Intelligent question and answer method and system based on deep learning and credit loan knowledge graph

Abstract

The application discloses an intelligent question and answer method and system based on deep learning and credit loan knowledge graph, which comprises the following steps: constructing a credit loan knowledge graph; designing and training a double-layer generative adversarial network model; semantic analysis of user queries; generation and evaluation of final answers; in the process of loaning and repaying, using the electric selling, insurance and collection records in the credit loan knowledge graph to interact with the user in real time; combining the answers generated by the double-layer generative adversarial network model to prewarn the complaint intention of the user, and performing text quality inspection according to the dialogue content; analyzing the dialogue text and generating a quality inspection report. Through the construction of the double-layer generative adversarial network model and the combination of the credit loan knowledge graph, accurate understanding of user queries and knowledge reasoning in the professional field are realized, so that the system can quickly convert the user queries into structured semantic information after receiving the user queries, and efficiently match and reason with the entities and relationships in the knowledge graph, thereby ensuring the accuracy and relevance of the answers.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent question answering, and relates to an intelligent question answering method and system based on deep learning and a credit loan knowledge graph. BACKGROUND

[0002] In the field of modern financial technology, credit loans, as an important financial product, are widely used in the financing needs of individuals and small and micro enterprises. However, with the increase in the number of users and the complexity of business, how to effectively manage user information, accurately assess credit risk, and provide personalized services in credit loan business has become a problem to be solved. Existing intelligent question answering systems usually rely on natural language processing technology and can understand and respond to user queries to some extent. However, the system faces several challenges when dealing with the financial field, especially credit loan business. First, the credit loan field involves a large number of professional terms and complex business logic, and traditional NLP (natural language processing) technology often cannot accurately understand and analyze these contents, resulting in inaccurate understanding of user intent and affecting the accuracy and relevance of the reply. Second, existing question answering systems are mostly trained based on general domain data and lack specific optimization for credit loan business, which cannot fully meet the special needs in this field. In addition, existing systems usually only provide standardized answers and lack the ability to analyze individual credit status and provide personalized services, which cannot meet the growing individual needs of users.

[0003] To improve these deficiencies, some research has begun to try to introduce knowledge graph technology into intelligent question answering systems. Knowledge graph organizes and manages knowledge through graph structure and can better understand the complex relationships between data. However, existing technologies still have several key problems when applying knowledge graph to credit loan intelligent question answering systems. First, how to effectively combine natural language processing with knowledge graph so that the system can understand complex natural language and conduct knowledge reasoning in the professional field. Second, existing knowledge graphs can only provide static knowledge and cannot be dynamically updated and reasoned based on real-time user data, limiting the level of system intelligence. Finally, existing systems lack comprehensive utilization of user credit data, which cannot achieve personalized credit assessment and risk management.

[0004] In summary, the existing technology in the application of credit loan intelligent question answering system mainly has the following problems: First, the system has understanding deviation when dealing with credit loan professional terms and complex business logic, affecting the quality of the reply. Second, it lacks specific optimization for credit loan business and cannot effectively cope with the special needs in the field. Finally, existing systems lack the ability to analyze individual credit status and provide personalized services, making it difficult to meet the individual needs of users. The existence of the above problems limits the application effect and service level of existing intelligent question answering systems in the credit loan field. Summary of the Invention

[0005] The purpose of this invention is to solve the problem of misunderstanding when dealing with professional terminology and complex business logic in credit lending, which affects the quality of responses, and to provide an intelligent question-answering method and system based on deep learning and credit lending knowledge graph.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] The intelligent question-answering method based on deep learning and credit loan knowledge graphs includes the following steps:

[0008] Construct a knowledge graph for credit loans, which includes entities and relationships in the processes of lead generation, lending, and repayment related to credit loan business;

[0009] Design and train a two-layer generative adversarial network (GAN) model, which includes a first generator, a second generator, a first discriminator, and a second discriminator. The first generator is used to generate a preliminary answer, and the second generator is used to generate a final answer based on the preliminary answer and user historical data. The two discriminators are used to evaluate the accuracy and relevance of the generated answer, respectively.

[0010] The semantic parsing of user queries utilizes a pre-trained natural language processing model to perform semantic parsing on the user-input query, and then passes the parsed result as input to the first generator to generate a preliminary answer in real time.

[0011] The generation and screening of preliminary answers: The first generator generates real-time preliminary answers based on the parsing results, matches the real-time preliminary answers with the credit loan knowledge graph, and evaluates their relevance through the first discriminator. The real-time preliminary answers that are highly related to the entities and relationships in the knowledge graph are then screened out, and the screened real-time preliminary answers are then passed to the second generator.

[0012] The generation and evaluation of the final answer: The second generator generates the final answer in real time based on the filtered preliminary answer in real time, combined with the user's historical credit data, repayment records and loan application history information. The final answer in real time is evaluated by the second discriminator to ensure its accuracy and usability.

[0013] During the lending and repayment process, the system utilizes telemarketing, consumer protection, and collection records from the credit loan knowledge graph to interact with users in real time via an online customer service robot; combined with answers generated by a two-layer generative adversarial network model, it provides early warnings of users' complaint intentions and performs text quality checks based on the dialogue content.

[0014] The system parses dialogue text and generates quality inspection reports. It uses a two-layer generative adversarial network model to parse dialogue text in the customer service system, extracts key fields and classifies them, and generates quality inspection reports related to the loan disbursement and repayment process based on the parsing results.

[0015] The construction of the credit loan knowledge graph specifically includes the following steps:

[0016] Identify and define key entities for multi-dimensional user profiles, including user credit data entity E. c User consumption behavior entity E s User social data entity E n Loan product entity E p Repayment Model Entity E r and lending institution entity E l The definition of each entity is determined based on its role in the credit lending business:

[0017]

[0018] Among them, E i For the weighted synthesis representation of entity i, α k Let A be a weighting factor, representing the importance of the k-th attribute to the entity. ik Let K be the value of the k-th attribute of the i-th entity. i The total number of attributes involved in entity i;

[0019] This involves linking multi-dimensional information from users' credit data, consumption behavior, and social data with loan products and repayment methods to construct relationships between entities, including the compatibility relationship between user credit data and loan products. cp The matching relationship between user consumption behavior and repayment mode R sr The influence relationship between user social data and credit data R nc :

[0020]

[0021] Among them, R ij Represents entity E i With another entity E j The strength of the relationship between them and Entity E i With another entity E j The embedding representation in the m-th dimension, β m Let M be the weight factor for the m-th dimension, where M is the total number of dimensions;

[0022] Assign weight parameters W to each entity relationship r Weight parameter Wr The weighting parameters are used to describe the importance of different data dimensions in user profiles. They are obtained through analysis and statistical calculation of historical data.

[0023]

[0024] Where γ is the relation R i The importance coefficient of relevant historical data, λ is an adjustment parameter that controls the convergence speed of relation weights, and n is the total number of relations;

[0025] In the multi-dimensional user profile credit loan knowledge graph, each entity is connected through relationships to construct the graph structure G(V,E) of the user profile, where V represents the set of entity nodes and E represents the set of edges between entities. The graph structure is represented in triplet form as (E... a ,R b E c ), where E a and E c Entities of different dimensions, R b The relationship between the two is as follows:

[0026]

[0027] Where A is the number of distinct entity combinations in the graph, δ i Let be the weight factor of the i-th entity combination, and × represents the Cartesian product operation. This ensures that the combination of entities and relationships forms a complete multi-dimensional user profile credit loan knowledge graph structure, while the weight of each entity in the graph can be reflected by accumulation.

[0028] Clustering algorithms are used to classify user profiles, grouping users' credit data, consumption behavior, and social data into corresponding user groups. Furthermore, the multi-dimensional user profile credit loan knowledge graph is hierarchically processed, enabling users to locate the relevant user group and corresponding loan product when searching.

[0029]

[0030] Among them, C k For the k-th cluster center, Let i be the feature vector of the i-th user in the j-th dimension. Let δ be the mean vector of the k-th cluster center in the j-th dimension. ij The degree of matching between users and cluster centers on the corresponding dimension;

[0031] By using data cleaning and ontology matching techniques, redundant and noisy data in the multi-dimensional user profile credit loan knowledge graph are eliminated, and semantic enhancement is performed on the multi-dimensional user profile credit loan knowledge graph.

[0032]

[0033] Among them, Sim(R) b η represents the semantic similarity of the relation. m Let m be the similarity weight factor for the m-th dimension. and Let i and j be the embedding representations of the i-th and j-th entities in the m-th dimension, respectively, and ||·|| be the vector modulus operation;

[0034] Ultimately, a multi-dimensional user profile credit loan knowledge graph is generated, encompassing user credit data, consumption behavior, and social data, and linked to loan products and repayment models.

[0035] The two-layer generative adversarial network model includes:

[0036] The first generator is used to generate preliminary answers related to the user's query. The first generator is trained based on the semantic information of the user's query and the entities and relationships in the credit loan knowledge graph.

[0037] The first discriminator is used to evaluate the relevance and authenticity of the preliminary answers generated by the first generator, match the preliminary answers with the credit loan knowledge graph, and determine whether they conform to the logic and relationships in the credit loan knowledge graph.

[0038] The second generator, based on the initial answer generated by the first generator, combines the user's historical data and needs to generate the final answer.

[0039] The second discriminator is used to evaluate the final answer generated by the second generator and determine whether the final answer is consistent with the user's historical data and credit status.

[0040] The generation and screening of the preliminary answers specifically includes the following steps:

[0041] Construct a first generator G1, which receives semantic information q from the user query and entities and relationships (E) from the multi-dimensional user profile credit loan knowledge graph. a ,R b E c As input, a preliminary answer a1 is generated through a multi-layer neural network:

[0042]

[0043] Where q is the semantic vector representation of the user query, and h i α is the historical context vector related to entities and relationships in the multi-dimensional user profile credit loan knowledge graph. i These are the weight coefficients based on the attention mechanism. and W represents the embedded representation of entities and relationships in the multi-dimensional user profile credit loan knowledge graph. h and W o Let b be the weight matrix of the first generator G1. h and b o For bias terms, || represents vector concatenation, || represents concatenation, ReLU is the activation function, and softmax is the normalization function;

[0044] The first discriminator D1 is trained to evaluate the relevance and authenticity of the initial answer a1 generated by the first generator G1. The initial answer a1 is matched with the relationships in the multi-dimensional user profile credit loan knowledge graph, and an evaluation score s1 is generated through the following scoring mechanism:

[0045]

[0046] Among them, W d Let b be the weight matrix of the discriminator. d Here, σ is the bias term, and σ is the Sigmoid activation function. and These are the embedded representations of entities and relationships related to the current query in the multi-dimensional user profile credit loan knowledge graph, respectively. j is the importance coefficient of the entity relationship, ⊙ represents the element-level multiplication operation, and Z is the normalization factor.

[0047] The generation and evaluation of the final answer specifically includes the following steps:

[0048] Construct a second generator G2, which is based on the initial answer a1 generated by the first generator G1, and combines the user's historical data h and requirements d to generate the final answer a2:

[0049]

[0050] Among them, W f Let b be the weight matrix of the second generator. f For the bias term, β k The weighting coefficients between user historical data and demand are: GRU (Gated Recurrent Unit) for fusing user historical data h and demand d; and tanh (Hyperbolic Tangent Activation Function).

[0051] Design and train a second discriminator D2, which is used to evaluate the personalized matching degree of the final answer a2 generated by the second generator G2. The final answer a2 is matched with the user's historical data h and credit status, and an evaluation score s2 is generated by the following formula:

[0052]

[0053] Among them, W g Let b be the weight matrix of the second discriminator. g For the bias term, δ n BiLSTM is a bidirectional long short-term memory network used to analyze the time-series correlation of user historical data, with weighting factors matched to user historical data.

[0054] By alternately training the first generator G1, the first discriminator D1, the second generator G2, and the second discriminator D2, the overall performance of the generative adversarial network is optimized, so that the generated final answer a2 reaches its optimal state.

[0055]

[0056] in, To generate the loss function for the adversarial network, p represents the expected value. g1 and p g2 These are the probability distributions for the answers generated by the first and second generators, respectively.

[0057] The aforementioned real-time interaction with users via online customer service chatbot specifically includes:

[0058] During the lending and repayment process, telemarketing records E are used in the credit loan knowledge graph based on multi-dimensional user profiles. tx Consumer protection record E cb and collection records E cs Real-time interaction is enabled, with telemarketing records, consumer protection records, and collection records representing user interaction behaviors and related historical data at different stages.

[0059] Using the final answer a2 combined with the user's real-time dialogue content c, the probability distribution P of the user's complaint intent is calculated through an attention mechanism. comp ;

[0060] Based on the probability distribution of complaint intent P comp Set threshold T comp When P comp ≥T comp At that time, the early warning module A is triggered. alert The early warning module generates a corresponding early warning signal and sends an alarm message to customer service personnel:

[0061]

[0062] Among them, A alert This is a warning signal, with a value of 1 indicating that a warning has been triggered and 0 indicating that a warning has not been triggered. T comp The preset threshold for the probability of a complaint intent;

[0063] The text quality inspection based on the dialogue content specifically refers to:

[0064] During text quality inspection, key information from user dialogues is extracted using the text analysis module and compared with entity E in the multi-dimensional user profile credit loan knowledge graph. i Relationship R j Compare and generate a quality inspection report Q. rep :

[0065]

[0066] Among them, W q Let b be the weight matrix of the quality inspection module. q For the bias term, γ ij For entity E i With relation R j The degree of matching in the dialogue text, where ReLU is the activation function, sum is the weighted summation operation, and Q is the Q-value. rep The generated text quality inspection report is used to assess the quality and compliance of user conversation content;

[0067] During the lending and repayment process, we continuously monitor the interaction records between users and customer service robots, and dynamically update the telemarketing records, consumer protection records, and collection records in the multi-dimensional user profile credit loan knowledge graph based on real-time dialogue content.

[0068] The parsing of the dialogue text and the generation of the quality inspection report specifically include:

[0069] The first generator of the two-layer generative adversarial network model is used to perform preliminary analysis of the dialogue text in the customer service system, and to identify and extract user intent, question type, relevant entities and relationships in the dialogue;

[0070] After extracting key information, the second generator of a two-layer generative adversarial network model is used to classify the information and map it to specific categories in the lending and repayment process.

[0071] After classification, the classification results are evaluated by the second discriminator of a two-layer generative adversarial network model. The system evaluates the degree of matching between the results and the entities and relationships in the multi-dimensional user profile credit loan knowledge graph, verifies the accuracy and relevance of the classification, and ensures that the classification results are consistent with the user's actual credit status, historical behavior data and related business rules.

[0072] Based on the classification and evaluation results, a quality inspection report related to the lending and repayment process is generated. The quality inspection report summarizes the key data and matching evaluation results involved in customer service dialogues and is used to evaluate the quality of the entire lending and repayment process.

[0073] After the quality inspection report is generated, the system will feed the report back to the business department. At the same time, based on the latest quality inspection report and the actual business situation, the system will dynamically update the data in the multi-dimensional user profile credit loan knowledge graph.

[0074] An intelligent question-answering system based on deep learning and credit loan knowledge graphs includes the following modules:

[0075] A credit loan knowledge graph construction module is used to construct a credit loan knowledge graph, which contains entities and relationships related to the credit loan business in the processes of lead generation, lending, and repayment.

[0076] A two-layer generative adversarial network (GAN) model design module is used to design and train a two-layer GAN model. The two-layer GAN model includes a first generator, a second generator, a first discriminator, and a second discriminator. The first generator is used to generate a preliminary answer, and the second generator is used to generate a final answer based on the preliminary answer and user historical data. The two discriminators are used to evaluate the accuracy and relevance of the generated answer, respectively.

[0077] The user query semantic parsing module is used for semantic parsing of user queries. It uses a pre-trained natural language processing model to perform semantic parsing on the user-input query, and passes the parsed result as input to the first generator to generate a real-time preliminary answer.

[0078] The preliminary answer generation and filtering module is used to generate and filter preliminary answers. The first generator generates real-time preliminary answers based on the parsing results, matches the real-time preliminary answers with the credit loan knowledge graph, and evaluates their relevance through the first discriminator. The real-time preliminary answers that are highly related to entities and relationships in the knowledge graph are filtered out, and then the filtered real-time preliminary answers are passed to the second generator.

[0079] The final answer generation and evaluation module is used to generate and evaluate the final answer. The second generator generates the real-time final answer based on the filtered real-time preliminary answer, combined with the user's historical credit data, repayment records and loan application history information. The real-time final answer is evaluated by the second discriminator to ensure its accuracy and usability.

[0080] The real-time interaction and early warning module is used to interact with users in real time through an online customer service robot during the lending and repayment process by utilizing telemarketing, consumer protection, and collection records in the credit loan knowledge graph; combined with the answers generated by the two-layer generative adversarial network model, it provides early warning of users' complaint intentions and performs text quality inspection based on the dialogue content.

[0081] The dialogue text parsing and quality inspection report generation module is used for parsing dialogue text and generating quality inspection reports. It uses a two-layer generative adversarial network model to parse the dialogue text in the customer service system, extract key fields and classify them, and generate quality inspection reports related to the loan disbursement and repayment process based on the parsing results.

[0082] An apparatus includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, performs the steps of the method as described in any of the preceding items.

[0083] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in any of the preceding claims.

[0084] Compared with the prior art, the present invention has the following beneficial effects:

[0085] The intelligent question-answering method based on deep learning and a credit loan knowledge graph in this invention constructs a two-layer generative adversarial network model, combining natural language processing technology with a multi-dimensional user profile credit loan knowledge graph. This achieves accurate understanding of user queries and knowledge reasoning within professional domains. The collaborative work of the first generator and the first discriminator enables the system to quickly transform user queries into structured semantic information and efficiently match and reason with entities and relationships in the knowledge graph. This not only ensures the accuracy and relevance of the initial answer but also provides users with authoritative answers based on professional knowledge. The further intervention of the second generator and the second discriminator adds a personalized touch to the initial answer by deeply mining multi-dimensional information such as credit records, consumption behavior, and social relationships in the user's historical data, making it closer to the user's actual needs and preferences.

[0086] This invention combines users' historical credit data, consumption behavior, and social data with an intelligent question-answering system. The system can analyze and update user profiles in real time and provide personalized credit assessments and loan suggestions based on users' real-time dialogue content and queries, ensuring that it accurately reflects users' credit status and behavioral characteristics at all times. Through dynamic training using a two-layer generative adversarial network, the system can promptly adjust the assessment model and recommendation strategy in response to changes in user behavior data and the external environment, improving the accuracy of credit assessment and the effectiveness of risk management.

[0087] The knowledge graph construction module in this invention not only supports the management of static data but also enhances the system's flexibility and adaptability through the online customer service robot module and the quality inspection report generation module. This enables dynamic updates to user profiles and the knowledge graph, ensuring the system can reflect users' current credit behavior and attitudes in real time, as well as market changes, policy adjustments, and user behavior trends, providing the system with a continuous stream of the latest knowledge resources. Furthermore, the quality inspection report optimizes the system's question-and-answer process, improving its continuous learning ability and business adaptability. Attached Figure Description

[0088] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0089] Figure 1 This is a flowchart of the intelligent question-answering method based on deep learning and credit loan knowledge graph of the present invention. Detailed Implementation

[0090] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0091] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0092] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0093] The present invention will now be described in further detail with reference to the accompanying drawings:

[0094] See Figure 1 The flowchart below shows the intelligent question-answering method based on deep learning and credit loan knowledge graph in this invention, which specifically includes the following steps:

[0095] S1, construct a credit loan knowledge graph, which includes entities and relationships in the process of attracting customers, lending, and repayment related to credit loan business.

[0096] S1.1, Identify and define the key entities of the multi-dimensional user profile, including the user credit data entity E. c User consumption behavior entity E s User social data entity E n Loan product entity E p Repayment Model Entity E r and lending institution entity E l The definition of each entity is determined based on its role in the credit lending business:

[0097]

[0098] Among them, E i For the weighted synthesis representation of entity i, α k Let A be a weighting factor, representing the importance of the k-th attribute to the entity. ik Let K be the value of the k-th attribute of the i-th entity. i This represents the total number of attributes involved in entity i.

[0099] S1.2 links multi-dimensional information such as user credit data, consumption behavior, and social data with loan products and repayment models to construct relationships between entities, including the compatibility relationship between user credit data and loan products. cp The matching relationship between user consumption behavior and repayment mode R sr The influence relationship between user social data and credit data R nc :

[0100]

[0101] Among them, R ij Represents entity E i With another entity E j The strength of the relationship, and Entity E i With another entity E j The embedding representation in the m-th dimension, β m Let M be the weight factor for the m-th dimension, where M is the total number of dimensions.

[0102] S1.3, assign weight parameters W to each entity relationship. r Weight parameter W r The weighting parameters are used to describe the importance of different data dimensions in user profiles. They are obtained through analysis and statistical calculation of historical data.

[0103]

[0104] Where γ is the relation R i The importance coefficient of the relevant historical data, λ is an adjustment parameter that controls the convergence speed of the relation weights, and n is the total number of relations.

[0105] S1.4, in the multi-dimensional user profile credit loan knowledge graph, each entity is connected through relationships to construct the user profile graph structure G(V,E), where V represents the set of entity nodes and E represents the set of relationship edges between entities. The graph structure is represented in triplet form as (E... a ,R b E c ), where E a and E c Entities of different dimensions, R b The relationship between the two is as follows:

[0106]

[0107] Where A is the number of distinct entity combinations in the graph, δ i Let be the weight factor of the i-th entity combination, and × denotes the Cartesian product operation. This ensures that the combination of entities and relationships forms a complete multi-dimensional user profile credit loan knowledge graph structure, while the weight of each entity in the graph can be reflected by accumulation.

[0108] S1.5 uses clustering algorithms to classify user profiles, grouping user credit data, consumption behavior, and social data into corresponding user groups. It also performs hierarchical processing on the multi-dimensional user profile credit loan knowledge graph, enabling users to locate the relevant user group and corresponding loan product when querying.

[0109]

[0110] Among them, C k For the k-th cluster center, Let i be the feature vector of the i-th user in the j-th dimension. Let δ be the mean vector of the k-th cluster center in the j-th dimension. ij This represents the degree of matching between users and cluster centers on the corresponding dimension.

[0111] S1.6 eliminates redundant and noisy data in the multi-dimensional user profile credit loan knowledge graph through data cleaning and ontology matching techniques, and performs semantic enhancement on the multi-dimensional user profile credit loan knowledge graph:

[0112]

[0113] Among them, Sim(R) b η represents the semantic similarity of the relation. m Let m be the similarity weight factor for the m-th dimension. and Let be the embedding representations of the i-th and j-th entities in the m-th dimension, respectively, and ||·|| be the vector modulus operation.

[0114] S1.7 ultimately generates a multi-dimensional user profile credit loan knowledge graph that includes user credit data, consumption behavior and social data, and is associated with loan products and repayment models.

[0115] S2. Design and train a two-layer generative adversarial network model, which includes a first generator, a second generator, a first discriminator, and a second discriminator. The first generator is used to generate a preliminary answer, and the second generator is used to generate a final answer based on the preliminary answer and user historical data. The two discriminators are used to evaluate the accuracy and relevance of the generated answer, respectively.

[0116] The two-layer generative adversarial network model includes:

[0117] S2.1, the first generator is used to generate preliminary answers related to the user query. The first generator is trained based on the semantic information of the user query and the entities and relationships in the credit loan knowledge graph.

[0118] S2.2, the first discriminator is used to evaluate the relevance and authenticity of the preliminary answer generated by the first generator, match the preliminary answer with the credit loan knowledge graph, and determine whether it conforms to the logic and relationships in the credit loan knowledge graph.

[0119] S2.3 The second generator generates the final answer based on the initial answer generated by the first generator, combined with the user's historical data and requirements.

[0120] S2.4, the second discriminator is used to evaluate the final answer generated by the second generator and determine whether the final answer is consistent with the user's historical data and credit status.

[0121] S3, semantic parsing of user queries, uses a pre-trained natural language processing model to perform semantic parsing on the user's input query, and passes the parsed result as input to the first generator to generate a preliminary answer in real time.

[0122] S4, Initial Answer Generation and Screening: The first generator generates real-time initial answers based on the parsing results. These initial answers are then matched against the credit loan knowledge graph, and a first discriminator assesses their relevance. Initial answers highly relevant to entities and relationships in the knowledge graph are screened out, and then passed to the second generator. The initial answer generation and screening process specifically includes the following steps:

[0123] S4.1, Construct the first generator G1. The first generator G1 receives the semantic information q from the user query and the entities and relationships (E) in the multi-dimensional user profile credit loan knowledge graph. a ,R b E c As input, a preliminary answer a1 is generated through a multi-layer neural network:

[0124]

[0125] Where q is the semantic vector representation of the user query, and h i α is the historical context vector related to entities and relationships in the multi-dimensional user profile credit loan knowledge graph. i These are the weight coefficients based on the attention mechanism. and W represents the embedded representation of entities and relationships in the multi-dimensional user profile credit loan knowledge graph. h and W o Let b be the weight matrix of the first generator G1. h and b o For bias terms, represents vector concatenation, || represents concatenation, ReLU is the activation function, and softmax is the normalization function.

[0126] S4.2, Train the first discriminator D1. The first discriminator D1 is used to evaluate the relevance and authenticity of the preliminary answer a1 generated by the first generator G1. It matches the preliminary answer a1 with the relationships in the multi-dimensional user profile credit loan knowledge graph and generates an evaluation score s1 through the following scoring mechanism:

[0127]

[0128] Among them, W d Let b be the weight matrix of the discriminator. d Here, σ is the bias term, and σ is the Sigmoid activation function. and These are the embedded representations of entities and relationships related to the current query in the multi-dimensional user profile credit loan knowledge graph, respectively. jis the importance coefficient of the entity relationship, ⊙ represents the element-level multiplication operation, and Z is the normalization factor.

[0129] S5, Final Answer Generation and Evaluation: The second generator, based on the filtered real-time preliminary answer and combining it with the user's historical credit data, repayment records, and loan application history, generates a real-time final answer. This real-time final answer is then evaluated by a second discriminator to ensure its accuracy and usability. The final answer generation and evaluation specifically includes the following steps:

[0130] S5.1, Construct a second generator G2. The second generator G2 is based on the preliminary answer a1 generated by the first generator G1, and combines the user's historical data h and requirements d to generate the final answer a2:

[0131]

[0132] Among them, W f Let b be the weight matrix of the second generator. f For the bias term, β k Here, represents the weighting coefficient between user historical data and demand; GRU is a gated recurrent unit used to fuse user historical data h and demand d; and tanh is the hyperbolic tangent activation function.

[0133] S5.2, Design and train the second discriminator D2. The second discriminator D2 is used to evaluate the personalized matching degree of the final answer a2 generated by the second generator G2. The final answer a2 is matched with the user's historical data h and credit status, and the evaluation score s2 is generated by the following formula:

[0134]

[0135] Among them, W g Let b be the weight matrix of the second discriminator. g For the bias term, δ n BiLSTM is a bidirectional long short-term memory network used to analyze the time-series correlation of user historical data, with weighting factors matched to user historical data.

[0136] S5.3 optimizes the overall performance of the generative adversarial network by alternately training the first generator G1, the first discriminator D1, the second generator G2, and the second discriminator D2, so that the generated final answer a2 reaches its optimal state.

[0137]

[0138] in, To generate the loss function for the adversarial network, p represents the expected value. g1 and p g2These are the probability distributions for the answers generated by the first and second generators, respectively.

[0139] S6 utilizes telemarketing, consumer protection, and collection records from the credit loan knowledge graph during the lending and repayment process to interact with users in real time via an online customer service robot; combined with answers generated by a two-layer generative adversarial network model, it provides early warnings of users' complaint intentions and performs text quality checks based on the dialogue content.

[0140] The aforementioned real-time interaction with users via online customer service chatbot specifically includes:

[0141] S6.1, during the lending and repayment process, based on the telemarketing records E in the multi-dimensional user profile credit loan knowledge graph. tx Consumer protection record E cb and collection records E cs Real-time interaction is enabled, with telemarketing records, consumer protection records, and collection records representing user interaction behaviors and related historical data at different stages.

[0142] S6.2, using the final answer a2 combined with the user's real-time dialogue content c, calculate the probability distribution P of the user's complaint intent through an attention mechanism. comp .

[0143] S6.3, based on the probability distribution P of the complaint intent comp Set threshold T comp When P comp ≥T comp At that time, the early warning module A is triggered. alert The early warning module generates a corresponding early warning signal and sends an alarm message to customer service personnel:

[0144]

[0145] Among them, A alert This is a warning signal, with a value of 1 indicating that a warning has been triggered and 0 indicating that a warning has not been triggered. T comp This is a preset threshold for the probability of complaining intent.

[0146] The text quality inspection based on the dialogue content specifically refers to:

[0147] S6.4, during text quality inspection, key information from user dialogues is extracted through the text analysis module and compared with entity E in the multi-dimensional user profile credit loan knowledge graph. i Relationship R j Compare and generate a quality inspection report Q. rep :

[0148]

[0149] Among them, W qLet b be the weight matrix of the quality inspection module. q For the bias term, γ ij For entity E i With relation R j The degree of matching in the dialogue text, where ReLU is the activation function, sum is the weighted summation operation, and Q is the Q-value. rep The generated text quality inspection report is used to assess the quality and compliance of user conversation content.

[0150] S6.5 continuously monitors the interaction records between users and customer service robots during the loan disbursement and repayment process, and dynamically updates the telemarketing records, consumer protection records, and collection records in the multi-dimensional user profile credit loan knowledge graph based on real-time dialogue content.

[0151] S7, Dialogue Text Parsing and Quality Inspection Report Generation, utilizes a two-layer generative adversarial network model to parse dialogue text in the customer service system, extract key fields, classify them, and generate quality inspection reports related to the loan disbursement and repayment processes based on the parsing results. Specifically, it includes:

[0152] S7.1 uses the first generator of a two-layer generative adversarial network model to perform preliminary parsing of the dialogue text in the customer service system, and identifies and extracts user intent, question type, related entities and relationships in the dialogue.

[0153] S7.2 After extracting key information, the second generator of the two-layer generative adversarial network model is used to classify the information and map it to specific categories in the lending and repayment process.

[0154] S7.3 After classification is completed, the classification results are evaluated by the second discriminator of the two-layer generative adversarial network model. The system evaluates the degree of matching between the results and the entities and relationships in the multi-dimensional user profile credit loan knowledge graph, verifies the accuracy and relevance of the classification, and ensures that the classification results are consistent with the user's actual credit status, historical behavior data and related business rules.

[0155] S7.4, based on the classification assessment results, generates a quality inspection report related to the lending and repayment process. The quality inspection report summarizes the key data and matching assessment results involved in the customer service dialogue, and is used to evaluate the quality of the entire lending and repayment process.

[0156] S7.5 After the quality inspection report is generated, the system will feed the quality inspection report back to the business department. At the same time, based on the latest quality inspection report and the actual business situation, the system will dynamically update the data in the multi-dimensional user profile credit loan knowledge graph.

[0157] One embodiment of the present invention is an intelligent question-answering system based on deep learning and a credit loan knowledge graph, comprising the following modules:

[0158] A credit loan knowledge graph construction module is used to construct a credit loan knowledge graph, which contains entities and relationships related to the credit loan business in the processes of lead generation, lending, and repayment.

[0159] A two-layer generative adversarial network (GAN) model design module is used to design and train a two-layer GAN model. The two-layer GAN model includes a first generator, a second generator, a first discriminator, and a second discriminator. The first generator is used to generate a preliminary answer, and the second generator is used to generate a final answer based on the preliminary answer and user historical data. The two discriminators are used to evaluate the accuracy and relevance of the generated answer, respectively.

[0160] The user query semantic parsing module is used for semantic parsing of user queries. It uses a pre-trained natural language processing model to perform semantic parsing on the user-input query, and passes the parsed result as input to the first generator to generate a real-time preliminary answer.

[0161] The preliminary answer generation and filtering module is used to generate and filter preliminary answers. The first generator generates real-time preliminary answers based on the parsing results, matches the real-time preliminary answers with the credit loan knowledge graph, and evaluates their relevance through the first discriminator. The real-time preliminary answers that are highly related to entities and relationships in the knowledge graph are filtered out, and then the filtered real-time preliminary answers are passed to the second generator.

[0162] The final answer generation and evaluation module is used to generate and evaluate the final answer. The second generator generates the real-time final answer based on the filtered real-time preliminary answer, combined with the user's historical credit data, repayment records and loan application history information. The real-time final answer is evaluated by the second discriminator to ensure its accuracy and usability.

[0163] The real-time interaction and early warning module is used to interact with users in real time through an online customer service robot during the lending and repayment process by utilizing telemarketing, consumer protection, and collection records in the credit loan knowledge graph; combined with the answers generated by the two-layer generative adversarial network model, it provides early warning of users' complaint intentions and performs text quality inspection based on the dialogue content.

[0164] The dialogue text parsing and quality inspection report generation module is used for parsing dialogue text and generating quality inspection reports. It uses a two-layer generative adversarial network model to parse the dialogue text in the customer service system, extract key fields and classify them, and generate quality inspection reports related to the loan disbursement and repayment process based on the parsing results.

[0165] This invention presents an intelligent question-answering method based on deep learning and a credit loan knowledge graph, specifically including identifying the user's credit data E. c Consumer behavior E s Social data E n and loan product E p Repayment Mode E r This application identifies key entities and generates representations for each entity using a weighted synthesis formula. Furthermore, it proposes a formula for constructing relationship strength. By constructing strong relationships between entities, the accuracy of user profiles is improved. Therefore, this invention first comprehensively analyzes the compatibility of users' credit, consumption behavior, social data, and loan products and repayment methods, enabling efficient data matching. Furthermore, the invention's protected aspects include: the algorithmic formula for user profile construction and the multi-dimensional association modeling method for credit loan knowledge graphs.

[0166] This proposal suggests a two-layer recurrent generative adversarial network (GAN) model for generating real-time answers relevant to user queries. First, a pre-trained natural language processing (NLP) model is used to semantically parse the user query. The parsed result is then input into a first generator to produce an initial answer. The first discriminator is used to evaluate the relevance of the initial answer a1, using the following scoring mechanism. The second generator combines the user's historical data and needs to generate the final answer a2, and the second discriminator evaluates the personalized matching degree of the answer. This method involves the first generator generating a preliminary answer based on the user's query semantic information, the second generator generating a personalized final answer based on the user's historical data and needs, and the first and second discriminators evaluating the relevance and matching degree of the generated answers. The proposed two-layer generative adversarial network model exhibits excellent flexibility and intelligence in handling complex query scenarios, and further optimizes the quality of answer generation through alternating training of the generative adversarial network. The key protection point of this invention lies in the design of the two-layer generative adversarial network model and its application in the credit lending field. This improved scheme protects the design of the two-layer generative adversarial network model, the various algorithms in the generation process, and the related discrimination mechanisms.

[0167] When the probability of complaint intent P comp Exceeding the preset threshold T comp At that time, warning signal A is triggered. alert It also generates quality inspection reports to assess the quality and compliance of user interactions. This part protects a real-time complaint intent analysis and early warning algorithm based on user interaction, as well as a quality inspection report generation mechanism. It is proposed to build an online customer service robot system that can perform semantic analysis and complaint intent identification based on the user's real-time dialogue content, and trigger a complaint early warning mechanism when necessary. In addition, the interactive customer service system proposed in this invention is based on multi-dimensional analysis of telemarketing, consumer protection and collection records, which can dynamically update user profiles, generate quality inspection reports and provide real-time feedback to relevant business departments.

[0168] One embodiment of the present invention provides a terminal device. This terminal device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0169] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention.

[0170] The device / terminal equipment may be a desktop computer, laptop, handheld computer, cloud server, or other computing device. The device / terminal equipment may include, but is not limited to, a processor and a memory.

[0171] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0172] The memory can be used to store the computer program and / or module. The processor implements various functions of the device / terminal equipment by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0173] If the modules / units integrated in the device / terminal equipment are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0174] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent question-answering method based on deep learning and a credit loan knowledge graph, characterized in that, The method comprises the following steps: constructing a credit loan knowledge graph, constructing a knowledge graph containing entities and relationships in the process of lead generation, loaning, and repayment related to credit loan business; designing and training a double-layer generative adversarial network model, the double-layer generative adversarial network model comprising a first generator, a second generator, a first discriminator, and a second discriminator; the first generator is used to generate a preliminary answer, and the second generator is used to generate a final answer based on the preliminary answer in combination with historical data of a user; two discriminators are respectively used to evaluate the accuracy and relevance of the generated answer; semantic analysis of a user query, using a pre-trained natural language processing model to perform semantic analysis on a query input by the user, and passing the analyzed result as input to the first generator to generate a real-time preliminary answer; generation and screening of the preliminary answer, the first generator generating a real-time preliminary answer according to the analysis result, matching the real-time preliminary answer with the credit loan knowledge graph, and evaluating the relevance thereof by the first discriminator to screen out real-time preliminary answers highly relevant to entities and relationships in the knowledge graph, and then passing the screened real-time preliminary answers to the second generator; generation and evaluation of the final answer, the second generator generating a real-time final answer based on the screened real-time preliminary answers in combination with historical credit data, repayment records, and loan application history information of the user, and the real-time final answer being evaluated by the second discriminator to ensure its accuracy and practicability; in the process of loaning and repayment, using the credit loan knowledge graph to generate real-time interaction between an online customer service robot and the user by using the records of electric marketing, consumer protection, and debt collection in the credit loan knowledge graph; combining the answers generated by the double-layer generative adversarial network model to prewarn the user's complaint intention, and performing text quality inspection according to the dialogue content; analysis of the dialogue text and generation of the quality inspection report, using the double-layer generative adversarial network model to analyze the dialogue text in the customer service system, extracting key fields and performing classification processing, and generating a quality inspection report related to the loaning and repayment process according to the analysis result. 2.The intelligent question answering method based on deep learning and credit loan knowledge graph of claim 1, wherein, The credit loan knowledge graph is constructed, specifically comprising the following steps: Identify and define key entities of multi-dimensional user portrait, including user credit data entity E c , user consumption behavior entity E s , user social data entity E n , loan product entity E p , repayment mode entity E r and loan institution entity E l , the definition of each entity is determined according to the role of the corresponding entity in the credit business: （1） wherein, E i is a weighted aggregate representation of the entity i , α k is a weight factor indicating the importance of the k th attribute to the entity, A ik is the i th attribute value of the k th entity, K i is the total number of attributes involved in the entity i . Correlate multi-dimensional information of credit data, consumption behavior and social data of a user with loan products and repayment modes, build relationships between entities, including the adaptability relationship between user credit data and loan products R cp , the matching relationship between user consumption behavior and repayment mode R sr and the influence relationship between user social data and credit data R nc : （2） wherein, R ij representing an entity E i a relationship strength between the entity E j and another entity, and are respectively an embedding representation of the entity E i and another entity E j in the first m dimension, β m is a weight factor for the first m dimension, M is the total number of dimensions; assigning a weight parameter to each entity relationship W r , the weight parameter W r The weight parameter is obtained by analyzing and statistically calculating historical data to describe the importance of different data dimensions in the user portrait. （3） wherein, γ i for the relationship R i a significance coefficient of the relevant historical data, λ for adjusting parameters, controlling the convergence speed of the relationship weight, n for the total number of relationships; In the multi-dimensional user portrait credit knowledge graph, each entity is connected through the relationship to construct a graph structure of the user portrait G ( V , E ), wherein, V denotes a set of entity nodes, E denotes a set of relationship edges between entities, and the graph structure is represented in the form of a triple ( E a , R b , E c ), wherein, E a and E c are entities of different dimensions, R b is the association relationship between the two. （4） wherein, A is the number of different entity combinations in the graph, δ i is the weight factor of the i th entity combination, and × denotes the Cartesian product operation, which ensures that the combination of entities and relationships forms a complete multi-dimensional user portrait credit knowledge graph structure, and the weight of each entity in the graph can be reflected by the cumulative manner. classify the user portraits by using a clustering algorithm, cluster the credit data, consumption behavior, and social data of the user into corresponding user groups, and perform hierarchical processing on the multi-dimensional user portrait credit loan knowledge graph, so that the user query is positioned to the corresponding user group and the corresponding loan product: （5） wherein, C k is the k th cluster center, is the i th user's feature vector in the j th dimension, is the k th cluster center's mean vector in the j th dimension, δ ij is the degree of matching between the user and the cluster center in the corresponding dimension; eliminate redundant and noisy data in the multi-dimensional user portrait credit loan knowledge graph by using data cleaning and ontology matching technology, and perform semantic enhancement on the multi-dimensional user portrait credit loan knowledge graph: （6） wherein, Sim R b is the semantic similarity of the relation, η m is the similarity weight factor of the m th dimension, and are the embedding representations of the i th and j th entity on the m th dimension, respectively, and ||·|| is the vector norm operation.​ finally generate a multi-dimensional user portrait credit loan knowledge graph covering the credit data, consumption behavior, and social data of the user, and associated with the loan product and repayment mode entities. 3.The intelligent question answering method based on deep learning and credit loan knowledge graph of claim 1, wherein, The double-layer generative adversarial network model comprises: the first generator is used to generate a preliminary answer related to the user query, and the first generator is trained according to the semantic information of the user query and the entities and relationships in the credit loan knowledge graph; The first discriminator is used for evaluating the relevance and authenticity of the preliminary answer generated by the first generator, matching the preliminary answer with the credit loan knowledge graph, and judging whether the preliminary answer conforms to the logic and relationship in the credit loan knowledge graph; The second generator generates the final answer based on the preliminary answer generated by the first generator and in combination with the historical data and demand of the user; The second discriminator is used for evaluating the final answer generated by the second generator and judging whether the final answer is consistent with the historical data and credit status of the user. 4.The intelligent question answering method based on deep learning and credit loan knowledge graph of claim 1, wherein, The generation and screening of the preliminary answer specifically include the following steps: constructing the first generator G 1, the first generator G 1 receiving semantic information of a user query q and entities and relationships in the multi-dimensional user portrait credit knowledge graph E a , R b , E c as input, generating a preliminary answer through a multi-layer neural network a 1: （7） wherein, q is a semantic vector representation of a user query, h i is a historical context vector related to entities and relations in the multi-dimensional user portrait credit knowledge graph, α i is a weight coefficient based on an attention mechanism, E xi , R yi and E zi are embedding representations of entities and relations in the multi-dimensional user portrait credit knowledge graph, respectively, W h and W o is a weight matrix of the first generator G 1, b h and b o are bias terms, represents a vector concatenation operation, || represents a connection operation, and ReLU is an activation function, soft max is a normalization function; Training the first discriminator D 1,the first discriminator D 1 is used to evaluate the relevance and authenticity of the preliminary answer G 1 generated by the first generator a 1, and match the preliminary answer a 1 with the relationship in the multi-dimensional user portrait credit knowledge graph, and generate an evaluation score through the following scoring mechanism s 1: （8） wherein, W d is a weight matrix of the discriminator, b d is a bias term, σ is a Sigmoid activation function, E xj , R yj and E zj are embedding representations of entities and relations in the multi-dimensional user portrait credit knowledge graph related to the current query, respectively, γ j is an importance coefficient of the entity relation, denotes an element-level multiplication operation, Z is a normalization factor. 5.The intelligent question answering method based on deep learning and credit loan knowledge graph of claim 1, wherein, The generation and evaluation of the final answer specifically include the following steps: build a second generator G 2, the second generator G 2, the first generator G 1, the preliminary answer generated a 1, and in combination with the user's historical data h and needs d generate a final answer a 2: （9） wherein, W f is a weight matrix for the second generator, b f is a bias term, β k is a weight coefficient between the user history data and the demand, GRU is a gated recurrent unit for fusing the user's history data h and the demand d , and tanh is a hyperbolic tangent activation function. design and train the second discriminator D 2, the second discriminator D 2, the second discriminator G 2, the final answer generated by the second generator a 2, the final answer generated by the second generator a 2, the final answer generated by the second generator h 2, the final answer generated by the second generator s 2: （10） wherein, W g is a weight matrix for the second discriminator, b g is a bias term, δ n is a weight factor matching the user history data, BiLSTM is a bidirectional long short-term memory network for analyzing time series correlation of the user history data; by alternating training the first generator G 1, the first discriminator D 1, the second generator G 2, and the second discriminator D 2, optimizing the overall performance of the generative adversarial network such that the generated final answer a 2 reaches an optimum: （11） wherein, a loss function for the generative adversarial network, denotes an expected value, p g1 and p g2 are probability distributions generated by the first and second generators, respectively. 6.The intelligent question answering method based on deep learning and credit loan knowledge graph of claim 1, wherein, The real-time interaction between the online customer service robot and the user specifically includes: The electric selling record in the credit loan knowledge graph based on multi-dimensional user portrait in the process of loaning and repaying E tx , the insurance record E cb and the collection record E cs Real-time interaction is performed, and the electric selling record, the insurance record and the collection record respectively represent the interaction behaviors and related historical data of the user in different stages; Utilize the final answer a 2 Combine the user's real-time conversation content c, and calculate the probability distribution of the user's complaint intention through the attention mechanism P comp ; According to the complaint intention probability distribution P comp Setting a threshold value T comp When P comp ≥ T comp , triggering the early warning module A alert , the early warning module generates a corresponding early warning signal and sends an alarm information to the customer service personnel: （12） wherein, A alert is a pre-warning signal, and the value 1 indicates triggering a pre-warning, and the value 0 indicates not triggering a pre-warning, T comp is a preset complaint intention probability threshold value; The text inspection according to the dialogue content specifically includes: When performing text quality inspection, the text analysis module extracts key information in the user dialogue and compares it with entities in the multi-dimensional user portrait credit knowledge graph E i and relationships R j to generate a quality inspection report Q rep : （13） wherein, W q is a weight matrix for the quality inspection module, b q is a bias term, γ ij is an entity E i and a relationship R j a matching degree in the dialogue text, ReLU is an activation function, and sum is a weighted summation operation, Q rep is a generated text quality inspection report for evaluating the quality and compliance of the user dialogue content; In the process of loaning and repayment, the interaction record between the user and the customer service robot is continuously monitored, and the dynamic update of the electric sales record, the record of protection against sales, and the record of collection in the multi-dimensional user portrait credit loan knowledge graph is performed according to the real-time dialogue content. 7.The intelligent question answering method based on deep learning and credit loan knowledge graph of claim 1, wherein, The analysis of the dialogue text and the generation of the quality inspection report specifically include: The first generator of the double-layer generative adversarial network model is used to preliminarily analyze the dialogue text in the customer service system, identify and extract the user intent, problem type, related entity and relationship in the dialogue; After the key information is extracted, the second generator of the double-layer generative adversarial network model is used to classify the information and map it to a specific category in the loaning and repayment process; After the classification is completed, the second discriminator of the double-layer generative adversarial network model is used to evaluate the classification result, the system evaluation result and the matching degree of the entity and relationship in the multi-dimensional user portrait credit loan knowledge graph are verified, the accuracy and relevance of the classification are verified, and the classification result is consistent with the actual credit status, historical behavior data and related business rules of the user; On the basis of the classification evaluation result, a quality inspection report related to the loaning and repayment process is generated, the quality inspection report summarizes the key data involved in the customer service dialogue and the matching degree evaluation result, and is used to evaluate the quality of the entire loaning and repayment process; After the generation of the quality inspection report, the system feeds back the quality inspection report to the business department, and dynamically updates the data in the multi-dimensional user portrait credit loan knowledge graph according to the latest quality inspection report and the actual business situation.

8. An intelligent question-answering system based on deep learning and credit loan knowledge graph, characterized in that, The following modules are included: A credit loan knowledge graph construction module is used to construct a credit loan knowledge graph, which contains entities and relationships in the process of lead generation, loaning and repayment related to credit loan business; A double-layer generative adversarial network model design module is used to design and train a double-layer generative adversarial network model, which includes a first generator, a second generator, a first discriminator and a second discriminator; the first generator is used to generate a preliminary answer, the second generator is used to generate a final answer based on the preliminary answer in combination with the historical data of the user; and the two discriminators are respectively used to evaluate the accuracy and relevance of the generated answer. A user query semantic analysis module for semantic analysis of user queries, which uses a pre-trained natural language processing model to perform semantic analysis on user input queries, and passes the analyzed results as input to a first generator to generate real-time preliminary answers; A preliminary answer generation and screening module for generating and screening preliminary answers, which uses a first generator to generate real-time preliminary answers based on the analysis results, matches the real-time preliminary answers with the credit loan knowledge graph, and uses a first discriminator to evaluate their relevance, screens out real-time preliminary answers that are highly relevant to entities and relationships in the knowledge graph, and then passes the screened real-time preliminary answers to a second generator; A final answer generation and evaluation module for generating and evaluating final answers, which uses a second generator to generate real-time final answers based on the screened real-time preliminary answers, in combination with the user's historical credit data, repayment records, and loan application history information, and uses a second discriminator to evaluate the real-time final answers to ensure their accuracy and usefulness; A real-time interaction and early warning module for real-time interaction with users through online customer service robots during the loan and repayment process, using the credit loan knowledge graph for telemarketing, consumer protection, and debt collection records; early warning of user complaints based on the answers generated by the double-layer generative adversarial network model, and text quality inspection based on the conversation content; A conversation text analysis and quality inspection report generation module for analyzing conversation text and generating quality inspection reports, which uses a double-layer generative adversarial network model to analyze conversation text in the customer service system, extracts key fields and performs classification processing, and generates quality inspection reports related to the loan and repayment process based on the analysis results.

9. An apparatus comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein, The processor executes the computer program to implement the steps of the method of any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the steps of the method of any one of claims 1-7.

Images