Financial business processing method and system, storage medium and computer program product

Abstract

The invention discloses a financial business processing method and system, a storage medium and a computer program product. Comprising the steps of obtaining a target financial business text input by a target object; determining a business type to which the text belongs, and analyzing the target financial business text according to a target knowledge graph corresponding to the business type in combination with a target model and a general large model corresponding to the business type to obtain a first output result and a second output result; determining a target private knowledge vector from an enterprise knowledge vector library, wherein the similarity between the target private knowledge vector and the feature vector of the target financial service text is higher than a first threshold value, and taking the target private knowledge vector as a third output result; and determining a target output result of the target financial service text according to the first output result, the second output result and the third output result. The technical problem that the accuracy of the processing result of the complex financial service is poor due to the fact that the related financial service system is difficult to process the complex financial service is solved.

Description

Technical Field

[0001] The present application relates to the technical field of financial data processing, and in particular, to a financial business processing method and system, a storage medium, and a computer program product. Background Art

[0002] With the vigorous development of global financial technology, especially the rapid progress of artificial intelligence, big data and natural language processing technology, intelligent investment advisor machines, as an important innovation in financial services, have become a key tool for financial institutions to enhance their competitiveness and expand new business models.

[0003] Currently, most intelligent investment advisors are implemented using natural language processing and machine learning technologies. NLP (Natural Language Processing) is used to understand and generate human language, so as to have natural and fluent conversations with customers, answer customer questions, and provide personalized investment advice; machine learning technology is used to learn investment rules through a large amount of historical data, predict market trends, and conduct risk assessment, so as to achieve more accurate investment decisions.

[0004] However, when faced with the complexity and professionalism of the financial field, intelligent investment advisors still reveal some limitations. For example, although the widespread application of models has enhanced the machine's natural language understanding and generation capabilities, its knowledge data illusion problem cannot be ignored, that is, the model may produce misleading analysis based on deviations or errors in the training data, which is particularly critical for the financial industry that requires precise decision-making. In addition, the model's ability to integrate institutional privatization knowledge is limited, making it difficult to meet the financial institutions' needs for personalized services.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] The embodiments of the present application provide a financial business processing method and system, a storage medium, and a computer program product to at least solve the technical problem that the relevant financial business system has difficulty in processing complex financial business, resulting in poor accuracy of processing results for complex financial business.

[0007] According to one aspect of an embodiment of the present application, a financial business processing method is provided, including: obtaining a target financial business text input by a target object; determining a business type to which the target financial business text belongs, determining a target model and a target knowledge graph corresponding to the business type from a preset model set and a preset knowledge graph set, analyzing the target financial business text according to the target knowledge graph and in combination with the target model and a preset general large model, respectively, to obtain a first output result and a second output result of the target financial business text; performing feature extraction on the target financial business text to obtain a corresponding feature vector, and determining a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value from a preset enterprise knowledge vector library, and using the target private knowledge vector as a third output result of the target financial business text, wherein the enterprise knowledge database includes multiple private knowledge vectors corresponding to multiple private knowledge entries; determining a target output result of the target financial business text according to the first output result, the second output result, and the third output result.

[0008] Optionally, determining the business type to which the target financial business text belongs includes: preprocessing the target financial business text, wherein the preprocessing includes at least one of the following: text cleaning, word segmentation, and stemming; extracting features from the preprocessed target financial business text to obtain a corresponding feature vector; analyzing the feature vector using a pre-trained classification model to obtain the business type to which the target financial business text belongs; wherein the classification model is obtained by iteratively training an initial learning model using a first training sample set and a first sample label set, wherein the first training sample set includes a plurality of historical financial business texts as training samples, and the first sample label set includes a business type label corresponding to each historical financial business text.

[0009] Optionally, feature extraction is performed on the preprocessed financial business text to obtain a corresponding feature vector, including: determining the meaning of each word in the preprocessed financial business text, and determining the degree of association between the meaning of each word and the business type; when the degree of association between the target word and the business type is greater than a preset second threshold value, determining that the target word is a keyword; converting each target word into a word vector, and performing weighted summation of the word vectors of each target word according to the degree of association between each target word and the business type to obtain a feature vector.

[0010] Optionally, the process of constructing a business type knowledge graph includes: obtaining multiple record texts related to the business type; for each record text, preprocessing the record text, and identifying at least one entity and the relationship between each entity from the preprocessed record text; and constructing a target knowledge graph with each entity as a node and the relationship between each entity as an edge.

[0011] Optionally, the target financial business text is analyzed according to the target knowledge graph and in combination with the target model and the preset general large model to obtain a first output result and a second output result of the target financial business text, including: for the target model and any one of the models in the general large model, using the model to perform entity recognition on the target financial business text to obtain multiple target entities; matching the multiple target entities with the business type knowledge graph to obtain other entities that have a relationship with each target entity; and using a result path composed of each target entity, each other entity, and the target relationship between each target entity and other entities as the first output result of the target financial business text.

[0012] Optionally, before determining a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value from a preset enterprise knowledge vector library, the method further includes: obtaining identity information of the target object, wherein the type of the identity information includes at least one of the following: an international mobile equipment identity code, a media access control address; determining a matching result between the identity information and a preset authorized identity information library, wherein the authorized identity information library includes identity information of multiple objects that are allowed to access the enterprise knowledge vector library; when the matching result is a successful match, determining a target private knowledge vector whose similarity with the feature vector is higher than a first threshold value from the preset enterprise knowledge vector library; when the matching result is a failed match, feeding back corresponding search failure prompt information to the target object, wherein the search failure prompt information is used to prompt the target object that it does not have access rights to the enterprise knowledge vector library.

[0013] Optionally, determining a target output result of a target financial business text based on the first output result, the second output result, and the third output result includes: using a pre-trained deep learning fusion network to use a self-attention mechanism to adjust the self-attention weights of the first output result, the second output result, and the third output result, and fusing the first output result, the second output result, and the third output result according to the self-attention weights to obtain the target output result of the target financial business text.

[0014] According to another aspect of an embodiment of the present application, a financial business processing system is also provided, including: an acquisition module, used to acquire a target financial business text input by a target object; a first output result determination module, used to determine the business type to which the target financial business text belongs, determine the target model and target knowledge graph corresponding to the business type from a preset model set and a preset knowledge graph set, analyze the target financial business text according to the target knowledge graph, and respectively combine the target model and the preset general large model to obtain a first output result and a second output result of the target financial business text; a second output result determination module, used to extract features from the target financial business text to obtain a corresponding feature vector, and determine a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value from a preset enterprise knowledge vector library, and use the target private knowledge vector as a third output result of the target financial business text, wherein the enterprise knowledge database includes multiple private knowledge vectors corresponding to multiple private knowledge entries; a result feedback module, used to determine the target output result of the target financial business text according to the first output result, the second output result and the third output result.

[0015] According to another aspect of an embodiment of the present application, a non-volatile storage medium is further provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned financial business processing method by running the computer program.

[0016] According to another aspect of an embodiment of the present application, a computer program product is also provided, which includes a stored computer program, wherein the computer program implements the above-mentioned financial business processing method when executed by a processor.

[0017] In an embodiment of the present application, a target financial business text input by a target object is obtained; the business type corresponding to the target financial business text is determined, and a target model and a target knowledge graph corresponding to the business type are determined from a preset model set and a preset knowledge graph set, and the target financial business text is analyzed according to the target knowledge graph, and respectively combined with the target model and the preset general large model to obtain a first output result and a second output result of the target financial business text; feature extraction is performed on the target financial business text to obtain a corresponding feature vector, and a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value is determined from a preset enterprise knowledge vector library, and the target private knowledge vector is used as the third output result of the target financial business text; the target output result of the target financial business text is determined according to the first output result, the second output result and the third output result. The above target output result can not only utilize the breadth of large-scale text data, but also tap the depth of private knowledge, so as to provide more accurate and personalized services to the target object, thereby solving the technical problem that the relevant financial business system is difficult to handle complex financial business, resulting in poor accuracy of the processing results for complex financial business. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0019] Figure 1 It is a hardware structure block diagram of a computer terminal (or mobile device) that can be optionally used to implement the financial business processing method according to an embodiment of the present application;

[0020] Figure 2 is a flowchart of an optional financial business processing method according to an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of the structure of an optional financial business processing system according to an embodiment of the present application;

[0022] Figure 4 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0024] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] In addition, the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set up between this system and the relevant users or organizations. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0026] Example 1

[0027] According to an embodiment of the present application, an embodiment of a financial business processing method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing a financial business processing method is shown. Figure 1As shown, the computer terminal 10 (or mobile device 10) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0029] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As involved in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0030] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the financial business processing method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the financial business processing method of the above-mentioned application is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0031] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0032] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0033] In the above operating environment, Figure 2 is a flowchart of an optional financial business processing method according to an embodiment of the present application, such as Figure 2 As shown, the method at least includes steps S202-S208, wherein:

[0034] Step S202: obtaining the target financial business text input by the target object.

[0035] In the technical solution provided in step S202, the target object may be a user of the intelligent business system, and the target financial business may be business information input by the user of the intelligent business system through an interactive interface, and the user may usually input his business information in voice or text form. If the question input by the target object is in voice form, the voice information may be converted into text form using voice recognition technology to obtain the financial business text of the target object.

[0036] Step S204, determine the business type to which the target financial business text belongs, determine the target model and target knowledge graph corresponding to the business type from the preset model set and the preset knowledge graph set, analyze the target financial business text based on the target knowledge graph and in combination with the target model and the preset general large model, and obtain the first output result and the second output result of the target financial business text.

[0037] The above business type can be understood as the financial business type to which the financial business text belongs, such as financial risk management, financial investment analysis, financial market forecasting and other specific types. The target model corresponding to the business type can be understood as: a model for processing tasks in a specific field; the target knowledge graph corresponding to the business type can be understood as: a structured knowledge base constructed by relevant knowledge in a specific field. The above first output result can be in the form of a vector to facilitate subsequent result fusion.

[0038] In the technical solution provided in step S204, since the general large model often consumes a lot of resources and maintenance costs when processing tasks in multiple fields, for this reason, the embodiment of the present application proposes that a large amount of annotated data in a specific field can be used in advance to train the small machine learning models corresponding to each field, and the general large model (such as the GPT large model) and the small machine learning models corresponding to each field form a model set. Subsequently, when processing tasks in a specific field, the small machine learning model corresponding to the field and the general large model are directly called, wherein the large-scale pre-training capability of the general large model enables it to understand financial texts, such as research reports, news, announcements, etc., extract key information, perform intelligent analysis, and even generate professional investment reports. The small machine learning model performs fine-grained learning and optimization for specific tasks, such as financial report analysis, and provides accurate interpretation of financial indicators and industry trend forecasts, so that it can respond faster and more accurately, reduce the processing of irrelevant information, and improve efficiency.

[0039] In addition, when the model processes a large amount of data information, because the massive data often contains a lot of noise, and the valuable information from the massive data is difficult to determine, the model training process will face the problem of knowledge data illusion, which will cause the model to over-rely on or misinterpret the deviations or errors in the training data during the training process, resulting in misleading analysis results in practical applications. For this reason, the embodiment of the present application proposes to supplement and correct the model's understanding ability by using the knowledge graph, reduce excessive reliance on large-scale text data, correct the knowledge data illusion, and improve the accuracy of the analysis results.

[0040] Therefore, the above step S204 can be understood as that after the intelligent business system identifies the business type corresponding to the target financial business text, it can select the target model and target knowledge graph corresponding to the business type from the associated multiple models (i.e., models designed for specific intentions) and pre-stored multiple knowledge graphs (knowledge graphs related to various fields), and use the selected target model and target knowledge graph to analyze the target financial business text faster and more accurately to obtain the first output result of the target financial business text.

[0041] As an optional implementation, in the technical solution provided in the above step S204, the intelligent business system may determine the business type corresponding to the target financial business text by the following method, including:

[0042] Step S11, preprocessing the target financial business text.

[0043] Among them, the above-mentioned preprocessing includes but is not limited to: cleaning data (to remove noise, outliers, and incomplete information), standardizing and formatting, word segmentation (i.e., splitting the target financial business text into independent lexical units), stemming (i.e., reducing words to their basic forms, removing tense, singular / plural changes, etc., to reduce the variation in word forms and unify the word representation), removing stop words (i.e., excluding common but low-information words such as 'of' and 'is' to reduce noise and focus on the key information in the text), etc. These preprocessing operations help improve the efficiency and accuracy of subsequent analysis.

[0044] Step S12: Extract features from the preprocessed target financial business text to obtain corresponding feature vectors.

[0045] Among them, the above-mentioned feature extraction process is to convert unstructured text into structured data for easy analysis by machine learning models. It can convert words into dense vectors through word embedding techniques (such as Word2Vec, GloVe, BERT, etc.) to retain semantic information; or, sentence vector encoding techniques can convert the entire sentence into a vector to reflect the overall semantics of the sentence.

[0046] Step S13: Analyze the feature vectors using a pre-trained classification model to obtain the business type to which the target financial business text belongs.

[0047] Specifically, the above-mentioned classification model is obtained by iteratively training an initial learning model using a first training sample set and a first sample label set. Among them, the above-mentioned initial learning model can be based on a deep learning architecture, such as a convolutional neural network (CNN), a recurrent neural network (RNN), an attention mechanism, or a Transformer model. The first training sample set includes multiple historical financial business texts as training samples, and the first sample label set includes the business type labels corresponding to each historical financial business text. By iteratively training the initial learning model using the first training sample set and the first sample label set, the trained model can identify different types of intents.

[0048] For example, if words such as 'risk assessment', 'credit default', and 'liquidity risk' frequently appear in the target financial business text, then it can be judged that the business type (intent) to which it belongs is related to financial risk management. Similarly, if keywords such as 'investment return', 'asset allocation', and'stock analysis' frequently appear in the target financial business text, then it can be judged that the business type (intent) to which it belongs is related to financial investment analysis; if keywords such as'market trend', 'economic forecast', and 'data analysis' frequently appear in the target financial business text, then it can be judged that the business type (intent) to which it belongs is closely related to financial market prediction.

[0049] It should be noted that different text types usually carry specific intentions. For example, a bank's risk management policy documents, credit assessment reports, etc., mainly serve financial risk management; investment strategy reports, stock market analysis reports serve financial investment analysis; economic forecast reports, market research reports serve financial market forecasts.

[0050] Optionally, in the technical solution provided in the above step S12, the method may include:

[0051] Step 1: Determine the meaning of each word in the preprocessed target financial business text, and determine the degree of correlation between the meaning of each word and the business type.

[0052] Specifically, the intelligent business system can first use dictionaries, corpus statistics or deep learning models (such as ELMo, BERT, etc.) to determine the specific meaning of each word in the context; then enrich the semantic similarity and keyword matching to determine the degree of association between the meaning of each word and the user's business type. For example, words such as "avoid", "control", and "monitor" are related to financial risk management; words such as "invest", "buy", and "sell" are linked to financial investment analysis; and words such as "forecast", "analyze", and "estimate" are related to financial market forecasts.

[0053] Step 2: When the correlation between the target word and the business type is greater than a preset second threshold value, the target word is determined to be a key word, that is, a word that is crucial for understanding the input financial business text.

[0054] Step 3: Convert each target word into a word vector, and perform weighted summation of the word vectors of each target word according to the degree of association between each target word and the business type to obtain a feature vector.

[0055] Among them, the above-mentioned word vector is a mathematical representation method for representing words as multidimensional vectors, which can capture the semantic information of words and their distribution characteristics in the corpus. Common word vector generation technologies include Word2Vec, GloVe or FastText. Therefore, based on the word vectors of key words, the intelligent business system can perform a weighted summation operation to obtain the final feature vector, where the weight coefficient is the quantitative value of the degree of association between each key word and the business type, ensuring that the feature vector more accurately represents the core semantic information of the target financial business text and accurately maps the user's intention. It should be noted that the sum of the weights of the word vectors of all target words is 1.

[0056] Optionally, the process of constructing the above target knowledge graph includes:

[0057] Step 1: Get multiple record texts related to the business type.

[0058] Specifically, all public and anonymized historical texts related to the business type may be obtained from a preset corpus, or all public and anonymized texts related to the business type may be obtained from the Internet.

[0059] Step 2: For each record text, preprocess the record text, and identify at least one entity and the relationship between the entities from the preprocessed record text.

[0060] Specifically, for each record text obtained as mentioned above, the record text can be preprocessed first, including but not limited to: text cleaning (removing irrelevant characters, correcting spelling errors), word segmentation (dividing the text into vocabulary units), part-of-speech tagging (determining the grammatical function of each word), etc., to improve the efficiency and accuracy of subsequent processing; then, use named entity recognition technology to identify at least one entity and the relationship between entities from the preprocessed record text.

[0061] Step 3: Build the target knowledge graph with each entity as a node and the relationship between entities as an edge.

[0062] It should be noted that in order to ensure the timeliness and accuracy of the graph, the intelligent business system can update the graph content regularly or dynamically in real time according to the actual application scenario, including adding new entities, modifying entity relationships, deleting expired information, etc. In addition, the intelligent system can analyze the strength and importance of the relationship between each entity and set corresponding weights for each edge in the knowledge graph, which helps to correct the relationship misunderstandings that may occur when large models process texts and improve the accuracy of analysis.

[0063] For example, taking the financial knowledge graph related to financial knowledge as an example, its construction process may include: first, obtaining relevant public, anonymized information texts (unstructured data) from financial web pages, news, reports, financial statements and other resources through distributed crawling; then, preprocessing these information texts, and using the improved domain named entity recognition algorithm, namely the Bi-LSTM (bidirectional long short-term memory network) + CRF (conditional random field) model to identify the entities in each information text, and extract the relationship between each entity through the relationship extraction algorithm. These entities are usually key concepts in the financial field, such as company names, stock codes, economic indicators, etc., and the relationship between each entity can be affiliation, competition, causality, etc.; finally, each entity is used as a node in the graph, and the relationship between entities is used as an edge to form a structured data representation, namely the knowledge graph.

[0064] In particular, the reason for using the domain named entity recognition algorithm in the above process is that named entity recognition is a key technology when constructing a knowledge graph corresponding to a business type, especially in the financial field, where high-quality labeled data is often difficult to obtain, especially in specific business scenarios. This limits the training efficiency and accuracy of the model, and may not be able to capture certain complex and important financial patterns. To this end, the embodiment of the present application proposes that the conditional random field CRF and the bidirectional long short-term memory network algorithm can be used as a basis to improve the recognition algorithm. Specifically:

[0065] (1) For the problem of less annotated corpus, active learning can be used to select the most valuable unlabeled data for manual recognition, thereby effectively improving the model performance when less manually annotated corpus is added.

[0066] (2) When there is little training data and the neural network training is insufficient, use the unlabeled text in the field to pre-train word vectors and use them as the input of the Bi-LSTM (bidirectional long short-term memory network) + CRF (conditional random field) model. Unlabeled text often exists in large quantities in the financial field. By fully mining the information in these texts, the lack of training data can be compensated. Pre-trained word vectors can capture the semantic and grammatical features of words in a specific field, providing richer information for the model.

[0067] Furthermore, the intelligent business system can obtain the first output result of the target model and the second output result of the general large model according to the following method. For the convenience of description, the target model and the general large model are collectively referred to as "models", and the first output result and the second output result are collectively referred to as "results". Specifically, the method includes the following steps:

[0068] Step S21, using the model to perform entity recognition on the target financial business text to obtain multiple target entities.

[0069] Step S22, matching multiple target entities with the knowledge graph corresponding to the business type to obtain other entities that have a relationship with each target entity.

[0070] Step S23, taking the result path composed of each target entity, each other entity, and the target relationship between each target entity and other entities as the result of the target financial business text.

[0071] The technical solution provided by the above steps S21-S23 can be understood as the intelligent business system calling the model to identify the domain entities of the target financial business text; then, the system uses graph algorithms (such as the shortest path algorithm, community discovery algorithm, clustering algorithm, etc.) to find other entities that have relationships with each target entity and the target relationship between each target entity and each other entity in the knowledge graph corresponding to the business type; finally, a result path consisting of the target entity, other entities and the target relationship between them is constructed, which clearly shows the result composition of the problem. By combining the model with the knowledge graph, it not only improves the intelligent business system's ability to understand and process specific domain knowledge, reduces the model's dependence on large-scale training data, and reduces knowledge data illusions; at the same time, it can also achieve a deep understanding and intelligent application of complex relationships and patterns, and provide users with more accurate, professional and personalized domain consultation.

[0072] For example, when the target financial business text is "Analyze the financial comparison between Company X and Company Y", since its business type is in the financial field, the question-answering model corresponding to this field is used to extract entities from the target financial business text, and the three target entities of "Company X", "Company Y" and "Financial comparison" are obtained; then, these three target entities are matched with the knowledge graph corresponding to the financial field, where the knowledge graph contains detailed financial data, historical records and analysis indicators, as well as the relationship network between companies. Therefore, through matching, we can find the nodes corresponding to "Company X", "Company Y", and "Financial Comparison", and find other nodes that have relationships with these three nodes (such as "Net Profit", "Revenue Growth Rate", etc.) and the target relationships between them (such as "Fast than", "Higher than", etc.); then, in order to ensure that the target entities and relationships do match the user's query intention, the system can also check whether "Financial Comparison" is correctly connected to the financial data of "Company X" and "Company Y" to avoid misleading results due to incorrect information in the knowledge graph or false positives in entity recognition; finally, the result path composed of the target entity, other entities and the target relationships between them, such as "Company X" has a higher net profit than "Company Y" and "Company Y" has a faster revenue growth rate than "Company X".

[0073] Step S206, perform feature extraction on the target financial business text to obtain a corresponding feature vector, and determine a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value from a preset enterprise knowledge vector library, and use the target private knowledge vector as the third output result of the target financial business text.

[0074] In the technical solution provided in step S206, it is considered that the existing general large model cannot provide users with private data within the enterprise, such as internal research reports, customer desensitized data, etc., when solving the problems raised by enterprise users, resulting in the inability to provide personalized services to users. To this end, the embodiment of the present application proposes to find a target private knowledge vector in the enterprise knowledge vector library whose similarity with the feature vector is higher than the first threshold value as the third output result. Among them, the above-mentioned enterprise knowledge vector library can be understood as a private knowledge vector library of a certain financial institution itself, so it includes multiple private knowledge items related to the financial institution and multiple private knowledge vectors corresponding to each private knowledge item (storing private knowledge items in vector form can retain the semantic information of knowledge, so that the most obscure internal knowledge can also be efficiently retrieved and understood), wherein private knowledge items include but are not limited to: internal reports, customer data, proprietary model results, etc. In addition, the process of extracting the feature vector of the target financial business text by the intelligent business system can refer to steps S11-S12, which will not be repeated here.

[0075] Specifically, the above-mentioned vector search process can be understood as follows: the intelligent business system can use a preset search algorithm (such as inverted index, approximate nearest neighbor search algorithm, etc.) to search for the private knowledge vector that is most similar to the feature vector of the target financial business text in the massive enterprise knowledge vector library, and use the most similar one or more private knowledge vectors as the third output result of the target financial business text.

[0076] In addition, when accessing the enterprise knowledge vector library, it is also necessary to follow the principles of data security and privacy protection, that is, the identity information of the target object needs to be verified to ensure that only authorized users and systems can access these private data and protect the user's sensitive information.

[0077] Therefore, before the intelligent business system determines the target private knowledge vector whose similarity with the feature vector is higher than the preset first threshold value from the preset enterprise knowledge vector library, it can also obtain the identity information of the target object, wherein the type of identity information includes but is not limited to: international mobile equipment identity code, media access control address, etc.; determine the matching result between the identity information and the preset authorized identity information library, wherein the authorized identity information library includes the identity information of multiple objects allowed to access the enterprise knowledge vector library; when the matching result is a successful match, determine the target private knowledge vector whose similarity with the feature vector is higher than the preset first threshold value from the preset enterprise knowledge vector library; when the matching result is a failed match, feedback the corresponding search failure prompt information to the target object, wherein the search failure prompt information is used to prompt the target object that it does not have access rights to the enterprise knowledge vector library.

[0078] It should be noted that the intelligent business system needs to update the enterprise knowledge vector library regularly or in real time according to business development and market changes to ensure the timeliness of privatized knowledge.

[0079] Step S208: determining a target output result of the target financial business text according to the first output result, the second output result and the third output result.

[0080] In the technical solution provided in step S208, after obtaining the first output result (public knowledge in a specific field, which can be a numerical value, classification result or a more complex structure), the second output result (public knowledge, which can be a text description or a series of vectors) and the third output result (private knowledge), the intelligent business system can also fuse these two results, so as to utilize the breadth of large-scale text data and the depth of private data, thereby providing more accurate and personalized services. Finally, the target output result obtained by fusion is fed back to the target object to realize a complete personalized question-answering process.

[0081] As an optional implementation, in the technical solution provided in the above step S208, the method may include: using a pre-trained deep learning fusion network to use a self-attention mechanism to adjust the self-attention weights of the first output result, the second output result, and the third output result, and fusing the first output result, the second output result, and the third output result according to the self-attention weights to obtain the target output result of the target financial business text.

[0082] Specifically, the above-mentioned deep learning fusion network includes a multi-layer neural network structure, and the output results (in vector form) from different sources are used as input to each layer of the network structure to map them to a unified vector representation space; then, the self-attention mechanism is used to calculate the self-attention weight of each output result. That is, for each output result (i.e., modality) corresponding to the input vector, its dot product with other modal vectors (i.e., input vectors corresponding to other output results) is calculated, and the Softmax function is applied to convert the dot product into a weight distribution, which is the self-attention weight, which reflects the degree of attention of the current modal vector to other modal vectors; then, the modal vectors are fused according to the self-attention weight to obtain a fusion vector that integrates all modal information. This process emphasizes that when fusion occurs, the network can automatically adjust the relative importance of information between different modalities according to business needs and the characteristics of input data; finally, the fusion vector representation will be further passed to the output layer of the network (such as a fully connected layer, a regression layer, or a classification layer) to generate the final target output answer (such as a financial decision recommendation or answer). In this process, the network can use the comprehensive information of the fusion vector to provide users with more personalized and accurate services.

[0083] It should be noted that by training the network to optimize the fusion process, supervised learning objectives such as input text-output results or user satisfaction ratings can be used to adjust the network parameters of the deep learning fusion network to improve the accuracy and relevance of the fusion results. In addition, backpropagation and gradient descent algorithms can be used to minimize the gap between the fusion results and the actual output or user feedback.

[0084] In the above process, by applying the self-attention mechanism to the multimodal fusion network (deep learning fusion network), the network can dynamically understand and adjust the information flow between different input modalities, thereby improving the accuracy and relevance of decision support when dealing with complex financial scenarios. This mechanism is particularly suitable for the financial field because it can handle the fusion of unstructured information (such as financial text) and structured information (such as financial data), while taking into account the time series characteristics of information, providing investors and financial institutions with more comprehensive and in-depth market insights. Through the fusion of the self-attention mechanism, the intelligent business system can process multimodal information more intelligently, reduce the impact of knowledge data illusion, and improve the reliability of decision-making.

[0085] Alternatively, the first output result, the second output result, and the third output result are fused according to preset weights to obtain the target output result of the target financial business text, wherein the preset weights can be customized according to the sources of the first output result, the second output result, and the third output result. For example, the weight of the first output result output by the small machine learning model in a specific field is set to 0.2, the weight of the second output result output by the large GPT model is set to 0.2, and the fourth weight of the third output result retrieved by vector is 0.6.

[0086] Alternatively, when the output result is a classification result or a discrete decision, a voting method can be used to determine the target output result. For example, for a binary classification problem, the second output result of the large model tends to be category A. If the third output result obtained by vector retrieval also tends to be category A, and the first output result of the small machine learning model tends to be category B, the final target output result can be determined to be category A according to the majority voting principle.

[0087] In addition, rule fusion can be performed, that is, specific rules can be formulated to fuse results from different sources. For example, if the credibility of the output result obtained by vector retrieval is higher than a certain threshold, the result will be used as the standard; if the output results of the general large model and the small machine learning model are consistent, the consistent result will be used, etc.

[0088] Through the above-mentioned diverse output result fusion methods, the output results found by vector retrieval are effectively fused with the deep understanding output of the general large model and the precise local information output of the small machine learning model, providing users with more accurate, comprehensive and unique answers to meet their knowledge needs in different scenarios.

[0089] Based on the solutions defined in the above steps S202 to S208, it can be known that the financial business processing method provided in the embodiment has the following advantages over the existing methods:

[0090] (1) The input text is analyzed by combining the knowledge graph and the model to supplement and correct the model's understanding ability, thereby helping the model to better understand the various target entities in the input text and the interactions between them, reducing excessive reliance on large-scale text data, correcting the knowledge data illusion, and improving the accuracy of the analysis results.

[0091] (2) The search algorithm is used to quickly retrieve the target private knowledge vector whose similarity with the feature vector is higher than the threshold value in the preset enterprise knowledge vector library, and the target private knowledge vector is used as the output result, thereby retaining the private data within the organization, thereby satisfying the personalized service of the target object.

[0092] (3) The target output result is obtained by locally and accurately fusing the first output result of the small machine learning model, the second output result of the general large model, and the third output result obtained by vector retrieval using a deep learning fusion network, so as to provide more accurate, comprehensive and unique results for the target object and meet the knowledge needs of users in different business scenarios.

[0093] Example 2

[0094] Based on Embodiment 1 of the present application, an embodiment of a financial business processing system is also provided, and the system executes the financial business processing method of the above embodiment when it is running. Figure 3 is a schematic diagram of the structure of an optional financial business processing system according to an embodiment of the present application, such as Figure 3 As shown, the financial business processing system at least includes an acquisition module 32, a first output result determination module 34, a second output result determination module 36 and a result feedback module 38, wherein:

[0095] An acquisition module 32 is used to acquire a target financial business text input by a target object;

[0096] The first output result determination module 34 is used to determine the business type to which the target financial business text belongs, determine the target model and target knowledge graph corresponding to the business type from the preset model set and the preset knowledge graph set, analyze the target financial business text according to the target knowledge graph and in combination with the target model and the preset general large model, and obtain the first output result and the second output result of the target financial business text;

[0097] A second output result determination module 36 is used to extract features from the target financial business text to obtain a corresponding feature vector, and determine a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value from a preset enterprise knowledge vector database, and use the target private knowledge vector as a third output result of the target financial business text, wherein the enterprise knowledge database includes a plurality of private knowledge vectors corresponding to a plurality of private knowledge items;

[0098] The result feedback module 38 is used to determine a target output result of the target financial business text according to the first output result, the second output result and the third output result.

[0099] It should be noted that the various modules in the above-mentioned financial business processing system can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0100] Example 3

[0101] According to an embodiment of the present application, a non-volatile storage medium is further provided, in which a program is stored, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the financial business processing method in Example 1.

[0102] Optionally, the device where the non-volatile storage medium is located implements the following steps by running the program: obtaining a target financial business text input by the target object; determining the business type to which the target financial business text belongs, determining a target model and a target knowledge graph corresponding to the business type from a preset model set and a preset knowledge graph set, analyzing the target financial business text based on the target knowledge graph and in combination with the target model and a preset general large model, to obtain a first output result and a second output result of the target financial business text; performing feature extraction on the target financial business text to obtain a corresponding feature vector, and determining a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value from a preset enterprise knowledge vector library, and using the target private knowledge vector as the third output result of the target financial business text, wherein the enterprise knowledge database includes multiple private knowledge vectors corresponding to multiple private knowledge entries; determining a target output result of the target financial business text based on the first output result, the second output result, and the third output result.

[0103] According to an embodiment of the present application, a computer program product is also provided, which includes a stored computer program, wherein when the computer program is executed by a processor, the financial business processing method in Example 1 is implemented.

[0104] Optionally, the computer program executes the following steps: obtaining a target financial business text input by a target object; determining a business type to which the target financial business text belongs, determining a target model and a target knowledge graph corresponding to the business type from a preset model set and a preset knowledge graph set, analyzing the target financial business text according to the target knowledge graph and in combination with the target model and a preset general large model, to obtain a first output result and a second output result of the target financial business text; performing feature extraction on the target financial business text to obtain a corresponding feature vector, and determining a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value from a preset enterprise knowledge vector library, and using the target private knowledge vector as a third output result of the target financial business text, wherein the enterprise knowledge database includes multiple private knowledge vectors corresponding to multiple private knowledge entries; determining a target output result of the target financial business text according to the first output result, the second output result, and the third output result.

[0105] According to an embodiment of the present application, a processor is also provided, which is used to run a program, wherein the financial business processing method in Example 1 is executed when the program is running.

[0106] Optionally, the following steps are executed when the program is running: obtaining a target financial business text input by a target object; determining the business type to which the target financial business text belongs, determining a target model and a target knowledge graph corresponding to the business type from a preset model set and a preset knowledge graph set, analyzing the target financial business text according to the target knowledge graph and in combination with the target model and a preset general large model, to obtain a first output result and a second output result of the target financial business text; performing feature extraction on the target financial business text to obtain a corresponding feature vector, and determining a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value from a preset enterprise knowledge vector library, and using the target private knowledge vector as a third output result of the target financial business text, wherein the enterprise knowledge database includes multiple private knowledge vectors corresponding to multiple private knowledge entries; determining a target output result of the target financial business text according to the first output result, the second output result, and the third output result.

[0107] According to an embodiment of the present application, an electronic device is also provided, wherein: Figure 4 is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application, such as Figure 4 As shown, the electronic device includes one or more processors; a memory for storing one or more programs, which, when the one or more programs are executed by the one or more processors, enables the one or more processors to run the programs, wherein the programs are configured to execute the financial business processing method in the above-mentioned embodiment 1 when running.

[0108] Optionally, the processor is configured to execute the following steps through a computer program: obtaining a target financial business text input by a target object; determining a business type to which the target financial business text belongs, determining a target model and a target knowledge graph corresponding to the business type from a preset model set and a preset knowledge graph set, analyzing the target financial business text based on the target knowledge graph and in combination with the target model and a preset general large model, to obtain a first output result and a second output result of the target financial business text; performing feature extraction on the target financial business text to obtain a corresponding feature vector, and determining a target private knowledge vector whose similarity with the feature vector is higher than a preset first threshold value from a preset enterprise knowledge vector library, and using the target private knowledge vector as the third output result of the target financial business text, wherein the enterprise knowledge database includes multiple private knowledge vectors corresponding to multiple private knowledge entries; determining a target output result of the target financial business text based on the first output result, the second output result, and the third output result.

[0109] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0110] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0111] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0112] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0113] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0114] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0115] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A financial business processing method, characterized in that: include: Get the target financial business text input by the target object; Determine the business type to which the target financial business text belongs, determine a target model and a target knowledge graph corresponding to the business type from a preset model set and a preset knowledge graph set, analyze the target financial business text according to the target knowledge graph and in combination with the target model and a preset general large model, and obtain a first output result and a second output result of the target financial business text; Performing feature extraction on the target financial business text to obtain a corresponding feature vector, and determining a target private knowledge vector having a similarity with the feature vector higher than a preset first threshold value from a preset enterprise knowledge vector database, and using the target private knowledge vector as a third output result of the target financial business text, wherein the enterprise knowledge database includes a plurality of private knowledge vectors corresponding to a plurality of private knowledge items; A target output result of the target financial service text is determined according to the first output result, the second output result and the third output result.

2. The method according to claim 1, characterized in that Determining the business type to which the target financial business text belongs includes: Preprocessing the target financial business text, wherein the preprocessing includes at least one of the following: text cleaning, word segmentation, and stemming; Performing feature extraction on the preprocessed target financial business text to obtain a corresponding feature vector; Analyzing the feature vector using a pre-trained classification model to obtain the business type to which the target financial business text belongs; The classification model is obtained by iteratively training the initial learning model using a first training sample set and a first sample label set, wherein the first training sample set includes multiple historical financial business texts as training samples, and the first sample label set includes a business type label corresponding to each of the historical financial business texts.

3. The method according to claim 2, characterized in that Performing feature extraction on the preprocessed financial business text to obtain the corresponding feature vector includes: Determining the meaning of each word in the preprocessed financial business text, and determining the degree of association between the meaning of each word and the business type; When the correlation between the target word and the business type is greater than a preset second threshold value, determining the target word as a key word; Each of the target words is converted into a word vector, and the word vectors of each of the target words are weighted and summed according to the degree of association between each of the target words and the business type to obtain the feature vector.

4. The method according to claim 1, characterized in that: The process of constructing the business type knowledge graph includes: Acquire multiple record texts related to the business type; For each of the record texts, preprocessing the record text, and identifying at least one entity and the relationship between the entities from the preprocessed record text; The target knowledge graph is constructed with each of the entities as a node and the relationship between the entities as an edge.

5. The method according to claim 4, characterized in that According to the target knowledge graph, the target financial business text is analyzed in combination with the target model and the preset general large model to obtain a first output result and a second output result of the target financial business text, including: For any one of the target model and the general large model, use the model to perform entity recognition on the target financial business text to obtain multiple target entities; Matching the multiple target entities with the business type knowledge graph to obtain other entities that have relationships with each of the target entities; The result path composed of each of the target entities, each of the other entities, and the target relationships between each of the target entities and the other entities is used as the output result of the target financial business text.

6. The method according to claim 1, characterized in that Before determining a target private knowledge vector having a similarity with the feature vector higher than a preset first threshold value from a preset enterprise knowledge vector library, the method further includes: Acquire the identity information of the target object, wherein the type of the identity information includes at least one of the following: an international mobile equipment identity code, a media access control address; Determine a matching result between the identity information and a preset authorized identity information library, wherein the authorized identity information library includes identity information of multiple objects that are allowed to access the enterprise knowledge vector library; When the matching result is a successful match, determining a target private knowledge vector having a similarity with the feature vector higher than the first threshold value from a preset enterprise knowledge vector library; In the case where the matching result is a matching failure, corresponding search failure prompt information is fed back to the target object, wherein the search failure prompt information is used to prompt the target object that it does not have access rights to the enterprise knowledge vector library.

7. The method according to claim 1, characterized in that Determining a target output result of the target financial service text according to the first output result, the second output result, and the third output result includes: The pre-trained deep learning fusion network uses a self-attention mechanism to adjust the self-attention weights of the first output result, the second output result, and the third output result, and fuses the first output result, the second output result, and the third output result according to the self-attention weights to obtain the target output result of the target financial business text.

8. A financial business processing system, characterized in that: include: An acquisition module is used to acquire a target financial business text input by a target object; A first output result determination module is used to determine the business type to which the target financial business text belongs, determine a target model and a target knowledge graph corresponding to the business type from a preset model set and a preset knowledge graph set, analyze the target financial business text according to the target knowledge graph and in combination with the target model and a preset general large model, and obtain a first output result and a second output result of the target financial business text; a second output result determination module, configured to extract features from the target financial business text to obtain a corresponding feature vector, and determine a target private knowledge vector from a preset enterprise knowledge vector database whose similarity to the feature vector is higher than a preset first threshold value, and use the target private knowledge vector as a third output result of the target financial business text, wherein the enterprise knowledge database includes a plurality of private knowledge vectors corresponding to a plurality of private knowledge items; A result feedback module is used to determine a target output result of the target financial business text according to the first output result, the second output result and the third output result.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores a computer program, wherein the device where the non-volatile storage medium is located executes the financial business processing method described in any one of claims 1 to 7 by running the computer program.

10. A computer program product, characterized in that include: A computer program, wherein when the computer program is executed by a processor, it implements the financial business processing method described in any one of claims 1 to 7.

Images