Operation Data Processing Method, Device, Electronic Device and Storage Medium

Abstract

The present disclosure provides an operation data processing method, which can be applied to the field of big data technology or the financial field. The operation data processing method includes: obtaining target operation data generated by a target user based on a transaction client within a first preset time period; inputting the target operation data into a pre-trained keyword extraction model to output target keywords corresponding to the target operation data; inputting the target keywords into a vectorization model to determine a target clustering cluster corresponding to the target keywords among at least two clustering clusters in the vectorization model; and determining vectorized data corresponding to the target operation data according to the target clustering cluster and a pre-designed calculation method. The present disclosure also provides an operation data processing device, equipment, and storage medium.

Description

Technical Field

[0001] The present disclosure relates to the field of big data technology or the financial field, and more particularly to an operation data processing method, apparatus, electronic device, and storage medium. Background Art

[0002] In a financial system, it is generally necessary to classify users in the financial system according to different characteristics, so as to be able to better provide targeted services for each category of users. Currently, before classifying users, it is necessary to vectorize the operation data of users. Common vectorization methods include directly using methods such as word2vec and tf-idf for vectorization after preprocessing the obtained operation data.

[0003] In the process of implementing the inventive concept of the present disclosure, the inventors found that there are at least the following problems in the related art: The accuracy of vectorizing operation data using the above vectorization methods is relatively low, which affects the classification results of users, and thus it is difficult to recommend financial products to users in a targeted manner. Summary of the Invention

[0004] In view of the above problems, the present disclosure provides an operation data processing method, apparatus, device, medium, and program product.

[0005] According to one aspect of the present disclosure, there is provided an operation data processing method, including:

[0006] Obtaining target operation data generated by a target user based on a trading client within a first preset time period;

[0007] Inputting the above target operation data into a pre-trained keyword extraction model, and outputting target keywords corresponding to the above target operation data;

[0008] Inputting the above target keywords into a vectorization model, and determining a target clustering cluster corresponding to the above target keywords among at least two clustering clusters in the above vectorization model; and

[0009] Determining vectorized data corresponding to the above target operation data according to the above target clustering cluster and a pre-designed calculation method.

[0010] According to an embodiment of the present disclosure, the determining vectorized data corresponding to the above target operation data according to the above target clustering cluster and a pre-designed calculation method includes:

[0011] Obtaining vectorized data of a target cluster center in the above target clustering cluster;

[0012] Determining the distance between the above target keyword and the above target cluster center to obtain a first distance;

[0013] Determine the distance between the above target keyword and the non-target cluster center to obtain a second distance, where the above non-target cluster center includes the cluster centers of the above at least two clustering clusters except the above target cluster center;

[0014] Determine the vectorized data corresponding to the above target operation data according to the vectorized data of the above target cluster center, the above first distance, the above second distance, and the number of clustering clusters.

[0015] According to an embodiment of the present disclosure, the above operation data processing method further includes,

[0016] Before inputting the above target operation data into a pre-trained keyword extraction model and outputting the target keyword corresponding to the above target operation data, preprocess the above target operation data to obtain the preprocessed target operation data;

[0017] Wherein, inputting the above target operation data into a pre-trained keyword extraction model and outputting the target keyword corresponding to the above target operation data includes:

[0018] Input the preprocessed target operation data into a pre-trained keyword extraction model and output the target keyword corresponding to the preprocessed target operation data.

[0019] According to an embodiment of the present disclosure, preprocessing the above target operation data to obtain the preprocessed target operation data includes:

[0020] Perform stop word removal, invalid word removal, and invalid English removal on the above target operation data to obtain the preprocessed target operation data.

[0021] According to an embodiment of the present disclosure, the above keyword extraction model is trained by the following method:

[0022] Obtain historical operation data generated by users in the financial system within a second preset time period to obtain first sample data;

[0023] Preprocess the above first sample data to obtain the preprocessed first sample data;

[0024] Label the preprocessed first sample data using a preset keyword set to obtain a training data set; and

[0025] Train the keyword extraction model to be trained using the above training data set to obtain the above keyword extraction model.

[0026] According to an embodiment of the present disclosure, the above vectorization model is trained by the following method:

[0027] Obtain the historical operation data generated by m users in the financial system during the third preset time period to obtain the second sample data, where m≥2;

[0028] Preprocess the above-mentioned second sample data to obtain the processed second sample data;

[0029] For each of the above m users, input the processed second sample data corresponding to the user into the above keyword extraction model, output the keywords corresponding to the user, and finally obtain m groups of keywords;

[0030] Cluster the above m groups of keywords to obtain n clustering clusters, where 2≤n≤m;

[0031] For each of the above m users, determine the vectorized data corresponding to the historical operation data of the user according to the above n clustering clusters, and finally obtain m vectorized data; and

[0032] Determine the above vectorization model according to the above n clustering clusters and the above m vectorized data.

[0033] According to an embodiment of the present disclosure, the keyword located at the cluster center in the above clustering cluster represents the historical operation data of one of the above m users;

[0034] For each of the above m users, determining the vectorized data corresponding to the historical operation data of the user according to the above n clustering clusters, and finally obtaining m vectorized data includes:

[0035] For each of the above n clustering clusters, vectorize the keyword located at the cluster center in the i-th clustering cluster to obtain the i-th cluster center vectorized data, where 1≤i≤n;

[0036] Use the above i-th cluster center vectorized data and a pre-designed calculation method to determine the non-cluster center vectorized data corresponding to the keyword located at the non-cluster center in the i-th clustering cluster, where the keyword located at the non-cluster center includes the keyword other than the keyword located at the cluster center in the i-th clustering cluster, and finally obtain n cluster center vectorized data and m-n non-cluster center vectorized data;

[0037] Determine the above m vectorized data according to the above n cluster center vectorized data and the above m-n non-cluster center vectorized data.

[0038] According to an embodiment of the present disclosure, using the above i-th cluster center vectorized data and a pre-designed calculation method to determine the non-cluster center vectorized data corresponding to the keyword located at the non-cluster center in the i-th clustering cluster:

[0039] Determine the distance between the keyword located at a non-cluster center in the i-th clustering cluster and the cluster center in the i-th clustering cluster to obtain a third distance;

[0040] Determine the distance between the keyword located at a non-cluster center in the i-th clustering cluster and other cluster centers to obtain a fourth distance, where the other cluster centers include the cluster centers in the n clustering clusters except for the cluster center in the i-th clustering cluster;

[0041] Determine the non-cluster center vectorization data corresponding to the keyword located at a non-cluster center in the i-th clustering cluster according to the cluster center vector data of the i-th clustering cluster, the third distance, the fourth distance, and the number of clustering clusters.

[0042] According to an embodiment of the present disclosure, the above operation data processing method further includes:

[0043] Obtain the transaction data generated by the target user based on the transaction client within the first preset time period;

[0044] Input the transaction data and the vectorization data into a product recommendation model, and output a financial product corresponding to the target user.

[0045] Another aspect of the present disclosure provides an operation data processing device, including:

[0046] A first acquisition module, configured to acquire target operation data generated by a target user based on a transaction client within a first preset time period;

[0047] A first input / output module, configured to input the target operation data into a pre-trained keyword extraction model and output a target keyword corresponding to the target operation data;

[0048] A first determination module, input the target keyword into a vectorization model, and determine a target clustering cluster corresponding to the target keyword in at least two clustering clusters in the vectorization model; and

[0049] A second determination module, configured to determine vectorization data corresponding to the target operation data according to the target clustering cluster and a pre-designed calculation method.

[0050] Another aspect of the present disclosure provides an electronic device, including: one or more processors; a memory, configured to store one or more programs, where when the one or more programs are executed by the one or more processors, the one or more processors execute the above operation data processing method.

[0051] Another aspect of the present disclosure further provides a computer-readable storage medium, on which executable instructions are stored, and when the instructions are executed by a processor, the processor executes the above operation data processing method.

[0052] Another aspect of the present disclosure also provides a computer program product, including a computer program which, when executed by a processor, implements the above-mentioned operation data processing method.

[0053] According to an embodiment of the present disclosure, by inputting target operation data generated by a target user on a transaction client into a pre-trained keyword extraction model, target keywords corresponding to the target operation data are obtained, and then the target keywords are input into a vectorization model to determine a target cluster corresponding to the target keywords in at least two clusters in the vectorization model; then, according to the target cluster and a pre-designed calculation method, vectorized data corresponding to the target operation data is determined. At least partially overcome the technical problem of low accuracy when vectorizing operation data in the prior art. It achieves the technical effect of using a model for keyword extraction and vectorization conversion, with a more refined and complex process, and improves the accuracy of vectorizing operation data on the basis of sacrificing the time cost of vectorization. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Through the following description of the embodiments of the present disclosure with reference to the drawings, the above content and other objects, features and advantages of the present disclosure will become clearer. In the drawings:

[0055] Figure 1 Schematically shows an application scenario diagram of an operation data processing method, device, equipment, medium and program product according to an embodiment of the present disclosure;

[0056] Figure 2 Schematically shows a flowchart of an operation data processing method according to an embodiment of the present disclosure;

[0057] Figure 3 Schematically shows a flowchart of an operation data processing method according to another embodiment of the present disclosure;

[0058] Figure 4 Schematically shows a schematic diagram of a residual network connection according to an embodiment of the present disclosure;

[0059] Figure 5 Schematically shows a schematic diagram of a transformer sub-module structure according to an embodiment of the present disclosure;

[0060] Figure 6 Schematically shows a schematic diagram of a multi-head attention mechanism according to an embodiment of the present disclosure;

[0061] Figure 7 Schematically shows a schematic diagram of a bert model according to an embodiment of the present disclosure;

[0062] Figure 8Schematically shows a schematic diagram of the input layer of the BERT model according to an embodiment of the present disclosure;

[0063] Figure 9 Schematically shows a flowchart of a method for training a vectorization model according to an example of the present disclosure;

[0064] Figure 10 Schematically shows a structural block diagram of an operation data processing device according to an embodiment of the present disclosure; and

[0065] Figure 11 Schematically shows a block diagram of an electronic device suitable for implementing an operation data processing method according to an embodiment of the present disclosure. Detailed implementation manners

[0066] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, obviously, one or more embodiments can also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present disclosure.

[0067] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising", etc. used herein indicate the presence of the described features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0068] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0069] In the case of using expressions such as "at least one of A, B, and C", generally, it should be interpreted according to the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include, but not be limited to, a system having only A, only B, only C, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.).

[0070] In the financial system, it is generally necessary to classify users in the financial system according to different characteristics, so as to provide targeted services to each category of users. At present, before classifying users, it is necessary to vectorize the user's operation data. The commonly used vectorization methods include pre-processing the obtained operation data and directly using word2vec and tf-idf methods for vectorization.

[0071] In the process of realizing the inventive concept disclosed herein, the inventors discovered that there are at least the following problems in the related technology: since the operation data generated by the user on the trading client has a poor correlation before and after, the accuracy of the existing vectorization method is low, and the vectorized vector cannot accurately express the user's operation behavior, which affects the classification results of the user, making it difficult to recommend financial products to the user in a targeted manner.

[0072] In view of this, the present disclosure aims at the above technical problems, by inputting the user's operation data into a pre-trained keyword extraction model to extract keywords, and then vectorizing the extracted keywords using a pre-trained vectorization model to obtain vectorized data corresponding to the operation data. This method uses a model to perform keyword extraction and vectorization conversion, and the process is more sophisticated and complex. It improves the accuracy of vectorization on the basis of sacrificing the time cost of vectorization, thereby overcoming the technical problem of low accuracy in the prior art.

[0073] Specifically, an embodiment of the present disclosure provides an operation data processing method, including: obtaining target operation data generated by a target user based on a transaction client within a first preset time period; inputting the above target operation data into a pre-trained keyword extraction model, and outputting a target keyword corresponding to the above target operation data; inputting the above target keyword into a vectorization model, and determining a target cluster corresponding to the above target keyword in at least two clusters in the above vectorization model; and determining the vectorized data corresponding to the above target operation data based on the above target cluster and a preset calculation method.

[0074] It should be noted that the operation data processing method and device provided in the embodiments of the present disclosure can be used in the field of big data technology or the financial field. The operation data processing method and device provided in the embodiments of the present disclosure can also be used in any field other than the field of big data technology and the financial field. The application field of the operation data processing method and device provided in the embodiments of the present disclosure is not limited.

[0075] In the technical solution of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0076] In the technical solution of the present disclosure, the processing of data, such as acquisition, collection, storage, use, processing, transmission, provision, disclosure, and application, complies with the provisions of relevant laws and regulations, takes necessary confidentiality measures, and does not violate public order and good customs.

[0077] Figure 1 FIG. schematically shows an application scenario diagram of an operation data processing method, apparatus, device, medium, and program product according to an embodiment of the present disclosure.

[0078] As Figure 1 shown, the application scenario 100 according to this embodiment may include a network, terminal devices, and a server. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0079] Users can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications may be installed on the terminal devices 101, 102, 103, such as financial applications, shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0080] The terminal devices 101, 102, 103 may be various electronic devices with a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, and desktop computers, etc.

[0081] The server 105 may be a server providing various services, such as a background management server that supports the websites browsed by users using the terminal devices 101, 102, 103 (only for example). The background management server may analyze and process data such as received user requests, and feedback the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0082] It should be noted that the operation data processing method provided by the embodiments of the present disclosure can generally be executed by the server 105. Correspondingly, the operation data processing device provided by the embodiments of the present disclosure can generally be disposed in the server 105. The operation data processing method provided by the embodiments of the present disclosure can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the terminal devices 101, 102, 103, and / or the server 105. Correspondingly, the operation data processing device provided by the embodiments of the present disclosure can also be disposed in a server or a server cluster different from the server 105 and capable of communicating with the terminal devices 101, 102, 103, and / or the server 105. Alternatively, the operation data processing method provided by the embodiments of the present disclosure can also be executed by the terminal devices 101, 102, or 103, or can also be executed by other terminal devices different from the terminal devices 101, 102, or 103. Correspondingly, the operation data processing device provided by the embodiments of the present disclosure can also be disposed in the terminal devices 101, 102, or 103, or can be disposed in other terminal devices different from the terminal devices 101, 102, or 103.

[0083] For example, the operation data can be originally stored in any one of the terminal devices 101, 102, or 103 (e.g., the terminal device 101, but not limited thereto), or stored on an external storage device and can be imported into the terminal device 101. Then, the terminal device 101 can execute the operation data processing method provided by the embodiments of the present disclosure locally, or send the operation data to other terminal devices, servers, or server clusters, and the other terminal devices, servers, or server clusters that receive the operation data execute the operation data processing method provided by the embodiments of the present disclosure.

[0084] It should be understood that Figure 1 the numbers of terminal devices, networks, and servers in

[0085] are merely illustrative. According to the implementation requirements, there can be any number of terminal devices, networks, and servers. Figure 1 are merely illustrative. According to the implementation requirements, there can be any number of terminal devices, networks, and servers. Figures 2 to 9 The operation data processing method of the embodiments of the present disclosure will be described in detail below based on the

[0086] Figure 2 scenario described.

[0087] As Figure 2 shown, the operation data processing method of this embodiment includes operations S210 to S240, and this operation data processing method can be executed by a server or a terminal device.

[0088] In operation S210, obtain the target operation data generated by the target user based on the transaction client within the first preset time period.

[0089] According to an embodiment of the present disclosure, the first preset time period may be operation data generated by the user within a certain time period. For example, the first preset time period may include one day, one month, one quarter, one year, etc. The transaction client may include any financial client that can conduct transactions. For example, the transaction client may include a mobile banking client, a fund client, etc.

[0090] According to an embodiment of the present disclosure, the target operation data may be operation data generated by the target user operating the transaction client within the first preset time period. For example, the target operation data may be operation data generated by the target user operating the mobile banking client within one day. For example, the operation data may specifically include credit investigation application record query, domestic remittance|transfer remittance - remittance home page, credit card|query credit card list, credit card|apply for a card|apply for a card major list page, personal loan|apply for Rong E Jie|apply for credit consumer loan home page, mobile banking|credit investigation application record query, my account|balance query|my account list.

[0091] In operation S220, input the above target operation data into a pre-trained keyword extraction model, and output the target keywords corresponding to the above target operation data.

[0092] According to an embodiment of the present disclosure, the target operation data is operation data generated by the target user operating on the transaction client, and the context relationship is not as clear as the meaning of a sentence or a paragraph. And there are many invalid words in this type of data and few keywords reflecting the theme. Therefore, by using a keyword extraction model to extract keywords from the text first and removing some irrelevant words, the target keywords are obtained.

[0093] In operation S230, input the above target keywords into a vectorization model, and determine the target clustering cluster corresponding to the above target keywords among at least two clustering clusters in the above vectorization model.

[0094] In operation S240, determine the vectorized data corresponding to the above target operation data according to the above target clustering cluster and a preset calculation method.

[0095] According to an embodiment of the present disclosure, the vectorization model is obtained through clustering. When determining the vectorized data corresponding to the target operation data, it is necessary to determine which clustering cluster the target operation data belongs to, and then determine the vectorized data corresponding to the target operation data according to the preset calculation method.

[0096] According to an embodiment of the present disclosure, by inputting the target operation data generated by the target user on the trading client into a pre-trained keyword extraction model, a target keyword corresponding to the target operation data is obtained. Then, the target keyword is input into a vectorization model to determine a target cluster corresponding to the target keyword among at least two cluster clusters in the vectorization model. Thereafter, vectorized data corresponding to the target operation data is determined according to the target cluster and a pre-designed calculation method. At least partially, it overcomes the technical problem of low accuracy when vectorizing operation data in the prior art. It achieves the technical effect of using a model for keyword extraction and vectorization conversion, with a more refined and complex process, and improving the accuracy of vectorizing operation data at the cost of vectorization time.

[0097] According to an embodiment of the present disclosure, determining the vectorized data corresponding to the target operation data according to the target cluster and the pre-designed calculation method includes: obtaining the vectorized data of the target cluster center in the target cluster; determining the distance between the target keyword and the target cluster center to obtain a first distance; determining the distance between the target keyword and non-target cluster centers to obtain a second distance, where the non-target cluster centers include the cluster centers other than the target cluster center among the at least two cluster clusters; and determining the vectorized data corresponding to the target operation data according to the vectorized data of the target cluster center, the first distance, the second distance, and the number of cluster clusters.

[0098] According to an embodiment of the present disclosure, the operation data processing method further includes: preprocessing the target operation data to obtain processed target operation data before inputting the target operation data into the pre-trained keyword extraction model and outputting the target keyword corresponding to the target operation data; where inputting the target operation data into the pre-trained keyword extraction model and outputting the target keyword corresponding to the target operation data includes: inputting the processed target operation data into the pre-trained keyword extraction model and outputting the target keyword corresponding to the processed target operation data.

[0099] According to an embodiment of the present disclosure, the target operation data may be historical operation data generated by the target user on the trading client within one day. There may be duplicate operation data for the target customer at different times before and after. Then, the duplicate operation data is preprocessed. For example, if the target user performs transfer and remittance operations at different times within one day, only one transfer and remittance operation data is retained in the processed target operation data.

[0100] According to an embodiment of the present disclosure, preprocessing the above-mentioned target operation data to obtain the processed target operation data includes: performing stop word removal, invalid word removal, and invalid English removal on the above-mentioned target operation data to obtain the processed target operation data.

[0101] According to an embodiment of the present disclosure, stop word removal processing may include removing symbols such as "|", "-", etc.; invalid word removal processing may include removing "mobile banking", "login", "create", "modify", etc.; invalid English removal processing includes removing "json", "android", "ios", "token", etc.

[0102] In one embodiment, for example, the target operation data includes: Customer Service - Balance Change Reminder Preprocessing | ICBC Messenger - Balance Change Reminder - My Messenger, Investment and Financial Management - Purchase of Financial Products | Finance - Financial Product List, Account Management - Query Transaction Details | Account Details Large C Branch Details List json(android), Free Inquiry - Detail Details |, Precious Metals Home Page - Market Query |, | Create or Modify Token Page, Personal Loan - Apply for Rong E Loan | Apply for Credit Consumption Loan First Page. Then, the processed target operation data obtained after preprocessing the target operation data may include: Customer Service, Balance Change Reminder Preprocessing, ICBC Messenger, Balance Change Reminder, My Messenger, Investment and Financial Management, Purchase of Financial Products, Finance, Financial Product List, Account Management, Query Transaction Details, Account Details Large C Branch Details List, Free Inquiry, Detail Details, Precious Metals Home Page, Market Query, Personal Loan, Apply for Rong E Loan, Apply for Credit Consumption Loan First Page.

[0103] Figure 3 Schematically shows a flowchart of an operation data processing method according to another embodiment of the present disclosure.

[0104] As Figure 3 shown, the operation data processing method of this embodiment includes operations S310 to S380.

[0105] In operation S310, obtain the target operation data generated by the target user based on the transaction client within the first preset time period.

[0106] In operation S320, preprocess the target operation data to obtain the processed target operation data.

[0107] In operation S330, input the processed target operation data into a pre-trained keyword extraction model, and output the target keywords corresponding to the target operation data.

[0108] In operation S340, the target keyword is input into the vectorization model, and the target cluster corresponding to the target keyword is determined among at least two cluster clusters in the vectorization model.

[0109] In operation S350, the vectorized data of the target cluster center in the target cluster is obtained.

[0110] In operation S360, the distance between the target keyword and the target cluster center is determined to obtain a first distance.

[0111] In operation S370, the distance between the target keyword and the non-target cluster center is determined to obtain a second distance, where the non-target cluster center includes the cluster centers other than the target cluster center among at least two cluster clusters.

[0112] In operation S380, the vectorized data corresponding to the target operation data is determined according to the vectorized data of the target cluster center, the first distance, the second distance, and the number of cluster clusters.

[0113] According to an embodiment of the present disclosure, the above keyword extraction model is trained by the following method: obtaining historical operation data generated by users in the financial system during a second preset time period to obtain first sample data; preprocessing the above first sample data to obtain processed first sample data; labeling the processed first sample data with a preset keyword set to obtain a training data set; and training the keyword extraction model to be trained with the above training data set to obtain the above keyword extraction model.

[0114] According to an embodiment of the present disclosure, the users in the financial system may be existing users in the financial system.

[0115] According to an embodiment of the present disclosure, the second preset time period may be operation data generated by a user within a certain time period. For example, the second preset time may include one day, one month, one quarter, one year, etc.

[0116] In one of the embodiments, the historical operation data generated by the user in the financial system within one day, that is, the first sample data, is shown in Table 1:

[0117] Table 1

[0118]

[0119]

[0120] According to an embodiment of the present disclosure, preprocessing the first sample data includes removing operation data corresponding to users with fewer than a preset number of operation items within a day, and removing duplicate operation data of the same user at different times within a day. At the same time, stop word removal, invalid word removal, and invalid English word removal can also be performed on the first sample data to obtain the processed first sample data. For example, the processed first sample data can be as shown in Table 2:

[0121] Table 2

[0122]

[0123]

[0124] According to an embodiment of the present disclosure, using a preset keyword set to label the processed first sample data above to obtain a training data set. For example, it can include: manually labeling the processed first sample data, and the amount of manually labeled data can be 10,000 pieces. Then these 10,000 pieces of data can be used as the training data set. Then divide the 10,000 pieces of data into a training set and a test set according to a ratio of 8:2, use the training set to train the keyword extraction model to be trained, and use the test set for testing. It should be noted that the data is input as a whole into the model for training, the data has multiple corresponding labels, and the model is also a multi-label classification model accordingly.

[0125] According to an embodiment of the present disclosure, since there are many invalid sentences and few keywords reflecting the theme in the operation data generated by users in the financial system, if the entire text is directly clustered, the clustering effect is not good. To achieve a more ideal text clustering effect, before text clustering, the present disclosure first extracts keywords from the text to remove some unimportant words. The present disclosure uses the Bert model to extract keywords from the text, and the extracted keywords are equivalent to the labeled keywords, and then clustering is performed according to the keywords.

[0126] In one embodiment, the preset keyword set may include: remittance, consumer loan, balance change reminder, wealth management, holding amount, loan, domestic remittance, transaction details, fund, holding, deposit, credit card, detailed details, precious metal, yield, transfer, apply for a card, customer service, free inquiry, cross-bank, current, my account, Rong E Jie, ICBC Messenger, happy life version, fixed term, balance, ICBC, investment and financial management, credit investigation application, personal loan, balance inquiry, investment, purchase of wealth management, risk level.

[0127] Using the preset keyword set to label the processed first sample data, the obtained training data set can be as shown in Table 3:

[0128] Table 3

[0129]

[0130]

[0131] According to an embodiment of the present disclosure, the keyword extraction model may adopt a BERT model and a Transformer model. The BERT model is a currently popular pre-trained model based on a language model.

[0132] The most important part of the BERT model is the Transformer structure. The Transformer structure abandons the traditional convolutional neural network or recurrent neural network. The entire network is composed of an attention mechanism. More precisely, it is composed of a multi-head self-attention mechanism and a fully connected feed-forward neural network. Both of the two sub-modules adopt a residual connection and layer normalization (LN), which speeds up the network convergence rate. The residual connection makes the transmission of gradients more efficient, and gradients are the key to learning. Without gradients or gradient dispersion, it is difficult for model parameters to change, while excessive gradients or gradient explosion may cause the model to fail to converge.

[0133] Figure 4 A schematic diagram of a residual network connection according to an embodiment of the present disclosure is schematically shown.

[0134] As Figure 4 shown, this structure decomposes the objective function that originally needs to be learned into two parts: f(x) + x. The part belonging to x is directly passed into the subsequent network through an identity mapping, and the remaining f(x) part is learned through a residual mapping. Because as the network deepens, if it can be trained to converge, it will be more effective than a shallow network. However, as the propagation distance of the gradient increases, its magnitude significantly decreases, and the decrease in the gradient leads to a negligible change in the parameters. Because the derivative of addition is 1, the residual connection can propagate the gradient of the network to the deep layer, which keeps the gradient within an acceptable range, updates the parameters quickly, and speeds up the network convergence rate.

[0135] For layer normalization, normalization operations are performed on the dimension of the sequence length, and at the same time, it is not necessary to calculate the moving mean and variance of the mini-batch, saving time overhead to a certain extent.

[0136] Figure 5 A schematic diagram of the Transformer sub-module structure according to an embodiment of the present disclosure is schematically shown.

[0137] As Figure 5As shown, the feedforward neural network is actually two 1*1 convolutional operations. First, for the features in the C dimension, they are weighted by a convolutional kernel to become a single channel, and then M convolutional kernels make its output become M-dimensional. Then, ReLU activation is performed on it. After that, it returns to a vector in the C dimension through the second 1*1 convolutional operation without an activation function. By using convolution to replace the simple fully connected layer, while increasing the non-linear expression of the model, there is no clear requirement for the dimension of the input vector. For ordinary fully convolutional networks, 1*1 convolution also has the functions of reducing parameters and computational complexity and magnifying and shrinking the original feature vector.

[0138] Its input is the feature vector in the C dimension of the previous layer, which can be intuitively understood as an interpretation of C different features of the original text. The meaning of "multi-head" in the multi-head self-attention mechanism is that the C-dimensional features are divided into h parts for operation, where h is a hyperparameter, usually taken as 8. It can be understood as comparing in 8 subspaces divided from the original C-dimensional space and performing attention operations. Note that the parameters in the attention mechanism of the 8 heads are not shared like those in the single-head attention. Therefore, the multi-head attention mechanism can complete attention operations from different perspectives.

[0139] Figure 6 Schematically shows a schematic diagram of the multi-head attention mechanism according to an embodiment of the present disclosure.

[0140] As Figure 6 shown, after the Q / K / V vectors respectively pass through a linear mapping, immediately follows the attention mechanism. In the ordinary attention mechanism, usually the output of each step of the decoder is used as Q / K / V for all the outputs of the encoder, while the Q / K / V of the self-attention mechanism are all the outputs of the encoder, indicating that the network finds the most relevant words for each word in the input sentence in the original input sentence and outputs their weighted sum. The implementation formula is as follows:

[0141]

[0142] The d in the formula k represents the dimension after linear mapping. Dividing by d k makes the data after the dot product of QK T still follow a Gaussian distribution with a mean of 0 and a variance of 1, avoiding the dot product sum from being too large so that the gradient of the softmax function decreases, and then leading to gradient disappearance. First, the vectors between the multi-heads are concatenated into the original C-dimensional vector, and then through residual connection and layer normalization, the obtained result is sent into two 1*1 convolutions, whose actual utility is equivalent to a fully connected feedforward neural network. The formula is as follows:

[0143] FFN(x) = max(0, xW1 + b)W2 + b2

[0144] The variability of the input vector is realized. In the first layer, there is a ReLU activation function. In the second layer, there is no activation function, only a linear transformation is performed. Finally, after residual connection and layer normalization, the forward propagation of a sub-layer result is completed.

[0145] Replacing traditional recurrent neural networks or convolutional neural networks with a fully connected network of self-attention mechanisms has the following advantages:

[0146] (1) Compared with the limitation of serial operation required by recurrent neural networks, when performing forward or backward propagation, it relies on data from the previous time step. All connections in the transformer are essentially fully connected layers with different meanings. The benefits for accelerating the training process are self-evident.

[0147] (2) For very long sequences, although LSTM (Long Short-Term Memory Network) is used, recurrent neural networks still have a significant problem of gradient dispersion. However, in this regard, the transformer based on fully connected layers does not have the problem of affecting the convergence and speed of training due to the sequence length.

[0148] (3) Not only in terms of training speed, but also some recent studies show that as a text feature extractor, the transformer can extract better-quality features than LSTM.

[0149] (4) For convolutional neural networks, there is a problem that they cannot be parallel computed and are limited by the size of the convolutional kernel, and still cannot effectively capture long-distance features. Although stacking depth and dilated convolutions can solve this problem to a certain extent, it is still not as simple, effective, and elegant as the transformer. Instead, integrating the convolutional neural network into the basic sub-layer of the transformer is a more appropriate way.

[0150] Figure 7 A schematic diagram of the bert model according to an embodiment of the present disclosure is schematically shown.

[0151] As Figure 7 shown, the bert is composed of multiple layers of bidirectional transformer encoders 710. Since the bert model is only composed of the encoder module of the transformer, the bert model can be parallel, which helps to accelerate the training speed. Therefore, this model can often be stacked deeper.

[0152] Figure 8 A schematic diagram of the input layer of the bert model according to an embodiment of the present disclosure is schematically shown.

[0153] As Figure 8As shown, the input layer 800 vector of the BERT model consists of three parts, namely, the word vector 810 (token embedding) representing the meaning of the sentence, the segment vector 820 (segment embedding) representing whether the sentence belongs to the previous or next sentence, and the position vector 830 (position embedding), which are simply added together. Different from the representation of the position vector in Transformer, the position representation here is trained along with the training of the network during the pre-training stage. And BERT only uses word vectors to process Chinese. Based on the characteristics of pre-training with a large amount of Chinese corpora, the knowledge that can be represented by the tokenization of the complex BERT model has already been implicitly learned in the above-mentioned encoding layer and hidden layer.

[0154] According to an embodiment of the present disclosure, there are 10,000 pieces of training data for the BERT model, and there are 35 types of labels, that is to say, the model classifies 35 types in total.

[0155] The training of the BERT model may include the following two training processes. The first stage is the pre-training of the BERT model. The pre-training tasks of BERT are different from other language models. One of them is called the masked language model (MLM). In order to train the deep bidirectional representation, a certain percentage of the input words are randomly masked, and then the model is required to correctly predict the masked words through its context. This process is called the masked language model. At this time, the task has a certain similarity with the CBOW (continuous bag of words model) in Word2Vec, and both can be attributed to the cloze test of the article. However, CBOW is a bag-of-words model based on n-gram, while BERT is not limited to n-gram alone. Due to the reason of Transformer, it can pay attention to the whole article, and the tools used are different.

[0156] In the masked language model, one word is randomly masked with a probability of 15% and marked as "[MASK]". Then, a classification prediction is made for the "[MASK]" position. However, if the "[MASK]" mark is always used during the pre-training stage, this mark will not be used during the model fine-tuning. The difference in masking between the two may lead to a decrease in the performance of the model. Therefore, BERT has made an improvement, that is, while selecting 15% of the words for masking, only 80% of the selected words are actually replaced with the "[MASK]" mark, 10% of the probability is randomly replaced with another word, and the remaining 10% of the probability is not replaced. In each iteration of the MLM model, only 15% of the words are trained, which determines that the model needs more iterations during the pre-training process to complete the encoding of all words. For a fine-tunable model that can be reused, the increased time cost is not so important.

[0157] The second stage is the next sentence prediction task. This is mainly considered for some question-answering and natural language inference tasks that require an understanding of the relationship between the upper and lower sentences. This is also the reason why there is a segment vector representation in the input layer, which respectively represents a vector for the upper sentence and the lower sentence. There is a 50% probability that the lower sentence is the next sentence of the upper sentence, and a 50% probability that it is a randomly selected sentence. The implementation method is as Figures 3 - 6 shown. Both the upper and lower sentences end with the "[SEP]" token, and it is determined whether they are upper and lower sentences through the "[CLS]" token at the head.

[0158] According to an embodiment of the present disclosure, the above vectorization model is trained by the following method: obtaining historical operation data generated by m users in the financial system during a third preset time period to obtain second sample data, where m≥2; preprocessing the above second sample data to obtain processed second sample data; for each of the m users, inputting the processed second sample data corresponding to the user into the above keyword extraction model, outputting keywords corresponding to the user, and finally obtaining m groups of keywords; clustering the above m groups of keywords to obtain n clustering clusters, where 2≤n≤m; for each of the m users, determining vectorized data corresponding to the historical operation data of the user according to the above n clustering clusters, and finally obtaining m vectorized data; and determining the above vectorization model according to the above n clustering clusters and the above m vectorized data.

[0159] According to an embodiment of the present disclosure, the m users in the financial system can be all existing users in the financial system or some users.

[0160] According to an embodiment of the present disclosure, the third preset time period can be operation data generated by a user within a certain time period. For example, the second preset time can include one day, one month, one quarter, one year, etc. It should be noted that the third preset time period and the second preset time period can be the same or different.

[0161] According to an embodiment of the present disclosure, the second sample data and the first sample data can be the same or different.

[0162] In one embodiment, if the second sample data and the first sample data are the same, the processed second sample data, that is, the data in Table 2 above. Then, for each of the m users, inputting the processed second sample data corresponding to the user into the above keyword extraction model, outputting keywords corresponding to the user, and finally obtaining m groups of keywords includes: sequentially inputting the data in Table 2 into the keyword extraction model to extract keywords corresponding to each user, as shown in Table 4.

[0163] Table 4

[0164]

[0165]

[0166]

[0167] According to an embodiment of the present disclosure, the K-means clustering algorithm can be used to cluster m groups of keywords.

[0168] The steps of K-means may include:

[0169] (1) Randomly select K points as the clustering centers of K classes, denoted by k i ;

[0170] (2) Traverse all data points p j , and find the closest clustering center point k j to p i by calculating the distance. At this time, it can be said that the jth data belongs to the ith class;

[0171] (3) Calculate the center points of all data in the ith class respectively as the new clustering center points of this class;

[0172] (4) Repeat steps (2) and (3) until the clustering centers of each class no longer change.

[0173] The original K-means algorithm initially randomly selects K points in the dataset as the clustering centers, while K-means++ selects K clustering centers according to the following idea: assuming that n initial clustering centers have been selected (0 < n < K), then when selecting the (n + 1)th clustering center: the points farther away from the current n clustering centers will have a higher probability of being selected as the (n + 1)th clustering center. When selecting the first clustering center (n = 1), it is also done by a random method.

[0174] The initialization process of the K-Means++ algorithm is as follows:

[0175] Randomly select a sample point in the dataset as the first initialized clustering center and select the remaining clustering centers:

[0176] (1) Calculate the distance between each sample point in the sample and the initialized clustering centers, and select the shortest distance among them, denoted as d_i;

[0177] (2) Select the sample with the maximum distance as the new clustering center with a probability, and repeat the above process until k clustering centers are all determined;

[0178] (3) For the k initialized cluster centers, use the K-Means algorithm to calculate the final cluster centers.

[0179] According to an embodiment of the present disclosure, cluster m groups of keywords to obtain 24 clusters.

[0180] According to an embodiment of the present disclosure, the keywords located at the cluster center in the above clusters represent the historical operation data of one of the above m users; for each of the above m users, according to the above n clusters, determine the vectorized data corresponding to the historical operation data of the user, and finally obtain m vectorized data, including: for each of the above n clusters, vectorize the keywords located at the cluster center in the i-th cluster to obtain the i-th cluster center vectorized data, where 1 ≤ i ≤ n; use the above i-th cluster center vectorized data and a pre-designed calculation method to determine the non-cluster center vectorized data corresponding to the keywords located at non-cluster center positions in the i-th cluster, where the keywords located at non-cluster center positions include the keywords other than the keywords located at the cluster center in the i-th cluster, and finally obtain n cluster center vectorized data and m - n non-cluster center vectorized data; determine the above m vectorized data according to the above n cluster center vectorized data and the above m - n non-cluster center vectorized data.

[0181] According to an embodiment of the present disclosure, the above uses the above i-th cluster center vectorized data and a pre-designed calculation method to determine the non-cluster center vectorized data corresponding to the keywords located at non-cluster center positions in the i-th cluster: determine the distance between the keywords located at non-cluster center positions in the i-th cluster and the cluster center in the i-th cluster to obtain a third distance; determine the distance between the keywords located at non-cluster center positions in the i-th cluster and other cluster centers to obtain a fourth distance, where the other cluster centers include the cluster centers other than the cluster center in the i-th cluster among the above n clusters; determine the non-cluster center vectorized data corresponding to the keywords located at non-cluster center positions in the i-th cluster according to the cluster center vector data of the i-th cluster, the above third distance, the above fourth distance, and the number of clusters.

[0182] According to an embodiment of the present disclosure, the cluster center of each cluster represents the operation data of a certain user. The vector of the cluster center of each cluster is the result of word2vec of the keywords extracted by the user using bert, and the vector dimension is 256 dimensions. Therefore, in the case of clustering to obtain 24 clusters, 24 256-dimensional word vectors are generated. The generation method of the vectors of other customers not at the cluster center is as follows:

[0183] The word vector of each cluster center d i is N i , and for other users d ij in this cluster from this cluster center di The distance is S ij and the distance to other cluster centers is S ij ’, then for other users d that are not cluster centers ij the word vector N is calculated using the following formula:

[0184]

[0185] Figure 9 Schematically shows a flowchart of a method for training a vectorization model according to an example of the present disclosure.

[0186] As Figure 9 shown, the vectorization model method of this embodiment includes operation S901 to operation S910.

[0187] In operation S901, obtain the historical operation data generated by m users in the financial system within a third preset time period to obtain second sample data, where m≥2.

[0188] In operation S902, preprocess the second sample data to obtain the preprocessed second sample data.

[0189] In operation S903, input the preprocessed second sample data into a keyword extraction model, output the keywords corresponding to the users, and finally obtain m groups of keywords.

[0190] In operation S904, cluster the m groups of keywords to obtain 24 clustering clusters.

[0191] In operation S905, for each of the 24 clustering clusters, perform operations S906 to S909, and finally obtain m vectorized data.

[0192] In operation S906, use word2vec to vectorize the keywords located at the cluster center in the clustering cluster to obtain cluster center vector data.

[0193] In operation S907, determine the distance of the keywords located at non-cluster centers from the cluster center in the current clustering cluster to obtain a third distance.

[0194] In operation S908, determine the distance of the keywords located at non-cluster centers in the current clustering cluster from other cluster centers to obtain a fourth distance, where the other cluster centers include the cluster centers excluding the cluster center in the current clustering cluster.

[0195] In operation S909, determine the non-cluster center vector data corresponding to the keywords located at non-cluster centers in the current clustering cluster according to the cluster center vectorized data of the clustering cluster, the third distance, the fourth distance, and the number of clustering clusters.

[0196] In operation S910, a vectorization model is determined based on 24 clusters and m vectorized data.

[0197] According to an embodiment of the present disclosure, the vectorized data of the target user is obtained by using the above method, and users with a relatively high similarity to the target user are determined by using the vectorized data of the target user, thereby facilitating the classification of users.

[0198] For example, the user number of the target user is 10003. The target operation data corresponding to the input number 10003 is input, and the corresponding vectorized data is obtained. Then, the mean square error or the mean absolute error is used to judge the similarity between the vectorized data of the target user and the vectorized data of other users. The results are shown in Table 5:

[0199] Table 5

[0200]

[0201]

[0202]

[0203] According to an embodiment of the present disclosure, the above operation data processing method further includes: obtaining the transaction data generated by the target user based on the transaction client within the above first preset time period; inputting the transaction data and the vectorized data into a product recommendation model, and outputting a financial product corresponding to the target user.

[0204] According to an embodiment of the present disclosure, the vectorized data can be applied to the recommendation of financial products for users. The previously obtained 256-dimensional vectorized data can be used as an attribute feature for recommendation. For the recommendation of financial products, other attribute features are also needed for supplementation. The output of the recommendation model is the type of financial product. The addition of the attribute feature of the user operation data vector can make the recommendation of financial products more accurate because the vectors of similar customers are also similar.

[0205] In one of the embodiments, the input attributes of the recommendation model are shown in Table 6.

[0206] Table 6

[0207]

[0208]

[0209] The operation data processing method provided according to the embodiments of the present disclosure improves the problem of low accuracy of traditional methods for data with low relatedness. The improved method is more precise and complex in processing. Compared with traditional methods, the time cost of vectorization is greater, but the accuracy is higher. Since the vectorization scenario does not require real-time performance, the optimization scheme proposed in the embodiments of the present disclosure that sacrifices time cost to improve vector accuracy is also more real-time.

[0210] It should be noted that unless it is clearly stated that there is a sequential execution order between different operations in the flowcharts shown in the embodiments of the present disclosure, or there is a sequential execution order between different operations in the technical implementation, the execution order between multiple operations can be unordered, and multiple operations can also be executed simultaneously.

[0211] Based on the above operation data processing method, the present disclosure also provides an operation data processing device. The following will be combined with Figure 10 to describe this device in detail.

[0212] Figure 10 The structural block diagram of the operation data processing device according to the embodiments of the present disclosure is schematically shown.

[0213] As Figure 10 shown, the operation data processing device 1000 of this embodiment includes a first acquisition module 1010, a first input / output module 1020, a first determination module 1030, and a second determination module 1040.

[0214] The first acquisition module 1010 is used to acquire target operation data generated by a target user based on a transaction client within a first preset time period. In one embodiment, the first acquisition module 1010 can be used to execute operation S210 described above, which will not be elaborated here.

[0215] The first input / output module 1020 is used to input the above target operation data into a pre-trained keyword extraction model and output target keywords corresponding to the above target operation data. In one embodiment, the first input / output module 1020 can be used to execute operation S220 described above, which will not be elaborated here.

[0216] The first determination module 1030 is used to input the above target keywords into a vectorization model and determine a target clustering cluster corresponding to the above target keywords among at least two clustering clusters in the above vectorization model. In one embodiment, the first determination module 1030 can be used to execute operation S230 described above, which will not be elaborated here.

[0217] The second determination module 1040 is configured to determine the vectorized data corresponding to the target operation data according to the above-mentioned target clustering cluster and the pre-designed calculation method. In one embodiment, the second determination module 1040 may be configured to perform the operation S240 described above, which will not be elaborated herein.

[0218] According to an embodiment of the present disclosure, the second determination module includes: a first acquisition unit, a first determination unit, a second determination unit, and a third determination unit.

[0219] The first acquisition unit is configured to acquire the vectorized data of the target cluster center in the above-mentioned target clustering cluster.

[0220] The first determination unit is configured to determine the distance between the above-mentioned target keyword and the above-mentioned target cluster center to obtain a first distance.

[0221] The second determination unit is configured to determine the distance between the above-mentioned target keyword and a non-target cluster center to obtain a second distance, where the non-target cluster center includes the cluster centers other than the above-mentioned target cluster center in the above-mentioned at least two clustering clusters.

[0222] The third determination unit is configured to determine the vectorized data corresponding to the above-mentioned target operation data according to the vectorized data of the above-mentioned target cluster center, the above-mentioned first distance, the above-mentioned second distance, and the number of clustering clusters.

[0223] According to an embodiment of the present disclosure, the above-mentioned operation data processing device further includes: a preprocessing module.

[0224] The preprocessing module is configured to preprocess the above-mentioned target operation data to obtain the processed target operation data before inputting the above-mentioned target operation data into the pre-trained keyword extraction model and outputting the target keyword corresponding to the above-mentioned target operation data.

[0225] According to an embodiment of the present disclosure, the first input / output module is further configured to input the above-mentioned processed target operation data into the pre-trained keyword extraction model and output the target keyword corresponding to the above-mentioned processed target operation data.

[0226] According to an embodiment of the present disclosure, the above-mentioned preprocessing module is further configured to perform stop word removal processing, invalid word removal processing, and invalid English removal processing on the above-mentioned target operation data to obtain the above-mentioned processed target operation data.

[0227] According to an embodiment of the present disclosure, the above-mentioned operation data processing device further includes a keyword extraction model training module, and the keyword extraction model training module includes a first acquisition sub-module, a first preprocessing sub-module, a labeling sub-module, and a training sub-module.

[0228] The first acquisition sub-module is configured to acquire historical operation data generated by a user in the financial system within a second preset time period to obtain first sample data.

[0229] The first preprocessing sub-module is configured to preprocess the above first sample data to obtain the preprocessed first sample data.

[0230] The annotation sub-module is configured to annotate the preprocessed first sample data by using a preset keyword set to obtain a training data set.

[0231] The training sub-module is configured to train a keyword extraction model to be trained by using the above training data set to obtain the above keyword extraction model.

[0232] According to an embodiment of the present disclosure, the above operation data processing device further includes a vectorization model training module, and the vectorization model training module includes a second acquisition sub-module, a second preprocessing sub-module, an input-output sub-module, a clustering sub-module, a first determination sub-module, and a second determination sub-module.

[0233] The second acquisition sub-module is configured to acquire historical operation data generated by m users in the financial system within a third preset time period to obtain second sample data, where m≥2.

[0234] The second preprocessing sub-module is configured to preprocess the above second sample data to obtain the preprocessed second sample data.

[0235] The input-output sub-module is configured to, for each of the m users, input the preprocessed second sample data corresponding to the user into the above keyword extraction model, output keywords corresponding to the user, and finally obtain m groups of keywords.

[0236] The clustering sub-module is configured to cluster the above m groups of keywords to obtain n clustering clusters, where 2≤n≤m.

[0237] The first determination sub-module is configured to, for each of the m users, determine vectorized data corresponding to the historical operation data of the user according to the above n clustering clusters, and finally obtain m pieces of vectorized data.

[0238] The second determination sub-module is configured to determine the above vectorization model according to the above n clustering clusters and the above m pieces of vectorized data.

[0239] According to an embodiment of the present disclosure, the keyword located at the cluster center in the above clustering cluster represents the historical operation data of one of the m users.

[0240] According to an embodiment of the present disclosure, the above first determination sub-module includes: a vectorization unit, a fourth determination unit, and a fifth determination unit.

[0241] A vectorization unit, configured to vectorize the keywords located at the cluster center in the i-th cluster for each of the above-mentioned n clusters, to obtain the i-th cluster center vectorized data, where 1 ≤ i ≤ n.

[0242] A fourth determination unit, configured to use the above-mentioned i-th cluster center vectorized data and a pre-designed calculation method to determine the non-cluster center vectorized data corresponding to the keywords located at non-cluster centers in the i-th cluster, where the keywords located at non-cluster centers include the keywords other than the keywords located at the cluster center in the i-th cluster, and finally obtain n cluster center vectorized data and m - n non-cluster center vectorized data.

[0243] A fifth determination unit, configured to determine the above-mentioned m vectorized data according to the above-mentioned n cluster center vectorized data and the above-mentioned m - n non-cluster center vectorized data.

[0244] According to an embodiment of the present disclosure, the above-mentioned fourth determination unit includes: a first determination subunit, a second determination subunit, and a third determination subunit.

[0245] The first determination subunit is configured to determine the distance between the keywords located at non-cluster centers in the i-th cluster and the cluster center in the i-th cluster, to obtain a third distance.

[0246] The second determination subunit is configured to determine the distance between the keywords located at non-cluster centers in the i-th cluster and other cluster centers, to obtain a fourth distance, where the other cluster centers include the cluster centers other than the cluster center in the i-th cluster among the above-mentioned n clusters.

[0247] The third determination subunit is configured to determine the non-cluster center vectorized data corresponding to the keywords located at non-cluster centers in the i-th cluster according to the cluster center vector data of the i-th cluster, the above-mentioned third distance, the above-mentioned fourth distance, and the number of clusters.

[0248] According to an embodiment of the present disclosure, the above-mentioned operation data processing device further includes: a second acquisition module and a second input / output module.

[0249] The second acquisition module is configured to acquire the transaction data generated by the target user based on the above-mentioned transaction client within the above-mentioned first preset time period.

[0250] The second input / output module is configured to input the above-mentioned transaction data and the above-mentioned vectorized data into the product recommendation model, and output the financial products corresponding to the above-mentioned target user.

[0251] Any of a plurality of modules, sub-modules, units, and sub-units according to embodiments of the present disclosure, or at least part of the functions of any of them, may be implemented in one module. Any one or more of the modules, sub-modules, units, and sub-units according to embodiments of the present disclosure may be split into multiple modules for implementation. Any one or more of the modules, sub-modules, units, and sub-units according to embodiments of the present disclosure may be at least partially implemented as a hardware circuit, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-chip, a system-on-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or may be implemented by any other reasonable way of integrating or packaging circuits, such as hardware or firmware, or may be implemented in any one of the three implementation manners of software, hardware, and firmware, or in any appropriate combination of several of them. Alternatively, one or more of the modules, sub-modules, units, and sub-units according to embodiments of the present disclosure may be at least partially implemented as a computer program module, which may perform corresponding functions when the computer program module is run.

[0252] According to embodiments of the present disclosure, any of a plurality of modules among the first acquisition module 1010, the first input / output module 1020, the first determination module 1030, and the second determination module 1040 may be combined and implemented in one module, or any one of them may be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules may be combined with at least part of the functions of other modules and implemented in one module. According to embodiments of the present disclosure, at least one of the first acquisition module 1010, the first input / output module 1020, the first determination module 1030, and the second determination module 1040 may be at least partially implemented as a hardware circuit, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-chip, a system-on-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or may be implemented by any other reasonable way of integrating or packaging circuits, such as hardware or firmware, or may be implemented in any one of the three implementation manners of software, hardware, and firmware, or in any appropriate combination of several of them. Alternatively, at least one of the first acquisition module 1010, the first input / output module 1020, the first determination module 1030, and the second determination module 1040 may be at least partially implemented as a computer program module, which may perform corresponding functions when the computer program module is run.

[0253] It should be noted that the part of the operation data processing device in the embodiments of the present disclosure corresponds to the part of the operation data processing method in the embodiments of the present disclosure. For the description of the operation data processing device part, please refer to the operation data processing method part for details, and will not be elaborated here.

[0254] Figure 11A block diagram of an electronic device suitable for implementing an operation data processing method according to an embodiment of the present disclosure is schematically shown.

[0255] As Figure 11 shown, the electronic device 1100 according to an embodiment of the present disclosure includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage section 1108 into a random access memory (RAM) 1103. The processor 1101 may include, for example, a general-purpose microprocessor (e.g., CPU), an instruction set processor, and / or a related chipset, and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include on-board memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0256] In the RAM 1103, various programs and data required for the operation of the electronic device 1100 are stored. The processor 1101, the ROM 1102, and the RAM 1103 are connected to each other via a bus 1104. The processor 1101 performs various operations of the method flow according to an embodiment of the present disclosure by executing the programs in the ROM 1102 and / or the RAM 1103. It should be noted that the program may also be stored in one or more memories other than the ROM 1102 and the RAM 1103. The processor 1101 may also perform various operations of the method flow according to an embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0257] According to an embodiment of the present disclosure, the electronic device 1100 may further include an input / output (I / O) interface 1105, and the input / output (I / O) interface 1105 is also connected to the bus 1104. The electronic device 1100 may further include one or more of the following components connected to the I / O interface 1105: an input section 1106 including a keyboard, a mouse, etc.; an output section 1107 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1108 including a hard disk, etc.; and a communication section 1109 including a network interface card such as a LAN card, a modem, etc. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the I / O interface 1105 as needed. A removable medium 1111, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1110 as needed so that a computer program read from it can be installed into the storage section 1108 as needed.

[0258] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the foregoing embodiments; or may exist alone without being assembled into the device / apparatus / system. The foregoing computer-readable storage medium carries one or more programs, and when the one or more programs are executed, the methods according to the embodiments of the present disclosure are implemented.

[0259] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but not limited to: portable computer disks, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), portable compact disk read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. In the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 1102 and / or the RAM 1103 described above and / or one or more memories other than the ROM 1102 and the RAM 1103.

[0260] An embodiment of the present disclosure further includes a computer program product, which includes a computer program, and the computer program includes program code for executing the method shown in the flowchart. When the computer program product runs in a computer system, the program code is used to cause the computer system to implement the operation data processing method provided by the embodiment of the present disclosure.

[0261] When the computer program is executed by the processor 1101, the above functions defined in the system / apparatus of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the foregoing systems, apparatuses, modules, units, etc. may be implemented by computer program modules.

[0262] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices and magnetic storage devices. In another embodiment, the computer program may also be transmitted and distributed in the form of a signal on a network medium, and downloaded and installed through the communication part 1109, and / or installed from the removable medium 1111. The program code included in the computer program may be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.

[0263] In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 1109, and / or installed from the removable medium 1111. When the computer program is executed by the processor 1101, the above functions defined in the system of the embodiments of the present disclosure are executed. According to the embodiments of the present disclosure, the above-described systems, devices, apparatuses, modules, units, etc. can be implemented by computer program modules.

[0264] According to the embodiments of the present disclosure, the program code for executing the computer program provided by the embodiments of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The programming languages include, but are not limited to, such as Java, C++, python, the "C" language, or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., by connecting through the Internet using an Internet service provider).

[0265] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0266] Those skilled in the art can understand that the features recited in the various embodiments and / or claims of the present disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly recited in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features recited in the various embodiments and / or claims of the present disclosure can be combined and combined in various ways. All such combinations and / or combinations fall within the scope of the present disclosure.

[0267] The embodiments of the present disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the embodiments have been described separately above, this does not mean that the measures in each embodiment cannot be used advantageously in combination. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art can make various substitutions and modifications, and all such substitutions and modifications should fall within the scope of the present disclosure.

Claims

1. A method for processing operation data, comprising: Obtaining target operation data generated by a target user based on a transaction client within a first preset time period; Inputting the target operation data into a pre-trained keyword extraction model, and outputting target keywords corresponding to the target operation data; Inputting the target keywords into a vectorization model, and determining a target cluster corresponding to the target keywords among at least two cluster clusters in the vectorization model; And Determining vectorized data corresponding to the target operation data according to the target cluster and a preset calculation method; Wherein, the determining the vectorized data corresponding to the target operation data according to the target cluster and the preset calculation method includes: Obtaining vectorized data of a target cluster center in the target cluster; Determining the distance between the target keyword and the target cluster center to obtain a first distance; Determining the distance between the target keyword and a non-target cluster center to obtain a second distance, where the non-target cluster center includes cluster centers other than the target cluster center among the at least two cluster clusters; Determining vectorized data corresponding to the target operation data according to the vectorized data of the target cluster center, the first distance, the second distance, and the number of cluster clusters.

2. The method according to claim 1, further comprising, Before inputting the target operation data into the pre-trained keyword extraction model and outputting target keywords corresponding to the target operation data, preprocessing the target operation data to obtain processed target operation data; Among them, The inputting the target operation data into the pre-trained keyword extraction model and outputting target keywords corresponding to the target operation data includes: Inputting the processed target operation data into the pre-trained keyword extraction model, and outputting target keywords corresponding to the processed target operation data.

3. The method according to claim 2, wherein, The preprocessing the target operation data to obtain processed target operation data includes: Performing stop word removal, invalid word removal, and invalid English removal on the target operation data to obtain the processed target operation data.

4. The method according to claim 1, wherein, The keyword extraction model is trained by the following method: Obtaining historical operation data generated by users in a financial system within a second preset time period to obtain first sample data; Preprocessing the first sample data to obtain processed first sample data; Labeling the processed first sample data by using a preset keyword set to obtain a training data set; And Training a keyword extraction model to be trained by using the training data set to obtain the keyword extraction model.

5. The method according to claim 1, wherein The vectorization model is trained by the following method: Obtaining historical operation data generated by m users in a financial system within a third preset time period to obtain second sample data, where m≥2; Preprocessing the second sample data to obtain processed second sample data; For each of the m users, input the processed second sample data corresponding to the user into the keyword extraction model, output the keywords corresponding to the user, and finally obtain m groups of keywords; Cluster the m groups of keywords to obtain n clustering clusters, where 2 ≤ n ≤ m; For each of the m users, determine the vectorized data corresponding to the historical operation data of the user according to the n clustering clusters, and finally obtain m vectorized data; and Determine the vectorization model according to the n clustering clusters and the m vectorized data.

6. The method according to claim 5, wherein The keyword located at the cluster center in the clustering cluster represents the historical operation data of one of the m users; The step of, for each of the m users, determining the vectorized data corresponding to the historical operation data of the user according to the n clustering clusters, and finally obtaining m vectorized data includes: For each of the n clustering clusters, vectorize the keyword located at the cluster center in the i-th clustering cluster to obtain the i-th cluster center vectorized data, where 1 ≤ i ≤ n; Use the i-th cluster center vectorized data and a pre-designed calculation method to determine the non-cluster center vectorized data corresponding to the keyword located at the non-cluster center in the i-th clustering cluster, where the keyword located at the non-cluster center includes the keywords in the i-th clustering cluster except the keyword located at the cluster center, and finally obtain n cluster center vectorized data and m - n non-cluster center vectorized data; Determine the m vectorized data according to the n cluster center vectorized data and the m - n non-cluster center vectorized data.

7. The method according to claim 6, wherein, The step of using the i-th cluster center vectorized data and a pre-designed calculation method to determine the non-cluster center vectorized data corresponding to the keyword located at the non-cluster center in the i-th clustering cluster: Determine the distance between the keyword located at the non-cluster center in the i-th clustering cluster and the cluster center in the i-th clustering cluster to obtain a third distance; Determine the distance between the keyword located at the non-cluster center in the i-th clustering cluster and other cluster centers to obtain a fourth distance, where the other cluster centers include the cluster centers in the n clustering clusters except the cluster center in the i-th clustering cluster; Determine the non-cluster center vectorized data corresponding to the keyword located at the non-cluster center in the i-th clustering cluster according to the cluster center vector data of the i-th clustering cluster, the third distance, the fourth distance, and the number of clustering clusters.

8. The method according to claim 1, further comprising: Obtain the transaction data generated by the target user based on the transaction client within the first preset time period; Input the transaction data and the vectorized data into a product recommendation model, and output the financial products corresponding to the target user.

9. An operation data processing device, comprising: A first acquisition module, configured to acquire target operation data generated by a target user based on a transaction client within a first preset time period; A first input / output module, configured to input the target operation data into a pre-trained keyword extraction model, and output target keywords corresponding to the target operation data; The first determination module inputs the target keyword into a vectorization model, and determines a target clustering cluster corresponding to the target keyword among at least two clustering clusters in the vectorization model; And The second determination module is configured to determine vectorized data corresponding to the target operation data according to the target clustering cluster and a pre-designed calculation method; Wherein, the second determination module includes: a first acquisition unit, a first determination unit, a second determination unit, and a third determination unit; The first acquisition unit is configured to acquire vectorized data of a target cluster center in the target clustering cluster; The first determination unit is configured to determine the distance between the target keyword and the target cluster center to obtain a first distance; The second determination unit is configured to determine the distance between the target keyword and a non-target cluster center to obtain a second distance, where the non-target cluster center includes cluster centers other than the target cluster center among the at least two clustering clusters; The third determination unit is configured to determine vectorized data corresponding to the target operation data according to the vectorized data of the target cluster center, the first distance, the second distance, and the number of clustering clusters.

10. An electronic device, comprising: One or more processors; A storage device for storing one or more programs, Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are caused to execute the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, on which executable instructions are stored, and when the instructions are executed by a processor, the processor is caused to execute the method according to any one of claims 1 to 8.

12. A computer program product, comprising a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images