A Counterparty Recognition Method, Device and Electronic Device Based on Deep Learning

Abstract

The present invention discloses a counterparty recognition method, device and electronic device based on deep learning. The method includes: obtaining a bond information text, extracting elements from the bond information text to obtain the extracted element data; cutting the bond information text according to a preset structured rule index and the element data to form a spot trading order information; constructing a prompt-based classification model, training the classification model according to financial prior knowledge and the spot trading order information to generate a counterparty recognition model; obtaining a bond information text to be parsed, and preprocessing the bond information text to be parsed; inputting the preprocessed bond information text into the counterparty recognition model to obtain the type of the counterparty. The embodiments of the present invention can effectively extract various elements in spot trading; avoid the problem of conflicts among different hard-coded rules; and solve the problem of high rule maintenance cost.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a transaction counterparty identification method, device and electronic device based on deep learning. Background Art

[0002] Spot bond trading is the most common type of transaction in the field of financial fixed income. In the text understanding of spot bond trading, the identification of the counterparty of the spot bond trading is a very important issue. For the text understanding of spot bond trading, it is generally necessary to integrate a series of intelligent algorithms and business rule indexes to complete the comprehensive understanding of the spot bond text. A common spot bond trading corpus is shown below. "K bond, 0000123.IB 0.5Y 3.9AA institution issued to BB institution\n Y bond 000673.IB 2+1Y 3.55CC institution issued to DD institution requesting HH institution". To understand this text and provide convenient and intelligent operations for downstream tasks, this text needs to be structured into the structure shown in the following table.

[0003] Table 1 Structure of the text after structuring

[0004]

[0005] Structured text requires counterparty identification to identify the buyer, seller and bridge institution in each order.

[0006] The existing counterparty identification methods mainly adopt two schemes. One is a method based on deep learning. Based on the idea of information extraction, the counterparties in the text are extracted and classified to determine whether they are "our party, counterparty, bridge institution", etc. This type of method mainly includes information extraction models such as "Albert+CRF, LSTM+CRF, Bert+BiLSTM+CRF". The second is a method based on "business rules". The "institution name" information is extracted through the information extraction model, and then a proprietary logical index is formulated in combination with the "business rules" to judge the "institution name" to determine whether it is "our party, counterparty institution", or "bridge institution", etc.

[0007] First, the deep learning method based on "information extraction" can only use "text information" in model training and prediction, and cannot include "financial professional logic" outside the market. It cannot judge the direction of the opponent very well, and the accuracy rate is not high. Moreover, the same entity appears multiple times in the text, and the entity categories in different positions are not the same. The information extraction model is difficult to learn, resulting in the low accuracy of the model for entity recognition.

[0008] Secondly, a dedicated counterparty identification algorithm needs to be constructed, which requires a large number of training samples in this field and is difficult to perform cold start. In the case of lack of samples, the accuracy of the model cannot reach the commercial standard.

[0009] Thirdly, extraction is performed based on the algorithm model. By combining business rules to formulate logical indexes, the deficiencies of the pure algorithm model solution can be solved. However, the current logical indexes usually adopt hard-coded calculation methods. Two problems are brought about by this: a) It is difficult to maintain. For each judgment logic, corresponding processing logic needs to be added, and the maintenance cost is very high. b) The generalization ability of the hard-coding technology is insufficient, and conflicts are likely to occur between different logics, and the scalability is not strong.

[0010] The counterparty identification methods in the prior art have high maintenance costs and low entity identification accuracy.

[0011] Therefore, the prior art still needs to be improved and developed. Summary of the Invention

[0012] In view of the above deficiencies of the prior art, the present invention provides a counterparty identification method, device and electronic device based on deep learning, aiming to solve the problems of high maintenance cost and low entity identification accuracy in the counterparty identification methods in the prior art.

[0013] The technical solution of the present invention is as follows:

[0014] The first embodiment of the present invention provides a counterparty identification method based on deep learning. The method includes:

[0015] Obtain a bond information text, extract elements from the bond information text to obtain the extracted element data;

[0016] Cut the bond information text according to a preset structured rule index and element data to form spot trading order information;

[0017] Construct a prompt-based classification model, and train the classification model according to financial prior knowledge and the spot trading order information to generate a counterparty identification model;

[0018] Obtain a bond information text to be parsed, and preprocess the bond information text to be parsed;

[0019] Input the preprocessed bond information text into the counterparty identification model to obtain the type of the counterparty.

[0020] Further, the obtaining a bond information text, extracting elements from the bond information text to obtain the extracted element data includes:

[0021] Obtain the bond information text, and encode the bond information text to generate a sequence text;

[0022] Based on a deep neural network, extract elements from the sequence text to obtain the extracted element data.

[0023] Further, index according to the preset structured rules and the element data to cut the bond information text to form the spot trading order information, including:

[0024] Obtain the bond information text, cut the line elements, generate line-by-line text information and the elements included in each line of text information;

[0025] Obtain the information category of each line of text information;

[0026] Aggregate each line of information according to the information category to form the spot trading order information at the order dimension.

[0027] Further, before constructing a prompt-based classification model and training the classification model according to financial prior knowledge and the spot trading order information to generate a counterparty identification model, including:

[0028] Identify the spot trading order information to obtain the position information of the institution name in the spot trading order information.

[0029] Further, constructing a prompt-based classification model and training the classification model according to financial prior knowledge and the spot trading order information to generate a counterparty identification model, including:

[0030] Construct a prompt-based classification model;

[0031] Based on financial prior knowledge and the position information of the institution name, recombine the element data to generate a recombined corpus;

[0032] Based on the prompt method, splice the institution name to be predicted with the recombined corpus to generate a target corpus;

[0033] Input the target corpus into the classification model to train the classification model to generate a counterparty identification model.

[0034] Further, obtaining the bond information text to be parsed and preprocessing the bond information text to be parsed, including:

[0035] Obtain the bond information text to be parsed, extract elements from the bond information text to be parsed to obtain the extracted element data;

[0036] Cut the text of the bond information to be parsed according to the preset structured rule index and element data to form the spot trading order information to be parsed.

[0037] Further, inputting the preprocessed text of the bond information into the counterparty recognition model to obtain the type of the counterparty, including:

[0038] Input the preprocessed text of the bond information into the counterparty recognition model, and obtain the institution name output by the counterparty recognition model;

[0039] Obtain the type of the current visual party, and determine the counterparty type according to the type of the visual party and the institution name.

[0040] Another embodiment of the present invention provides a counterparty recognition device based on deep learning. The device includes:

[0041] An element extraction module, configured to obtain the text of the bond information, extract elements from the text of the bond information, and obtain the extracted element data;

[0042] A text cutting module, configured to cut the text of the bond information according to the preset structured rule index and element data to form spot trading order information;

[0043] A model training module, configured to build a prompt-based classification model, and train the classification model according to financial prior knowledge and the spot trading order information to generate a counterparty recognition model;

[0044] A text preprocessing module, configured to obtain the text of the bond information to be parsed and preprocess the text of the bond information to be parsed;

[0045] A counterparty recognition module, configured to input the preprocessed text of the bond information into the counterparty recognition model to obtain the type of the counterparty.

[0046] Another embodiment of the present invention provides an electronic device, where the electronic device includes at least one processor; and,

[0047] A memory communicatively connected to the at least one processor; wherein,

[0048] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned counterparty recognition method based on deep learning.

[0049] Another embodiment of the present invention also provides a non-volatile computer-readable storage medium storing computer-executable instructions, which, when executed by one or more processors, enable the one or more processors to execute the above-mentioned counterparty identification method based on deep learning.

[0050] Advantages: The embodiments of the present invention can effectively extract various elements in the spot bond trading; the classification algorithm replacement hard-coded scheme has stronger generalization ability than the existing hard-coded scheme, can solve more different types of problems, and avoid the problem of conflicts among different hard-coded rules; and solves the problem of high rule maintenance cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The present invention will be further described below in conjunction with the drawings and embodiments, where:

[0052] Figure 1 is a flowchart of a preferred embodiment of a counterparty identification method based on deep learning of the present invention;

[0053] Figure 2 is a schematic flowchart of the overall solution of a preferred embodiment of a counterparty identification method based on deep learning of the present invention;

[0054] Figure 3 is a schematic flowchart of order splitting of a preferred embodiment of a counterparty identification method based on deep learning of the present invention;

[0055] Figure 4 is a schematic diagram of a classification model of a preferred embodiment of a counterparty identification method based on deep learning of the present invention;

[0056] Figure 5 is a schematic diagram of functional modules of a preferred embodiment of a counterparty identification device based on deep learning of the present invention;

[0057] Figure 6 is a schematic diagram of the hardware structure of a preferred embodiment of an electronic device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] To make the objectives, technical solutions and effects of the present invention clearer and more definite, the present invention will be further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0059] The embodiments of the present invention will be introduced below in conjunction with the drawings.

[0060] In view of the above problems, the embodiments of the present invention provide a counterparty identification method based on deep learning. Please refer to Figure 1 ,Figure 1 The flowchart of a preferred embodiment of a transaction counterparty identification method based on deep learning of the present invention is shown in FIG. Figure 1 As shown, it includes:

[0061] Step S100: obtaining bond information text, extracting elements from the bond information text, and obtaining extracted element data;

[0062] Step S200: cutting the bond information text according to the preset structured rule index and element data to form current bond transaction order information;

[0063] Step S300: construct a prompt-based classification model, train the classification model according to financial prior knowledge and the current securities transaction order information, and generate a counterparty identification model;

[0064] Step S400: obtaining the bond information text to be parsed, and preprocessing the bond information text to be parsed;

[0065] Step S500: input the preprocessed bond information text into the counterparty identification model to obtain the type of counterparty.

[0066] When implementing it, Figure 2 The overall framework diagram of the embodiment of the present invention is shown in FIG. The deep algorithm model is adopted, and the main innovation is that the NER model is no longer used, but an innovative prompt-based classification model is proposed to identify the counterparty.

[0067] For unstructured text, we first use the information extraction algorithm model to extract elements from structured text, and extract the core elements in the text, such as "bond name, bond code, term, volume, institution name" and other core elements. Secondly, we build a structured rule index, cut multiple orders for the "multiple bonds" situation, and form the information of the current trading orders one by one. Thirdly, we build a prompt-based classification model, and train the model according to the "rules of the financial industry and the location where the institutional information appears" to determine whether the direction of the institution is "our side, the counterparty" or a "bridge" institution.

[0068] In one embodiment, the bond information text is obtained, and elements are extracted from the bond information text to obtain extracted element data, including:

[0069] Obtain bond information text, and encode the bond information text to generate sequence text;

[0070] The sequence text is subjected to element extraction based on a deep neural network to obtain extracted element data.

[0071] In specific implementation, a deep neural network is adopted to extract elements from unstructured text. Specifically, a deep learning model based on Bert+LSTM+CRF and specifically optimized for cash bonds (including but not limited to this model) is used to extract information such as "bond name, bond code, quantity, institution name" in the sequential text, and various core elements in the sequential text are obtained.

[0072] In one embodiment, the bond information text is cut according to a preset structured rule index and element data to form cash bond transaction order information, including:

[0073] Obtain the bond information text, cut the line elements, and generate line-by-line text information and the elements included in each line of text information;

[0074] Obtain the information category of each line of text information;

[0075] Aggregate each line of information according to the information category to form cash bond transaction order information at the order dimension.

[0076] In specific implementation, as Figure 3 shown, for the cash bond transaction text, this solution constructs a business index-based logical judgment module. The structured logic is divided into the following steps:

[0077] Cut the elements line by line, and cut each line and the elements extracted based on information extraction according to "\n" to form line-by-line text and the elements included in each line of information.

[0078] Judge the category of each line of information. Judge the information of each line as "order information, supplementary information, shared information, other information". Order information is defined as: the text information of the core cash bond transaction theme such as "bond name or bond code" in this line of information. Supplementary information is defined as: supplementary information without the core information of cash bond transactions in this line of information. Shared information is the public shared information shared to each order information in this line of information. Other information is some invalid text information, such as information without any valid entities.

[0079] Aggregate each line of information based on the rules. The rule logic is: with "order as the core", splice the "supplementary information" into the "order information" according to the "upward supplement" logic. The "shared information" is directly copied and supplemented to each "order information". Thus, one or more order information at the "order" dimension are formed.

[0080] In one embodiment, a prompt-based classification model is constructed. Before training the classification model according to financial prior knowledge and the cash bond transaction order information to generate a counterparty identification model, it includes:

[0081] Identify the spot trading order information and obtain the position information of the institution name in the spot trading order information.

[0082] In specific implementation, one or more order information are obtained, which includes core order elements, supplementary element information, and shared element information. In these element sequences, there is one or more "institution names". Identify the spot trading order information and obtain the position information of the institution name in the spot trading order information.

[0083] Based on the position information of these "institution names", comprehensively judge whether the "institution name" is the "buying institution", "selling institution", or "bridging institution" according to information such as the sender and recipient of this text.

[0084] In one embodiment, construct a prompt-based classification model, train the classification model according to financial prior knowledge and the spot trading order information, and generate a counterparty identification model, including:

[0085] Construct a prompt-based classification model;

[0086] Based on financial prior knowledge and the position information of the institution name, recombine the element data to generate recombined corpus;

[0087] Use a prompt-based method to splice the institution name to be predicted with the recombined corpus to generate target corpus;

[0088] Input the target corpus into the classification model, train the classification model, and generate a counterparty identification model.

[0089] In specific implementation, based on established financial rules, such as "Bond A 00001, 20 million, yield 3.14, XX Bank offers to YY Fund through ZZ Securities"; this solution extracts the elements [bond name, bond code, institution name, trading direction, request method, line break character "\n"] from the transaction corpus, and recombines them into a sentence according to the order in which these elements appear. For example, the above transaction corpus after extraction and combination is "Bond A 00001 XX Bank offers to YY Fund through ZZ Securities".

[0090] b). Prompt-based corpus. Since in a classification model, there is only one mutually exclusive category for a piece of corpus, it is impossible to determine from the extracted corpus "Bond A 00001 XX Bank issues to YY Fund requesting ZZ Securities" that XX Bank is the seller institution, YY Fund is the buyer institution, and ZZ Securities is the bridge institution. Therefore, this solution adopts a prompting method. By concatenating the institution name to be predicted with the corpus, the model is prompted to predict which institution name, thereby realizing the identification of counterparties. The concatenated corpus is "XX Bank # Bond A 00001 XX Bank issues to YY Fund requesting ZZ Securities" and "YY Fund # Bond A 00001 XX Bank issues to YY Fund requesting ZZ Securities". The above-mentioned corpus can prompt the model which institution is the object to be predicted, and the concatenated corpus is not the same, so there will be no result of multiple mutually exclusive labels for the same corpus, thus realizing the accuracy of the model.

[0091] In the embodiments of the present invention, a classification algorithm is used. Currently, Bert + [CLS] + dense is adopted; the pre-trained model includes but is not limited to BERT, but may include pre-trained models such as albert. Taking the Bert + [CLS] + dense model as an example, the network structure diagram is as Figure 4 shown.

[0092]

[0093] output = dense(BERT(h cls ; θ)) (Formula 2)

[0094] where dense(x) = W T x + b, W is the transformation matrix, b is the bias parameter, h is the hidden vector of the Transformer Encoder, θ is the parameter of the model, h cls is the [CLS] vector output in BERT, x is the input information of the model (the text after extracting and concatenating the prompt content mentioned above for b). M is the number of Transformer Encoders in the BERT model.

[0095] This solution adopts the above model structure to classify the concatenated input corpus and determine which one of the buyer, seller, or bridge institution the corpus and its corresponding institution belong to.

[0096] In one embodiment, the bond information text to be parsed is obtained, and the bond information text to be parsed is preprocessed, including:

[0097] The bond information text to be parsed is obtained, and the elements of the bond information text to be parsed are extracted to obtain the extracted element data;

[0098] The bond information text to be parsed is segmented according to the preset structured rule index and element data to form the current bond transaction order information to be parsed.

[0099] In the specific implementation, the bond information text to be parsed is obtained, and elements are extracted from the bond information text to be parsed. The extracted elements include but are not limited to data such as bond name, bond code, quantity, institution name, etc., and the bond information text to be parsed is cut according to the structured rule index to form one or more order information. The specific structured rule index is the same as the structured rule index in the above-mentioned training sample, and will not be repeated here.

[0100] In one embodiment, the pre-processed bond information text is input into the counterparty identification model to obtain the type of counterparty, including:

[0101] Input the preprocessed bond information text into the counterparty identification model to obtain the institution name output by the counterparty identification model;

[0102] Get the type of the current visual party, and determine the counterparty type based on the visual party type and institution name.

[0103] In specific implementation, when the counterparty identification model has determined the seller / buyer / bridge, the party / counterparty / bridge institution still needs to be determined based on the visual party. The visual party is the institution that sees the current bond transaction information. If the visual party is the buyer's institution, the counterparty is the seller's institution. If the visual party is the seller's institution, the counterparty is the buyer's institution. If the visual party is the bridge institution, both the buyer and the seller are counterparties.

[0104] The embodiment of the present invention is a method for identifying opponents in spot transactions, which combines financial logic and uses a prompt-based classification algorithm to identify opponents. It can be extended to other scenarios where it is necessary to identify financial transaction opponents.

[0105] The embodiment of the present invention provides a method for counterparty identification based on deep learning, which adopts a counterparty identification scheme built with a prompt-based classification algorithm and can effectively extract various elements in current securities transactions.

[0106] The solution that uses classification algorithms to replace hard coding has stronger generalization capabilities than the existing hard coding solution, can solve more different types of problems, avoid conflicts between different hard-coded rules; and solves the problem of high rule maintenance costs.

[0107] It should be noted that there is not necessarily a certain order among the above steps. Those of ordinary skill in the art can understand from the description of the embodiments of the present invention that in different embodiments, the above steps can have different execution orders, that is, they can be executed in parallel or exchanged, etc.

[0108] Another embodiment of the present invention provides a counterparty recognition device based on deep learning, as Figure 5 shown. The device 1 includes:

[0109] An element extraction module 11, configured to obtain a bond information text, extract elements from the bond information text, and obtain the extracted element data;

[0110] A text cutting module 12, configured to cut the bond information text according to a preset structured rule index and element data to form a spot trading order information;

[0111] A model training module 13, configured to construct a prompt-based classification model, and train the classification model according to financial prior knowledge and the spot trading order information to generate a counterparty recognition model;

[0112] A text preprocessing module 14, configured to obtain a bond information text to be parsed and preprocess the bond information text to be parsed;

[0113] A counterparty recognition module 15, configured to input the preprocessed bond information text into the counterparty recognition model to obtain the type of the counterparty.

[0114] For the specific implementation manner, refer to the method embodiments, which will not be elaborated here.

[0115] Another embodiment of the present invention provides an electronic device, as Figure 6 shown. The electronic device 10 includes:

[0116] One or more processors 110 and a memory 120, Figure 6 Taking one processor 110 as an example for introduction, the processor 110 and the memory 120 can be connected through a bus or other means, Figure 6 Taking the connection through the bus as an example.

[0117] The processor 110 is used to complete various control logics of the electronic device 10. It can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a single-chip microcomputer, an ARM (Acorn RISC Machine), or other programmable logic devices, discrete gate or transistor logic, discrete hardware controls, or any combination of these components. Additionally, the processor 110 can also be any conventional processor, microprocessor, or state machine. The processor 110 can also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other such configuration.

[0118] The memory 120, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions corresponding to the deep learning-based counterparty identification method in the embodiments of the present invention. The processor 110 executes various functional applications and data processing of the device 10 by running the non-volatile software programs, instructions, and units stored in the memory 120, that is, implements the deep learning-based counterparty identification method in the above method embodiments.

[0119] The memory 120 can include a program storage area and a data storage area. Among them, the program storage area can store an operating device and application programs required for at least one function; the data storage area can store data created according to the use of the device 10, etc. In addition, the memory 120 can include high-speed random access memory and can also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 120 optionally includes a memory remotely set relative to the processor 110, and these remote memories can be connected to the device 10 through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0120] One or more units are stored in the memory 120 and, when executed by one or more processors 110, execute the deep learning-based counterparty identification method in any of the above method embodiments. For example, execute the Figure 1 method steps S100 to step S500 described above.

[0121] The embodiments of the present invention provide a non-volatile computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by one or more processors. For example, execute the Figure 1 method steps S100 to step S500 described above.

[0122] By way of example, the non-volatile storage medium can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. The volatile memory can include random access memory (RAM) as an external cache memory. By way of illustration and not limitation, RAM can be obtained in many forms such as synchronous RAM (SRAM), dynamic RAM, (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), Synchl ink DRAM (SLDRAM), and direct Rambus (Rambus) RAM (DRRAM). The disclosed memory controls or memories of the operating environment described herein are intended to include one or more of these and / or any other suitable type of memory.

[0123] Another embodiment of the present invention provides a computer program product. The computer program product includes a computer program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions that, when executed by a processor, cause the processor to execute the deep learning-based counterparty identification method of the above method embodiment. For example, execute the method steps S100 to step S500 described above. Figure 1 in the method.

[0124] The embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0125] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform. Of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the related technology, can be embodied in the form of a software product. The computer software product can exist in a computer-readable storage medium such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of each embodiment or some parts of the embodiments.

[0126] Among other things, conditional language such as "able to", "can", "may", or "could" generally aims to convey that a particular embodiment can include (while other embodiments do not include) a particular feature, element, and / or operation, unless specifically stated otherwise or otherwise understood within the context in which it is used. Thus, such conditional language generally also aims to imply that the feature, element, and / or operation is in some way required for one or more embodiments or that one or more embodiments must include logic for determining whether the feature, element, and / or operation is included or will be performed in any particular embodiment, with or without input or prompting.

[0127] What has been described herein in the specification and drawings includes examples of methods and apparatuses for providing a deep learning-based counterparty identification. Of course, it is not possible to describe every conceivable combination of elements and / or methods for the purposes of describing the various features of the present disclosure, but it will be recognized that many additional combinations and permutations of the disclosed features are possible. Thus, it is evident that various modifications can be made to the present disclosure without departing from the scope or spirit thereof. In addition, or in the alternative, other embodiments of the present disclosure may be apparent from consideration of the specification and drawings and practice of the disclosure as presented herein. The intention is that the examples presented in the specification and drawings be considered illustrative in all respects and not restrictive. Although specific terms are employed herein, they are used in a generic and descriptive sense and not for purposes of limitation.

Claims

1. A counterparty identification method based on deep learning, characterized in that, The method includes: Obtain the bond information text, extract elements from the bond information text, and obtain the extracted element data; Cut the bond information text according to the preset structured rule index and element data to form the spot transaction order information; Construct a prompt-based classification model, and train the classification model according to financial prior knowledge and the spot transaction order information to generate a counterparty identification model; Obtain the bond information text to be parsed, and preprocess the bond information text to be parsed; Input the preprocessed bond information text into the counterparty identification model to obtain the type of the counterparty; The obtaining of the bond information text, extracting elements from the bond information text, and obtaining the extracted element data includes: Obtain the bond information text, and encode the bond information text to generate a sequence text; Extract elements from the sequence text based on a deep neural network to obtain the extracted element data; The cutting of the bond information text according to the preset structured rule index and element data to form the spot transaction order information includes: Obtain the bond information text, perform element cutting on the bond information text to generate line-by-line text information and the elements included in each line of text information; Obtain the information category of each line of text information; Aggregate each line of information according to the information category to form the spot transaction order information at the order dimension; The constructing of the prompt-based classification model, and training the classification model according to financial prior knowledge and the spot transaction order information to generate a counterparty identification model includes: Construct a prompt-based classification model; Based on financial prior knowledge and the position information of the institution name, recombine the element data to generate recombined corpus; Splice the institution name to be predicted with the recombined corpus based on the prompt method to generate target corpus; Input the target corpus into the classification model, and train the classification model to generate a counterparty identification model.

2. The method according to claim 1, wherein Before the constructing of the prompt-based classification model, and training the classification model according to financial prior knowledge and the spot transaction order information to generate a counterparty identification model, it includes: Identify the spot transaction order information, and obtain the position information of the institution name in the spot transaction order information.

3. The method according to claim 2, characterized in that, The obtaining of the bond information text to be parsed, and preprocessing the bond information text to be parsed includes: Obtain the bond information text to be parsed, extract elements from the bond information text to be parsed, and obtain the extracted element data; Cut the bond information text to be parsed according to the preset structured rule index and element data to form the spot transaction order information to be parsed.

4. The method according to claim 3, wherein The inputting of the preprocessed bond information text into the counterparty identification model to obtain the type of the counterparty includes: Input the preprocessed bond information text into the counterparty identification model, and obtain the institution name output by the counterparty identification model; Obtain the type of the current visual party, and determine the counterparty type according to the type of the visual party and the institution name.

5. A counterparty recognition device based on deep learning, characterized in that, The device includes: An element extraction module, configured to obtain a bond information text, extract elements from the bond information text, and obtain the extracted element data; A text cutting module, configured to cut the bond information text according to a preset structured rule index and element data to form spot trading order information; A model training module, configured to build a prompt-based classification model, and train the classification model according to financial prior knowledge and the spot trading order information to generate a counterparty identification model; A text preprocessing module, configured to obtain a bond information text to be parsed and preprocess the bond information text to be parsed; A counterparty identification module, configured to input the preprocessed bond information text into the counterparty identification model to obtain the type of the counterparty; The obtaining of the bond information text, extracting elements from the bond information text, and obtaining the extracted element data includes: Obtaining a bond information text, and encoding the bond information text to generate a sequence text; Based on a deep neural network, extracting elements from the sequence text to obtain the extracted element data; The cutting of the bond information text according to a preset structured rule index and element data to form spot trading order information includes: Obtaining a bond information text, cutting elements of the bond information text to generate line-by-line text information and elements included in each line of text information; Obtaining the information category of each line of text information; Aggregating each line of information according to the information category to form spot trading order information at the order dimension; The building of the prompt-based classification model, and training the classification model according to financial prior knowledge and the spot trading order information to generate a counterparty identification model includes: Building a prompt-based classification model; Based on financial prior knowledge and the position information of the institution name, recombining the element data to generate a recombined corpus; Based on a prompt method, splicing the institution name to be predicted with the recombined corpus to generate a target corpus; Inputting the target corpus into the classification model, and training the classification model to generate a counterparty identification model.

6. An electronic device, characterized in that, The electronic device includes at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the deep learning-based counterparty identification method according to any one of claims 1-4.

7. A non-volatile computer-readable storage medium, characterized in that, The non-volatile computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by one or more processors, the one or more processors can be enabled to execute the deep learning-based counterparty identification method according to any one of claims 1-4.

Images