A fund product name recognition method, device, equipment, medium and product

Abstract

The application provides a fund product name recognition method, which can be applied to the technical field of natural language processing and artificial intelligence. The fund product name recognition method comprises the following steps: performing entity recognition on a to-be-processed text by using a model to obtain a fund product name; determining a possible full name corresponding to the recognized fund product name; and determining a correct full name from the possible full names based on the recognized fund product name by using the model; wherein the model is obtained by training a label data through a fine-tuning stage, the label data comprises a full name of a sample fund product, a short name of the sample fund product and a sample simulation text, the short name of the sample fund product and the sample simulation text are generated by using the model based on the full name of the sample fund product in the fine-tuning stage. The application also provides a fund product name recognition device, equipment, medium and product.

Description

Technical Field

[0001] This application relates to the fields of natural language processing and artificial intelligence, and more specifically to a method, apparatus, device, medium, and product for recognizing fund product names. Background Technology

[0002] In the financial sector, fund products are frequently mentioned in financial-related business documents such as client visit records, public opinion information, and meeting minutes. The key to accurately associating these documents with corresponding fund products lies in precisely mapping the product names appearing in the text to standard names, thereby obtaining the corresponding product codes. This process is of great significance for the management and analysis of fund products and the conduct of related business.

[0003] Traditional fund product name recognition methods often rely on large amounts of labeled data for model training, requiring significant manpower and resources for data annotation, resulting in high costs. Furthermore, when new fund products are added, traditional models typically need to be retrained to recognize the new product names, exhibiting poor generalization ability. In addition, existing models suffer from poor accuracy and stability in fund product name recognition; for example, during model inference, they are prone to the "illusion" problem, generating non-existent or incorrect fund product name associations. Summary of the Invention

[0004] In view of the above problems, according to a first aspect of this application, a method for identifying fund product names is provided. The method includes: using a model to perform entity recognition from a text to be processed to obtain a fund product name; determining possible full names corresponding to the identified fund product name; and using the model to determine the correct full name from the possible full names based on the identified fund product name. The model is obtained through a fine-tuning phase based on labeled data, the labeled data including the full name of a sample fund product, the abbreviation of a sample fund product, and sample simulated text, wherein the abbreviation of the sample fund product and the sample simulated text are generated by the model based on the full name of the sample fund product during the fine-tuning phase.

[0005] According to an embodiment of this application, the fine-tuning stage includes: taking the full name of the sample fund product as input, generating an abbreviation of the sample fund product corresponding to the full name of the sample fund product using a model; generating sample simulated text including the name of the sample fund product, wherein the name of the sample fund product includes the abbreviation of the sample fund product using the model; using the sample simulated text and the name of the sample fund product corresponding to the sample simulated text as label data, training the model to obtain the name of the sample fund product from the sample simulated text through entity recognition; and using the full name of the sample fund product and the corresponding abbreviation of the sample fund product as label data, training the model to determine the correct full name from the possible full names of the sample fund product based on the abbreviation of the sample fund product.

[0006] According to an embodiment of this application, the step of taking the full name of the sample fund product as input and generating an abbreviation of the sample fund product corresponding to the full name of the sample fund product using a model specifically includes: generating possible abbreviations based on the full name using the model; and determining the abbreviation corresponding to the full name based on the possible abbreviations when the possible abbreviations meet preset rules.

[0007] According to an embodiment of this application, the preset rule includes: a sequential subset of the possible abbreviations corresponding to the full name; and / or, the possible abbreviations corresponding to different full names do not repeat.

[0008] According to an embodiment of this application, determining the possible full name corresponding to the identified fund product name specifically includes: calculating word vectors based on the fund product name using a word embedding model, and determining the possible full name through vector similarity matching.

[0009] According to an embodiment of this application, the step of using the model to determine the correct full name from the possible full names based on the identified fund product name specifically includes: the model using a constraint decoding method to infer the correct full name from the possible full names based on the fund product name identified by the entity.

[0010] According to an embodiment of this application, in the fine-tuning stage, a total loss function is determined based on entity recognition loss, abbreviation and full name matching loss, and rule loss; the entity recognition loss is used to measure the accuracy of the model in obtaining the fund product name from the simulated text by entity recognition, the abbreviation and full name matching loss is used to measure the accuracy of the model in determining the correct full name from possible full names based on the abbreviation, and the rule loss is used to penalize cases that do not conform to the preset rules.

[0011] A second aspect of this application provides a fund product name recognition device, comprising: an entity recognition module for performing entity recognition from a text to be processed using a model to obtain the fund product name; a possible full name determination module for determining the possible full name corresponding to the recognized fund product name; and a correct full name determination module for determining the correct full name from the possible full names based on the recognized fund product name using the model; wherein the model is obtained through a fine-tuning phase based on labeled data, the labeled data including the full name of a sample fund product, the abbreviation of a sample fund product, and sample simulated text, the abbreviation of the sample fund product and the sample simulated text being generated by the model based on the full name of the sample fund product during the fine-tuning phase.

[0012] According to embodiments of this application, the fund product name recognition device may further include an abbreviation sample generation module, a sample simulated text generation module, an entity recognition training module, and an abbreviation-to-full name matching training module. The abbreviation sample generation module can be used to generate an abbreviation of the sample fund product corresponding to its full name using a model, taking the full name of the sample fund product as input. The sample simulated text generation module can be used to generate sample simulated text including the sample fund product name, where the sample fund product name includes the abbreviation of the sample fund product. The entity recognition training module can be used to train the model to perform entity recognition from the sample simulated text to obtain the sample fund product name, using the sample simulated text and the corresponding sample fund product name as label data. The abbreviation-to-full name matching training module can be used to train the model to determine the correct full name from the possible full names of the sample fund product based on the abbreviation of the sample fund product, using the full name of the sample fund product and its corresponding abbreviation as label data.

[0013] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0014] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0015] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0016] The above-described one or more embodiments have the following advantages or beneficial effects: Through the fund product name recognition method of this application, in the fine-tuning stage, the model is trained based on label data including the full name of the sample fund product, the abbreviation of the sample fund product, and sample simulated text. Furthermore, the abbreviation and sample simulated text of the sample fund product are generated by the model during the fine-tuning stage. Specifically, in the fine-tuning stage, the model generates the abbreviation and sample simulated text based on the full name of the sample fund product. Thus, in the above-described fine-tuning stage, there is no need to rely on manpower and resources for complex data labeling work, effectively reducing the cost and difficulty of data engineering; and when a new fund product is added, only the full name of the new product needs to be input as the full name of the sample fund product, and the model can generate the corresponding abbreviation and sample simulated text during the fine-tuning stage, enabling the model to recognize the new product without retraining the model, thus exhibiting strong generalization ability; by using the full name, abbreviation, and sample simulated text of the sample fund product as label data, the model's entity recognition ability and ability to determine the correct full name are trained. Through these model training steps, the stability of the model's recognition ability is improved, and the reliability of the recognition results is enhanced. In the application phase, the trained model is used to perform entity recognition from the text to be processed, obtaining the fund product name and determining the corresponding possible full names. Then, the model is used to determine the correct full name from among the possible full names based on the fund product name. In this way, the corresponding fund product can be accurately associated with texts such as customer visit records, public opinion information, and various meeting contents, so as to obtain the product code of the fund product and improve the accuracy, reliability, and generalization of the fund product name recognition results. Attached Figure Description

[0017] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0018] Figure 1 The illustration shows an application scenario diagram of the fund product name recognition method, apparatus, device, medium and product according to embodiments of this application;

[0019] Figure 2 A flowchart illustrating a fund product name identification method according to an embodiment of this application is shown schematically;

[0020] Figure 3 The flowchart illustrating the fine-tuning stage of the fund product name recognition method according to an embodiment of this application is shown in the illustration.

[0021] Figure 4 This schematically illustrates a flowchart of a method for generating an abbreviation of a sample fund product corresponding to the full name of the sample fund product, according to an embodiment of this application;

[0022] Figure 5 This schematically illustrates a structural block diagram of a fund product name recognition device according to an embodiment of this application; and

[0023] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a fund product name recognition method according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application will now 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 this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0028] First, the technical terms used in this application are explained and clarified as follows:

[0029] The fine-tuning phase refers to the additional, targeted training process using a dataset specific to a particular domain or task, based on a pre-trained large language model. This helps the general large model to better adapt to the execution of a specific task.

[0030] The application phase refers to the stage where the model is deployed in a real-world environment to solve real business problems or provide services to users. In this stage, the model receives real input data and generates corresponding output results.

[0031] Word embedding model is a core technology in natural language processing. Its core goal is to represent words or symbols as dense, low-dimensional real-valued vectors (i.e., word vectors), which is equivalent to turning words in natural language into a numerical form that computers can better understand and calculate.

[0032] Constraint decoding is a technique that forces the model output to contain or conform to specific constraints during the text generation process, thus limiting the output to a specified set of options.

[0033] In the financial sector, financial business documents such as customer visit records, public opinion information, and various meeting contents often mention fund products. Accurately associating these documents with the fund products mentioned, and accurately mapping the product names in the documents to standard names to obtain the corresponding product codes, is of great significance for the management and analysis of fund products and the development of related businesses.

[0034] However, previous technologies for recognizing fund product names typically required a large amount of labeled data provided by humans for model training, which increased costs. Furthermore, as fund products are constantly updated, the model needs to be retrained for new fund products, failing to meet the generalization requirements. In addition, current models may make mistakes during inference, such as the "hallucination" problem, leading to unstable recognition capabilities.

[0035] Based on this, embodiments of this application provide a method for recognizing fund product names. The method includes: using a model to perform entity recognition from a text to be processed to obtain a fund product name; determining possible full names corresponding to the recognized fund product name; and using the model to determine the correct full name from the possible full names based on the recognized fund product name. The model is obtained through a fine-tuning phase based on labeled data, where the labeled data includes the full name of a sample fund product, the abbreviation of a sample fund product, and sample simulated text. The abbreviation of the sample fund product and the sample simulated text are generated by the model based on the full name of the sample fund product during the fine-tuning phase.

[0036] The fund product name recognition method provided in this application, during the fine-tuning stage, trains the model based on labeled data including the full name, abbreviation, and simulated text of the sample fund product. Specifically, the abbreviation and simulated text of the sample fund product are generated by the model during the fine-tuning stage. This eliminates the need for extensive data labeling work requiring manpower and resources, effectively reducing the cost and difficulty of data engineering. Furthermore, when a new fund product is added, only the full name of the new product needs to be input as the full name of the sample fund product; the model can then generate the corresponding abbreviation and simulated text during the fine-tuning stage, enabling the model to recognize the new product without retraining, thus exhibiting strong generalization ability. By using the full name, abbreviation, and simulated text of the sample fund product as labeled data, the model's entity recognition ability and ability to determine the correct full name are trained. This model training improves the stability of the model's recognition ability and the reliability of the recognition results. In the application phase, the trained model is used to perform entity recognition from the text to be processed, obtaining the fund product name and determining the corresponding possible full names. Then, the model is used to determine the correct full name from among the possible full names based on the fund product name. In this way, the corresponding fund product can be accurately associated with texts such as customer visit records, public opinion information, and various meeting contents, so as to obtain the product code of the fund product and improve the accuracy, reliability, and generalization of the fund product name recognition results.

[0037] It should be noted that the fund product name recognition method, apparatus, device, medium, and product provided in the embodiments of this application can be used in the fields of big data technology and artificial intelligence technology, as well as in the financial field, and can also be used in a variety of fields other than the fields of big data technology, artificial intelligence technology, and the financial field. The application fields of the fund product name recognition method, apparatus, device, medium, and product provided in the embodiments of this application are not limited.

[0038] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0039] Figure 1The illustration shows an application scenario diagram of the fund product name recognition method, apparatus, device, medium and product according to embodiments of this application.

[0040] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0041] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0042] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0043] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received 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 devices.

[0044] It should be noted that the fund product name recognition method provided in this application embodiment can generally be executed by server 105. Correspondingly, the fund product name recognition device provided in this application embodiment can generally be located in server 105. The fund product name recognition method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the fund product name recognition device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0045] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0046] Figure 2 A flowchart illustrating a fund product name identification method according to an embodiment of this application is shown.

[0047] like Figure 2 As shown, the fund product name identification method may include operations S210 to S230.

[0048] In operation S210, the model is used to perform entity recognition from the text to be processed to obtain the fund product name.

[0049] In this embodiment, the text to be processed is used as input, and entity recognition is performed using a model. The text to be processed is the financial-related business text to be processed. The fund product name is obtained through entity recognition by the model. The fund product name includes the abbreviation of the fund product and, optionally, the full name of the fund product. In this way, the fund product name entity recognition of the text to be processed is realized using a large model, and the fund product name is extracted from it to facilitate the subsequent determination of the full name of the fund product (here, if the entity recognition obtains the full name of the fund product, there is no need to determine the full name later, or when determining the full name, the full name of the fund product in the system is directly matched, thereby realizing functions such as determining the product code).

[0050] It should be noted that in common financial business documents, fund product names may be in abbreviated form, full form, or a combination of abbreviated and full form.

[0051] It is understandable that it is easier to identify the full name of a fund product (usually a standard name that corresponds to the product code) than to identify the abbreviation of the fund product by physical identification. It is also relatively easy to obtain the product code based on the full name. Therefore, the case where the fund product name includes the full name will not be discussed in detail here.

[0052] In operation S220, the possible full name corresponding to the identified fund product name is determined.

[0053] In this embodiment of the application, the fund product name includes the abbreviation of the fund product. Therefore, the abbreviation of the fund product extracted by entity recognition is used to determine the corresponding possible full name, which can be one or more.

[0054] For example, in operation S220, a word embedding model can be used to calculate word vectors based on the identified fund product name, and possible full names can be determined through vector similarity matching. Here, a word embedding model is used to calculate word vectors based on the fund product name extracted from entity recognition, and possible full names are determined through vector similarity matching. Multiple possible full names constitute a list of possible full names.

[0055] In operation S230, the model is used to determine the correct full name from the possible full names based on the identified fund product name.

[0056] In this embodiment, the model extracts the fund product name based on entity recognition and infers the correct full name from possible full names. That is, the fund product name (including abbreviations and optionally the full name) corresponds to the correct full name. In this embodiment, the model achieves the precise association between text related to fund products in financial business text (the text to be processed) and the corresponding fund products.

[0057] For example, in operation S230, the model can use a constrained decoding method to infer the correct full name from the possible full names based on the fund product name identified by the entity. Here, the model uses a constrained decoding method to determine the correct full name from the possible full names (i.e., the list of possible full names, which includes at least one possible full name) based on the identified fund product name (including the abbreviation and optionally the full name).

[0058] Optionally, after operation S230, the method further includes: retrieving the corresponding product code from the system based on the correct full name. In this way, the aforementioned fund product name recognition method achieves the accurate association between texts involving fund products in financial-related business texts (which are to be processed) and the corresponding fund products. Obtaining the corresponding product code facilitates the management and analysis of fund products.

[0059] Figure 3 The flowchart illustrating the fine-tuning stage of the fund product name identification method according to an embodiment of this application is shown.

[0060] like Figure 3 As shown, the fine-tuning stage in the fund product name recognition method of this embodiment includes operations S310 to S340.

[0061] In operation S310, the full name of the sample fund product is taken as input, and the model generates an abbreviation of the sample fund product corresponding to the full name of the sample fund product. Here, the model generates the corresponding abbreviation based on the full name of the sample fund product. In this embodiment, there is no restriction on the specific abbreviation generation rules; they can be inferred by a large model or added by the user.

[0062] In this embodiment of the application, the model is constructed based on a large language model of natural language processing.

[0063] In operation S320, the model is used to generate sample simulated text including the name of the sample fund product, wherein the name of the sample fund product includes the abbreviation of the sample fund product.

[0064] It should be noted that in common financial business documents, fund product names may be in abbreviated form, full form, or a combination of abbreviated and full form.

[0065] Understandably, identifying the full name of a fund product is easier than identifying its abbreviation (it is usually the standard name, corresponding to the product code). Therefore, the case where the sample fund product name includes the full name of the sample fund product will not be discussed in detail here.

[0066] In this embodiment of the application, a model is used to generate sample simulated text. The sample simulated text can be financial business texts such as news, announcements, customer visit records, and meeting minutes. These sample simulated texts mention the name of the sample fund product, and the sample simulated text includes the name of the sample fund product, and the name of the sample fund product includes the abbreviation of the sample fund product.

[0067] In operation S330, the simulated sample text and the corresponding sample fund product name are used as label data to train the model to perform entity recognition from the simulated sample text to obtain the sample fund product name. Here, the simulated sample text and the sample fund product name within it are used as label data. During model training, this label data improves the model's ability to identify entities with fund product names from financially relevant business texts.

[0068] The sample simulated text is generated by the model in operation S320. For example, the sample simulated text is a customer visit record that mentions the name of the sample fund product; the sample simulated text is public opinion information, such as news media reports and comments, user discussions on social media platforms, etc., which involve the name of the sample fund product; the sample simulated text is various meeting content, such as meeting discussion records or meeting speeches, which involve the name of the sample fund product.

[0069] In operation S340, the full name and corresponding abbreviation of the sample fund product are used as label data. The model is trained to determine the correct full name of the sample fund product from the possible full names based on the abbreviation. By using the full name and corresponding abbreviation of the sample fund product as label data to train the model, the model's ability to select the correct full name from the possible full name options based on the abbreviation is improved.

[0070] For example, as can be seen from the aforementioned operation S310, the abbreviation of the sample fund product is generated by the model based on the full name of the sample fund product. That is, the abbreviation and the full name are matched accordingly. This can improve the model's ability to select the correct full name from multiple possible full names based on the abbreviation, without relying on human or material resources for data annotation. It achieves the effects of low cost, low difficulty and high recognition ability. Furthermore, when adding a new fund product, there is no need for additional human or material resources to retrain the model. By inputting the full name of the new fund product, the model can have the ability to recognize the new fund product.

[0071] Figure 4 The flowchart illustrates a method for generating an abbreviation of a sample fund product corresponding to the full name of the sample fund product, according to an embodiment of this application, by taking the full name of the sample fund product as input and using a model.

[0072] like Figure 4 As shown, the method in this embodiment that takes the full name of the sample fund product as input and uses the model to generate the abbreviation of the sample fund product corresponding to the full name of the sample fund product may include operations S410 to S420, and operations S410 to S420 may at least partially execute operation S310.

[0073] In operation S410, the model is used to generate possible abbreviations based on the full name.

[0074] In this embodiment, the model uses the full name of the sample fund product as input to generate possible abbreviations, which can be one or more. The possible abbreviations may or may not correctly correspond to the full name as input.

[0075] In operation S420, when the possible abbreviations meet the preset rules, the abbreviation corresponding to the full name is determined based on the possible abbreviations.

[0076] In this embodiment, the preset rule can be a pre-defined restriction rule. The specific content of the preset rule can be designed according to actual needs, such as limiting the number of characters in the abbreviation, limiting the word order of the abbreviation, etc. One or more of these rules can be selected as specific preset rules according to actual needs. When the possible abbreviation generated in operation S410 meets the preset rule, the abbreviation is determined based on the possible abbreviation. Here, the possible abbreviation can usually be directly used as the abbreviation; or, one or more possible abbreviations can be determined as the abbreviation from multiple possible abbreviations that meet the preset rule. In this way, the model first generates possible abbreviations based on the full name, determines whether the possible abbreviations meet the preset rule, and then determines the correct corresponding abbreviation when the possible abbreviations meet the preset rule, so that the generated abbreviation better meets the preset rule.

[0077] Furthermore, the preset rules include: a sequential subset of the possible abbreviations corresponding to the full name; and / or, the possible abbreviations corresponding to different full names do not overlap.

[0078] In one embodiment of this application, the preset rule is a subset of the order of possible abbreviations of the corresponding full name. That is, the order of the possible abbreviations is the same as the order of the corresponding full name, and the possible abbreviations are a subset of the order of the corresponding full name. For example, if the corresponding full name is "a certain 90-day short-term bond", according to the preset rule of this embodiment, "a certain 90-day short-term bond" satisfies the preset rule, while "a certain short-term bond" does not satisfy the preset rule because the order of the words is different from the order of the corresponding full name.

[0079] In another embodiment of this application, the preset rule is that the possible abbreviations corresponding to different full names do not overlap, that is, the possible abbreviations corresponding to different full names (different fund products) do not overlap. For example, the full name of the first fund product is "a certain 90-day holding short-term bond", which may be abbreviated as "a certain 90-day short-term bond", and the full name of the second fund product is "a certain 30-day holding short-term bond", which may be abbreviated as "a certain 30-day short-term bond". In this case, the possible abbreviations corresponding to different full names do not overlap, thus satisfying the preset rule; however, if the full name of the first fund product is "a certain 90-day holding short-term bond", which may be abbreviated as "a certain short-term bond", and the full name of the second fund product is "a certain 30-day holding short-term bond", which may be abbreviated as "a certain short-term bond", then the possible abbreviations corresponding to different full names overlap, thus not satisfying the preset rule.

[0080] In another embodiment of this application, the preset rule is a sequential subset of possible abbreviations of the corresponding full names, and the possible abbreviations corresponding to different full names do not overlap. The specific rules can be referred to in the foregoing embodiments, and will not be repeated here.

[0081] In this way, the pre-defined rules for the sequential subsets conform to the patterns of fund product names, which helps the model generate more accurate abbreviations and thus improves the model's recognition ability after training. The pre-defined rule that the possible abbreviations corresponding to different full names do not repeat avoids the problem of the model-generated abbreviations lacking recognizability and being easily confused.

[0082] After operation S410, the following steps are also included: S430, when the possible abbreviation does not meet the preset rules, the possible abbreviation and the preset rules are fed back to the model, and the model is made to re-infer the possible abbreviation by adjusting the prompt process until a possible abbreviation that meets the preset rules is obtained.

[0083] In this embodiment, by repeatedly feeding back the results and pre-defined rules, and by dynamically adjusting the prompting process, the model can stably infer and generate possible abbreviations that conform to the pre-defined rules, thereby determining the abbreviation corresponding to the full name of the sample fund product.

[0084] It should be noted that financial-related business documents often contain a mixture of the full and abbreviations of fund products. To make the fund product name recognition method of this application more suitable for recognizing various product names, in this embodiment, the fund product name also includes the full name of the sample fund product.

[0085] Accordingly, in operation S320, the model is used to generate sample simulated text including the name of the sample fund product, wherein the name of the sample fund product includes the abbreviation and full name of the sample fund product.

[0086] In this way, the sample simulated text generated by the model is more consistent with the financial business texts encountered in real-world applications. Therefore, using the sample simulated text for model training can more effectively and comprehensively improve the model's recognition ability.

[0087] Furthermore, in the fine-tuning stage, a total loss function is determined based on entity recognition loss, abbreviation and full name matching loss, and rule loss; the entity recognition loss is used to measure the accuracy of the model in obtaining the fund product name from the simulated text by entity recognition, the abbreviation and full name matching loss is used to measure the accuracy of the model in determining the correct full name from possible full names based on the abbreviation, and the rule loss is used to penalize cases that do not conform to the preset rules.

[0088] In this embodiment, the total loss function determined based on entity recognition loss, abbreviation and full name matching loss, and rule loss is as follows:

[0089] ;

[0090] Where L is the total loss, For entity recognition loss, Loss of matching abbreviations with full names Loss due to rules, , , For weighting coefficients, and The proportion of each part of the loss in the total loss can be adjusted by adjusting the weight coefficients.

[0091] Specifically, The cross-entropy loss function is used to measure the accuracy of the model in extracting fund product names from entity recognition in text. The formula is as follows:

[0092] ;

[0093] Where N is the number of entity categories, For real labels, Predict probabilities for the model.

[0094] Specifically, The cross-entropy loss function is used to measure the accuracy of the model in determining the correct full name from possible full names based on the abbreviation. The formula is as follows:

[0095] ;

[0096] Where M is the number of possible universal options. To truly match the tags, Predict the matching probability for the model.

[0097] Specifically, The formula is used to penalize situations that do not conform to preset rules:

[0098] ;

[0099] Where S is the order matching degree between the abbreviation and the full name of the fund product (if the abbreviation is a subset of the order of the full name, then S=1; otherwise, S is a value between 0 and 1, which is determined according to the matching degree). This represents the penalty coefficient for the rule.

[0100] In this way, by designing the total loss function, entity recognition loss, abbreviation-to-full name matching loss, and rule loss are integrated, allowing the model to simultaneously focus on multiple key tasks during training: entity recognition, abbreviation-to-full name matching, and pre-defined rules. Simultaneously, a rule loss is introduced for the pre-defined rule that the abbreviation is a subset of the full name, penalizing cases where the pre-defined rule is not met. This guides the model to follow the pre-defined rules during learning, reducing erroneous outputs that do not conform to the rules, effectively suppressing the "illusion" problem, and improving the accuracy and standardization of the model's ability to recognize fund product names. Furthermore, the weight coefficients... , , The model can be dynamically adjusted based on different training stages and actual performance. For example, in the early stages of model training, the weight of the rule loss can be appropriately increased to help the model master the basic preset rules as soon as possible; in the later stages of model training, the weight coefficients can be adjusted based on the performance of entity recognition and abbreviation / full name matching tasks to make the model converge towards a better direction.

[0101] According to the fund product name recognition method provided in this application, during the fine-tuning stage, the model is trained based on labeled data including the full name, abbreviation, and simulated text of the sample fund products. Specifically, the abbreviation and simulated text of the sample fund products are generated by the model during the fine-tuning stage. In this way, model training does not rely on extensive data labeling work, effectively reducing the cost and difficulty of data engineering. When a new fund product is added, only the full name of the new product needs to be input, and the model can generate the corresponding abbreviation, enabling the model to recognize the new product without retraining, thus exhibiting strong generalization ability. By using the full name, abbreviation, and simulated text of the sample fund products as labeled data, the model's entity recognition ability and ability to determine the correct full name are trained. This model training improves the stability of the model's recognition ability and the reliability of the recognition results. In the application phase, the trained model is used to perform entity recognition from the text to be processed, obtaining the fund product name and determining the corresponding possible full names. Then, the model is used to determine the correct full name from among the possible full names based on the fund product name. In this way, the corresponding fund product can be accurately associated with texts such as customer visit records, public opinion information, and various meeting contents, so as to obtain the product code of the fund product and improve the accuracy, reliability, and generalization of the fund product name recognition results.

[0102] Specifically, the fund product name identification method provided in this application brings the following beneficial effects:

[0103] 1. In the fine-tuning phase, by designing preset rules, the possible abbreviations generated by the model based on the full names of sample fund products can meet the preset rules. When a possible abbreviation meets the preset rules, the corresponding abbreviation is determined based on the possible abbreviation, eliminating the need for data annotation and reducing the cost and difficulty of data engineering. The preset rules guide the model towards a better convergence. The preset rules include: a possible abbreviation being a sequential subset of the corresponding full name, and / or, the possible abbreviations corresponding to different full names not being repeated. The preset rule of sequential subsets conforms to the patterns of fund product names, which helps the model generate more accurate abbreviations, thereby improving the model's recognition ability after training. The preset rule of non-repetition between possible abbreviations corresponding to different full names avoids the problem of the model-generated possible abbreviations lacking recognizability and being easily confused.

[0104] 2. In the application phase, the fund product name is obtained by entity recognition in the text to be processed, and then the corresponding possible full name is determined. Based on the identified fund product name, the correct full name is determined from the possible full names. This achieves the function of accurately associating the text involving fund products in the financial business text to be processed with the corresponding fund products. It enables the identification of fund products in financial business texts such as customer visit records, public opinion information, and various meeting contents in the financial field, which is beneficial to the management and analysis of fund products and the development of related businesses.

[0105] 3. In the fine-tuning phase, a total loss function is designed to integrate entity recognition loss, abbreviation and full name matching loss, and rule loss, enabling the model to focus on multiple key tasks simultaneously during training. Simultaneously, a rule loss is introduced to address the pre-defined rule that the abbreviation of a fund product is a subset of the full name. This rule penalizes cases where the pre-defined rule is not met, guiding the model to follow it during learning, reducing erroneous outputs that do not conform to the pre-defined rule, effectively suppressing the "illusion" problem, and improving the accuracy and standardization of the model's product name recognition ability. Furthermore, the weight coefficients α, β, and γ can be dynamically adjusted according to different training stages and actual performance, which is beneficial for the model to converge towards a better direction.

[0106] Based on the aforementioned fund product name recognition method, this application also provides a fund product name recognition device. The following will be combined with... Figure 5 The device is described in detail.

[0107] Figure 5 The diagram illustrates the structure of a fund product name recognition device according to an embodiment of this application.

[0108] like Figure 5As shown, the fund product name recognition device 500 in this embodiment includes an entity recognition module 510, a possible full name determination module 520, and a correct full name determination module 530.

[0109] The entity recognition module 510 is used to perform entity recognition from the text to be processed using a model to obtain the fund product name. The model is trained on labeled data through a fine-tuning phase. The labeled data includes the full name of the sample fund product, the abbreviation of the sample fund product, and a simulated sample text. The abbreviation and the simulated sample text are generated by the model based on the full name of the sample fund product during the fine-tuning phase. In one embodiment, the entity recognition module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0110] The possible full name determination module 520 is used to determine the possible full name corresponding to the identified fund product name. In one embodiment, the possible full name determination module 520 can be used to perform the operation S220 described above, which will not be repeated here.

[0111] The correct full name determination module 530 is used to determine the correct full name from the possible full names based on the identified fund product name using the model. In one embodiment, the correct full name determination module 530 can be used to perform the operation S230 described above, which will not be repeated here.

[0112] According to embodiments of this application, the fund product name recognition device may further include an abbreviation sample generation module, a sample simulated text generation module, an entity recognition training module, and an abbreviation and full name matching training module.

[0113] The abbreviation sample generation module can be used to take the full name of the sample fund product as input and generate an abbreviation of the sample fund product corresponding to the full name of the sample fund product using a model. In one embodiment, the abbreviation sample generation module can be used to perform the operation S310 described above, which will not be repeated here.

[0114] The sample simulation text generation module can be used to generate sample simulation text including the name of the sample fund product using the model, wherein the sample fund product name includes the abbreviation of the sample fund product. In one embodiment, the sample simulation text generation module can be used to perform the operation S320 described above, which will not be repeated here.

[0115] The entity recognition training module can be used to train the model to perform entity recognition from the sample simulated text and the sample fund product name corresponding to the sample simulated text as label data. In one embodiment, the entity recognition training module can be used to perform the operation S330 described above, which will not be repeated here.

[0116] The abbreviation and full name matching training module can be used to train the model to determine the correct full name from the possible full names of the sample fund products based on the abbreviation of the sample fund products, using the full name and the corresponding abbreviation of the sample fund product as label data. In one embodiment, the abbreviation and full name matching training module can be used to perform the operation S340 described above, which will not be repeated here.

[0117] According to embodiments of this application, any multiple modules among the entity recognition module 510, the possible full name determination module 520, and the correct full name determination module 530 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the entity recognition module 510, the possible full name determination module 520, and the correct full name determination module 530 can be at least partially implemented as hardware circuitry, 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-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the entity recognition module 510, the possible full name determination module 520, and the correct full name determination module 530 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0118] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a fund product name recognition method according to an embodiment of this application.

[0119] like Figure 6As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0120] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0121] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0122] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0123] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0124] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the model-based fine-tuning product name recognition method provided in the embodiments of this application.

[0125] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0126] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0127] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0128] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0129] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0130] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A method for identifying fund product names, characterized in that, The method includes: The model is used to perform entity recognition from the text to be processed to obtain the fund product name; Determine the possible full name corresponding to the identified fund product name; and, The model is used to determine the correct full name from among the possible full names based on the identified fund product name; The model is trained on labeled data through a fine-tuning phase. The labeled data includes the full name of the sample fund product, the abbreviation of the sample fund product, and the sample simulated text. The abbreviation of the sample fund product and the sample simulated text are generated by the model based on the full name of the sample fund product during the fine-tuning phase. The fine-tuning phase includes: The model generates an abbreviation of the sample fund product corresponding to its full name, using the full name of the sample fund product as input. The model is used to generate sample simulated text including the name of the sample fund product, wherein the name of the sample fund product includes the abbreviation of the sample fund product; Using the simulated sample text and the corresponding sample fund product name as label data, the model is trained to perform entity recognition from the simulated sample text to obtain the sample fund product name; and... Using the full name and the corresponding abbreviation of the sample fund product as label data, the model is trained to determine the correct full name from the possible full names of the sample fund product based on the abbreviation of the sample fund product. The process of taking the full name of a sample fund product as input and using the model to generate an abbreviation of the sample fund product corresponding to the full name of the sample fund product includes: The model is used to generate possible abbreviations based on the full name; When the possible abbreviations satisfy preset rules, the abbreviation corresponding to the full name is determined based on the possible abbreviations; If the possible abbreviation does not meet the preset rules, the possible abbreviation and the preset rules are fed back to the model, and the model is made to re-infer the possible abbreviation by adjusting the prompt process until a possible abbreviation that meets the preset rules is obtained.

2. The method according to claim 1, characterized in that, The preset rules include: The possible abbreviations are sequential subsets of the corresponding full names; and / or, the possible abbreviations corresponding to different full names do not overlap.

3. The method according to claim 1, characterized in that, The determination and identification of the possible full names corresponding to the fund product names specifically includes: The word embedding model is used to calculate word vectors based on the fund product name, and the possible full name is determined by vector similarity matching.

4. The method according to claim 1, characterized in that, The method of using the model to determine the correct full name from the possible full names based on the identified fund product name specifically includes: The model uses a constrained decoding method to infer the correct full name from the possible full names based on the fund product name identified by the entity.

5. The method according to claim 2, characterized in that, In the fine-tuning phase, a total loss function is determined based on entity recognition loss, abbreviation and full name matching loss, and rule loss. The entity recognition loss is used to measure the accuracy of the model in obtaining the fund product name from the simulated text by entity recognition. The abbreviation and full name matching loss is used to measure the accuracy of the model in determining the correct full name from possible full names based on the abbreviation. The rule loss is used to penalize cases that do not conform to the preset rules.

6. A fund product name recognition device, characterized in that, The device includes: The entity recognition module is used to perform entity recognition from the text to be processed using a model to obtain the fund product name; A possible full name determination module is used to determine the possible full name corresponding to the identified fund product name; and, The correct full name determination module is used to determine the correct full name from the possible full names based on the identified fund product name using the model; The sample generation module is used to generate an abbreviation of the sample fund product corresponding to the full name of the sample fund product by taking the full name of the sample fund product as input and using model reasoning. The sample simulation text generation module is used to generate sample simulation text including the name of the sample fund product using the model, wherein the name of the sample fund product includes the abbreviation of the sample fund product; An entity recognition training module is used to train the model to perform entity recognition from the sample simulated text and the sample fund product name corresponding to the sample simulated text as label data to obtain the sample fund product name; The abbreviation and full name matching training module is used to use the full name of the sample fund product and the corresponding abbreviation of the sample fund product as label data to train the model to determine the correct full name of the sample fund product from the possible full names of the sample fund product based on the abbreviation of the sample fund product. The model is trained on labeled data through a fine-tuning phase. The labeled data includes the full name of the sample fund product, the abbreviation of the sample fund product, and the sample simulated text. The abbreviation of the sample fund product and the sample simulated text are generated by the model based on the full name of the sample fund product during the fine-tuning phase. The process of taking the full name of a sample fund product as input and using the model to generate an abbreviation of the sample fund product corresponding to the full name of the sample fund product includes: The model is used to generate possible abbreviations based on the full name; When the possible abbreviations satisfy preset rules, the abbreviation corresponding to the full name is determined based on the possible abbreviations; If the possible abbreviation does not meet the preset rules, the possible abbreviation and the preset rules are fed back to the model, and the model is made to re-infer the possible abbreviation by adjusting the prompt process until a possible abbreviation that meets the preset rules is obtained.

7. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 5.

9. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 5.

Images