Product name identification method, device, electronic device and medium

Abstract

Provided is a product name recognition method, device, electronic device and medium, which can be applied to the fields of big data technology and artificial intelligence technology. The method includes: obtaining a financial business text including a fund product abbreviation; using the financial business text as input, using a pre-built information extraction model to perform preliminary extraction of fund product abbreviations, and obtaining candidate fund abbreviations and corresponding matching possibilities; performing possibility screening based on the candidate fund abbreviations and corresponding matching possibilities, and obtaining preliminary fund abbreviations; obtaining a list of fund product full names including the full names of all fund products; using a name matching model, performing similarity matching between the preliminary fund abbreviations and the list of fund product full names, and obtaining matching results and similarity scores corresponding to each preliminary fund abbreviation; and obtaining the full name of the fund product contained in the financial business text based on the matching results and similarity scores.

Description

Technical Field

[0001] The present invention relates to the fields of big data technology and artificial intelligence technology, and more specifically to a product name recognition method, device, electronic device and medium. Background Art

[0002] In the field of financial services, especially in the field of investment management and analysis, accurate recognition of fund product names is extremely critical in customer records, investment advisor conversations, roadshow speeches and other related texts. These scenarios often involve a large number of fund product name mentions, and the recognition and extraction of these names are more challenging than other types of name recognition due to their conceptual and confusing nature, word-making phenomena and the randomness of abbreviations.

[0003] Furthermore, for the recognition of fund product names, the existing technologies mainly include the following methods, but each has its limitations: the first is the word vector similarity matching method. Although this method is rich in semantic representation, has high computational efficiency and strong scalability, it is limited in processing contextual information and does not fully express the semantics of rare words or vocabulary in specific fields, so it is difficult to accurately handle the particularity of fund product names; the second is the NLP model training recognition method. This method can understand the context and recognize named entities through training, but its effect is highly dependent on a large amount of labeled data. Given that the amount of fund product name data is relatively small and constantly updated, it is difficult for this method to converge to a good effect; the third is large model entity extraction. Although the large model has shown certain capabilities in entity extraction, it is difficult to effectively handle the confusing nature of fund product names due to the lack of sufficient professionalism, resulting in low recognition accuracy.

[0004] Therefore, there is currently a lack of a low-cost, generalizable, and accurate solution suitable for accurately extracting and identifying fund product names from complex financial texts. Summary of the invention

[0005] In view of the above problems, according to a first aspect of the present invention, a product name recognition method is provided, the method comprising: obtaining a financial business text including a fund product abbreviation; taking the financial business text as input, using a pre-built information extraction model to perform preliminary extraction of fund product abbreviations, and obtaining candidate fund abbreviations and corresponding matching possibilities; performing possibility screening based on the candidate fund abbreviations and corresponding matching possibilities, and obtaining preliminary fund abbreviations; obtaining a list of fund product full names including the full names of all fund products; using a name matching model, performing similarity matching between the preliminary fund abbreviations and the list of fund product full names, and obtaining matching results and similarity scores corresponding to each preliminary fund abbreviation; and obtaining the full names of the fund products contained in the financial business text based on the matching results and similarity scores.

[0006] According to some exemplary embodiments, the name matching model is constructed based on an improved minimum edit distance algorithm, which specifically includes: defining target characters, candidate characters and unfamiliar characters, wherein the target characters are represented by the set of all characters contained in the full name of the fund product, the candidate characters are represented by the set of all characters in the abbreviation of the fund product to be matched, and the unfamiliar characters are represented by characters that appear in the candidate character set but not in the target character set; defining editing costs based on the target characters, the candidate characters and the unfamiliar characters, which includes: setting the editing costs for characters that exist in both the target characters and the candidate characters to 1, setting the editing costs for unfamiliar characters to m, and setting the editing costs for characters with reversed order to n, where m and n are both empirically set values; and improving the minimum edit distance algorithm based on the editing costs to construct the name matching model.

[0007] According to some exemplary embodiments, the information extraction model is constructed based on a large language model of natural language processing, and the financial business text is used as input, and a pre-constructed information extraction model is used to perform preliminary extraction of fund product abbreviations to obtain candidate fund abbreviations and corresponding matching possibilities, specifically including: designing a prompt project based on the financial business text, including role setting, task description and output format guidance; and interacting with the information extraction model based on the prompt project to obtain candidate fund abbreviations and corresponding matching possibilities returned by the information extraction model.

[0008] According to some exemplary embodiments, the method further comprises: fine-tuning a large language model based on financial business text and corresponding entities and concepts to construct the information extraction model, wherein improving the accuracy of entity recognition and optimizing the understanding of specific financial terms are taken as fine-tuning goals.

[0009] According to some exemplary embodiments, the Start_With strategy of the chat annotation language is used to guide the output format of the information extraction model, so as to specify that the output of the information extraction model starts with a header row of a table.

[0010] According to some exemplary embodiments, obtaining the full name of the fund product contained in the financial business text based on the matching results and the similarity scores specifically includes: screening the matching results whose similarity scores exceed a first threshold to obtain a target matching result; in response to the presence of multiple target matching results, sorting the target matching results in descending order according to the similarity scores; and in response to the similarity score difference between the target matching result ranked first and the target matching result ranked second exceeding a second threshold, using the target matching result ranked first as the full name of the fund product contained in the financial business text.

[0011] According to some exemplary embodiments, the prompting project is interacted with the information extraction model to obtain the candidate fund abbreviations and the corresponding matching possibilities returned by the information extraction model, which specifically includes: inputting the prompting project into the information extraction model, using the information extraction model to analyze the context information in the financial business text to obtain the candidate fund abbreviations and correlation analysis results; and comparing the correlation analysis results with a preset correlation threshold to obtain the matching possibility.

[0012] According to a second aspect of the present invention, a product name recognition device is proposed, which includes: a financial business text acquisition module, which is used to: acquire a financial business text including a fund product abbreviation; an abbreviation extraction module, which is used to: take the financial business text as input, use a pre-built information extraction model to perform preliminary extraction of fund product abbreviations, and obtain candidate fund abbreviations and corresponding matching possibilities; a preliminary fund abbreviation screening module, which is used to: perform possibility screening based on the candidate fund abbreviations and the corresponding matching possibilities, and obtain preliminary fund abbreviations; a full name list acquisition module, which is used to: acquire a fund product full name list including the full names of all fund products; a similarity matching module, which is used to: use a name matching model to perform similarity matching between the preliminary fund abbreviations and the fund product full name list, and obtain matching results and similarity scores corresponding to each preliminary fund abbreviation; and a fund product full name acquisition module, which is used to: obtain the full name of the fund product contained in the financial business text based on the matching results and similarity scores.

[0013] According to some exemplary embodiments, the abbreviation extraction module may include a prompt engineering design unit and an interaction unit.

[0014] According to some exemplary embodiments, the prompt engineering design unit may be used to design a prompt engineering based on the financial business text, including role setting, task description, and output format guidance.

[0015] According to some exemplary embodiments, the interaction unit may be used to interact with the information extraction model based on the prompting project to obtain the candidate fund abbreviations and corresponding matching possibilities returned by the information extraction model.

[0016] According to some exemplary embodiments, the abbreviation extraction module may further include a context analysis unit and a possibility matching unit.

[0017] According to some exemplary embodiments, the context analysis unit may be used to input the prompt project into the information extraction model, and use the information extraction model to analyze the context information in the financial business text to obtain candidate fund abbreviations and correlation analysis results.

[0018] According to some exemplary embodiments, the possibility matching unit may be used to compare the correlation analysis result with a preset correlation threshold to obtain the matching possibility.

[0019] According to some exemplary embodiments, the similarity matching module may include a character definition unit, an edit cost definition unit, and a model improvement unit.

[0020] According to some exemplary embodiments, the character definition unit can be used to define target characters, candidate characters and unfamiliar characters, wherein the target characters are represented as the set of all characters contained in the full name of the fund product, the candidate characters are represented as the set of all characters in the abbreviation of the fund product to be matched, and the unfamiliar characters are represented as characters that appear in the candidate character set but not in the target character set.

[0021] According to some exemplary embodiments, the editing cost definition unit can be used to define the editing cost based on the target character, the candidate character and the unfamiliar character, including: setting the editing cost for characters that exist in both the target character and the candidate character to 1, setting the editing cost of the unfamiliar character to m, and setting the editing cost of the characters with reversed order to n, where m and n are both empirically set values.

[0022] According to some exemplary embodiments, the model improvement unit may be configured to improve a minimum edit distance algorithm based on the edit cost to construct the name matching model.

[0023] According to some exemplary embodiments, the similarity matching module may include a target matching result acquisition unit, a sorting unit, and a comparison unit.

[0024] According to some exemplary embodiments, the target matching result acquisition unit may be used to screen the matching results whose similarity scores exceed a first threshold to obtain the target matching result.

[0025] According to some exemplary embodiments, the sorting unit may be configured to sort the target matching results in descending order of similarity scores in response to the presence of multiple target matching results.

[0026] According to some exemplary embodiments, the comparison unit may be used to, in response to a similarity score difference between a target matching result ranked first and a target matching result ranked second exceeding a second threshold, use the target matching result ranked first as the full name of the fund product contained in the financial business text.

[0027] According to a third aspect of the present invention, there is provided an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the method as described above.

[0028] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the method as described above.

[0029] According to a fifth aspect of the present invention, there is provided a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method as described above is implemented.

[0030] One or more of the above embodiments have the following advantages or beneficial effects: According to the product name recognition method provided by the present invention, by dividing the entire recognition process into several stages, namely, preliminary extraction, possibility screening and similarity matching, the amount of data to be processed in each stage is effectively reduced. This staged processing can reduce unnecessary calculations, especially in the later similarity matching stage, where only a limited number of screened data sets are processed, thereby improving overall computing efficiency; automating the entire process from obtaining financial business texts to identifying and verifying the full name of fund products, thereby reducing the need for manual operations and significantly improving processing speed and accuracy; enabling users to quickly obtain the full name of fund products extracted from financial business texts, thereby achieving timely feedback, which means that users can obtain the information they need in a shorter time, enhancing the efficiency of decision support and improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The above contents and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0032] Figure 1 A schematic diagram showing an application scenario of a product name recognition method, device, equipment, and medium according to an embodiment of the present invention;

[0033] Figure 2 A flowchart of a method for identifying a product name according to an embodiment of the present invention is schematically shown;

[0034] Figure 3 A flowchart of a method for initially extracting a fund product abbreviation according to an embodiment of the present invention is schematically shown;

[0035] Figure 4 A flowchart of a method for performing context analysis of fund abbreviations using an information extraction model according to an embodiment of the present invention is schematically shown;

[0036] Figure 5 Schematically shows a flow chart of a method for constructing the name matching model according to a minimum edit distance algorithm of an embodiment of the present invention;

[0037] Figure 6 A flowchart schematically shows a method for screening the full names of fund products according to an embodiment of the present invention;

[0038] Fig. 7A A diagram schematically shows the calculation effect of constructing a name matching model using a minimum edit distance algorithm to perform first similarity matching according to an embodiment of the present invention;

[0039] Figure 7B A diagram schematically shows the calculation effect of constructing a name matching model using a minimum edit distance algorithm to perform second similarity matching according to an embodiment of the present invention;

[0040] Figure 7C A diagram schematically shows the calculation effect of using the minimum edit distance algorithm to construct a name matching model for performing third similarity matching according to an embodiment of the present invention;

[0041] Fig. 8A A diagram schematically shows the calculation effect of constructing a name matching model using an improved minimum edit distance algorithm to perform a first similarity match according to an embodiment of the present invention;

[0042] Figure 8B A diagram schematically shows the calculation effect of constructing a name matching model using an improved minimum edit distance algorithm to perform second similarity matching according to an embodiment of the present invention;

[0043] Figure 8C A diagram schematically shows the calculation effect of constructing a name matching model using an improved minimum edit distance algorithm to perform third similarity matching according to an embodiment of the present invention;

[0044] Fig. 9 A schematic diagram of a structural block diagram of a device for identifying a product name according to an embodiment of the present invention is shown;

[0045] Fig.10 The block diagram schematically shows an electronic device suitable for a product name recognition method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] Below, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of embodiments of the present invention. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of concepts of the present invention.

[0047] The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0048] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0049] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0050] In the technical solution of the present invention, the acquisition, storage and application of user personal information involved all comply with the provisions of relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good morals.

[0051] First, the technical terms described in this article are explained and described as follows.

[0052] The Levenshtein Distance algorithm is a method for calculating the similarity between two strings. It measures the minimum number of editing operations required to convert from one string to another, including inserting, deleting, and replacing characters. The basic idea of ​​this algorithm is to use dynamic programming to build a matrix, where each element in the matrix represents the minimum edit distance required to convert from a prefix of one string to a prefix of another string. The minimum edit distance is then calculated by filling the matrix and iterating.

[0053] Natural Language Processing (NLP) is an important branch of artificial intelligence that aims to enable computers to understand, interpret, and generate human language. NLP technology is widely used in various tasks, including text classification, sentiment analysis, machine translation, question-answering systems, information extraction, speech recognition, and language generation.

[0054] The Start_With strategy refers to a strategy used in the automated chat system. Its core idea is to match and trigger corresponding replies or actions based on the beginning of the message entered by the user.

[0055] In scenarios such as customer records, investment advisors, and roadshows, a large number of fund product names are mentioned. Therefore, extracting fund names can help understand the fund products that customers have mentioned or expressed interest in in past communications, and can help financial advisors or fund management companies better customize and recommend investment products that suit customer needs; understanding which fund products are highlighted in roadshows and other promotional activities can help analyze the company's market strategy and product positioning. In addition, by analyzing the conversation records between customers and investment advisors, financial service providers can identify customer preferences and provide more personalized services in future interactions.

[0056] However, if you want to accurately extract the name of a fund product from a text, it is more difficult and requires more specificity than other name recognition. First, the concept of a fund name is very strong, and it is often combined with named entities in other fields, which is very confusing; second, fund products often have coined words, and they are not universal entity names; finally, different people are arbitrary in abbreviating the same fund name. In addition, for some funds with similar names, many people often have a custom abbreviation that is not recognizable enough due to habitual reasons or limited knowledge of commonly used fund names, which can easily lead to confusion.

[0057] The above problems pose a great technical challenge to the extraction of fund product names. Conventional name extraction methods have their own shortcomings in this field:

[0058] Conventional method 1: word vector similarity matching method. This method has rich semantic representation, high computational efficiency, and strong scalability, but the context information is limited. For some rare words or words in specific fields, the semantic expression is insufficient, there is subjective influence, and it cannot handle the particularity of fund product names well.

[0059] Conventional method 2: NLP model training and recognition method. After NLP model training, the named entities can be recognized. This method has a certain ability to understand context, but pre-training requires a large amount of labeled data. The amount of fund product data itself is not large, and new fund names and abbreviations emerge in an endless stream, making it difficult to converge to a better effect.

[0060] Conventional method 3: Use a large model for entity extraction. The large model has certain entity extraction problems, but due to the lack of professionalism of the large model, it is difficult to effectively deal with the confusing nature of fund product names, which often results in low recognition accuracy.

[0061] Based on this, an embodiment of the present invention provides a product name recognition method, the method comprising: obtaining a financial business text including a fund product abbreviation; using the financial business text as input, using a pre-built information extraction model to perform preliminary extraction of fund product abbreviations, and obtaining candidate fund abbreviations and corresponding matching possibilities; performing possibility screening based on the candidate fund abbreviations and corresponding matching possibilities, and obtaining preliminary fund abbreviations; obtaining a list of fund product full names including the full names of all fund products; using a name matching model, performing similarity matching between the preliminary fund abbreviations and the list of fund product full names, and obtaining matching results and similarity scores corresponding to each preliminary fund abbreviation; and obtaining the full names of fund products contained in the financial business text based on the matching results and similarity scores. According to the product name recognition method provided by the present invention, by dividing the entire recognition process into several stages, namely preliminary extraction, possibility screening, and similarity matching, the amount of data that needs to be processed in each stage is effectively reduced. This phased processing can reduce unnecessary calculations, especially in the later similarity matching stage, by processing only a limited number of screened data sets, thereby improving overall computing efficiency; it automates the entire process from obtaining financial business texts to identifying and verifying the full names of fund products, thereby reducing the need for manual operations and significantly improving processing speed and accuracy; it enables users to quickly obtain the full names of fund products extracted from financial business texts, thereby achieving timely feedback, which means that users can obtain the information they need in a shorter time, enhancing the efficiency of decision support and improving user experience.

[0062] It should be noted that the product name recognition method, device, equipment and medium determined by the present invention can be used in the fields of big data technology and artificial intelligence technology, can also be used in the financial field, and can also be used in various fields other than the fields of big data technology and artificial intelligence technology and the financial field. The application fields of the product name recognition method, device, equipment and medium provided by the embodiments of the present invention are not limited.

[0063] In the technical solution of the present invention, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0064] Figure 1 The application scenario diagram of the product name recognition method, device, equipment, and medium according to an embodiment of the present invention is schematically shown.

[0065] like Figure 1 As shown, the application scenario 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0066] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only examples).

[0067] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0068] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0069] It should be noted that the product name recognition method provided in the embodiment of the present invention can generally be executed by the server 105. Accordingly, the product name recognition device provided in the embodiment of the present invention can generally be set in the server 105. The product name recognition method provided in the embodiment of the present invention can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the product name recognition device provided in the embodiment of the present invention can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0070] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0071] Figure 2 The flowchart of the product name recognition method according to an embodiment of the present invention is schematically shown.

[0072] like Figure 2 As shown, the product name recognition method 200 of this embodiment may include operations S210 to S260.

[0073] In operation S210, a financial business text including a fund product abbreviation is obtained. The financial business text may include customer records, investment advisor dialogues, roadshow speeches, investment reports, market analysis, etc. These texts are all business communication texts related to finance and investment, and usually cover detailed discussions, suggestions, product introductions, and market strategies related to investment products, and usually use the fund product abbreviation.

[0074] For example, client records are records of communications between clients and fund managers or sales representatives, including mentions of fund products; investment advisory records are fund recommendations or discussions involved when the investment advisor provides consulting services; and roadshow materials are introductory materials or speeches used in fund roadshows, which usually detail the features and advantages of some fund products.

[0075] In operation S220, the financial business text is used as input, and a pre-built information extraction model is used to perform preliminary extraction of fund product abbreviations to obtain candidate fund abbreviations and corresponding matching possibilities.

[0076] In an embodiment of the present invention, the information extraction model is constructed based on a large language model of natural language processing.

[0077] Figure 3 The flowchart of the method for preliminary extraction of fund product abbreviations according to an embodiment of the present invention is schematically shown.

[0078] like Figure 3 As shown, the method for preliminary extraction of fund product abbreviations in this embodiment may include operations S310 to S320, and operations S310 to S320 may at least partially execute operation S220.

[0079] In operation S310, a prompt project is designed based on the financial business text, including role setting, task description and output format guidance.

[0080] In an embodiment of the present invention, role setting can be used to set roles in a natural language processing model, mainly to enable the model to better understand the context and expected behavior of its execution tasks. For example, the role of the model can be set as "investment analyst", so that the model will interpret and process input data in this specific context.

[0081] Role setting is usually achieved through predefined contexts or guiding language expressions, which can inspire the model to apply its prior knowledge in a specific field and improve the relevance and accuracy of processing. For example, "assuming you are a fund investment person" or "assuming you are a fund marketing person".

[0082] In an embodiment of the present invention, the task description may clearly indicate the specific task that the model needs to complete. In this embodiment, the task may be to extract the abbreviation of a fund product from a financial business text.

[0083] In an embodiment of the present invention, the output format guide can specify the specific format of the model output to ensure that the output data is structured and easy to process later. For example, the model can be specified to return results in a table format.

[0084] Exemplarily, the Start_With strategy of the chat annotation language can be used to guide the output format of the information extraction model to specify that the output of the information extraction model starts with the title row of the table. Specifically, the output format can be clearly specified in the output instruction of the model. For example, when training or configuring the model, the output format is set to a table, and the titles of each column are clearly specified, such as "Fund abbreviation", "Matching possibility", etc., so that the use of the Start_With strategy can ensure that each model output content starts with the table title row, such as "| Fund abbreviation | Matching possibility |".

[0085] In addition, the prompting project can also include other content to further effectively extract and evaluate fund product abbreviations from complex and diverse financial texts. For example, the input content can be described to the large model, such as "the input is a customer visit record" or "the input is a roadshow meeting minutes".

[0086] In operation S320, the prompting project interacts with the information extraction model to obtain candidate fund abbreviations and corresponding matching possibilities returned by the information extraction model.

[0087] In an embodiment of the present invention, after the prompt engineering design is completed, the pre-built information extraction model can be used to process actual financial business texts, which involves inputting text data and related prompts (role settings, task descriptions, and output format guidance) into the model, and the model performs tasks based on these inputs.

[0088] In an embodiment of the present invention, the model will analyze the input financial business text, identify the fund product abbreviations in the text, and evaluate the matching possibility of each abbreviation based on some internal algorithm (such as text similarity analysis, entity recognition or context analysis). The results output by the model will be presented in the format specified in the prompt project, and a matching possibility level or score will be marked next to each identified fund abbreviation, indicating the possibility that the abbreviation may be a known fund product.

[0089] In an embodiment of the present invention, during the training process of the information extraction model, the large language model can be fine-tuned based on the financial business text and the corresponding entities and concepts to construct the information extraction model, wherein improving the accuracy of entity recognition and optimizing the understanding of specific financial terms are used as fine-tuning goals. During fine-tuning, model parameters can be adjusted to meet specific needs in the financial field. For example, the learning rate can be adjusted, the training cycle for a specific field can be increased, and financial-specific pre-training data can be used.

[0090] Figure 4 The flowchart of the method for performing context analysis of fund abbreviations using an information extraction model according to an embodiment of the present invention is schematically shown.

[0091] like Figure 4 As shown, the method of using the information extraction model to perform context analysis on fund abbreviations in this embodiment may include operations S410 to S420.

[0092] In operation S410, the prompt project is input into the information extraction model, and the context information in the financial business text is analyzed using the information extraction model to obtain candidate fund abbreviations and correlation analysis results.

[0093] In an embodiment of the present invention, the information extraction model can use natural language analysis techniques such as context parsing and semantic analysis to identify the fund product abbreviations mentioned in the text. Specifically, the model can evaluate the contextual relevance of each abbreviation, that is, check the context in which the abbreviation is used in the text to determine its degree of relevance to investment strategies, market conditions, or other financial concepts. Based on the context analysis, the model can generate a list of candidate fund abbreviations, accompanied by the relevance analysis results of each abbreviation, which can include relevance scores or other metrics to indicate the strength of association between each abbreviation and its context.

[0094] In operation S420, the correlation analysis result is compared with a preset correlation threshold to obtain the matching possibility.

[0095] In an embodiment of the present invention, the setting of the correlation threshold can be based on historical data analysis, professional knowledge or model training results to determine whether the correlation of a fund abbreviation is high enough to confirm its possibility as a valid match. Exemplarily, the matching possibility can be divided into three levels: high, medium and low.

[0096] Return to reference Figure 2 In operation S230, possibility screening is performed based on the candidate fund abbreviations and the corresponding matching possibilities to obtain a preliminary fund abbreviation. For example, the candidate fund abbreviations with a high matching possibility can be used as the preliminary fund abbreviation.

[0097] In operation S240, a fund product full name list including full names of all fund products is obtained. Fund registration and approval information published by financial regulatory agencies of various countries can be obtained to obtain comprehensive fund product full name resources.

[0098] In operation S250, the name matching model is used to perform similarity matching between the preliminary selected fund abbreviations and the fund product full name list to obtain a matching result and a similarity score corresponding to each preliminary selected fund abbreviation.

[0099] In an embodiment of the present invention, a minimum edit distance algorithm can be used for similarity matching. The purpose of the minimum edit distance algorithm is to find out how many strings need to be changed between two strings to become consistent. The algorithm uses a dynamic programming algorithm strategy. The problem has an optimal substructure, and the minimum distance contains the sub-minimum edit distance. Assuming there are strings a and b, where i and j are the character subscripts of a and b respectively, the shortest edit distance required to convert the first i character sequence of string a into the first j character sequence of string b is the following formula:

[0100] (1)

[0101] The similarity calculation formula is:

[0102] (2)

[0103] The algorithm has three operations: deleting characters, inserting characters, and replacing characters. Each operation brings an editing cost of cost=1. Assuming that three fund name abbreviations are extracted: 90-day short bond, 90-day medium bond, and 90-day short holding, the algorithm effect of matching the real fund name "Fengrui 90-day short bond" is as follows: Figures 7A-7C shown.

[0104] Through calculation, we know that Fig. 7A The first similarity match in , the horizontal sequence is the product abbreviation, the vertical sequence is the product full name, length(full name)>length(abbreviation), then: distance=4,radio=1-4 / max(9,5)=0.5555; Figure 7B The second similarity match in , the horizontal sequence is the product abbreviation, the vertical sequence is the product full name, length(full name)>length(abbreviation), then: distance=5,radio=1-5 / max(9,5)=0.4444; Figure 7C For the third similarity match in , the horizontal sequence is the product abbreviation, the vertical sequence is the product full name, length(full name)>length(abbreviation), then: distance=4,radio=1-5 / max(9,5)=0.4444.

[0105] From the above three cases, we can see that the original minimum edit distance algorithm has the following problems:

[0106] 1. If there are characters in the text that do not exist in the full name of the fund, the similarity should be lower, after all, the characters in the abbreviation should be extracted from the full name, but the algorithm result shows that the distance between radio (90-day short-term debt) and radio (90-day short-term holding) is very close;

[0107] 2. The order of the characters in the abbreviation in the text should be consistent with the full name. If the order is inconsistent, it is most likely not the abbreviation of the product. However, the algorithm result shows that radio (90-day short bond) and radio (90-day medium bond) are very close. In complex scenarios where the names of multiple funds are very similar, there cannot be sufficient distinction.

[0108] In the embodiment of the present invention, from the above analysis, it can be seen that the daily abbreviation of the fund name has two characteristics: first, all the characters in the abbreviation are the same as those in the full name; second, the order of the characters in the abbreviation is consistent with the full name. Therefore, based on the above characteristics, the name matching model can be constructed using the improved minimum edit distance algorithm, and the difference between the two string sequences can be calculated based on the improved minimum edit distance algorithm.

[0109] Figure 5 The flowchart of the method for constructing the name matching model according to the minimum edit distance algorithm of an embodiment of the present invention is schematically shown.

[0110] like Figure 5 As shown, the method for constructing the name matching model using the minimum edit distance algorithm of this embodiment may include operations S510 to S530, and operations S510 to S530 may at least partially execute operation S250.

[0111] In operation S510, target characters, candidate characters and unfamiliar characters are defined, wherein the target characters are represented as the set of all characters contained in the full name of the fund product, the candidate characters are represented as the set of all characters in the abbreviation of the fund product to be matched, and the unfamiliar characters are represented as characters that appear in the candidate character set but not in the target character set.

[0112] In an embodiment of the present invention, different types of characters involved in the name matching process can be clearly distinguished and defined based on the abbreviation and full name of the fund, thereby allowing the model to more accurately evaluate and process different types of character matching situations when performing matching.

[0113] In operation S520, an editing cost is defined based on the target character, the candidate characters and the unfamiliar character, including: setting the editing cost of characters that exist in both the target character and the candidate characters to 1, setting the editing cost of the unfamiliar character to m, and setting the editing cost of the reversed character to n, where m and n are both empirically set values.

[0114] In the embodiment of the present invention, for characters that exist in both the target characters and the candidate characters, the editing cost is set to 1, indicating that this is a basic editing operation; the editing cost of unfamiliar characters is set to m, reflecting the potential mismatch risk introduced by such characters. If the order of characters in the candidate and the target is reversed, the editing cost is n, which takes into account the complexity of the change of the character order. Exemplarily, the values ​​of m and n can be set to 3 and 2 respectively.

[0115] In operation S530, the minimum edit distance algorithm is improved based on the edit cost to construct the name matching model. The formula of the name matching model is as follows:

[0116] (3)

[0117] Similarity calculation:

[0118] (4)

[0119] For example, the name matching model is constructed by using the improved minimum edit distance algorithm, and the similarity matching is calculated as follows: Figure 8A to Figure 8C As shown, Fig. 8A For the first similarity match in , the horizontal sequence is the product abbreviation, the vertical sequence is the product full name, length(full name)>length(abbreviation), then: distance=4,radio=1-4 / max(9,5)=0.5555. Compared with the original algorithm, the normal fund abbreviation has no effect; Figure 8B For the second similarity match in , the horizontal sequence is the product abbreviation, the vertical sequence is the product full name, length(full name)>length(abbreviation), then: distance=4,radio=1-7 / max(9,5)=0.2222. Compared with the original algorithm, the similarity score of the fund abbreviation with unfamiliar characters is significantly reduced; Figure 8C For the third similarity match in , the horizontal sequence is the product abbreviation, the vertical sequence is the product full name, length(full name)>length(abbreviation), then: distance=4,radio=1-7 / max(9,5)=0.2222. Compared with the original algorithm, the similarity score of the fund abbreviation with inconsistent character order is significantly reduced.

[0120] Therefore, by introducing a customized editing cost to improve the traditional minimum edit distance algorithm, the algorithm can more accurately evaluate the impact of different editing operations when processing fund name matching, thereby improving the overall matching accuracy.

[0121] Return to reference Figure 2 In operation S260, based on the matching result and the similarity score, the full name of the fund product contained in the financial business text is obtained.

[0122] Figure 6 The flowchart of the method for screening the full names of fund products according to an embodiment of the present invention is schematically shown.

[0123] like Figure 6 As shown, the method for screening the full names of fund products in this embodiment may include operations S610 to S630.

[0124] In operation S610, matching results whose similarity scores exceed a first threshold are screened to obtain target matching results.

[0125] In operation S620, in response to the existence of a plurality of target matching results, the target matching results are sorted in descending order of similarity scores.

[0126] In operation S630, in response to the similarity score difference between the first-ranked target matching result and the second-ranked target matching result exceeding a second threshold, the first-ranked target matching result is used as the full name of the fund product included in the financial service text.

[0127] In the embodiment of the present invention, the second threshold is used to evaluate the difference in scores between the two results with the highest scores. If the difference between the score of the first place and the score of the second place exceeds the second threshold, the result of the first place is considered to be significantly better than the other results and can be used as the final fund full name.

[0128] According to the product name recognition method provided by the present invention, the amount of data that needs to be processed in each stage is effectively reduced by dividing the entire recognition process into several stages: preliminary extraction, possibility screening, and similarity matching. This staged processing can reduce unnecessary calculations, especially in the later similarity matching stage, where only a limited number of screened data sets are processed, thereby improving overall computing efficiency; automating the entire process from obtaining financial business texts to identifying and verifying the full name of fund products, thereby reducing the need for manual operations and significantly improving processing speed and accuracy; enabling users to quickly obtain the full name of fund products extracted from financial business texts, thereby achieving timely feedback, which means that users can obtain the information they need in a shorter time, enhancing the efficiency of decision support, and improving user experience. Specifically, the following beneficial effects are brought about:

[0129] 1. By accurately distinguishing target characters, candidate characters, and unfamiliar characters, and setting specific editing costs, the improved minimum edit distance algorithm can more accurately handle subtle differences in name matching, making it particularly effective in identifying and processing complex or similar fund names, and improving the overall matching accuracy;

[0130] 2. Through precisely designed prompt engineering, the model can better understand the task requirements and output format, thereby improving the efficiency and accuracy when processing specific queries. This strategy helps to adjust the behavior of the model in a targeted manner to adapt to specific data processing needs;

[0131] 3. Screen and confirm the final fund product name by setting and applying thresholds to ensure that only the most relevant candidates are selected as the final result. This data-driven decision-making process reduces the possibility of mismatching and ensures high accuracy;

[0132] 4. By fine-tuning the large language model in a targeted manner, the model's accuracy in identifying financial entities (such as fund names, market indices, financial institutions, etc.) can be significantly improved. This is because the fine-tuning process uses training data specific to the financial field, allowing the model to better understand the naming and expressions unique to the financial field;

[0133] 5. The Start_With strategy can ensure that the output of the information extraction model is always presented in a consistent format, such as each output starting with a table header row, thereby reducing the format adjustment and data cleaning work required during subsequent processing.

[0134] Based on the above product name recognition method, the present invention also provides a product name recognition device. Fig. 9 The device is described in detail.

[0135] Fig. 9 The structure block diagram of the device for identifying a product name according to an embodiment of the present invention is schematically shown.

[0136] like Fig. 9 As shown, according to this embodiment, the product name identification device 900 includes a financial business text acquisition module 910, an abbreviation extraction module 920, a preliminary fund abbreviation screening module 930, a full name list acquisition module 940, a similarity matching module 950 and a fund product full name acquisition module 960.

[0137] The financial service text acquisition module 910 may be used to acquire the financial service text including the fund product abbreviation. In one embodiment, the financial service text acquisition module 910 may be used to perform the operation S210 described above, which will not be described in detail here.

[0138] The abbreviation extraction module 920 can be used to take the financial business text as input, use the pre-built information extraction model to perform preliminary extraction of fund product abbreviations, and obtain candidate fund abbreviations and corresponding matching possibilities. In one embodiment, the abbreviation extraction module 920 can be used to perform the operation S220 described above, which will not be repeated here.

[0139] The preliminary fund abbreviation screening module 930 can be used to perform possibility screening based on the candidate fund abbreviations and the corresponding matching possibilities to obtain preliminary fund abbreviations. In one embodiment, the preliminary fund abbreviation screening module 930 can be used to perform the operation S230 described above, which will not be repeated here.

[0140] The full name list acquisition module 940 may be used to acquire a fund product full name list including the full names of all fund products. In one embodiment, the full name list acquisition module 940 may be used to perform the operation S240 described above, which will not be described in detail herein.

[0141] The similarity matching module 950 can be used to use a name matching model to perform similarity matching between the preliminary selected fund abbreviations and the fund product full name list to obtain a matching result and a similarity score corresponding to each preliminary selected fund abbreviation. In one embodiment, the similarity matching module 950 can be used to perform the operation S250 described above, which will not be repeated here.

[0142] The similarity matching module 960 may be used to obtain the full name of the fund product contained in the financial service text based on the matching result and the similarity score. In one embodiment, the similarity matching module 960 may be used to perform the operation S260 described above, which will not be described in detail here.

[0143] According to an embodiment of the present invention, the abbreviation extraction module 920 may include a prompt engineering design unit and an interaction unit.

[0144] The prompt engineering design unit may be used to design a prompt engineering based on the financial business text, including role setting, task description and output format guidance. In one embodiment, the prompt engineering design unit may be used to perform the operation S310 described above, which will not be described in detail here.

[0145] The interaction unit may be used to interact with the information extraction model based on the prompting project to obtain the candidate fund abbreviations and corresponding matching possibilities returned by the information extraction model. In one embodiment, the interaction unit may be used to perform the operation S320 described above, which will not be described in detail here.

[0146] According to an embodiment of the present invention, the abbreviation extraction module 920 may further include a context analysis unit and a possibility matching unit.

[0147] The context analysis unit can be used to input the prompt project into the information extraction model, and use the information extraction model to analyze the context information in the financial business text to obtain the candidate fund abbreviation and the correlation analysis result. In one embodiment, the context analysis unit can be used to perform the operation S410 described above, which will not be repeated here.

[0148] The possibility matching unit may be used to compare the correlation analysis result with a preset correlation threshold to obtain the matching possibility. In one embodiment, the possibility matching unit may be used to perform the operation S420 described above, which will not be described in detail here.

[0149] According to an embodiment of the present invention, the similarity matching module 950 may include a character definition unit, an edit cost definition unit and a model improvement unit.

[0150] The character definition unit can be used to define target characters, candidate characters and unfamiliar characters, wherein the target characters are represented by the set of all characters included in the full name of the fund product, the candidate characters are represented by the set of all characters in the abbreviation of the fund product to be matched, and the unfamiliar characters are represented by characters that appear in the candidate character set but do not appear in the target character set. In one embodiment, the character definition unit can be used to perform the operation S510 described above, which will not be described in detail here.

[0151] The editing cost definition unit can be used to define editing costs based on the target character, the candidate character and the unfamiliar character, including: setting the editing cost of characters that exist in both the target character and the candidate character to 1, setting the editing cost of the unfamiliar character to m, and setting the editing cost of the reversed character to n, where m and n are both empirically set values. In one embodiment, the editing cost definition unit can be used to perform the operation S520 described above, which will not be repeated here.

[0152] The model improvement unit may be used to improve the minimum edit distance algorithm based on the edit cost to construct the name matching model. In one embodiment, the model improvement unit may be used to perform the operation S530 described above, which will not be described in detail here.

[0153] According to an embodiment of the present invention, the similarity matching module 960 may include a target matching result acquisition unit, a sorting unit and a comparison unit.

[0154] The target matching result acquisition unit may be used to screen the matching results whose similarity scores exceed the first threshold to obtain the target matching results. In one embodiment, the target matching result acquisition unit may be used to perform the operation S610 described above, which will not be described in detail here.

[0155] The sorting unit may be used to sort the target matching results in descending order of similarity scores in response to the presence of multiple target matching results. In one embodiment, the sorting unit may be used to perform the operation S620 described above, which will not be described in detail here.

[0156] The comparison unit may be used to respond to the difference in similarity scores between the first-ranked target matching result and the second-ranked target matching result exceeding a second threshold, and use the first-ranked target matching result as the full name of the fund product contained in the financial service text. In one embodiment, the comparison unit may be used to perform the operation S630 described above, which will not be described in detail herein.

[0157] According to an embodiment of the present invention, any multiple modules among the financial service text acquisition module 910, the abbreviation extraction module 920, the preliminary fund abbreviation screening module 930, the full name list acquisition module 940, the similarity matching module 950 and the fund product full name acquisition module 960 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the financial service text acquisition module 910, the abbreviation extraction module 920, the preliminary fund abbreviation screening module 930, the full name list acquisition module 940, the similarity matching module 950 and the fund product full name acquisition module 960 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or by any one of the three implementation methods of software, hardware and firmware or by an appropriate combination of any of them. Alternatively, at least one of the financial service text acquisition module 910, the abbreviation extraction module 920, the preliminary fund abbreviation screening module 930, the full name list acquisition module 940, the similarity matching module 950 and the fund product full name acquisition module 960 can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be executed.

[0158] Fig.10 The block diagram schematically shows an electronic device suitable for a product name recognition method according to an embodiment of the present invention.

[0159] like Fig.10 As shown, the electronic device 1000 according to an embodiment of the present invention includes a processor 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage part 1008 into a random access memory (RAM) 1003. The processor 1001 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1001 may also include an onboard memory for caching purposes. The processor 1001 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0160] In RAM 1003, various programs and data required for the operation of electronic device 1000 are stored. Processor 1001, ROM 1002 and RAM 1003 are connected to each other via bus 1004. Processor 1001 performs various operations of the method flow according to the embodiment of the present invention by executing the program in ROM 1002 and / or RAM 1003. It should be noted that the program can also be stored in one or more memories other than ROM 1002 and RAM 1003. Processor 1001 can also perform various operations of the method flow according to the embodiment of the present invention by executing the program stored in the one or more memories.

[0161] According to an embodiment of the present invention, the electronic device 1000 may further include an input / output (I / O) interface 1005, which is also connected to the bus 1004. The electronic device 1000 may further include one or more of the following components connected to the I / O interface 1005: an input portion 1006 including a keyboard, a mouse, etc.; an output portion 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 1008 including a hard disk, etc.; and a communication portion 1009 including a network interface card such as a LAN card, a modem, etc. The communication portion 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage portion 1008 as needed.

[0162] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiment; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present invention is implemented.

[0163] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present invention, the computer-readable storage medium may include the ROM 1002 and / or RAM 1003 described above and / or one or more memories other than ROM 1002 and RAM 1003.

[0164] The embodiment of the present invention also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method provided by the embodiment of the present invention.

[0165] The computer program executes the above functions defined in the system / device of the embodiment of the present invention when it is executed by the processor 1001. According to the embodiment of the present invention, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0166] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 1009, and / or installed from the removable medium 1011. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0167] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1009, and / or installed from the removable medium 1011. When the computer program is executed by the processor 1001, the above functions defined in the system of the embodiment of the present invention are performed. According to the embodiment of the present invention, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0168] According to an embodiment of the present invention, the program code for executing the computer program provided by the embodiment of the present invention can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0169] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0170] The embodiments of the present invention are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present invention. Although each embodiment is described above, it does not mean that the measures in each embodiment cannot be used in combination advantageously. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.

Claims

1. A method for identifying a product name, characterized in that: The method comprises: Obtain financial business text including fund product abbreviation; Taking the financial business text as input, using a pre-built information extraction model to perform preliminary extraction of fund product abbreviations, and obtaining candidate fund abbreviations and corresponding matching possibilities, wherein the information extraction model is constructed based on a large language model of natural language processing, and the information extraction model analyzes context information in the financial business text through a preset prompting project; Perform possibility screening based on the candidate fund abbreviations and corresponding matching possibilities to obtain the initial fund abbreviations; Get a list of fund product names including the full names of all fund products; Using the name matching model, similarity matching is performed between the shortlisted fund abbreviations and the full name list of the fund products to obtain a matching result and a similarity score corresponding to each shortlisted fund abbreviation; and Based on the matching result and the similarity score, the full name of the fund product contained in the financial business text is obtained.

2. The method according to claim 1, characterized in that: The name matching model is constructed based on the improved minimum edit distance algorithm, which specifically includes: Define target characters, candidate characters and unfamiliar characters, where the target characters are the set of all characters included in the full name of the fund product, the candidate characters are the set of all characters in the abbreviation of the fund product to be matched, and the unfamiliar characters are the characters that appear in the candidate character set but not in the target character set; Defining editing costs based on the target character, the candidate characters, and the unfamiliar characters, including: setting the editing costs of characters that exist in both the target character and the candidate characters to 1, setting the editing costs of unfamiliar characters to m, and setting the editing costs of reversed characters to n, where m and n are both empirically set values; and A minimum edit distance algorithm is improved based on the edit cost to construct the name matching model.

3. The method according to claim 1 or 2, characterized in that: The method of using the financial business text as input and using a pre-built information extraction model to perform preliminary extraction of fund product abbreviations to obtain candidate fund abbreviations and corresponding matching possibilities specifically includes: Designing a prompt project based on the financial business text, including role setting, task description and output format guidance; and Based on the interaction between the prompt project and the information extraction model, the candidate fund abbreviations and corresponding matching possibilities returned by the information extraction model are obtained.

4. The method according to claim 3, characterized in that The method further comprises: The large language model is fine-tuned based on the financial business text and the corresponding entities and concepts to build the information extraction model, wherein improving the accuracy of entity recognition and optimizing the understanding of specific financial terms are set as fine-tuning goals.

5. The method according to claim 3, characterized in that: The Start_With strategy of the chat annotation language is used to guide the output format of the information extraction model, so as to specify that the output of the information extraction model starts with the header row of the table.

6. The method according to claim 1, characterized in that The obtaining, based on the matching result and the similarity score, the full name of the fund product contained in the financial business text specifically includes: Filter the matching results whose similarity scores exceed a first threshold to obtain a target matching result; In response to the existence of multiple target matching results, sorting the target matching results in descending order of similarity scores; and In response to a similarity score difference between the first-ranked target matching result and the second-ranked target matching result exceeding a second threshold, the first-ranked target matching result is used as the full name of the fund product included in the financial service text.

7. The method according to claim 3, characterized in that The step of interacting with the information extraction model based on the prompting project to obtain the candidate fund abbreviations and corresponding matching possibilities returned by the information extraction model specifically includes: Inputting the prompt project into the information extraction model, using the information extraction model to analyze the context information in the financial business text, and obtaining the candidate fund abbreviations and correlation analysis results; and The correlation analysis result is compared with a preset correlation threshold to obtain the matching possibility.

8. A product name recognition device, characterized in that: The device comprises: The financial business text acquisition module is used to: acquire the financial business text including the abbreviation of the fund product; The abbreviation extraction module is used to: take the financial business text as input, use a pre-built information extraction model to perform preliminary extraction of fund product abbreviations, and obtain candidate fund abbreviations and corresponding matching possibilities, wherein the information extraction model is constructed based on a large language model of natural language processing, and the information extraction model analyzes context information in the financial business text through a preset prompting project; A preliminary fund abbreviation screening module is used to: perform possibility screening based on the candidate fund abbreviations and the corresponding matching possibilities to obtain the preliminary fund abbreviations; A full name list acquisition module is used to: acquire a fund product full name list including the full names of all fund products; A similarity matching module is used to: use a name matching model to perform similarity matching between the shortlisted fund names and the full name list of fund products, and obtain a matching result and a similarity score corresponding to each shortlisted fund name; and The fund product full name acquisition module is used to obtain the fund product full name contained in the financial business text based on the matching result and the similarity score.

9. An electronic device, comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images