is terminated.

[0091] In at least one embodiment of the present invention, the electronic device preprocesses the user information to obtain a plurality of structured features further comprising:

[0092] Identifying unstructured information in the user information based on a collection source of the user information;

[0093] Performing data cleaning on the unstructured information based on preset characters to obtain text information;

[0094] Extracting text sequence and time sequence from the text information;

[0095] The text sequence is subjected to text analysis based on preset features, and the time series is subjected to time analysis based on the preset features to obtain the multiple structural features.

[0096] The unstructured information may be information obtained from the non-relational database.

[0097] The preset characters may include, but are not limited to: punctuation marks, emoticons, non-Chinese characters, modal particles, etc.

[0098] The text sequence refers to the information related to communication in the text information, and the time sequence refers to the information related to time in the text information.

[0099] The preset features can be set according to actual needs. The preset features include statistical dimension features, time dimension features and trend dimension features. For example, the preset features for text analysis of the text sequence may include the total number of text words, the ratio of the total number of text words to the total number of expressions, the number of appearances of marketing words, and the marketing words with the highest frequency in the statistical dimension features; the preset features for text analysis of the text sequence may include the marketing words with the highest number of expressions in the most recent and most distant times, and the marketing words with the lowest number of expressions in the most recent and most distant times in the time dimension features; the preset features for text analysis of the text sequence may include the ratio of the number of times a marketing word is expressed in the last n days to the number of times a marketing word is expressed in the last m days in the trend dimension features. The preset features for time analysis of the text sequence may include the average communication interval duration, the maximum communication time interval, and the minimum communication time interval in the statistical dimension features; the preset features for time analysis of the text sequence may include whether the most recent communication time is in the morning, afternoon, or evening, whether the most recent communication time is on the weekend, whether the farthest communication time is in the morning, afternoon, or evening, and whether the farthest communication time is on the weekend in the time dimension features; the preset features for time analysis of the text sequence may include the average communication interval duration of the last i times and the average communication interval duration of the last j times in the trend dimension features, etc.

[0100] By cleaning the preset characters in the unstructured information, the amount of information in the text information can be reduced, thereby improving the generation efficiency of the multiple structured features. Furthermore, by extracting the text sequence and the time series, the text information can be analyzed in combination with the text dimension and the time dimension, thereby obtaining comprehensive structured features.

[0101] 103 . Select a first feature pair and a second feature pair based on feature similarity between any two structured features from the plurality of structured features.

[0102] In at least one embodiment of the present invention, the first feature pair refers to any two structural features whose feature similarity is greater than or equal to a first preset similarity threshold, and the second feature pair refers to any two structural features whose feature similarity is less than or equal to a second preset similarity threshold, wherein the first preset similarity threshold is greater than the second preset similarity threshold. The number of features in the first feature pair and the second feature pair can be multiple.

[0103] In at least one embodiment of the present invention, the electronic device selects the first feature pair and the second feature pair based on the feature similarity of any two structured features among the plurality of structured features, including:

[0104] Calculating the feature similarity based on the first vector and the second vector generated by constructing any two structured features;

[0105] Determine any two structural features whose feature similarity is greater than or equal to a first preset similarity threshold as the first feature pair;

[0106] Any two structural features whose feature similarity is less than or equal to a second preset similarity threshold are determined as the second feature pair, and the first preset similarity threshold is greater than the second preset similarity threshold.

[0107] Specifically, the calculation formula of the feature similarity is: Among them, S A,B Indicates the feature similarity, A i and B i Indicates the vector representation of the i-th dimension in the first vector and the second vector respectively.

[0108] Through the above implementation, the first feature pair and the second feature pair can be quickly screened out.

[0109] 104, identifying the feature importance of each structured feature based on the pre-trained classification model.

[0110] In at least one embodiment of the present invention, the classification model can be used to identify whether the user to be tested is a potential user. The present invention does not limit the specific model result of the classification model.

[0111] The feature importance refers to the contribution of each structural feature to the recognition of the classification model.

[0112] In at least one embodiment of the present invention, the electronic device identifies the feature importance of each structured feature based on the pre-trained classification model, including:

[0113] Performing an amplification process on the classification model to obtain multiple processing models;

[0114] Calculating a ratio of the number of features of the plurality of structured features to the number of models of the plurality of processing models;

[0115] Selecting features from the plurality of structured features based on the ratio and inputting them into the plurality of processing models for parallel processing respectively, so as to obtain an output category of each structured feature by the classification model;

[0116] Identifying the prediction accuracy of the classification model for each structured feature based on the comparison results of the labeled categories of the plurality of structured features and the output category;

[0117] The prediction accuracy is normalized to obtain the feature importance of each structured feature.

[0118] The number of structured features input by the electronic device to each processing model for category recognition is equal to the ratio.

[0119] The annotation category refers to the category that is pre-annotated for the multiple structured features. It can be understood that the annotation categories corresponding to the multiple structured features corresponding to the same user are the same.

[0120] By amplifying the classification model, the efficiency of the multiple processing models in generating the output categories can be improved, thereby improving the recognition efficiency of the feature importance. Further, by comparing the labeled categories of the multiple structured features with the output categories, the prediction accuracy of the classification model for each structured feature can be accurately quantified, thereby improving the accuracy of the feature importance.

[0121] 105 , fusing the first feature pairs according to the feature importance to obtain a fused feature, merging the second feature pairs to obtain a merged feature, and generating a plurality of target features according to the fused feature and the merged feature.

[0122] In at least one embodiment of the present invention, the fused feature refers to a feature generated after the first feature pair is fused, and the number of the fused features may be multiple, and the similarity between each fused feature is greater than the second preset similarity threshold and less than the first preset similarity threshold. The merged feature refers to a feature generated after the second feature pair is merged, and the number of the merged features may be multiple, and the similarity between each merged feature is greater than the second preset similarity threshold and less than the first preset similarity threshold. The multiple target features include the fused feature and the merged feature.

[0123] In at least one embodiment of the present invention, the electronic device fuses the first feature pair according to the feature importance to obtain a fused feature, merges the second feature pair to obtain a merged feature, and generates multiple target features according to the fused feature and the merged feature, including:

[0124] Calculating a first importance ratio of a first left feature in the first feature pair based on the feature importance, and calculating a second importance ratio of a first right feature in the first feature pair;

[0125] Based on the first importance ratio and the second importance ratio, a weighted sum calculation is performed on the first left feature and the first right feature to obtain the fused feature;

[0126] Performing mutually exclusive merging processing on the second left feature in the second feature pair and the second right feature in the second feature pair to obtain the merged feature;

[0127] The fused features and the combined features are determined as the multiple target features, and the feature similarity of any two target features among the multiple target features is greater than the second preset similarity threshold and less than the first preset similarity threshold.

[0128] The first importance ratio and the second importance ratio can be used to reasonably fuse the first left feature and the first right feature, thereby improving the accuracy of the fused feature; and by mutually exclusive merging the second left feature and the second right feature, it can be avoided that the merged feature cannot be accurately generated due to feature merging, thereby improving the accuracy of the multiple target features; at the same time, by setting the feature similarity of any two target features among the multiple target features to be greater than the second preset similarity threshold and less than the first preset similarity threshold, it can achieve comprehensive fusion of the multiple structured features and improve the adequacy of feature fusion.

[0129] Specifically, the generation formula of the fusion feature is: Among them, X ab represents the fusion feature, ω a represents the feature importance of the first left feature, ω b represents the feature importance of the first right feature, X a represents the first left feature, X b Represents the first right feature.

[0130] Specifically, the electronic device performs mutually exclusive merging processing on the second left feature in the second feature pair and the second right feature in the second feature pair, and the merged feature obtained includes:

[0131] For the second left feature, identifying any feature associated with the second left feature from the second right feature as a primary screening feature, and constructing a feature cluster based on the second left feature and the primary screening feature;

[0132] Identify a feature that is associated with both the second left feature and the primary screening feature from the second right feature as an intermediate feature;

[0133] Counting the number of times that the intermediate feature and the feature values ​​in the feature cluster are different from each other and are the configuration values ​​as the number of contradictions between the intermediate feature and the feature cluster;

[0134] If the number of contradictions is less than a preset contradiction threshold, merging the intermediate features into the feature cluster;

[0135] Each merged feature cluster is fused based on the feature importance to obtain the merged feature.

[0136] Among them, the initial screening feature refers to any feature in the second right feature that is associated with the second left feature. When it is detected that the feature value corresponding to any feature in the second right feature and the feature value corresponding to the second left feature are not the configuration values ​​at the same time, it is determined that the any feature has an associated relationship with the second left feature.

[0137] The preset conflict threshold can be set according to actual needs.

[0138] By identifying any feature associated with the second left feature from the second right feature as a primary screening feature, the feature cluster can be quickly constructed, and then by detecting the number of contradictions between the intermediate feature and the feature cluster, the relationship between the intermediate feature and the feature cluster can be accurately identified, thereby improving the generation accuracy of the merged feature.

[0139] 106. Perform recognition processing on the multiple target features based on the classification model to obtain a classification result of the user to be tested.

[0140] It should be emphasized that in order to further ensure the privacy and security of the above classification results, the above classification results can also be stored in a blockchain node.

[0141] In at least one embodiment of the present invention, the classification result includes, but is not limited to: the user to be tested is a potential customer, the user to be tested is not a potential customer, and the like.

[0142] In at least one embodiment of the present invention, the classification model includes a feature processing layer and an output layer, and the electronic device performs recognition processing on the multiple target features based on the classification model to obtain a classification result of the user to be tested, including:

[0143] Processing the multiple target features based on the feature processing layer to obtain a user score;

[0144] Comparing the user rating with a preset score in the output layer;

[0145] If the user score is greater than or equal to the preset score, the classification result is determined as the user to be tested being a potential customer; or

[0146] If the user score is less than the preset score, the classification result is determined as the user to be tested is not a potential customer.

[0147] Through the above implementation, since the multiple target features are features generated by fully mining and fusing the multiple structured features, the analysis of the classification result based on the multiple target features can improve the accuracy of the classification result.

[0148] It can be seen from the above technical solutions that the present invention can not only eliminate the influence of interference data on features by preprocessing the user information, but also convert the unstructured information in the user information into structured features, thereby improving the analysis efficiency. Furthermore, the classification model can accurately identify the feature importance, thereby improving the fusion accuracy of the first feature pair. At the same time, the present invention can effectively realize the fusion of the first feature pair and the second feature pair separately by combining the feature importance with the first feature pair and merging the second feature pair, thereby improving the feature validity of the multiple target features, thereby improving the accuracy of the classification result.

[0149] like Figure 2 , which is a functional module diagram of a preferred embodiment of a user classification device based on multivariate data fusion of the present invention. The user classification device based on multivariate data fusion 11 includes a collection unit 110, a preprocessing unit 111, a selection unit 112, a recognition unit 113 and a generation unit 114. The module / unit referred to in the present invention refers to a series of computer-readable instruction segments that can be acquired by the processor 13 and can perform fixed functions, which are stored in the memory 12. In this embodiment, the functions of each module / unit will be described in detail in subsequent embodiments.

[0150] The collecting unit 110 collects user information of the user to be tested.

[0151] In at least one embodiment of the present invention, the users to be tested may be a plurality of consumer users mined from a big data platform.

[0152] The user information includes data information of the user to be tested in multiple different dimensions. For example, the user information may include, but is not limited to: user basic attributes (such as age, gender, education level, occupation, etc.), user static behavior data (such as user's product ownership), user dynamic behavior data (such as APP click status, etc.) and voice communication data, text communication data and other information in different dimensions.

[0153] In at least one embodiment of the present invention, the collecting unit 110 collects relevant information about the user to be tested from multiple collecting sources as the user information.

[0154] The preprocessing unit 111 preprocesses the user information to obtain a plurality of structured features.

[0155] In at least one embodiment of the present invention, the multiple structured features include features obtained after data cleaning of structured information, and the multiple structured features also include features obtained after statistical feature, time feature and trend feature construction analysis of unstructured information.

[0156] In at least one embodiment of the present invention, the preprocessing unit 111 preprocesses the user information to obtain a plurality of structured features including:

[0157] Identifying structured information in the user information based on a collection source of the user information;

[0158] Building a user portrait wide table based on the structured information;

[0159] Identify abnormal data of field information corresponding to each wide table field in the user portrait wide table;

[0160] Filtering out target users corresponding to the abnormal data from the users to be tested;

[0161] Detecting whether the target user is an abnormal user based on the abnormal data;

[0162] If the target user is not the abnormal user, the abnormal data corresponding to the target user is replaced to obtain the multiple structured features.

[0163] The collection sources may include relational databases, non-relational databases and other platforms.

[0164] The structured information generally refers to information extracted from the relational database.

[0165] The user portrait wide table refers to a data table obtained by associating dimension data stored in different data tables with the user to be tested as the primary key. The user portrait wide table stores user information of multiple users to be tested.

[0166] The abnormal data refers to the field information in each wide table field whose deviation from the field information is greater than a preset deviation. The preset deviation can be determined according to actual needs.

[0167] The abnormal user refers to a user containing the abnormal data.

[0168] By constructing the user portrait wide table, the abnormal data can be accurately identified, thereby improving the accuracy of determining the target user. Furthermore, when it is detected that the target user is not the abnormal user, the influence of the abnormal data of the target user on the multiple structured features can be avoided by replacing the abnormal data.

[0169] Specifically, the preprocessing unit 111 performs replacement processing on the abnormal data corresponding to the target user, and obtains the multiple structured features including:

[0170] If the data type of the abnormal data corresponding to the target user is character type, then obtaining the mode information in the wide table field where the abnormal data is located from the user portrait wide table, and replacing the abnormal data based on the mode information to obtain the multiple structured features; or

[0171] If the data type of the abnormal data corresponding to the target user is a numerical type, the abnormal data is replaced based on the configuration value to obtain the multiple structured features.

[0172] The configuration value may be 0, or may be the average value in the wide table field where the abnormal data is located.

[0173] In other embodiments, if the target user is the abnormal user, the target user is removed from the users to be tested, and the identification of the target user is terminated.

[0174] In at least one embodiment of the present invention, the preprocessing unit 111 preprocesses the user information to obtain a plurality of structured features, further comprising:

[0175] Identifying unstructured information in the user information based on a collection source of the user information;

[0176] Performing data cleaning on the unstructured information based on preset characters to obtain text information;

[0177] Extracting text sequence and time sequence from the text information;

[0178] The text sequence is subjected to text analysis based on preset features, and the time series is subjected to time analysis based on the preset features to obtain the multiple structural features.

[0179] The unstructured information may be information obtained from the non-relational database.

[0180] The preset characters may include, but are not limited to: punctuation marks, emoticons, non-Chinese characters, modal particles, etc.

[0181] The text sequence refers to the information related to communication in the text information, and the time sequence refers to the information related to time in the text information.

[0182] The preset features can be set according to actual needs. The preset features include statistical dimension features, time dimension features and trend dimension features. For example, the preset features for text analysis of the text sequence may include the total number of text words, the ratio of the total number of text words to the total number of expressions, the number of appearances of marketing words, and the marketing words with the highest frequency in the statistical dimension features; the preset features for text analysis of the text sequence may include the marketing words with the highest number of expressions in the most recent and most distant times, and the marketing words with the lowest number of expressions in the most recent and most distant times in the time dimension features; the preset features for text analysis of the text sequence may include the ratio of the number of times a marketing word is expressed in the last n days to the number of times a marketing word is expressed in the last m days in the trend dimension features. The preset features for time analysis of the text sequence may include the average communication interval duration, the maximum communication time interval, and the minimum communication time interval in the statistical dimension features; the preset features for time analysis of the text sequence may include whether the most recent communication time is in the morning, afternoon, or evening, whether the most recent communication time is on the weekend, whether the farthest communication time is in the morning, afternoon, or evening, and whether the farthest communication time is on the weekend in the time dimension features; the preset features for time analysis of the text sequence may include the average communication interval duration of the last i times and the average communication interval duration of the last j times in the trend dimension features, etc.

[0183] By cleaning the preset characters in the unstructured information, the amount of information in the text information can be reduced, thereby improving the generation efficiency of the multiple structured features. Furthermore, by extracting the text sequence and the time series, the text information can be analyzed in combination with the text dimension and the time dimension, thereby obtaining comprehensive structured features.

[0184] The selection unit 112 selects a first feature pair and a second feature pair based on feature similarity between any two structured features among the plurality of structured features.

[0185] In at least one embodiment of the present invention, the first feature pair refers to any two structural features whose feature similarity is greater than or equal to a first preset similarity threshold, and the second feature pair refers to any two structural features whose feature similarity is less than or equal to a second preset similarity threshold, wherein the first preset similarity threshold is greater than the second preset similarity threshold. The number of features in the first feature pair and the second feature pair can be multiple.

[0186] In at least one embodiment of the present invention, the selecting unit 112 selects the first feature pair and the second feature pair based on the feature similarity of any two structured features among the plurality of structured features, including:

[0187] Calculating the feature similarity based on the first vector and the second vector generated by constructing any two structured features;

[0188] Determine any two structural features whose feature similarity is greater than or equal to a first preset similarity threshold as the first feature pair;

[0189] Any two structural features whose feature similarity is less than or equal to a second preset similarity threshold are determined as the second feature pair, and the first preset similarity threshold is greater than the second preset similarity threshold.

[0190] Specifically, the calculation formula of the feature similarity is: Among them, S A,B Indicates the feature similarity, A i and B i Indicates the vector representation of the i-th dimension in the first vector and the second vector respectively.

[0191] Through the above implementation, the first feature pair and the second feature pair can be quickly screened out.

[0192] The identification unit 113 identifies the feature importance of each structural feature based on a classification model that has been trained in advance.

[0193] In at least one embodiment of the present invention, the classification model can be used to identify whether the user to be tested is a potential user. The present invention does not limit the specific model result of the classification model.

[0194] The feature importance refers to the contribution of each structural feature to the recognition of the classification model.

[0195] In at least one embodiment of the present invention, the identification unit 113 identifies the feature importance of each structural feature based on the pre-trained classification model, including:

[0196] Performing an amplification process on the classification model to obtain multiple processing models;

[0197] Calculating a ratio of the number of features of the plurality of structured features to the number of models of the plurality of processing models;

[0198] Selecting features from the plurality of structured features based on the ratio and inputting them into the plurality of processing models for parallel processing respectively, so as to obtain an output category of each structured feature by the classification model;

[0199] Identifying the prediction accuracy of the classification model for each structured feature based on the comparison results of the labeled categories of the plurality of structured features and the output category;

[0200] The prediction accuracy is normalized to obtain the feature importance of each structured feature.

[0201] The number of structured features input by the recognition unit 113 to each processing model for category recognition is equal to the ratio.

[0202] The annotation category refers to the category that is pre-annotated for the multiple structured features. It can be understood that the annotation categories corresponding to the multiple structured features corresponding to the same user are the same.

[0203] By amplifying the classification model, the efficiency of the multiple processing models in generating the output categories can be improved, thereby improving the recognition efficiency of the feature importance. Further, by comparing the labeled categories of the multiple structured features with the output categories, the prediction accuracy of the classification model for each structured feature can be accurately quantified, thereby improving the accuracy of the feature importance.

[0204] The generating unit 114 fuses the first feature pairs according to the feature importance to obtain a fused feature, merges the second feature pairs to obtain a merged feature, and generates a plurality of target features according to the fused feature and the merged feature.

[0205] In at least one embodiment of the present invention, the fused feature refers to a feature generated after the first feature pair is fused, and the number of the fused features may be multiple, and the similarity between each fused feature is greater than the second preset similarity threshold and less than the first preset similarity threshold. The merged feature refers to a feature generated after the second feature pair is merged, and the number of the merged features may be multiple, and the similarity between each merged feature is greater than the second preset similarity threshold and less than the first preset similarity threshold. The multiple target features include the fused feature and the merged feature.

[0206] In at least one embodiment of the present invention, the generating unit 114 fuses the first feature pair according to the feature importance to obtain a fused feature, merges the second feature pair to obtain a merged feature, and generates multiple target features according to the fused feature and the merged feature, including:

[0207] Calculating a first importance ratio of a first left feature in the first feature pair based on the feature importance, and calculating a second importance ratio of a first right feature in the first feature pair;

[0208] Based on the first importance ratio and the second importance ratio, a weighted sum calculation is performed on the first left feature and the first right feature to obtain the fused feature;

[0209] Performing mutually exclusive merging processing on the second left feature in the second feature pair and the second right feature in the second feature pair to obtain the merged feature;

[0210] The fused features and the combined features are determined as the multiple target features, and the feature similarity of any two target features among the multiple target features is greater than the second preset similarity threshold and less than the first preset similarity threshold.

[0211] The first importance ratio and the second importance ratio can be used to reasonably fuse the first left feature and the first right feature, thereby improving the accuracy of the fused feature; and by mutually exclusive merging the second left feature and the second right feature, it can be avoided that the merged feature cannot be accurately generated due to feature merging, thereby improving the accuracy of the multiple target features; at the same time, by setting the feature similarity of any two target features among the multiple target features to be greater than the second preset similarity threshold and less than the first preset similarity threshold, it can achieve comprehensive fusion of the multiple structured features and improve the adequacy of feature fusion.

[0212] Specifically, the generation formula of the fusion feature is: Among them, X ab represents the fusion feature, ω a represents the feature importance of the first left feature, ω b represents the feature importance of the first right feature, X a represents the first left feature, X b Represents the first right feature.

[0213] Specifically, the generating unit 114 performs mutually exclusive merging processing on the second left feature in the second feature pair and the second right feature in the second feature pair, and the obtained merged feature includes:

[0214] For the second left feature, identifying any feature associated with the second left feature from the second right feature as a primary screening feature, and constructing a feature cluster based on the second left feature and the primary screening feature;

[0215] Identify a feature that is associated with both the second left feature and the primary screening feature from the second right feature as an intermediate feature;

[0216] Counting the number of times that the intermediate feature and the feature values ​​in the feature cluster are different from each other and are the configuration values ​​as the number of contradictions between the intermediate feature and the feature cluster;

[0217] If the number of contradictions is less than a preset contradiction threshold, merging the intermediate features into the feature cluster;

[0218] Each merged feature cluster is fused based on the feature importance to obtain the merged feature.

[0219] Among them, the initial screening feature refers to any feature in the second right feature that is associated with the second left feature. When it is detected that the feature value corresponding to any feature in the second right feature and the feature value corresponding to the second left feature are not the configuration values ​​at the same time, it is determined that the any feature has an associated relationship with the second left feature.

[0220] The preset conflict threshold can be set according to actual needs.

[0221] By identifying any feature associated with the second left feature from the second right feature as a primary screening feature, the feature cluster can be quickly constructed, and then by detecting the number of contradictions between the intermediate feature and the feature cluster, the relationship between the intermediate feature and the feature cluster can be accurately identified, thereby improving the generation accuracy of the merged feature.

[0222] The recognition unit 113 performs recognition processing on the multiple target features based on the classification model to obtain a classification result of the user to be tested.

[0223] It should be emphasized that in order to further ensure the privacy and security of the above classification results, the above classification results can also be stored in a blockchain node.

[0224] In at least one embodiment of the present invention, the classification result includes, but is not limited to: the user to be tested is a potential customer, the user to be tested is not a potential customer, and the like.

[0225] In at least one embodiment of the present invention, the classification model includes a feature processing layer and an output layer, and the recognition unit 113 performs recognition processing on the multiple target features based on the classification model to obtain the classification result of the user to be tested, including:

[0226] Processing the multiple target features based on the feature processing layer to obtain a user score;

[0227] Comparing the user rating with a preset score in the output layer;

[0228] If the user score is greater than or equal to the preset score, the classification result is determined as the user to be tested being a potential customer; or

[0229] If the user score is less than the preset score, the classification result is determined as the user to be tested is not a potential customer.

[0230] Through the above implementation, since the multiple target features are features generated by fully mining and fusing the multiple structured features, the analysis of the classification result based on the multiple target features can improve the accuracy of the classification result.

[0231] It can be seen from the above technical solutions that the present invention can not only eliminate the influence of interference data on features by preprocessing the user information, but also convert the unstructured information in the user information into structured features, thereby improving the analysis efficiency. Furthermore, the classification model can accurately identify the feature importance, thereby improving the fusion accuracy of the first feature pair. At the same time, the present invention can effectively realize the fusion of the first feature pair and the second feature pair separately by combining the feature importance with the first feature pair and merging the second feature pair, thereby improving the feature validity of the multiple target features, thereby improving the accuracy of the classification result.

[0232] like Figure 3 , which is a schematic diagram of the structure of an electronic device of a preferred embodiment of the user classification method based on multivariate data fusion of the present invention.

[0233] In one embodiment of the present invention, the electronic device 1 includes, but is not limited to, a memory 12, a processor 13, and computer-readable instructions stored in the memory 12 and executable on the processor 13, such as a user classification program based on multivariate data fusion.

[0234] Those skilled in the art will understand that the schematic diagram is merely an example of the electronic device 1 and does not constitute a limitation on the electronic device 1. The electronic device 1 may include more or fewer components than shown in the diagram, or a combination of certain components, or different components. For example, the electronic device 1 may also include input and output devices, network access devices, buses, etc.

[0235] The processor 13 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor 13 is the computing core and control center of the electronic device 1, and uses various interfaces and lines to connect various parts of the entire electronic device 1, and executes the operating system of the electronic device 1 and various installed applications, program codes, etc.

[0236] Exemplarily, the computer-readable instructions may be divided into one or more modules / units, which are stored in the memory 12 and executed by the processor 13 to complete the present invention. The one or more modules / units may be a series of computer-readable instruction segments capable of completing specific functions, which are used to describe the execution process of the computer-readable instructions in the electronic device 1. For example, the computer-readable instructions may be divided into a collection unit 110, a preprocessing unit 111, a selection unit 112, a recognition unit 113, and a generation unit 114.

[0237] The memory 12 can be used to store the computer-readable instructions and / or modules, and the processor 13 implements various functions of the electronic device 1 by running or executing the computer-readable instructions and / or modules stored in the memory 12, and calling the data stored in the memory 12. The memory 12 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the electronic device, etc. The memory 12 can include non-volatile and volatile memories, such as: a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other storage devices.

[0238] The memory 12 may be an external memory and / or an internal memory of the electronic device 1. Furthermore, the memory 12 may be a memory in a physical form, such as a memory stick, a TF card (Trans-flash Card), and the like.

[0239] If the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also instruct the relevant hardware to complete it through computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage medium. When the computer-readable instructions are executed by the processor, the steps of the above-mentioned various method embodiments can be implemented.

[0240] The computer-readable instructions include computer-readable instruction codes, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer-readable instruction code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM).

[0241] The blockchain referred to in the present invention is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the blockchain underlying platform, platform product service layer, and application service layer.

[0242] Combination Figure 1 The memory 12 in the electronic device 1 stores computer-readable instructions to implement a user classification method based on multivariate data fusion, and the processor 13 can execute the computer-readable instructions to implement:

[0243] Collect user information of the user to be tested;

[0244] Preprocessing the user information to obtain multiple structured features;

[0245] Selecting a first feature pair and a second feature pair based on feature similarity between any two structured features among the plurality of structured features;

[0246] Identify the feature importance of each structural feature based on the pre-trained classification model;

[0247] According to the feature importance, the first feature pair is fused to obtain a fused feature, the second feature pair is fused to obtain a combined feature, and a plurality of target features are generated according to the fused feature and the combined feature;

[0248] The multiple target features are identified and processed based on the classification model to obtain a classification result of the user to be tested.

[0249] Specifically, the specific implementation method of the processor 13 for the above-mentioned computer readable instructions can refer to Figure 1 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0250] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0251] The computer-readable storage medium stores computer-readable instructions, wherein the computer-readable instructions are used to implement the following steps when executed by the processor 13:

[0252] Collect user information of the user to be tested;

[0253] Preprocessing the user information to obtain multiple structured features;

[0254] Selecting a first feature pair and a second feature pair based on feature similarity between any two structured features among the plurality of structured features;

[0255] Identify the feature importance of each structural feature based on the pre-trained classification model;

[0256] According to the feature importance, the first feature pair is fused to obtain a fused feature, the second feature pair is fused to obtain a combined feature, and a plurality of target features are generated according to the fused feature and the combined feature;

[0257] The multiple target features are identified and processed based on the classification model to obtain a classification result of the user to be tested.

[0258] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0259] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0260] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0261] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or devices mentioned can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0262] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A user classification method based on multivariate data fusion, characterized in that: The user classification method based on multivariate data fusion includes: Collect user information of the user to be tested; Preprocessing the user information to obtain a plurality of structured features includes: identifying unstructured information in the user information based on a source of collection of the user information; performing data cleaning on the unstructured information based on preset characters to obtain text information; extracting a text sequence and a time sequence from the text information; performing text analysis on the text sequence based on preset features, and performing time analysis on the time sequence based on the preset features to obtain the plurality of structured features; Selecting a first feature pair and a second feature pair based on feature similarity between any two structured features among the plurality of structured features; Identify the feature importance of each structural feature based on the pre-trained classification model; The first feature pair is fused according to the feature importance to obtain a fused feature, the second feature pair is merged to obtain a merged feature, and a plurality of target features are generated according to the fused feature and the merged feature, including: calculating the first importance ratio of the first left feature in the first feature pair in the first feature pair based on the feature importance, and calculating the second importance ratio of the first right feature in the first feature pair in the first feature pair; performing a weighted sum calculation on the first left feature and the first right feature based on the first importance ratio and the second importance ratio to obtain the fused feature; performing mutually exclusive merging processing on the second left feature in the second feature pair and the second right feature in the second feature pair to obtain the merged feature; determining the fused feature and the merged feature as the plurality of target features, wherein the feature similarity of any two target features in the plurality of target features is greater than the second preset similarity threshold and less than the first preset similarity threshold; The multiple target features are identified and processed based on the classification model to obtain a classification result of the user to be tested.

2. The user classification method based on multivariate data fusion as claimed in claim 1, characterized in that: The preprocessing of the user information to obtain a plurality of structured features includes: Identifying structured information in the user information based on a collection source of the user information; Building a user portrait wide table based on the structured information; Identify abnormal data of field information corresponding to each wide table field in the user portrait wide table; Filtering out target users corresponding to the abnormal data from the users to be tested; Detecting whether the target user is an abnormal user based on the abnormal data; If the target user is not the abnormal user, the abnormal data corresponding to the target user is replaced to obtain the multiple structured features.

3. The user classification method based on multivariate data fusion as claimed in claim 1, characterized in that: The selecting the first feature pair and the second feature pair based on the feature similarity of any two structured features among the plurality of structured features comprises: Calculating the feature similarity based on the first vector and the second vector generated by constructing any two structured features; Determine any two structural features whose feature similarity is greater than or equal to a first preset similarity threshold as the first feature pair; Any two structural features whose feature similarity is less than or equal to a second preset similarity threshold are determined as the second feature pair, and the first preset similarity threshold is greater than the second preset similarity threshold.

4. The user classification method based on multivariate data fusion as claimed in claim 1, characterized in that: The feature importance of each structural feature identified based on the pre-trained classification model includes: Performing an amplification process on the classification model to obtain multiple processing models; Calculating a ratio of the number of features of the plurality of structured features to the number of models of the plurality of processing models; Selecting features from the plurality of structured features based on the ratio and inputting them into the plurality of processing models for parallel processing respectively, so as to obtain an output category of each structured feature by the classification model; Identifying the prediction accuracy of the classification model for each structured feature based on the comparison results of the labeled categories of the plurality of structured features and the output category; The prediction accuracy is normalized to obtain the feature importance of each structured feature.

5. The user classification method based on multivariate data fusion as claimed in claim 1, characterized in that: The mutually exclusive merging process is performed on the second left feature in the second feature pair and the second right feature in the second feature pair to obtain the merged feature, which includes: For the second left feature, identifying any feature associated with the second left feature from the second right feature as a primary screening feature, and constructing a feature cluster based on the second left feature and the primary screening feature; Identify a feature that is associated with both the second left feature and the primary screening feature from the second right feature as an intermediate feature; Counting the number of configuration values ​​that are different between the intermediate feature and the feature values ​​in the feature cluster as the number of contradictions between the intermediate feature and the feature cluster; If the number of contradictions is less than a preset contradiction threshold, merging the intermediate features into the feature cluster; Each merged feature cluster is fused based on the feature importance to obtain the merged feature.

6. A user classification device based on multivariate data fusion, characterized in that: The user classification device based on multivariate data fusion comprises: A collection unit, used to collect user information of the user to be tested; A preprocessing unit is used to preprocess the user information to obtain a plurality of structured features, including: identifying unstructured information in the user information based on the collection source of the user information; performing data cleaning on the unstructured information based on preset characters to obtain text information; extracting a text sequence and a time sequence from the text information; performing text analysis on the text sequence based on preset features, and performing time analysis on the time sequence based on the preset features to obtain the plurality of structured features; A selection unit, configured to select a first feature pair and a second feature pair based on feature similarity between any two structured features among the plurality of structured features; A recognition unit, used to identify the feature importance of each structural feature based on a pre-trained classification model; A generating unit is used to fuse the first feature pair according to the feature importance to obtain a fused feature, merge the second feature pair to obtain a merged feature, and generate multiple target features according to the fused feature and the merged feature, including: calculating the first importance ratio of the first left feature in the first feature pair in the first feature pair based on the feature importance, and calculating the second importance ratio of the first right feature in the first feature pair in the first feature pair; performing weighted sum calculation on the first left feature and the first right feature based on the first importance ratio and the second importance ratio to obtain the fused feature; performing mutually exclusive merging processing on the second left feature in the second feature pair and the second right feature in the second feature pair to obtain the merged feature; determining the fused feature and the merged feature as the multiple target features, and the feature similarity of any two target features in the multiple target features is greater than the second preset similarity threshold and less than the first preset similarity threshold; The recognition unit is further used to perform recognition processing on the multiple target features based on the classification model to obtain a classification result of the user to be tested.

7. An electronic device, characterized in that: The electronic device comprises: a memory storing computer-readable instructions; and A processor executes the computer-readable instructions stored in the memory to implement the user classification method based on multivariate data fusion as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions are executed by a processor in an electronic device to implement the user classification method based on multivariate data fusion as described in any one of claims 1 to 5.

Images