Data processing method and device, equipment, storage medium and computer program product

Abstract

A data processing method, device, equipment, storage medium and computer program product can be applied to various fields or scenes such as cloud technology, artificial intelligence, block chain, Internet of Vehicles, intelligent transportation, smart home and the like. The method comprises: obtaining initial text features and statistical features of a target object, the initial text features being determined according to object text data and transaction associated text data of the target object, and the statistical features being determined according to historical transaction data of the target object; processing the initial text features to determine text sparse features and text dense features of the target object; processing the statistical features and the text sparse features to determine first fusion features of the target object, and processing the statistical features and the text dense features to determine second fusion features of the target object; and determining a business recognition result of the target object according to the first fusion features and the second fusion features. The embodiments of the present application can effectively improve the recognition accuracy of the business category of the target object.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a data processing method, a data processing device, a computer device, a computer readable storage medium and a computer program product. BACKGROUND

[0002] The business category of the target object can be distinguished according to the information filled in by the target object when the target object is registered. For example, the name of the target object is "aa restaurant" according to the information filled in by the target object when the target object is registered, and it can be known that the business category of the target object is catering. However, when the information of the target object is incomplete or the information content is incorrect, the business category of the target object is not accurately distinguished by the above method, and it is also difficult to obtain the real business category when the target object is out of range or the business scope does not match the actual situation. SUMMARY

[0003] The embodiments of the present application provide a data processing method, device, equipment, storage medium and computer program product, which can effectively improve the identification accuracy of the business category of the target object.

[0004] In one aspect, the embodiments of the present application provide a data processing method, and the method comprises:

[0005] obtaining initial text features and statistical features of a target object, wherein the initial text features are determined according to object text data and transaction associated text data of the target object, and the statistical features are determined according to historical transaction data of the target object;

[0006] processing the initial text features to determine text sparse features and text dense features of the target object;

[0007] processing the statistical features and the text sparse features to determine first fusion features of the target object, and processing the statistical features and the text dense features to determine second fusion features of the target object;

[0008] determining a business identification result of the target object according to the first fusion features and the second fusion features.

[0009] In another aspect, the embodiments of the present application provide a data processing device, and the device comprises:

[0010] an obtaining unit, configured to obtain initial text features and statistical features of a target object, wherein the initial text features are determined according to object text data and transaction associated text data of the target object, and the statistical features are determined according to historical transaction data of the target object;

[0011] a processing unit, configured to process the initial text features to determine text sparse features and text dense features of the target object;

[0012] The processing unit is further configured to process the statistical features and the text sparse features to determine first fusion features of the target object, and process the statistical features and the text dense features to determine second fusion features of the target object.

[0013] The processing unit is further configured to determine a service recognition result of the target object according to the first fusion features and the second fusion features.

[0014] Correspondingly, an embodiment of the present application provides a computer device, which comprises a processor, a communication interface and a memory, the processor, the communication interface and the memory are connected with each other, wherein the memory stores a computer program, and the processor is configured to invoke the computer program to execute the data processing method in any possible implementation manner.

[0015] Correspondingly, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the data processing method provided in the embodiment of the present application.

[0016] Correspondingly, an embodiment of the present application further provides a computer program product, which comprises a computer program or computer instructions, and the computer program or computer instructions are executed by a processor to implement the steps of the data processing method provided in the embodiment of the present application.

[0017] Correspondingly, an embodiment of the present application further provides a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium, a processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the data processing method provided in the embodiment of the present application.

[0018] The embodiment of the application obtains the text sparse feature and the text dense feature of the target object based on the initial text feature of the target object, processes the statistical feature and the text sparse feature of the target object to determine the first fusion feature of the target object, processes the statistical feature and the text dense feature of the target object to determine the second fusion feature of the target object, and determines the business recognition result of the target object according to the first fusion feature and the second fusion feature. The initial text feature determined according to the object text data and the transaction associated text data of the target object and the statistical feature determined according to the historical transaction data of the target object can obtain the real business category of the target object. Meanwhile, the initial text feature and the statistical feature are used to describe the target object from multiple dimensions, which can effectively improve the recognition accuracy of the business category of the target object. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0020] Figure 1 A system architecture schematic diagram of a data processing system provided by the embodiment of the application is provided.

[0021] Figure 2 A flowchart of a data processing method provided by the embodiment of the application is provided. Figure 1

[0022] Figure 3 A flowchart of a data processing method provided by the embodiment of the application is provided. Figure 2

[0023] Figure 4 A structure schematic diagram of a BERT model provided by the embodiment of the application is provided.

[0024] Figure 5 A structure schematic diagram of an intent recognition model provided by the embodiment of the application is provided.

[0025] Figure 6 A structure schematic diagram of a data processing device provided by the embodiment of the application is provided.

[0026] Figure 7 A structure schematic diagram of a computer device provided by the embodiment of the application is provided. DETAILED DESCRIPTION

[0027] ​​With reference to the drawings and the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0028] It should be noted that the terms "first", "second", etc. described in the embodiments of the present application are only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the technical features limited by "first", "second", etc. can be explicitly or implicitly included at least one of the features.

[0029] The present application proposes a data processing method, which can effectively improve the recognition accuracy of the business category of the target object, and can be applied to various fields or scenes such as cloud technology, artificial intelligence, block chain, Internet of vehicles, intelligent transportation, smart home, etc.

[0030] In an embodiment, the data processing method can be implemented based on machine learning technology in artificial intelligence technology. Machine learning (ML) is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is a subject that studies how computers simulate or implement human learning behavior to obtain new knowledge or skills, reorganize existing knowledge structure to continuously improve performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its application is widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rule-based learning.

[0031] In an implementable embodiment, the data processing method can also be implemented based on cloud technology and / or blockchain technology. Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc. applied based on a cloud computing business model, and can form a resource pool for on-demand use in a flexible and convenient manner. Cloud computing technology will become an important support. The background service of a technical network system requires a large amount of computing and storage resources, such as video websites, picture websites and more portals. With the high development and application of the Internet industry, in the future, every item may have its own identification mark and needs to be transmitted to the background system for logical processing. Different levels of data will be processed separately, and various industry data will need strong system support, which can only be realized through cloud computing. Blockchain (blockchain or block chain) is a string of text records (also known as blocks) connected and protected by cryptography. Each block contains the encrypted hash of the previous block, the corresponding timestamp, and transaction data (usually represented by the hash value calculated by the Merkle tree algorithm), which makes the block content difficult to tamper with. The distributed ledger connected by blockchain technology can effectively record transactions between two parties and permanently verify the transaction.

[0032] The data processing method provided by the embodiments of the present application can specifically involve one or more of cloud storage, cloud database, and big data in cloud technology. For example, data (such as initial text features, statistical features, etc.) required for executing the data processing method is obtained from the cloud database. For another example, the data required for executing the data processing method can be stored in the form of blocks on the blockchain; the data generated by executing the data processing method (such as business identification results) can be stored in the form of blocks on the blockchain; in addition, the computer device executing the data processing method can be a node device in the blockchain network.

[0033] Please refer to Figure 1 , Figure 1 for a system architecture diagram of a data processing system provided by the embodiments of the present application. Figure 1 The system architecture shown in the figure can be used to implement the data processing method proposed by the embodiments of the present application. As shown in Figure 1 , the network architecture includes a computer device 10 and a plurality of terminals 11 (three are shown as an example in the figure).

[0034] The computer device 10 can be a server or a terminal with data processing functions. The server can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and basic cloud computing services such as big data and artificial intelligence platforms. The terminal can be a smartphone, a tablet computer, a notebook computer, a desktop computer, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, and the like, but is not limited thereto. Figure 1 The terminal 11 is connected to the computer device 10 through a network.

[0035] Figure 1 The system architecture can implement the data processing method provided by the embodiments of the present application. The implementation process of the method generally includes: the computer device 10 obtains the initial text features and statistical features of the target object sent by the terminal 11. The initial text features are determined according to the object text data and transaction associated text data of the target object, and the statistical features are determined according to the historical transaction data of the target object. The computer device 10 processes the initial text features to determine the text sparse features and text dense features of the target object, processes the statistical features and the text sparse features to determine the first fusion features of the target object, and processes the statistical features and the text dense features to determine the second fusion features of the target object. According to the first fusion features and the second fusion features, the business recognition result of the target object is determined. According to the initial text features determined according to the object text data and transaction associated text data of the target object and the statistical features determined according to the historical transaction data of the target object, the real business category of the target object can be obtained. At the same time, the initial text features and the statistical features are used to describe the target object from multiple dimensions, which improves the recognition accuracy of the business category of the target object.

[0036] In an embodiment, the computer device 10 can send the business recognition result of the target object to the terminal 11. The terminal 11 can determine whether the target object has a risk situation such as out-of-scope operation or inconsistent operation scope with the actual situation according to the business category registered by the target object and the business category included in the business recognition result, to perform risk assessment on the target object. For example, the business category registered by the target object is food wholesale, and the business category included in the business recognition result is clothing, which belongs to inconsistent operation scope with the actual situation.

[0037] It can be understood that the system architecture diagram described in the embodiments of the present application is used to more clearly illustrate the technical solutions of the embodiments of the present application, and does not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of system architecture and the appearance of new business scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0038] The data processing method provided by the embodiments of the present application is briefly introduced above, and the specific implementation manner of the data processing method is described in detail below.

[0039] Please refer to Figure 2 , Figure 2 The flowchart of a data processing method provided by the embodiments of the present application Figure 1 The data processing method described in the embodiments of the present application can be executed by the computer device 10 in Figure 1 , and the data processing method includes but is not limited to the following steps:

[0040] S201, obtaining initial text features and statistical features of a target object, wherein the initial text features are determined according to object text data and transaction associated text data of the target object, and the statistical features are determined according to historical transaction data of the target object.

[0041] The target object is an object that needs to determine a business category, and the object can be a merchant. The business category is used to represent the industry to which the target object belongs, such as that the business category of “aa restaurant” is “catering”, the business category of “xx suit customization” is “clothing”, and the business category of “yy hot pot seasoning” is “seasoning”, etc.

[0042] The object text data includes business registration information of the target object, such as the name and registered address of the target object. The transaction associated text data includes text associated with transactions of the target object, such as the name of the goods sold by the target object, evaluation information, complaint information, note information, the name and introduction of the platform of the transaction (the platform of the transaction can be a public account, an application, etc.), and the customer public opinion report of the target object. It should be noted that the object text data and the transaction associated text data involved need to obtain the permission or consent of the transaction object and the target object, and the use and processing need to comply with relevant laws, regulations and standards of countries and regions. The initial text features are determined according to the object text data and the transaction associated text data of the target object, for example, the initial text features include each text in the object text data and the transaction associated text data.

[0043] The historical transaction data includes transaction amount (which can be the average amount in a certain period of time), average age and gender ratio of the transaction object, active time of the transaction, category of the sold goods, and the like. It should be noted that the historical transaction data involves obtaining the permission or consent of the transaction object and the target object, and the use and processing need to comply with relevant laws, regulations and standards of the country and region. The statistical features are determined according to the historical transaction data of the target object, for example, the transaction amount is 5000, the average age and the male and female gender ratio of the transaction object are 18 and 1 / 1.2 respectively, and the category of the sold goods is clothes (each category of goods can be represented by a number, such as 1 for clothes and 2 for shoes), and the statistical features can be represented as [5000, 18, 1 / 1.2, 1].

[0044] S202, processing the initial text features to determine the text sparse features and the text dense features of the target object.

[0045] In the text sparse features, the values of a small number of elements are not zero, and the values of a large number of elements are zero. In the text dense features, the values of a large number of elements are not zero, and the values of a small number of elements are zero.

[0046] In an embodiment, a text encoding network can be used to encode and process the initial text features to obtain the text dense features of the target object. The text encoding network can include a recurrent neural network (RNN), a long short-term memory (LSTM), a bidirectional encoder representations from transformers (BERT) model, and the like.

[0047] In an embodiment, the initial text feature includes a plurality of subtext features, which can be text corresponding to an attribute, the attribute being a name of the target object, a registered address, a name of a commodity sold by the target object, evaluation information, complaint information, note information, and a name and introduction of a transaction platform, etc. For example, the attribute is evaluation information of a commodity sold, and the text corresponding to the attribute is "this dress is really beautiful", and the subtext feature is "this dress is really beautiful". The encoding corresponding to the subtext feature can be determined according to whether the subtext feature matches the business category. For example, the subtext feature is "this dress is really beautiful", and the business category is "clothing", and the subtext feature matches the business category. When the subtext feature matches the business category, the encoding corresponding to the subtext feature is determined to be 1, and when the subtext feature does not match the business category, the encoding corresponding to the subtext feature is determined to be 0. Finally, according to the matching results of each subtext feature included in the initial text feature and the business category, the text sparse feature of the target object can be obtained. For example, the initial text feature includes subtext features corresponding to the name of the target object and the name of the commodity sold by the target object, and if the name of the commodity sold by the target object matches the business category in the name of the target object and the name of the commodity sold by the target object, the text sparse feature of the target object can be [0, 1].

[0048] In an available embodiment, the text dense feature can be one-hot encoded to map the text dense feature to the text sparse feature, or the text sparse feature can be embedded by using an embedding method to map the text sparse feature to the text dense feature.

[0049] In an available embodiment, the text dense feature can be one-hot encoded to map the text dense feature to the text sparse feature, or the text sparse feature can be embedded by using an embedding method to map the text sparse feature to the text dense feature.

[0050] In an embodiment, the first fusion feature of the target object can be obtained by fusing the statistical feature and the text sparse feature, and the first fusion feature is a sparse feature. The fusion processing can be in a splicing manner. When the statistical feature and the text sparse feature are vectors, the statistical feature and the text sparse feature can be concatenated to obtain the first fusion feature of the target object. When the statistical feature and the text sparse feature are matrices, if the dimensions of the statistical feature and the text sparse feature are the same, the statistical feature and the text sparse feature can be spliced into a matrix, and if the dimensions of the statistical feature and the text sparse feature are different, the statistical feature and the text sparse feature can be spliced into a matrix by using a zero padding operation, and the matrix is the first fusion feature of the target object.

[0051] In an embodiment, the second fusion feature of the target object can be obtained by fusing the statistical feature and the text dense feature. The fusion processing can be in a splicing manner. When the statistical feature and the text dense feature are vectors, the statistical feature and the text dense feature can be spliced to obtain the second fusion feature of the target object. When the statistical feature and the text dense feature are matrices, if the dimensions of the statistical feature and the text dense feature are the same, the statistical feature and the text dense feature can be spliced into a matrix; if the dimensions of the statistical feature and the text dense feature are different, the statistical feature and the text dense feature can be spliced into a matrix by zero padding, and the matrix is the second fusion feature of the target object.

[0052] In S204, a service identification result of the target object is determined according to the first fusion feature and the second fusion feature.

[0053] In an embodiment, the first fusion feature and the second fusion feature can be fused, and then a classifier is used to classify the fused features to obtain the service identification result of the target object. The service identification result includes the service category of the target object and / or the probability that the target object belongs to each service category. When the service identification result includes the probability that the target object belongs to each service category, the service category corresponding to the maximum probability can be taken as the service category of the target object. The classifier can be a Naive Bayes classification algorithm, a K-Nearest Neighbor (kNN) classification algorithm, a support vector machine, a linear classifier, maximum entropy, a neural network, etc.

[0054] In the embodiments of the present application, the text sparse feature and the text dense feature of the target object are obtained based on the initial text feature of the target object, the first fusion feature of the target object is determined by processing the statistical feature and the text sparse feature of the target object, the second fusion feature of the target object is determined by processing the statistical feature and the text dense feature of the target object, and the service identification result of the target object is determined according to the first fusion feature and the second fusion feature. According to the above scheme, the real service category of the target object can be obtained based on the initial text feature determined according to the object text data and the transaction associated text data of the target object and the statistical feature determined according to the historical transaction data of the target object. Meanwhile, the initial text feature and the statistical feature are used to describe the target object from multiple dimensions, which can effectively improve the identification accuracy of the service category of the target object.

[0055] Please refer to Figure 3 , Figure 3 A flowchart of a data processing method provided by the embodiments of the present application Figure 2 The data processing method described in the embodiments of the present application can be implemented by Figure 1The data processing method is executed by the computer device 10 in the embodiment, and the data processing method includes but is not limited to the following steps:

[0056] S301, obtaining initial text features and statistical features of the target object, wherein the statistical features include numerical features and category features.

[0057] The numerical features are determined according to historical transaction data related to numerical values of the target object, such as transaction amounts, average ages of transaction objects, male and female sex ratios, and active times of transactions included in the historical transaction data. It should be noted that the historical transaction data related to numerical values of the target object needs to obtain permission or consent of the transaction object and the target object, and the use and processing need to comply with relevant laws, regulations and standards of relevant countries and regions. Assuming that the transaction amount is 5000, the average age of the transaction object and the male and female sex ratio are 18 and 1 / 1.2, and the active time of the transaction is 18-20, the numerical features can be represented as [5000, 18, 1 / 1.2, 18-20].

[0058] The category features are determined according to historical transaction data related to categories of the target object, such as categories of sold goods in the historical transaction data. It should be noted that the historical transaction data related to categories of the target object needs to obtain permission or consent of the transaction object and the target object, and the use and processing need to comply with relevant laws, regulations and standards of relevant countries and regions. Assuming that the category of the sold goods is shoes (each category of goods can be represented by a number, such as 1 for clothes and 2 for shoes), the category features can be represented as [2].

[0059] The initial text features can refer to the related description of S201, and the embodiment will not be repeated.

[0060] S302, determining text sparse features of the target object according to the keyword set and the initial text features, wherein the keywords in the keyword set are used to indicate business categories.

[0061] The keywords in the keyword set can be used to indicate business categories, such as "education", "training", "examination", and "question", which usually indicate education. The keyword set can be obtained by setting keywords of each business category, or can be obtained by statistical analysis of related data of each business category.

[0062] Specifically, for any business category, determining whether the initial text feature includes any subtext feature matching the keywords corresponding to the business category in the keyword set, the matching refers to determining whether the subtext feature includes the keywords corresponding to the business category in the keyword set, when the subtext feature includes the keywords corresponding to the business category in the keyword set, it is determined that the matching result is that the subtext feature matches the business category, and when the subtext feature does not include the keywords corresponding to the business category in the keyword set, it is determined that the matching result is that the subtext feature does not match the business category. For example, the subtext feature in the initial text feature is "the paper quality of this test paper is very good", the keywords corresponding to the education category in the keyword set include "education", "training", "examination", "question", and "test paper", and it can be obtained that the matching result is that the subtext feature matches the education category. Further, when the matching result is that the subtext feature matches any business category, it is determined that the coding value of the element corresponding to the subtext feature is a first coding value (which can be artificially set, for example, 1), and when the matching result is that the subtext feature does not match the keywords corresponding to any business category, it is determined that the coding value of the element corresponding to the subtext feature is a second coding value (which can be artificially set, for example, 0). According to the coding value of the element corresponding to each subtext feature and the position of the element corresponding to each subtext feature set, the text coding feature corresponding to the any business category can be obtained.

[0063] Further, the text coding features corresponding to each business category can be fused to obtain the text sparse feature of the target object. The fusion processing can be splicing processing, for example, the text sparse feature of the target object can be obtained by splicing the text coding features corresponding to each business category.

[0064] 303、processing the initial text feature to determine the text dense feature of the target object.

[0065] In an embodiment, the computer device determines the reference text feature according to the subtext feature included in the initial text feature and the sentence start identifier corresponding to the subtext feature, the sentence start identifier being used to identify the sentence start of the subtext feature, so that the reference text feature can be obtained by adding the sentence start identifier corresponding to the subtext feature before the subtext feature. For example, the sentence start identifier [CLS] corresponding to the subtext feature is added before the subtext feature “the paper quality of this test paper is good”, and the reference text feature includes the “position of the element corresponding to the subtext feature [CLS]”. Further, the reference text feature is encoded to obtain an encoding result. The encoding process calculates the mutual relationship of each word in the reference text feature with all the words in the sentence, and then uses the mutual relationship to adjust the weight of each word to obtain the text encoding representation corresponding to each word, so that the text encoding representation not only contains the semantic of the word itself, but also contains the relationship between other words and the word. The text encoding features corresponding to each sentence start identifier can be obtained from the encoding result to represent the semantic of the whole sentence of the subtext feature corresponding to each sentence start identifier. The text encoding features corresponding to each sentence start identifier are fused to obtain the text dense feature of the target object. The fusion process can be splicing, such as splicing the text encoding features corresponding to each sentence start identifier into the text dense feature of the target object.

[0066] In a specific implementation, the BERT model is used to encode the reference text feature to obtain an encoding result. The BERT model pre-trains the bidirectional deep representation (Embedding) by jointly adjusting the bidirectional converter (Transformer) in all layers. Referring to Figure 4 , the input of the BERT model is the reference text feature, and the reference text feature can include the split identifier [SEP] for splitting two sentences in the subtext feature in addition to the sentence start identifier before the subtext feature. For example, the initial text feature includes the subtext feature A “the paper quality of this test paper is good, but the clarity is not very high”, and the subtext feature B “aa bookstore”, and the reference text feature can include “[CLS] the paper quality of this test paper is good [SEP] but the clarity is not very high [CLS] aa bookstore”. In the network structure of the BERT model nThe input encoding is obtained by obtaining word encoding features (encoding words into feature vectors), segmentation encoding features (used to distinguish two sentences), and position encoding features (encoding position information of words into feature vectors) of the reference text features, and adding the word embedding encoding, segmentation encoding, and position encoding of the reference text features. The BERT model encodes and outputs the encoding result of the reference text features. The text features corresponding to each sentence initial identifier are obtained from the encoding result and fused to obtain the text dense features of the target object.

[0067] In an embodiment, the model parameters in the BERT model are trained based on the training data of the business identification model. The BERT model has good generalization ability for features not learned in the training data, which can effectively improve the identification accuracy of the business category of the target object.

[0068] In S304, the numerical features, the category features, and the text sparse features are processed to determine the first fusion features of the target object.

[0069] In an implementation, the computer device performs binning processing on the numerical features to obtain binning features corresponding to the numerical features. For example, the continuous age is binned into five intervals: 0-20 years old, 20-40 years old, 40-60 years old, 60-80 years old, and 80-100 years old. When the average age of the transaction object recorded in the numerical features is 18 years old, it belongs to the first bin, and the binning feature can be represented as [1, 0, 0, 0, 0]. For another example, the continuous time is binned into four intervals: 0-6 hours, 6-12 hours, 12-18 hours, and 18-24 hours. When the active time of the transaction recorded in the numerical features is 18-20 hours, it belongs to the fourth bin, and the binning feature can be represented as [0, 0, 0, 1]. The computer device performs one-hot encoding on the category features to obtain one-hot features corresponding to the category features. One-hot encoding is also called one-hot encoding. For example, the categories of the sold goods include clothes, shoes, and hats, which are represented as 0, 1, and 2 in the category features, respectively. Therefore, clothes, shoes, and hats are represented as [0, 0, 1], [0, 1, 0], and [1, 0, 0] in the one-hot features, respectively.

[0070] Further, the binning features, the one-hot features, and the text sparse features are processed to obtain cross-multiplication fusion features. Specifically, any two features among the binning features, the one-hot features, and the text sparse features are cross-multiplied (also called vector outer product) to obtain cross-multiplication features. The cross-multiplication fusion features are obtained by fusing the cross-multiplication features. The fusion processing can be splicing processing of the cross-multiplication features to obtain the cross-multiplication fusion features.

[0071] In an embodiment, the first fusion feature of the target object is obtained by performing linear transformation on the fused binning feature, one-hot feature, text sparse feature and cross-multiplication fusion feature. Specifically, the binning feature, one-hot feature, text sparse feature fusion and cross-multiplication fusion feature are fused by splicing to obtain the fused feature, and then a generalized linear model is used to perform linear transformation on the fused feature to obtain the first fusion feature of the target object. The generalized linear model is shown in the following formula (1):

[0072] y = w T [x, φ(x)] + b (1)

[0073] wherein φ(x) represents the cross-multiplication fusion feature, x represents the feature obtained by fusing the binning feature, one-hot feature and text sparse feature, w and b represent linear parameters, and T represents transposition.

[0074] In an embodiment, the linear parameters w and b in the generalized linear model are obtained by training the service identification model based on training data, which can enable the linear model to capture high-order correlations between sparse features, memorize features that have appeared together in training data, explore co-occurrence in training data, and effectively improve the recognition accuracy of the service category of the target object.

[0075] S305, processing the numerical feature, the category feature and the text dense feature to determine the second fusion feature of the target object.

[0076] In an embodiment, the category feature is processed based on a dense matrix to obtain a low-dimensional dense feature corresponding to the category feature. Specifically, the one-hot encoding corresponding to the category feature is cross-multiplied with the dense matrix to obtain the low-dimensional dense feature corresponding to the category feature. Taking the category feature including the category of the sold goods and the geographic location area where the transaction object is located as an example, the geographic location area where the transaction object is located needs to obtain the permission or consent of the transaction object, and the use and processing need to comply with relevant laws, regulations and standards of relevant countries and regions. If the category of the sold goods includes {clothes, shoes}, the geographic location area where the transaction object is located includes {north, south}, and the category feature is [2 (indicating belonging to shoes), 1 (indicating belonging to north)], the dense matrix is represented as W 2×dim , the dimension of the dense matrix is 4 (total number of categories, 2+2) x dim, dim is the dimension of the preset low-dimensional dense feature, the one-hot feature [0, 1, 1, 0] corresponding to the category feature can be obtained, and the one-hot feature [0, 1, 1, 0] corresponding to the category feature is cross-multiplied with the dense matrix W 4×dim to obtain [w 2,1 , w 2,2 , …, w 2,dim ]+[w3,1 , w 3,2 ,..., w 3,dim ]. Wherein, [w 2,1 , w 2,2 ,..., w 2,dim ] represents the category of the sold goods belongs to the low-dimensional dense feature corresponding to shoes, [w 3,1 , w 3,2 ,..., w 3,dim ] represents the geographical location area where the transaction object is located belongs to the low-dimensional dense feature corresponding to the north, and the low-dimensional dense feature corresponding to the category feature is obtained by fusing (which can be splicing) the low-dimensional dense feature corresponding to the category of the sold goods belongs to the shoes and the low-dimensional dense feature corresponding to the geographical location area where the transaction object is located belongs to the north.

[0077] In an embodiment, the matrix parameters of the dense matrix are obtained by training the business identification model based on the training data, so that each row of the dense matrix can accurately represent its corresponding category, thereby obtaining effective low-dimensional dense features and improving the recognition accuracy of the business category of the target object.

[0078] Further, the extreme values in the numerical features can be removed, normalized, and the like, and then the numerical features, the low-dimensional dense features, and the text dense features are fused to obtain the initial fusion features of the target object, and the fusion processing can be splicing the numerical features, the low-dimensional dense features, and the text dense features. The initial fusion features of the target object are encoded to obtain the second fusion features of the target object, and the encoding processing can be realized by inputting the initial fusion features into a hidden layer, and the hidden layer can be L (a positive integer) layers. The calculation of each layer of the hidden layer can use the following formula (2):

[0079] a l+1＝ f(W l a l +b l ) (2)

[0080] Wherein, a l represents the activation value of the lth layer in the L layer, b l represents the bias of the lth layer in the L layer, W l represents the weight of the lth layer in the L layer, and represents the activation function. It can be understood that the activation value of the first layer in the L layer is the initial fusion feature.

[0081] In an embodiment, the bias and weight in the hidden layer are obtained by training the business identification model based on the training data, so that the second fusion features obtained based on the hidden layer have more rich expression ability compared with the initial fusion features, and the recognition accuracy of the business category of the target object can be improved.

[0082] S306、According to the first fusion feature and the second fusion feature, determine the service identification result of the target object.

[0083] In a specific implementation, the first fusion feature and the second fusion feature can be spliced, and then a linear classifier (for example, a softmax classifier, a sigmoid classifier) is used to classify the fused features to obtain the service identification result of the target object, which includes the probability of the target object belonging to each service category. The formula is as shown in the following formula (3):

[0084]

[0085] Wherein, σ represents the Softmax function, W Wide T [x, φ(x)] + b represents the first fusion feature of the target object, W Deep l a l +b l represents the second fusion feature of the target object.

[0086] The above BERT model, linear model, hidden layer and dense matrix of the application can be obtained by training the service identification model as shown in the following formula (1) based on the training data. Figure 5 The service identification model includes a BERT model, a linear model, a hidden layer, a dense layer and a linear classifier, and the dense layer includes a dense matrix. The model parameters of the BERT model, the linear parameters of the linear model, the bias and weight of the hidden layer, and the matrix parameters of the dense matrix in the dense layer of the service identification model can be obtained by random initialization.

[0087] In an embodiment, the process of training the service identification model based on the training data includes but is not limited to the following steps:

[0088] I. Obtain training data, which includes training text features and training statistical features of a plurality of training objects, and the real service category of each training object.

[0089] The training text features of each training object are determined according to the object text data and the transaction-related text data of each training object. The training statistical features of each training object include training numerical features and training category features. The training numerical features of each training object are determined according to the historical transaction data related to the numerical value of each training object. The training category features of each training object are determined according to the historical transaction data related to the category of each training object. It should be noted that the historical transaction data, the object text data and the transaction-related text data of the training object need to obtain the permission or consent of the corresponding transaction object and the training object, and the use and processing need to comply with the relevant laws, regulations and standards of the country and region.

[0090] II. Determine the predicted business category of each training object according to the training text features and the training statistical features of each training object, and the business identification model.

[0091] Specifically, ① according to the keyword set and the training text features of each training object, the training text sparse features of each training object are obtained. ② According to the training subtext features included in the training text features of each training object and the corresponding first character identifiers of the training subtext features, the training reference text features of each training object are determined; the BERT model in the business recognition model is used to encode and process the training reference text features of each training object, and the training text feature codes corresponding to each first character identifier are obtained; the training text coding features corresponding to each first character identifier are fused to obtain the training text dense features of each training object. ③ The training numerical features are binned to obtain the training binned features corresponding to the training numerical features; the training category features are one-hot encoded to obtain the training one-hot features corresponding to the training category features; the training binned features, the training one-hot features and the training text sparse features are processed to obtain the training cross-multiplication fusion features, which can be specifically obtained by cross-multiplying (also called vector outer product) any two of the training binned features, the training one-hot features and the training text sparse features, and then fusing the training cross-multiplication features to obtain the training cross-multiplication fusion features; the training binned features, the training one-hot features, the training text sparse features and the training cross-multiplication fusion features are fused and then linearly transformed by the linear model in the business recognition model to obtain the first training fusion features of each training object. ④ The training category features are densely processed by the dense layer in the business recognition model to obtain the low-dimensional dense features corresponding to the training category features; the training numerical features, the low-dimensional dense features and the training text dense features are fused to obtain the training initial fusion features of each training object; the training initial fusion features are encoded by the hidden layer in the business recognition model to obtain the second training fusion features of each training object. ⑤ The first training fusion features and the second training fusion features of each training object are fused and then input into the linear classifier in the business recognition model to obtain the predicted business category of each training object.

[0092] III. The predicted loss parameter is determined according to the predicted business category and the real business category of each training object.

[0093] In an embodiment, the loss function of the business recognition model can be obtained, and the predicted loss parameter is obtained by inputting the predicted business category and the real business category of each training object into the loss function. The loss function can be a square loss function, an exponential loss function, a logarithmic loss function, etc.

[0094] IV. The parameters of the BERT model, the linear model, the hidden layer and the dense layer in the business recognition model are adjusted according to the loss parameter to obtain the adjusted BERT model, the adjusted linear model, the adjusted hidden layer and the adjusted dense layer.

[0095] In an embodiment, the parameters of the BERT model, the linear model, the hidden layer and the dense layer in the business identification model are adjusted according to the prediction loss parameter obtained each time, and whether the prediction loss parameter obtained after the parameter adjustment meets the training stop condition is determined, and when the training stop condition is met, the training is stopped. The meeting of the training stop condition can mean that the prediction loss parameter obtained after the parameter adjustment is less than or equal to a preset loss threshold, or a specified number of iteration times is reached, for example, when the business identification model has undergone 1000 iterations, it is determined that the training stop condition is met.

[0096] In an embodiment, the BERT model obtained by stopping the training can be applied to encode the reference text features of the target object to obtain an encoding result. The linear model obtained by stopping the training can be applied to linearly transform the features fused after the binning features, the one-hot features, the text sparse features and the cross-multiplication fusion features of the target object to obtain the first fusion features of the target object. The dense matrix in the dense layer obtained by stopping the training can be applied to densely process the category features of the target object to obtain the low-dimensional dense features corresponding to the category features. The hidden layer obtained by stopping the training can be applied to encode the initial fusion features of the target object to obtain the second fusion features of the target object.

[0097] Ensemble learning is to train each model separately and then combine the results of the models, which requires each independent model to be learned well enough to benefit from the subsequent combination, so the magnitude of each model is relatively larger. However, the joint training adopted in the present application enables the business identification model to combine the memory capability of the linear model and the generalization capability of the deep model (the BERT model, the hidden layer, etc.), thereby improving the overall performance of the model. The linear model can accurately remember and judge the business category when encountering features that have been encountered, and the deep model has sufficient generalization capability when encountering features that have not been encountered. In addition, training the cross-multiplication fusion features can improve the memory capability of the linear model.

[0098] In the embodiments of the present application, the text sparse features and the text dense features of the target object can be determined by using the keyword set and / or the initial text features, the first fusion features of the target object can be determined by processing the numerical features, the category features and the text sparse features, the second fusion features of the target object can be determined by processing the numerical features, the category features and the text dense features, and the business identification result of the target object can be determined according to the first fusion features and the second fusion features. Through the above scheme, the real transaction data of the target object can be used for modeling to obtain the real business category of the target object. At the same time, the initial text features, the numerical features and the category features are used to describe the target object from multiple dimensions, and the memory capability of the linear model and the generalization capability of the deep model (the BERT model, the hidden layer, etc.) are combined, which can effectively improve the identification accuracy of the business category of the target object.

[0099] It can be understood that in the specific embodiments of the present application, related data such as object text data, transaction associated text data and historical transaction data are involved. When the above embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of countries and regions.

[0100] The above describes the method of the embodiments of the present application in detail. In order to facilitate better implementation of the above scheme of the embodiments of the present application, correspondingly, the device of the embodiments of the present application is provided below. Please refer to Figure 6 , Figure 6 is a structural schematic diagram of a data processing device provided by the embodiments of the present application. The data processing device 60 can include:

[0101] The acquisition unit 601 is configured to acquire initial text features and statistical features of a target object, wherein the initial text features are determined according to object text data and transaction associated text data of the target object, and the statistical features are determined according to historical transaction data of the target object.

[0102] The processing unit 602 is configured to process the initial text features, determine text sparse features and text dense features of the target object.

[0103] The processing unit 602 is further configured to process the statistical features and the text sparse features to determine first fusion features of the target object, and process the statistical features and the text dense features to determine second fusion features of the target object.

[0104] The processing unit 602 is further configured to determine a business recognition result of the target object according to the first fusion features and the second fusion features.

[0105] In an embodiment, the acquisition unit 601 is specifically configured to acquire a keyword set, wherein keywords in the keyword set are used to indicate business categories.

[0106] The processing unit 602 is configured to determine text encoding features corresponding to each business category according to the keyword set and the initial text features, and perform fusion processing on the text encoding features corresponding to each business category to obtain the text sparse features of the target object.

[0107] In an embodiment, the processing unit 602 is specifically configured to: for any business category, determine whether a subtext feature included in the initial text feature matches a keyword corresponding to the any business category in the keyword set; and determine an encoding value of an element corresponding to the subtext feature according to a matching result, and determine a text encoding feature corresponding to the any business category according to the encoding value.

[0108] In an embodiment, the processing unit 602 is specifically configured to: determine a reference text feature according to a subtext feature included in the initial text feature and a first character identifier corresponding to the subtext feature; perform encoding processing on the reference text feature to obtain an encoding result; obtain a text encoding feature corresponding to each first character identifier from the encoding result; and perform fusion processing on the text encoding features corresponding to the each first character identifier to obtain the text dense feature of the target object.

[0109] In an embodiment, the statistical feature includes a numerical feature and a category feature, and the processing unit 602 is specifically configured to: perform binning processing on the numerical feature to obtain a binning feature corresponding to the numerical feature; perform one-hot encoding on the category feature to obtain a one-hot feature corresponding to the category feature; perform processing on the binning feature, the one-hot feature, and the text sparse feature to obtain a cross-multiplication fusion feature; and perform linear transformation on the binning feature, the one-hot feature, the text sparse feature, and the cross-multiplication fusion feature after fusion to obtain the first fusion feature of the target object.

[0110] In an embodiment, the processing unit 602 is specifically configured to: perform cross-multiplication on any two features in the binning feature, the one-hot feature, and the text sparse feature to obtain a cross-multiplication feature; and perform fusion processing on each cross-multiplication feature to obtain a cross-multiplication fusion feature.

[0111] In an embodiment, the statistical feature includes a numerical feature and a category feature, and the processing unit 602 is specifically configured to: perform dense processing on the category feature based on a dense matrix to obtain a low-dimensional dense feature corresponding to the category feature; perform fusion processing on the numerical feature, the low-dimensional dense feature, and the text dense feature to obtain an initial fusion feature of the target object; and perform encoding processing on the initial fusion feature of the target object to obtain a second fusion feature of the target object.

[0112] It can be understood that the functions of the functional units of the data processing apparatus described in the embodiments of the present application can be specifically implemented according to the methods in the above method embodiments, and the specific implementation process can refer to the related description of the above method embodiments, which will not be described here.

[0113] The embodiment of the application obtains the text sparse feature and the text dense feature of the target object based on the initial text feature of the target object, processes the statistical feature and the text sparse feature of the target object to determine the first fusion feature of the target object, processes the statistical feature and the text dense feature of the target object to determine the second fusion feature of the target object, and determines the business recognition result of the target object according to the first fusion feature and the second fusion feature. The initial text feature determined according to the object text data and the transaction associated text data of the target object and the statistical feature determined according to the historical transaction data of the target object can obtain the real business category of the target object, and the initial text feature and the statistical feature are used to describe the target object from multiple dimensions, which can effectively improve the recognition accuracy of the business category of the target object.

[0114] As shown in Figure 7 , Figure 7 is a structural schematic diagram of a computer device provided by the embodiment of the application, and the internal structure of the computer device 70 is as shown in Figure 7 , which comprises one or more processors 701, a memory 702, and a communication interface 703. The processor 701, the memory 702, and the communication interface 703 can be connected through a bus 704 or other means, and the embodiment of the application takes the connection through the bus 704 as an example.

[0115] The processor 701 (also called CPU (Central Processing Unit, central processor)) is the computing core and control core of the computer device 70, which can parse various instructions in the computer device 70 and process various data of the computer device 70, for example: the CPU can be used to parse the on-off instruction sent by the user to the computer device 70 and control the computer device 70 to perform the on-off operation; for another example: the CPU can transmit various interactive data between the internal structures of the computer device 70, and the like. The communication interface 703 can optionally include a standard wired interface, a wireless interface (such as Wi-Fi, a mobile communication interface, etc.), and is controlled by the processor 701 to receive and send data. The memory 702 is a memory device in the computer device 70, which is used to store computer programs and data. It can be understood that the memory 702 here can include the built-in memory of the computer device 70, and of course can also include the expansion memory supported by the computer device 70. The memory 702 provides a storage space that stores the operating system of the computer device 70, which can include but is not limited to: Windows system, Linux system, etc., and the application does not limit this. Specifically, the processor 701 executes the following operations by running the computer program stored in the memory 702:

[0116] obtaining initial text features and statistical features of a target object, the initial text features being determined according to object text data and transaction-related text data of the target object, and the statistical features being determined according to historical transaction data of the target object;

[0117] processing the initial text features to determine text sparse features and text dense features of the target object;

[0118] processing the statistical features and the text sparse features to determine first fusion features of the target object, and processing the statistical features and the text dense features to determine second fusion features of the target object;

[0119] determining a business recognition result of the target object according to the first fusion features and the second fusion features.

[0120] In an embodiment, the processor 701 is specifically configured to: obtain a keyword set, keywords in the keyword set being used to indicate business categories; determine text encoding features corresponding to respective business categories according to the keyword set and the initial text features; and perform fusion processing on the text encoding features corresponding to the respective business categories to obtain the text sparse features of the target object.

[0121] In an embodiment, the processor 701 is specifically configured to: for any business category, determine whether a subtext feature included in the initial text features matches a keyword corresponding to the any business category in the keyword set; determine an encoding value of an element corresponding to the subtext feature according to a matching result, and determine a text encoding feature corresponding to the any business category according to the encoding value.

[0122] In an embodiment, the processor 701 is specifically configured to: determine reference text features according to subtext features included in the initial text features and sentence head identifiers corresponding to the subtext features; perform encoding processing on the reference text features to obtain an encoding result; obtain text encoding features corresponding to respective sentence head identifiers from the encoding result; and perform fusion processing on the text encoding features corresponding to the respective sentence head identifiers to obtain the text dense features of the target object.

[0123] In an embodiment, the statistical features include numerical features and category features, and the processor 701 is specifically configured to: perform binning processing on the numerical features to obtain binning features corresponding to the numerical features; perform one-hot encoding on the category features to obtain one-hot features corresponding to the category features; perform processing on the binning features, the one-hot features, and the text sparse features to obtain cross-multiplication fusion features; and perform linear transformation on the binning features, the one-hot features, the text sparse features, and the cross-multiplication fusion features after fusion to obtain the first fusion features of the target object.

[0124] In an embodiment, the processor 701 is specifically configured to: perform cross-multiplication on any two features in the binning features, the one-hot features, and the text sparse features to obtain cross-multiplication features; and perform fusion processing on each cross-multiplication feature to obtain cross-multiplication fusion features.

[0125] In an embodiment, the statistical features include numerical features and category features, and the processor 701 is specifically configured to: perform dense processing on the category features based on a dense matrix to obtain low-dimensional dense features corresponding to the category features; perform fusion processing on the numerical features, the low-dimensional dense features, and the text dense features to obtain initial fusion features of the target object; and perform encoding processing on the initial fusion features of the target object to obtain second fusion features of the target object.

[0126] In a specific implementation, the processor 701, the memory 702, and the communication interface 703 described in the embodiments of the present application can perform the implementation modes of the computer device described in the data processing method provided by the embodiments of the present application, and can also perform the implementation modes described in the data processing apparatus provided by the embodiments of the present application, which will not be described here.

[0127] In the embodiments of the present application, the text sparse features and the text dense features of the target object are obtained based on the initial text features of the target object, the first fusion features of the target object are determined by processing the statistical features and the text sparse features of the target object, the second fusion features of the target object are determined by processing the statistical features and the text dense features of the target object, and the business recognition result of the target object is determined according to the first fusion features and the second fusion features; the above scheme can obtain the real business category of the target object by using the initial text features determined according to the object text data and the transaction associated text data of the target object and the statistical features determined according to the historical transaction data of the target object, and can effectively improve the recognition accuracy of the business category of the target object by using the initial text features and the statistical features to describe the target object from multiple dimensions.

[0128] The embodiment of the present application further provides a computer readable storage medium, wherein instructions are stored in the computer readable storage medium, and when the instructions are executed on a computer, the computer executes the data processing method according to the embodiment of the present application. The specific implementation manner can refer to the foregoing description, and will not be described here.

[0129] The embodiment of the present application further provides a computer program product, which comprises a computer program or computer instructions, and when the computer program or computer instructions are executed by a processor, the steps of the data processing method according to the embodiment of the present application are implemented. The specific implementation manner can refer to the foregoing description, and will not be described here.

[0130] The embodiment of the present application further provides a computer program, which comprises computer instructions stored in a computer readable storage medium, and a processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the data processing method according to the embodiment of the present application. The specific implementation manner can refer to the foregoing description, and will not be described here.

[0131] It should be noted that, for each of the foregoing method embodiments, in order to simply describe, each is described as a combination of a series of actions, but those skilled in the art should know that the present application is not limited to the action sequence described, because according to the present application, some steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily necessary for the present application.

[0132] Those of ordinary skill in the art can understand that all or part of the steps of the various methods in the above embodiments can be completed by a program instructing relevant hardware, and the program can be stored in a computer readable storage medium, and the storage medium can include a flash disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0133] The above only describes some embodiments of the present application, and of course cannot limit the scope of the rights of the present application, so equivalent changes made according to the claims of the present application are still within the scope of the present application.

Claims

1. A data processing method, characterized in that: The method comprises: Obtaining initial text features and statistical features of a target object, wherein the initial text features are determined based on object text data and transaction-related text data of the target object, and the statistical features are determined based on historical transaction data of the target object; the statistical features include numerical features and categorical features; Processing the initial text features to determine the text sparse features of the target object; Determining reference text features based on subtext features included in the initial text features and sentence start identifiers corresponding to the subtext features; encoding the reference text features to obtain encoding results; obtaining text encoding features corresponding to each sentence start identifier from the encoding results; fusing the text encoding features corresponding to each sentence start identifier to obtain text dense features of the target object; The numerical features are binned to obtain bin features corresponding to the numerical features; the categorical features are one-hot encoded to obtain one-hot features corresponding to the categorical features; the bin features, the one-hot features, and the text sparse features are processed to obtain a cross-product fusion feature; the bin features, the one-hot features, the text sparse features, and the cross-product fusion feature are fused and linearly transformed to obtain a first fusion feature of the target object; Performing dense processing on the category features based on a dense matrix to obtain low-dimensional dense features corresponding to the category features; performing fusion processing on the numerical features, the low-dimensional dense features and the text dense features to obtain initial fused features of the target object; performing encoding processing on the initial fused features of the target object to obtain second fused features of the target object; A service identification result of the target object is determined according to the first fusion feature and the second fusion feature.

2. The method according to claim 1, characterized in that Processing the initial text features to determine the text sparse features of the target object includes: Acquire a keyword set, where keywords in the keyword set are used to indicate a business category; Determining text coding features corresponding to each business category based on the keyword set and the initial text features; The text encoding features corresponding to each of the business categories are fused to obtain the text sparse features of the target object.

3. The method according to claim 2, characterized in that Determining text coding features corresponding to respective business categories based on the keyword set and the initial text features includes: For any business category, determining whether a subtext feature included in the initial text feature matches a keyword corresponding to any business category in the keyword set; The coding value of the element corresponding to the sub-text feature is determined according to the matching result, and the text coding feature corresponding to any business category is determined according to the coding value.

4. The method according to any one of claims 1 to 3, characterized in that The processing of the binning feature, the one-hot feature, and the text sparse feature to obtain a cross-product fusion feature includes: Performing a cross product on any two of the binning feature, the one-hot feature, and the text sparse feature to obtain a cross product feature; The cross-product features are fused to obtain the cross-product fusion features.

5. A data processing device, characterized in that: The device comprises: an acquisition unit, configured to acquire initial text features and statistical features of a target object, wherein the initial text features are determined based on object text data and transaction-related text data of the target object, and the statistical features are determined based on historical transaction data of the target object; the statistical features include numerical features and categorical features; a processing unit configured to: process the initial text features to determine the sparse text features of the target object; determine reference text features based on subtext features included in the initial text features and sentence start identifiers corresponding to the subtext features; encode the reference text features to obtain an encoding result; obtain text encoding features corresponding to respective sentence start identifiers from the encoding result; and fuse the text encoding features corresponding to respective sentence start identifiers to obtain the dense text features of the target object; The processing unit is further used to: perform binning processing on the numerical features to obtain binning features corresponding to the numerical features, perform one-hot encoding on the category features to obtain one-hot features corresponding to the category features, process the binning features, the one-hot features and the text sparse features to obtain cross-product fusion features, fuse the binning features, the one-hot features, the text sparse features and the cross-product fusion features, and perform linear transformation to obtain a first fusion feature of the target object; perform dense processing on the category features based on a dense matrix to obtain a low-dimensional dense feature corresponding to the category features, perform fusing processing on the numerical features, the low-dimensional dense features and the text dense features to obtain an initial fusion feature of the target object, and perform encoding processing on the initial fusion feature of the target object to obtain a second fusion feature of the target object; The processing unit is further configured to determine a service identification result of the target object according to the first fusion feature and the second fusion feature.

6. A computer device, characterized in that: The computer device includes a memory, a communication interface and a processor, and the memory, the communication interface and the processor are interconnected; the memory stores a computer program, and the processor calls the computer program stored in the memory to implement the data processing method described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the data processing method according to any one of claims 1 to 4 is implemented.

8. A computer program product, characterized in that The computer program product comprises a computer program or computer instructions, and when the computer program or computer instructions are executed by a processor, the data processing method according to any one of claims 1 to 4 is implemented.

Images