Call-out report generation method and device, storage medium and electronic equipment

Abstract

The invention provides an exhaustion report generation method and device, a storage medium and electronic equipment, and is applied to the technical field of artificial intelligence. According to the method, the key attribute information is extracted by acquiring the basic information, the risk information and the transaction information of the target customer, and the knowledge set related to the attributes is found by using the pre-constructed knowledge base and the vector index database. And then, calculating the similarity between the key attributes and the associated knowledge by using a cross coding algorithm, and sorting according to the similarity to obtain the most relevant knowledge. And the sorting result is used for constructing cue words for inputting a large language model, so that a more accurate and practical exhaustion note report is generated. According to the method and the system, the accuracy and the practicability of the exhaustion report are remarkably improved by combining specific customer information and rich associated knowledge.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a due diligence report generation method, device, storage medium and electronic device. Background Art

[0002] In financial institutions, conducting due diligence on customers is the core of anti-money laundering work. This is not only a key step in implementing the customer due diligence system, but also an important line of defense for predicting money laundering crimes and preventing financial risks. Among them, the writing of due diligence reports is a complex and tedious task, which usually includes three steps: first, collecting basic information, risk information and transaction information of customers; second, analyzing and evaluating the identity and behavior of customers, and conducting risk rating and control; finally, writing a report to describe the customer situation and complete the due diligence.

[0003] However, due to the uneven capabilities of grassroots staff, when faced with a large number of due diligence clients or complex analytical tasks, the analysis and evaluation are often not in-depth enough, the quality of due diligence reports is poor, and it is difficult to predict illegal activities in a timely manner.

[0004] Therefore, how to improve the accuracy and practicality of due diligence reports has become a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of the above problems, the present invention provides a due diligence report generation method, device, storage medium and electronic device that overcome the above problems or at least partially solve the above problems. The technical solution is as follows:

[0006] A method for generating a due diligence report, comprising:

[0007] Obtain basic customer information, customer risk information and customer transaction information of target customers;

[0008] Extracting key attribute information from the basic customer information, customer risk information and customer transaction information;

[0009] Based on the pre-built knowledge base and vector index base, obtaining an associated knowledge set corresponding to the key attribute information, wherein the associated knowledge set includes at least one associated knowledge;

[0010] Calculate the similarity between the key attribute information and each of the associated knowledge in the associated knowledge set by a cross coding algorithm;

[0011] Sorting each of the associated knowledge in descending order of similarity to obtain an associated knowledge sorting result;

[0012] Constructing a model prompt word based on the key attribute information and the associated knowledge sorting result;

[0013] The model prompt words are input into the large language model to generate a due diligence report on the target customer.

[0014] Optionally, obtaining the associated knowledge set corresponding to the key attribute information based on the pre-built knowledge base and vector index library includes:

[0015] Extracting a first preset number of first associated knowledge similar to the key attribute information from the knowledge base using the BM25 algorithm;

[0016] Extracting a second preset number of second associated knowledge similar to the key attribute information from the vector index library using a Bi-Encode algorithm;

[0017] Each of the first associated knowledge and each of the second associated knowledge are concatenated to obtain an associated knowledge set corresponding to the key attribute information.

[0018] Optionally, the calculating the similarity between the key attribute information and each of the associated knowledge in the associated knowledge set by using a cross coding algorithm includes:

[0019] Obtaining a keyword vector set of the key attribute information, wherein the keyword vector set includes keyword vectors corresponding to each keyword extracted from the key attribute information;

[0020] Obtaining a first knowledge vector set of the associated knowledge set, wherein the first knowledge vector set includes the knowledge vectors of each of the associated knowledge in the associated knowledge set;

[0021] The keyword vector set is concatenated with each of the knowledge vectors and then input into a large language model, so that the large language model can calculate the similarity between the keyword vector set and each of the knowledge vectors.

[0022] Optionally, constructing a model prompt word based on the key attribute information and the associated knowledge sorting result includes:

[0023] Substitute the keyword set corresponding to the key attribute information and the second knowledge vector set corresponding to the associated knowledge ranking result into a pre-built customer due diligence report prompt template to obtain model prompt words, wherein the keyword set includes each keyword extracted from the key attribute information, and the second knowledge vector set includes the knowledge vector of each associated knowledge in the associated knowledge ranking result.

[0024] Optionally, the knowledge base is constructed based on anti-money laundering expertise, banking domain knowledge and historical due diligence reports, and the vector index library is constructed by vectorizing all the information in the knowledge base using a text2vec model trained from the historical due diligence reports.

[0025] Optionally, the basic customer information includes the natural attributes, account attributes and social attributes of the target customer, the customer risk information includes the target customer's judicial record, violations and suspicious circumstances, and the customer transaction information includes the target customer's large-value transaction information, transaction frequency information, and transaction channel information.

[0026] Optionally, extracting key attribute information from the customer basic information, customer risk information and customer transaction information includes:

[0027] Extracting the target user's age, gender, nationality and place of residence from the natural attributes as first key attribute information;

[0028] From the account attributes, extract whether the target user has opened a RMB bank settlement account, whether the target user has opened a foreign exchange account, whether the target user has opened a free trade zone account, the number of accounts opened and closed, and the account status as the second key attribute information;

[0029] Extracting the target user's industry, company position and position information from the social attributes as the third key attribute information;

[0030] Extracting from the judicial records whether the target user has been judicially frozen, queried, or deducted by the competent authority as the fourth key attribute information;

[0031] Extracting the target user's dishonesty record and overdue record from the illegal behaviors as the fifth key attribute information;

[0032] Extracting suspicious transaction records of the target user in suspicious situations as the sixth key attribute information;

[0033] Extracting the number of large-value transactions and the proportion of large-value transactions of the target user from the large-value transaction information as the seventh key attribute information;

[0034] Extracting the target user's monthly transaction frequency, income-expenditure ratio, and main transaction time points from the transaction frequency information as the eighth key attribute information;

[0035] The main transaction channel of the target user is extracted from the transaction channel information as the ninth key attribute information.

[0036] A due diligence report generation device includes: a customer information acquisition unit, a key attribute information extraction unit, an associated knowledge set acquisition unit, a similarity calculation unit, an associated knowledge ranking result acquisition unit, a model prompt word construction unit and a due diligence report generation unit.

[0037] The customer information acquisition unit is used to obtain basic customer information, customer risk information and customer transaction information of the target customer;

[0038] The key attribute information extraction unit is used to extract key attribute information from the customer basic information, customer risk information and customer transaction information;

[0039] The associated knowledge set obtaining unit is used to obtain an associated knowledge set corresponding to the key attribute information based on a pre-built knowledge base and a vector index base, wherein the associated knowledge set includes at least one associated knowledge;

[0040] The similarity calculation unit is used to calculate the similarity between the key attribute information and each of the associated knowledge in the associated knowledge set by using a cross-coding algorithm;

[0041] The associated knowledge ranking result obtaining unit is used to sort the associated knowledge in descending order of similarity to obtain an associated knowledge ranking result;

[0042] The model prompt word construction unit is used to construct the model prompt word based on the key attribute information and the associated knowledge sorting result;

[0043] The due diligence report generating unit is used to input the model prompt words into the large language model to generate a due diligence report on the target customer.

[0044] A computer-readable storage medium having a program stored thereon, wherein when the program is executed by a processor, the due diligence report generation method described in any one of the above items is implemented.

[0045] An electronic device, comprising at least one processor, and at least one memory and a bus connected to the processor; wherein the processor and the memory communicate with each other via the bus; and the processor is used to call program instructions in the memory to execute any of the above-mentioned due diligence report generation methods.

[0046] By means of the above technical scheme, the present invention provides a method, device, storage medium and electronic device for generating a due diligence report, which obtains the basic information, risk information and transaction information of the target customer; extracts key attribute information from the basic information, risk information and transaction information of the target customer; obtains the associated knowledge set corresponding to the key attribute information based on the pre-built knowledge base and vector index library, wherein the associated knowledge set includes at least one associated knowledge; calculates the similarity between the key attribute information and each associated knowledge in the associated knowledge set by a cross-coding algorithm; sorts each associated knowledge in descending order of similarity to obtain the sorting result of the associated knowledge; constructs a model prompt word based on the sorting result of the key attribute information and the associated knowledge; inputs the model prompt word into the large language model to generate a due diligence report for the target customer. The present invention obtains the basic information, risk information and transaction information of the target customer, extracts the key attribute information, and uses the pre-built knowledge base and vector index library to find the knowledge set related to these attributes. Subsequently, the cross-coding algorithm is used to calculate the similarity between the key attribute and the associated knowledge, and sorts them by similarity to obtain the most relevant knowledge. This ranking result is used to construct the prompt words for input into the large language model, thereby generating a more accurate and practical due diligence report. The present invention significantly improves the accuracy and practicality of the due diligence report by combining specific customer information with rich associated knowledge.

[0047] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0049] Figure 1 A schematic diagram showing a flow chart of an implementation of a method for generating a due diligence report provided in an embodiment of the present invention;

[0050] Figure 2 A flow chart showing another implementation of a method for generating a due diligence report provided in an embodiment of the present invention;

[0051] Figure 3 A flow chart showing another implementation of a method for generating a due diligence report provided in an embodiment of the present invention;

[0052] Figure 4A system logic block diagram of a due diligence report generation method provided by an embodiment of the present invention is shown;

[0053] Figure 5 A schematic diagram of the structure of a due diligence report generating device provided by an embodiment of the present invention is shown;

[0054] Figure 6 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0055] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0056] In financial institutions, customer due diligence is a key part of anti-money laundering work, aimed at predicting money laundering behavior and preventing financial risks. The writing of due diligence reports is a complex part of the process, which includes three steps: collecting information, analyzing and evaluating, and writing reports. However, due to the different capabilities of grassroots staff and the large number of customers, the quality of reports is often poor, which affects the timely prediction of money laundering crimes. Therefore, it has become an urgent need to develop a due diligence report generation technology that combines search aggregation and GPT models.

[0057] The GPT (Generative Pre-trained Transformer) model is a large pre-trained model that uses deep learning technology to generate natural language that humans can understand. Unlike small models used for specific tasks, the GPT model has the characteristics of large scale and pre-training, and can be trained on massive amounts of general data, thereby significantly improving the generalization, versatility and practicality of artificial intelligence. Therefore, through GPT technology, combining the basic information of customers, risk information and transaction information to generate due diligence analysis reports can greatly reduce the workload and improve due diligence efficiency.

[0058] However, the data quality used by the GPT model during training varies, which may lead to unclear logic and confusing content in the generated report. Due diligence reports must be authentic, accurate, and comprehensive, and be able to conduct risk analysis based on key attributes. Therefore, when generating due diligence reports, search aggregation technology needs to be used to provide accurate and timely auxiliary information to the GPT model to ensure that the generated report is compliant.

[0059] Current solutions usually only store and classify key customer attributes, but lack the introduction of professional knowledge, cannot generate complete reports, lack data depth and breadth, and are not very practical. Based on user portraits, this invention introduces professional domain knowledge, uses vector libraries and knowledge bases to perform reasoning and analysis on customer attributes, and generates corresponding report conclusions, which significantly improves the flexibility and practicality of the application.

[0060] like Figure 1 As shown, a flowchart of an implementation of a method for generating a due diligence report provided by an embodiment of the present invention is provided. The method may include:

[0061] S100. Obtain basic customer information, customer risk information, and customer transaction information of target customers.

[0062] Among them, basic customer information is used to identify the customer's identity and background, usually including the customer's name, address, contact information, identification number (such as ID number or passport number), occupation, employer information and other basic identity data.

[0063] Among them, customer risk information is used to assess the financial risks that customers may bring, especially money laundering risks. Risk information may include the customer's source of income, financial status, credit score, whether there is any connection with high-risk countries or regions, whether it is on the sanctions list or monitoring list, and any past suspicious activity records.

[0064] Among them, customer transaction information involves the customer's financial activity records, including account transaction history, transaction amount, transaction frequency, transaction object, and cross-border transaction status. This information is used to analyze the customer's transaction behavior patterns to identify abnormal activities or potential money laundering.

[0065] S110. Extract key attribute information from the customer basic information, customer risk information and customer transaction information.

[0066] Key attribute information refers to the most important elements for assessing customer identity, risk and behavior extracted from basic customer information, customer risk information and customer transaction information. Key attribute information helps financial institutions to conduct risk assessment and decision-making effectively.

[0067] In order to conduct a comprehensive customer profile analysis, the embodiment of the present invention extracts key attributes from the three dimensions of customer basic information, risk information and transaction information. Each key attribute will be recorded in the form of a keyword to construct a key attribute set, expressed as: query = Key1, Key2, ..., Key k …Key K}, k∈(1,K), where Key krepresents the kth key attribute of the customer, and K represents the total number of keywords in the key attribute set.

[0068] Optionally, basic customer information may include the target customer's natural attributes, account attributes, and social attributes, customer risk information may include the target customer's judicial record, violations, and suspicious circumstances, and customer transaction information may include the target customer's large-value transaction information, transaction frequency information, and transaction channel information.

[0069] Among them, the basic information of customers is comprehensively described through natural attributes, account attributes and social attributes. Natural attributes include the customer's age, gender, nationality and place of residence. Account attributes involve the customer's account status, such as whether to open a RMB bank settlement account, a foreign exchange account, a free trade zone account, and the number and status of account openings and closings. Social attributes cover the customer's industry, corporate position and position information.

[0070] Among them, customer risk information is used to assess customer risk behavior through judicial records, violations and suspicious situations. Judicial records refer to the records of judicial freezing, inquiry or deduction of customers by the competent authorities to assess their illegal risks. Violations include records of dishonesty and overdue records, which are used to assess customers' dishonesty or overdue risks. Suspicious situations involve records of suspicious transactions in the past.

[0071] Among them, customer transaction information is analyzed through large-value transactions, transaction frequency and transaction channels to deeply understand customers' transaction habits and behaviors. Specific key indicators include the number of large-value transactions, the proportion of large-value transactions, the monthly transaction frequency, the proportion of income and expenditure, the main transaction time points and the main transaction channels.

[0072] The embodiment of the present invention constructs a comprehensive portrait of the customer through the above information.

[0073] Optional, in the above Figure 1 On the basis of one or more corresponding embodiments, in another optional embodiment provided by the embodiment of the present invention, extracting key attribute information from the customer basic information, customer risk information and customer transaction information may specifically include:

[0074] The target user's age, gender, nationality and place of residence are extracted from the natural attributes as the first key attribute information.

[0075] From the account attributes, whether the target user has opened a RMB bank settlement account, whether he has opened a foreign exchange account, whether he has opened a free trade zone account, the number of accounts opened and closed, and the account status are extracted as the second key attribute information.

[0076] The target user's industry, company position and position information are extracted from the social attributes as the third key attribute information.

[0077] The fourth key attribute information is extracted from the judicial records to determine whether the target user has been judicially frozen, inquired, or deducted by the competent authority.

[0078] The target user's dishonesty record and overdue record are extracted from the violation behavior as the fifth key attribute information.

[0079] In suspicious situations, the suspicious transaction behavior records of the target user that have occurred are extracted as the sixth key attribute information.

[0080] The number of large-value transactions and the proportion of large-value transactions of the target user are extracted from the large-value transaction information as the seventh key attribute information.

[0081] The target user's monthly transaction frequency, income-expenditure ratio and main transaction time are extracted from the transaction frequency information as the eighth key attribute information.

[0082] The main transaction channel of the target user is extracted from the transaction channel information as the ninth key attribute information.

[0083] The embodiments of the present invention can construct a comprehensive and accurate customer portrait by extracting diversified key attribute information in different dimensions, which plays an important role in promoting risk management and personalized services. In natural attributes, understanding the customer's age, gender, nationality and place of residence provides a basis for market segmentation and customized services. In account attributes, analyzing the customer's account type and status helps to evaluate the diversity and stability of financial behavior. In social attributes, in-depth understanding of the customer's occupation and position information can identify potential economic activities and risk exposure. In the analysis of judicial records and violations, potential legal and credit risks are identified to prevent the impact of bad behavior on institutions. In transaction information, paying attention to large transactions, transaction frequency and channel use can reveal customers' financial behavior patterns and preferences, help detect abnormal activities, and optimize service strategies. Combining these key attribute information, financial institutions can better predict customer needs, identify risks and provide more tailored solutions.

[0084] S120. Based on a pre-built knowledge base and a vector index base, obtain an associated knowledge set corresponding to the key attribute information, wherein the associated knowledge set includes at least one associated knowledge.

[0085] Among them, the knowledge base is an organized collection of information, containing a large amount of structured and unstructured data, usually used to store knowledge about a specific field. The knowledge base can include information in various forms such as text, images, and videos, which are processed and organized for easy retrieval and use. In customer profiling analysis, the knowledge base may contain information about customer behavior, market trends, laws and regulations, etc.

[0086] Among them, the vector index library is a technology for efficient retrieval and management of data, which is particularly useful when processing high-dimensional data. By converting information into mathematical vector form, the vector index library can quickly find the records most similar to the query. In customer portraits, key attribute information can be converted into vectors, and the vector index library can be used for rapid matching and retrieval to help find relevant knowledge.

[0087] The associated knowledge set is a set of information obtained by querying the knowledge base or vector index library, which is related to the input key attribute information. In customer analysis, the associated knowledge set may contain information such as market research, risk analysis, or behavior patterns related to specific customer characteristics, providing a context to understand customer behavior and needs.

[0088] Among them, associated knowledge refers to information directly related to a specific key attribute in the associated knowledge set. Associated knowledge provides specific insights or data to help explain or predict customer behavior. For example, for a customer with a high transaction frequency, associated knowledge may include analysis of the market impact or potential risks of the transaction trend.

[0089] Optionally, the knowledge base is constructed based on anti-money laundering expertise, banking domain knowledge, and historical due diligence reports, and the vector index library is constructed by vectorizing all the information in the knowledge base using a text2vec model trained from historical due diligence reports.

[0090] Among them, anti-money laundering expertise involves information on laws and regulations, compliance requirements, and money laundering case analysis. Banking domain knowledge covers expertise in areas such as banking operations, financial products, customer service, and risk management. Historical due diligence reports include due diligence reports on past customers, providing cases and data from actual operations. The embodiment of the present invention integrates anti-money laundering expertise, banking domain knowledge, and historical due diligence reports to form a multi-level knowledge base that can support complex data analysis and decision-making processes.

[0091] In order to implement semantic-based asymmetric search, it is necessary to vectorize the information in the knowledge base. In this embodiment of the present invention, key attributes extracted from historical due diligence reports are used to create a training set. During the training process, each due diligence report (document t ) and its corresponding key attribute (Key t ) are combined, and these combinations are given an initial similarity score (for example: 5 points), that is, {document t ,key t,5}. Next, by calculating the cosine similarity between the newly matched key attributes and the original key attributes, different similarity combinations are generated. The goal of this step is to train a text2vec model through diversified training data so that it can accurately convert text into vector representation.

[0092] After the text2vec model is trained, all the information in the knowledge base is input into the model for vectorization. Through this vectorization, each piece of information in the knowledge base is transformed into a high-dimensional vector. These vectors are organized into a vector index library for efficient similarity search and information retrieval.

[0093] Specifically, the embodiment of the present invention uses the BM25 method and the Bi-Encode algorithm to search and extract related knowledge based on the key attribute set, wherein BM25 is a probability-based retrieval function whose main function is to evaluate the relevance of each document from a document collection based on an input query (here, the key attribute set). Bi-Encode is an algorithm involving two independent encoders (usually deep learning models) for generating a vector representation of the input text for similarity calculation in a high-dimensional vector space.

[0094] S130. Calculate the similarity between the key attribute information and each associated knowledge in the associated knowledge set by using a cross-coding algorithm.

[0095] Among them, the cross encoding algorithm is a method for calculating the similarity of two inputs (usually texts).

[0096] Similarity is a measure of the degree of similarity between two inputs (such as text). In text processing, similarity is usually measured by calculating the distance or angle between two text vectors (generated by encoding algorithms). Similarity can be calculated based on cosine similarity, Euclidean distance, or dot product.

[0097] The embodiment of the present invention calculates the similarity between key attributes and associated knowledge through a cross-coding algorithm, and can evaluate the relevance of each associated knowledge with key attribute information, thereby selecting the most relevant knowledge for further analysis or decision support.

[0098] S140 , sorting each piece of related knowledge in descending order of similarity to obtain a related knowledge sorting result.

[0099] Among them, the result of the associated knowledge ranking refers to the ordered list obtained after sorting a set of associated knowledge according to their similarity to a key attribute. Specifically, this ranking result shows the correlation between each associated knowledge item and the key attribute, arranged from the most relevant to the least relevant. Such a ranking result is used to identify the most important and useful information so that it can be given priority in further analysis or decision-making.

[0100] Specifically, according to the calculated similarity values ​​{score1, score2, …, score n+m}, sort all the related knowledge. Put the related knowledge with higher similarity in front and the related knowledge with lower similarity in the back, so as to get a list of related knowledge arranged from high to low in similarity.

[0101] The embodiment of the present invention sorts each associated knowledge from large to small according to the similarity score. The refined sorting helps to improve the quality of due diligence report generation and ensure that the report content is more relevant and accurate.

[0102] S150, constructing a model prompt word based on the key attribute information and the associated knowledge sorting results.

[0103] Model prompts are a series of information inputs designed specifically for generating due diligence reports on target customers. Model prompts are inputs that combine key attribute information of customers with prioritized associated knowledge to guide the big language model in generating due diligence reports. Model prompts ensure that the big language model can fully understand the background and characteristics of customers, thereby generating more accurate and useful analysis results.

[0104] S160: Input the model prompt words into the large language model to generate a due diligence report on the target customer.

[0105] Among them, the big language model is an artificial intelligence model based on deep learning, which is specially designed to understand and generate natural language text. The big language model can be a GPT model, an LDA (Latent Dirichlet Allocation) model, or a PLSA (Probabilistic Latent Semantic Analysis) model.

[0106] Optionally, the large language model used in the embodiment of the present invention can select the open source ChatGPT model, and train and fine-tune it in combination with professional knowledge in the financial field to enhance its performance in the financial field.

[0107] Among them, the due diligence report, also known as the due diligence report, is a document generated after an in-depth investigation and analysis of the client. The purpose of the due diligence report is to comprehensively evaluate the status, potential risks and opportunities of the target and help decision makers make wise choices.

[0108] For example: By inputting the model prompt words into the GPT model, an example of the due diligence report generated by the embodiment of the present invention may be: "Customer profile: Zhang San is an 88-year-old male customer involved in judicial inquiries, which indicates that he may be related to certain legal issues. Risk assessment shows that he is an elderly customer, a person involved in judicial inquiries, and a credit card user, and is assessed to be at a high money laundering risk level. In addition, Zhang San has total assets of 10 million. Asset overview: The survey shows that Zhang San's total assets are 10 million. However, the specific source and nature of the assets are not yet clear. In order to more accurately understand his asset status, it is recommended to verify it through relevant documents provided by Zhang San (such as bank statements, financial statements, etc.). Customer background investigation: Due to the involvement of judicial inquiries, an in-depth understanding of Zhang San's personal background is required. It is recommended to conduct the following investigation: Find out whether Zhang San has been involved in or is currently involved in any judicial case, including court records, police reports, etc. In addition, investigate his credit card usage to check whether there is any overdue repayment or abnormal credit limit usage. It is also necessary to analyze his consumption habits and behavior patterns to determine whether there is a money laundering risk."

[0109] The due diligence report generation process provided by the embodiment of the present invention combines advanced data processing and artificial intelligence technologies. By constructing model prompt words, the processed and sorted key information is input into the large language model to generate a detailed and insightful due diligence report, thereby improving the quality of the due diligence report and the accuracy of decision-making.

[0110] The present invention provides a method for generating a due diligence report, which includes: obtaining basic customer information, customer risk information and customer transaction information of a target customer; extracting key attribute information from the basic customer information, customer risk information and customer transaction information; obtaining an associated knowledge set corresponding to the key attribute information based on a pre-built knowledge base and vector index library, wherein the associated knowledge set includes at least one associated knowledge; calculating the similarity between the key attribute information and each associated knowledge in the associated knowledge set by a cross-coding algorithm; sorting each associated knowledge in order of similarity from large to small to obtain an associated knowledge sorting result; constructing a model prompt word based on the key attribute information and the associated knowledge sorting result; inputting the model prompt word into a large language model to generate a due diligence report for the target customer. The present invention extracts key attribute information by obtaining the basic information, risk information and transaction information of the target customer, and uses the pre-built knowledge base and vector index library to find a knowledge set related to these attributes. Subsequently, a cross-coding algorithm is used to calculate the similarity between the key attribute and the associated knowledge, and sorting by similarity to obtain the most relevant knowledge. This sorting result is used to construct a prompt word input into the large language model, thereby generating a more accurate and practical due diligence report. The present invention significantly improves the accuracy and practicality of due diligence reports by combining specific customer information with rich associated knowledge.

[0111] Optional, based on Figure 1 The method shown, such as Figure 2 As shown, a flowchart of another implementation of the method for generating a due diligence report provided by an embodiment of the present invention is shown, and step S120 may specifically include:

[0112] S200, using the BM25 algorithm to extract a first preset number of first associated knowledge similar to the key attribute information from the knowledge base.

[0113] Specifically, in order to extract knowledge with semantic similarity to key attribute information from the knowledge base, the embodiment of the present invention first segmented the query according to keywords, then used the BM25 algorithm to calculate the similarity score between each knowledge and each document in the knowledge base, and extracted the top n related knowledge with higher similarity.

[0114] The similarity calculation formula of the BM25 algorithm is as follows:

[0115]

[0116] Among them, W k is the weight of the kth keyword, d is the document in the knowledge base, R(Key k ,d) is the correlation between the keyword and document d, and K is the number of keywords in the key attribute set.

[0117] S210: Use the Bi-Encode algorithm to extract a second preset number of second associated knowledge similar to the key attribute information from the vector index library.

[0118] Specifically, in order to understand the semantics of the key attribute set and perform analytical reasoning, the embodiment of the present invention uses the Bi-Encode algorithm to extract a second preset amount of associated knowledge with similar semantics from the vector index library.

[0119] Furthermore, the embodiment of the present invention first inputs the query into the text2vec model and converts it into a vector representation, which is recorded as a token. query = {quetoken1,quetoken2,…,quetoken l}, where l is the length of the output vector of the text2vec model, and quetoken1 is the lth token value of the query. Next, the cosine similarity formula is used to calculate the similarity between the query vector and the knowledge vector in the vector library to extract m related knowledge with high similarity. The knowledge in the vector library is represented by token doc ={doctoken1,doctoken2,…,doctoken l}, where doctoken l is the lth token value of the document. The formula for calculating cosine similarity is as follows:

[0120]

[0121] in, It is the sum of the products of the elements at the corresponding positions of the query vector and the document vector, which is used to calculate the dot product of the query vector and the document vector; quetoken j is the jth token value of the query vector, indicating the jth vector component of the query; doctoken j is the jth token value of the document vector, representing the jth vector component of the document; l is the length of the vector, that is, the dimension of the vector output by the text2vec model. The query and document vectors have the same length; is the Euclidean norm (or length) of the query vector, obtained by taking the square root of the sum of the squares of each component of the query vector; is the Euclidean norm (or length) of the document vector, obtained by taking the square root of the sum of the squares of each component of the document vector.

[0122] S220: Concatenate each first associated knowledge with each second associated knowledge to obtain an associated knowledge set corresponding to the key attribute information.

[0123] Specifically, the embodiment of the present invention integrates the n associated knowledge obtained by searching with the BM25 algorithm and the m associated knowledge extracted with the Bi-Encode algorithm into a complete associated knowledge set: document = {document1, document2, ..., document n+m}, which contains all the knowledge associated with key attribute information.

[0124] The embodiment of the present invention uses the BM25 algorithm and the Bi-Encode algorithm to extract the first and second associated knowledge similar to the key attribute information from the knowledge base and the vector index library respectively, and splices and integrates the two to build a comprehensive and comprehensive associated knowledge set. This method combines the advantages of traditional text retrieval technology and modern vector semantic matching, which not only improves the accuracy of knowledge acquisition related to key attribute information, but also expands the coverage of knowledge, thereby providing richer information support for generating due diligence reports.

[0125] Optional, based on Figure 1 The method shown, such as Figure 3 As shown, a flowchart of another implementation of the due diligence report generation method provided by an embodiment of the present invention is shown, and step S130 may specifically include:

[0126] S300: Obtain a keyword vector set of key attribute information, wherein the keyword vector set includes keyword vectors corresponding to each keyword extracted from the key attribute information.

[0127] Specifically, the embodiment of the present invention can input the key attribute set query into the text2vec model for processing to generate a keyword vector set. This keyword vector set contains the vectors corresponding to each keyword extracted from the key attribute information, represented as tokens. query = {quetoken1,quetoken2,…,quetoken l}, where each quetoken represents the vector feature of a keyword and l is the total length after vectorization.

[0128] S310: Obtain a first knowledge vector set of an associated knowledge set, wherein the first knowledge vector set includes knowledge vectors of each associated knowledge in the associated knowledge set.

[0129] Specifically, the embodiment of the present invention associates each associated knowledge in the associated knowledge set, namely {document1, document2, ..., document n+m}, and input them into the text2vec model one by one for vectorization processing to obtain the first knowledge vector set. This vector set consists of the knowledge vector of each associated knowledge in the associated knowledge set. Specifically, the vector of the i-th associated knowledge is represented as Where i represents the index of associated knowledge ranging from 1 to (n+m), and each knowledge vector contains l features.

[0130] S320: concatenate the keyword vector set with each knowledge vector respectively and input the concatenated vectors into the large language model, so that the large language model can calculate the similarity between the keyword vector set and each knowledge vector respectively.

[0131] Specifically, the embodiment of the present invention concatenates the keyword vector set with each knowledge vector respectively, and then inputs the concatenation results into the large language model, so that the model calculates the similarity between the keyword vector set and each knowledge vector.

[0132] Furthermore, the embodiment of the present invention vectorizes the keyword token query Related knowledge tokenklg i They are concatenated one by one and then input into the GPT model. The purpose is to evaluate the semantic similarity between these keywords and each associated knowledge, so as to find the knowledge most relevant to the keywords, that is, to solve:

[0133] score i = argmax doci∈D P(doc i |query)

[0134] Among them, score i is the similarity between the key attribute set and the i-th associated knowledge, doc i is the input associated knowledge, D is the vector library; through score i By transformation, we can get:

[0135]

[0136] Since P(query) has the same value as each associated knowledge, the probability of associated knowledge appearing in the knowledge base is P(doc i ) are considered equal, so we can calculate P(query|doc i ) to obtain the maximum similarity, so score i can be transformed into:

[0137] score i =∑log(softmax(out doc ))

[0138] where out doc This is the predicted value output by the GPT model. i part.

[0139] The embodiment of the present invention can capture the most important elements in the due diligence report by converting key attribute information into a keyword vector set. These keyword vectors are combined with the knowledge vectors of the associated knowledge set to form a comprehensive information matrix. When these spliced ​​vectors are input into the large language model, the model can evaluate the similarity between keywords and different knowledge, so as to automatically identify and highlight the most relevant information in the report in terms of generating a due diligence report, thereby improving the accuracy and depth of the report.

[0140] Optional, in the above Figure 1 On the basis of one or more corresponding embodiments, in another optional embodiment provided by the embodiment of the present invention, constructing a model prompt word based on key attribute information and the result of sorting related knowledge may specifically include:

[0141] The keyword set corresponding to the key attribute information and the second knowledge vector set corresponding to the associated knowledge ranking result are substituted into the pre-built customer due diligence report prompt template to obtain the model prompt words, wherein the keyword set includes each keyword extracted from the key attribute information, and the second knowledge vector set includes the knowledge vector of each associated knowledge in the associated knowledge ranking result.

[0142] When constructing the model prompt words, you first need to convert the key attribute information into a set of keywords, which are the core elements extracted from the key attribute information. At the same time, the sorting results of the associated knowledge are converted into a second knowledge vector set, which contains the specific knowledge vectors of the sorted associated knowledge. Then, substitute this information into the pre-built customer due diligence report prompt template to generate the model prompt words. The construction pattern of the prompt words is as follows: "Combining the customer's key attributes with the associated knowledge information, write a due diligence report. The customer's key attributes are {key1, key2,…, key k}, its associated knowledge is: {document1,document2,…,document n+m}. Where k represents the number of keywords extracted from customer key attributes, n represents the amount of knowledge obtained through keyword search, and m represents the amount of knowledge obtained through semantic search.

[0143] The embodiment of the present invention extracts key attribute information as a keyword set and converts the result of the associated knowledge ranking into a second knowledge vector set, and can substitute this information into a pre-built customer due diligence report prompt template to generate model prompt words. The embodiment of the present invention improves the accuracy and relevance of the report by systematically integrating the customer's core attributes and associated knowledge, and uses templated prompt words to guide the model to generate content, so that the report can better reflect the customer's specific situation and background knowledge, thereby improving the efficiency and accuracy of due diligence report generation.

[0144] In order to facilitate understanding of the system logic of the due diligence report generation method provided in the embodiment of the present invention, Figure 4 To explain: Figure 4 The figure shows a system logic block diagram of the due diligence report generation method provided by an embodiment of the present invention. First, in the customer due diligence system, the key attributes of the customer are extracted. Then, the search aggregation algorithm is used to combine these attributes with the knowledge in the fields of anti-money laundering and finance. Then, the GPT model is used to realize the automatic generation of the due diligence report. In order to ensure the accuracy and comprehensiveness of the report, when searching for related knowledge, the system constructs a knowledge base and a vector library from three sources: anti-money laundering expertise, banking field knowledge, and historical due diligence reports. A preliminary knowledge search is performed using the BM25 algorithm and the Bi-Encode algorithm, and then the Cross-Encode algorithm is used to finely sort the related knowledge. Finally, the customer's key attributes are combined with the related knowledge, prompt words are constructed and input into the GPT model, thereby generating the final due diligence conclusion.

[0145] Although operations are depicted in a particular order, this should not be understood as requiring that the operations be performed in the particular order shown or in a sequential order.Multitasking and parallel processing may be advantageous under certain circumstances.

[0146] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0147] Corresponding to the above method embodiment, the embodiment of the present invention also provides a due diligence report generation device, whose structure is as follows: Figure 5 As shown, it includes: a customer information obtaining unit 10, a key attribute information extraction unit 20, an associated knowledge set obtaining unit 30, a similarity calculation unit 40, an associated knowledge ranking result obtaining unit 50, a model prompt word construction unit 60 and a due diligence report generation unit 70.

[0148] The customer information obtaining unit 10 is used to obtain basic customer information, customer risk information and customer transaction information of target customers.

[0149] The key attribute information extraction unit 20 is used to extract key attribute information from the customer basic information, customer risk information and customer transaction information.

[0150] The associated knowledge set obtaining unit 30 is used to obtain an associated knowledge set corresponding to the key attribute information based on a pre-built knowledge base and a vector index base, wherein the associated knowledge set includes at least one associated knowledge.

[0151] The similarity calculation unit 40 is used to calculate the similarity between the key attribute information and each associated knowledge in the associated knowledge set by using a cross coding algorithm.

[0152] The associated knowledge ranking result obtaining unit 50 is used to sort each associated knowledge in descending order of similarity to obtain an associated knowledge ranking result.

[0153] The model prompt word construction unit 60 is used to construct the model prompt word based on the key attribute information and the associated knowledge sorting result.

[0154] The due diligence report generating unit 70 is used to input the model prompt words into the large language model to generate a due diligence report on the target customer.

[0155] Optionally, the associated knowledge set acquisition unit 30 is specifically used to use the BM25 algorithm to extract a first preset number of first associated knowledge similar to the key attribute information in the knowledge base; use the Bi-Encode algorithm to extract a second preset number of second associated knowledge similar to the key attribute information in the vector index library; splice each first associated knowledge and each second associated knowledge to obtain an associated knowledge set corresponding to the key attribute information.

[0156] Optionally, the similarity calculation unit 40 is specifically used to obtain a keyword vector set of key attribute information, wherein the keyword vector set includes keyword vectors corresponding to each keyword extracted from the key attribute information; obtain a first knowledge vector set of an associated knowledge set, wherein the first knowledge vector set includes knowledge vectors of each associated knowledge in the associated knowledge set; and concatenate the keyword vector set with each knowledge vector and input them into the large language model, so that the large language model solves the similarity between the keyword vector set and each knowledge vector.

[0157] Optionally, the model prompt word construction unit 60 is specifically used to substitute the keyword set corresponding to the key attribute information and the second knowledge vector set corresponding to the associated knowledge ranking result into the pre-built customer due diligence report prompt template to obtain the model prompt words, wherein the keyword set includes each keyword extracted from the key attribute information, and the second knowledge vector set includes the knowledge vector of each associated knowledge in the associated knowledge ranking result.

[0158] Optionally, the knowledge base is constructed based on anti-money laundering expertise, banking domain knowledge, and historical due diligence reports, and the vector index library is constructed by vectorizing all the information in the knowledge base using a text2vec model trained from historical due diligence reports.

[0159] Optionally, basic customer information includes the target customer's natural attributes, account attributes and social attributes, customer risk information includes the target customer's judicial record, violations and suspicious circumstances, and customer transaction information includes the target customer's large-value transaction information, transaction frequency information, and transaction channel information.

[0160] Optionally, the key attribute information extraction unit 20 is specifically used to extract the target user's age, gender, nationality and place of residence from the natural attributes as the first key attribute information; extract whether the target user has opened a RMB bank settlement account, whether he has opened a foreign exchange account, whether he has opened a free trade zone account, the number of accounts opened and closed, and the account status from the account attributes as the second key attribute information; extract the target user's industry, corporate position and position information from the social attributes as the third key attribute information; extract whether the target user has been judicially frozen, inquired, or deducted by the competent authority from the judicial records as the fourth key attribute information; extract the target user's dishonest record and overdue record from the violation behavior as the fifth key attribute information; extract the target user's suspicious transaction behavior record from the suspicious situation as the sixth key attribute information; extract the target user's large transaction number and large transaction proportion from the large transaction information as the seventh key attribute information; extract the target user's monthly transaction frequency, income and expenditure proportion and main transaction time points from the transaction frequency information as the eighth key attribute information; extract the target user's main transaction channel from the transaction channel information as the ninth key attribute information.

[0161] The present invention provides a due diligence report generation device, which is used to: obtain basic customer information, customer risk information and customer transaction information of a target customer; extract key attribute information from the basic customer information, customer risk information and customer transaction information; obtain an associated knowledge set corresponding to the key attribute information based on a pre-built knowledge base and vector index library, wherein the associated knowledge set includes at least one associated knowledge; calculate the similarity between the key attribute information and each associated knowledge in the associated knowledge set by a cross-coding algorithm; sort each associated knowledge in order of similarity from large to small to obtain the associated knowledge sorting result; construct a model prompt word based on the key attribute information and the associated knowledge sorting result; input the model prompt word into a large language model to generate a due diligence report for the target customer. The present invention obtains the basic information, risk information and transaction information of the target customer, extracts the key attribute information, and uses the pre-built knowledge base and vector index library to find the knowledge set related to these attributes. Subsequently, a cross-coding algorithm is used to calculate the similarity between the key attribute and the associated knowledge, and sorts them by similarity to obtain the most relevant knowledge. This sorting result is used to construct the prompt word input into the large language model, thereby generating a more accurate and practical due diligence report. The present invention significantly improves the accuracy and practicality of due diligence reports by combining specific customer information with rich associated knowledge.

[0162] Regarding the device in the above embodiment, the specific manner in which each unit performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0163] The due diligence report generating device includes a processor and a memory. The above-mentioned customer information obtaining unit 10, key attribute information extraction unit 20, associated knowledge set obtaining unit 30, similarity calculation unit 40, associated knowledge ranking result obtaining unit 50, model prompt word construction unit 60 and due diligence report generating unit 70 are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0164] The processor contains a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and the kernel parameters can be adjusted to obtain the basic information, risk information and transaction information of the target customer, extract key attribute information, and use the pre-built knowledge base and vector index library to find the knowledge set related to these attributes. Subsequently, the cross-coding algorithm is used to calculate the similarity between the key attributes and the associated knowledge, and sort them by similarity to obtain the most relevant knowledge. This sorting result is used to construct the prompt words input into the large language model, thereby generating a more accurate and practical due diligence report.

[0165] An embodiment of the present invention provides a computer-readable storage medium having a program stored thereon, which implements the due diligence report generating method when executed by a processor.

[0166] An embodiment of the present invention provides a processor, which is used to run a program, wherein the due diligence report generation method is executed when the program is running.

[0167] like Figure 6 As shown, an embodiment of the present invention provides an electronic device 1000, which includes at least one processor 1001, at least one memory 1002 connected to the processor 1001, and a bus 1003; wherein the processor 1001 and the memory 1002 communicate with each other through the bus 1003; the processor 1001 is used to call the program instructions in the memory 1002 to execute the above-mentioned due diligence report generation method. The electronic device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0168] The present invention also provides a computer program product, which, when executed on an electronic device, is suitable for executing a program that initializes the steps of a due diligence report generation method.

[0169] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatuses, electronic devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable device to generate a machine, so that the instructions executed by the processor of the computer or other programmable device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0170] In a typical configuration, an electronic device includes one or more processors (CPU), a memory, and a bus. The electronic device may also include an input / output interface, a network interface, and the like.

[0171] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip. The memory is an example of a computer-readable medium.

[0172] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0173] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0174] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0175] In the description of the present invention, it should be understood that the terms "up", "down", "front", "back", "left" and "right" etc. indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the positions or elements referred to must have specific directions, be constructed and operate in specific directions. Therefore, they should not be understood as limitations of the present invention.

[0176] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. It should also be noted that the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0177] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0178] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for generating a due diligence report, characterized in that: include: Obtain basic customer information, customer risk information and customer transaction information of target customers; Extracting key attribute information from the basic customer information, customer risk information and customer transaction information; Based on the pre-built knowledge base and vector index base, obtaining an associated knowledge set corresponding to the key attribute information, wherein the associated knowledge set includes at least one associated knowledge; Calculate the similarity between the key attribute information and each of the associated knowledge in the associated knowledge set by a cross coding algorithm; Sorting each of the associated knowledge in descending order of similarity to obtain an associated knowledge sorting result; Constructing a model prompt word based on the key attribute information and the associated knowledge sorting result; The model prompt words are input into the large language model to generate a due diligence report on the target customer.

2. The method according to claim 1, characterized in that: The obtaining of the associated knowledge set corresponding to the key attribute information based on the pre-built knowledge base and vector index base includes: Extracting a first preset number of first associated knowledge similar to the key attribute information from the knowledge base using the BM25 algorithm; Extracting a second preset number of second associated knowledge similar to the key attribute information from the vector index library using a Bi-Encode algorithm; Each of the first associated knowledge and each of the second associated knowledge are concatenated to obtain an associated knowledge set corresponding to the key attribute information.

3. The method according to claim 1, characterized in that The calculating the similarity between the key attribute information and each of the associated knowledge in the associated knowledge set by a cross coding algorithm includes: Obtaining a keyword vector set of the key attribute information, wherein the keyword vector set includes keyword vectors corresponding to each keyword extracted from the key attribute information; Obtaining a first knowledge vector set of the associated knowledge set, wherein the first knowledge vector set includes the knowledge vectors of each of the associated knowledge in the associated knowledge set; The keyword vector set is concatenated with each of the knowledge vectors and then input into a large language model, so that the large language model can calculate the similarity between the keyword vector set and each of the knowledge vectors.

4. The method according to claim 1, characterized in that: The constructing of model prompt words based on the key attribute information and the associated knowledge sorting result includes: Substitute the keyword set corresponding to the key attribute information and the second knowledge vector set corresponding to the associated knowledge ranking result into a pre-built customer due diligence report prompt template to obtain model prompt words, wherein the keyword set includes each keyword extracted from the key attribute information, and the second knowledge vector set includes the knowledge vector of each associated knowledge in the associated knowledge ranking result.

5. The method according to claim 1, characterized in that The knowledge base is constructed based on anti-money laundering expertise, banking domain knowledge and historical due diligence reports, and the vector index library is constructed by vectorizing all the information in the knowledge base using the text2vec model trained from the historical due diligence reports.

6. The method according to any one of claims 1 to 5, characterized in that The basic customer information includes the natural attributes, account attributes and social attributes of the target customer, the customer risk information includes the target customer's judicial record, violations and suspicious circumstances, and the customer transaction information includes the target customer's large-value transaction information, transaction frequency information, and transaction channel information.

7. The method according to claim 6, characterized in that The key attribute information extracted from the customer basic information, customer risk information and customer transaction information includes: Extracting the target user's age, gender, nationality and place of residence from the natural attributes as first key attribute information; From the account attributes, extract whether the target user has opened a RMB bank settlement account, whether the target user has opened a foreign exchange account, whether the target user has opened a free trade zone account, the number of accounts opened and closed, and the account status as the second key attribute information; Extracting the target user's industry, company position and position information from the social attributes as the third key attribute information; Extracting from the judicial records whether the target user has been judicially frozen, queried, or deducted by the competent authority as the fourth key attribute information; Extracting the target user's dishonesty record and overdue record from the illegal behaviors as the fifth key attribute information; Extracting suspicious transaction records of the target user in suspicious situations as the sixth key attribute information; Extracting the number of large-value transactions and the proportion of large-value transactions of the target user from the large-value transaction information as the seventh key attribute information; Extracting the target user's monthly transaction frequency, income-expenditure ratio, and main transaction time points from the transaction frequency information as the eighth key attribute information; The main transaction channel of the target user is extracted from the transaction channel information as the ninth key attribute information.

8. A due diligence report generating device, characterized in that: include: Customer information acquisition unit, key attribute information extraction unit, related knowledge set acquisition unit, similarity calculation unit, related knowledge ranking result acquisition unit, model prompt word construction unit and due diligence report generation unit, The customer information acquisition unit is used to obtain basic customer information, customer risk information and customer transaction information of the target customer; The key attribute information extraction unit is used to extract key attribute information from the customer basic information, customer risk information and customer transaction information; The associated knowledge set obtaining unit is used to obtain an associated knowledge set corresponding to the key attribute information based on a pre-built knowledge base and a vector index base, wherein the associated knowledge set includes at least one associated knowledge; The similarity calculation unit is used to calculate the similarity between the key attribute information and each of the associated knowledge in the associated knowledge set by using a cross-coding algorithm; The associated knowledge ranking result obtaining unit is used to sort the associated knowledge in descending order of similarity to obtain an associated knowledge ranking result; The model prompt word construction unit is used to construct the model prompt word based on the key attribute information and the associated knowledge ranking result; The due diligence report generating unit is used to input the model prompt words into the large language model to generate a due diligence report on the target customer.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the due diligence report generating method according to any one of claims 1 to 7 is implemented.

10. An electronic device, comprising at least one processor, and at least one memory and a bus connected to the processor; wherein: The processor and the memory communicate with each other via the bus; The processor is used to call the program instructions in the memory to execute the due diligence report generation method according to any one of claims 1 to 7.

Images