WO2019071904A1

WO2019071904A1 - Bayesian network-based question-answering apparatus, method and storage medium

Info

Publication number: WO2019071904A1
Application number: PCT/CN2018/077344
Authority: WO
Inventors: 徐国强
Original assignee: 深圳壹账通智能科技有限公司
Priority date: 2017-10-13
Filing date: 2018-02-27
Publication date: 2019-04-18
Also published as: CN107807968A; CN107807968B

Abstract

A Bayesian network-based question-answering method, a question-answering apparatus and a computer-readable storage medium. The method comprises: receiving and parsing a question input by a user by means of a client, so as to recognize in the question a target parameter representing user intent and an attribute parameter associated with the target parameter (S10); inputting the target parameter and the attribute parameter into a pre-trained Bayesian network model, and using a directed acyclic graph and a conditional probability table set of the Bayesian network model to infer a value of the target parameter (S20); returning the value of the target parameter inferred by the Bayesian network model to the user (S30). The present method performs causal inference of a question, and on the basis of an inferred result, answers a question put forth by a user.

Description

Question and answer device, method and storage medium based on Bayesian network

This application claims the priority of the Chinese Patent Application filed on October 13, 2017, the Chinese Patent Office, the application number is 201710955002.X, and the invention name is "Bayesian network-based question answering device, method and storage medium". The content is incorporated into the application by reference.

Technical field

The present application relates to the field of human-computer interaction technologies, and in particular, to a Bayesian network-based question answering device, method, and computer readable storage medium.

Background technique

Human-computer interaction is the science of studying the interaction between systems and users. Among them, the system can be a variety of machines, but also computerized systems and software. Various artificial intelligence systems can be realized through human-computer interaction, for example, an intelligent customer service system, a voice control system, and the like. The intelligent question answering system is a typical application of human-computer interaction. When the customer asks a question, the intelligent question answering system automatically answers the answer to the question to the user. However, in the existing intelligent question answering system, the answers are mostly obtained by searching texts or knowledge bases, and most of them do not have deep reasoning ability.

Summary of the invention

The application provides a Bayesian network-based question answering device, method and computer readable storage medium, the main purpose of which is to enable the intelligent question answering process to have deep reasoning ability.

To achieve the above object, the present application provides a Bayesian network-based question answering device, the device comprising: a memory, a processor, and a memory-based Bayesian network-based quiz program stored on the memory, the Bayesian-based Bayesian When the network quiz program is executed by the processor, the following steps are implemented:

Parameter extraction step: receiving and parsing a question input by the user through the client, to identify a target parameter representing the user's intention and an attribute parameter associated with the target parameter from the question;

Inference step: inputting the target parameter and the attribute parameter into a pre-trained Bayesian network model, and inferring the value of the target parameter by using the directed acyclic graph and the conditional probability table set of the Bayesian network model;

Answer generation step: feedback the value of the target parameter inferred by the Bayesian network model to the user.

In addition, to achieve the above object, the present application further provides a Bayesian network-based question and answer method, the method comprising:

The answer generation step: feeding back the value of the target parameter inferred by the Bayesian network model to the user.

In addition, in order to achieve the above object, the present application further provides a computer readable storage medium on which a Bayesian network-based question answering program is stored, and the Bayesian network-based question answering program is processed. The steps of the Bayesian network-based question and answer method as described above are implemented when the device is executed.

Compared with the prior art, the Bayesian network-based question answering device, method and computer readable storage medium proposed by the present application can perform causal reasoning on user input questions through a Bayesian network model, and based on the inference result Answer the questions posed by the user. Enhance the user interaction experience by understanding user needs through natural dialogue.

DRAWINGS

1 is a schematic diagram of a preferred embodiment of a question and answer device based on a Bayesian network;

2 is a block diagram of a question and answer procedure based on a Bayesian network in FIG. 1;

2a is a schematic diagram of an undirected acyclic graph in a Bayesian network model;

Figure 2b is a schematic diagram of a directed acyclic graph in a Bayesian network model;

Figure 2c is a schematic diagram of a set of probability tables in a Bayesian network model;

3 is a flow chart of a preferred embodiment of a Bayesian network based question and answer method according to the present application;

FIG. 4 is a flowchart of a specific configuration of the Bayesian network model in the Bayesian network-based question and answer method of the present application.

The implementation, functional features, and advantages of the present application will be further described with reference to the accompanying drawings.

Detailed ways

It is understood that the specific embodiments described herein are merely illustrative of the application and are not intended to be limiting.

The application provides a question and answer device 1 based on a Bayesian network. Referring to FIG. 1, a schematic diagram of a preferred embodiment of a question and answer apparatus 1 based on a Bayesian network is provided.

In the present embodiment, the Bayesian network-based question answering device 1 may be an electronic device having a computing function such as a smart phone, a tablet computer, an e-book reader, or a portable computer.

The Bayesian network-based question answering device 1 includes a memory 11, a processor 12, a display 13, a communication bus 14, and a network interface 15. The device obtains business data from a service database over a network.

The memory 11 includes a memory and at least one type of readable storage medium. The memory provides a cache for the operation of the mobile terminal; the readable storage medium may be a non-volatile storage medium such as a flash memory, a hard disk, a multimedia card, a card type memory, or the like. In some embodiments, the readable storage medium may be an internal storage unit of the Bayesian network based question answering device 1, such as the hard disk of the Bayesian network based question answering device 1. In other embodiments, the readable storage medium may also be an external storage device of the Bayesian network-based question answering device 1, such as a plug-in type provided on the Bayesian network-based question answering device 1. Hard disk, smart memory card (SMC), Secure Digital (SD) card, flash card, etc.

In the present embodiment, the readable storage medium of the memory 11 is generally used to store application software and historical service data installed in the Bayesian network-based question answering device 1, such as a Bayesian network-based question answering program 10 , customer history default data, etc. The memory 11 can also be used to temporarily store data that has been output or is about to be output.

The processor 12, in some embodiments, may be a Central Processing Unit (CPU), microprocessor or other data processing chip for running program code or processing data stored in the memory 11, such as performing Bayesian based The Q&A program 10 of the network to implement any of the following Bayesian network-based question and answer methods.

The display 13 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch sensor, or the like in some embodiments. The display 13 is used to display the results of the processing in the Bayesian network based question answering device 1 and the visualized user interface.

Communication bus 14 is used to implement connection communication between these components.

The network interface 15 is mainly used to connect to a server and perform data communication with the server.

Preferably, the Bayesian network based question answering device 1 may further comprise a user interface, including a standard wired interface and a wireless interface. The optional user interface may include an input unit such as a keyboard, a voice input device such as a microphone, a device having a voice recognition function, a voice output device such as an audio, a headphone, and the like.

Preferably, when the Bayesian network based question answering device 1 is a mobile electronic device, such as a mobile phone, at least one type of sensor, such as a light sensor, a motion sensor, and other sensors, may also be included. Specifically, the light sensor includes an ambient light sensor and a proximity sensor, wherein the ambient light sensor can adjust the brightness of the display panel according to the brightness of the ambient light, and the proximity sensor can turn off the display panel and/or the backlight when the mobile phone moves to the ear. As a kind of motion sensor, the accelerometer sensor can detect the magnitude of acceleration in all directions (usually three axes). When it is stationary, it can detect the magnitude and direction of gravity. It can be used to identify the gesture of the mobile phone (such as horizontal and vertical screen switching, related Game, magnetometer attitude calibration), vibration recognition related functions (such as pedometer, tapping), etc.; as well as fingerprint sensors, pressure sensors, iris sensors, molecular sensors, gyroscopes, barometers, hygrometers, Other sensors such as thermometers and infrared sensors will not be described here.

Figure 1 shows only a Bayesian network based question answering device 1 having components 11-14 and a Bayesian network based question answering program 10, but it should be understood that not all illustrated components may be implemented and may be substituted. Implement more or fewer components.

2 is a block diagram of a preferred embodiment of the Bayesian network based challenge program 10 of FIG.

In this embodiment, the Bayesian network-based question answering program 10 can be divided into a parameter extraction module 110, an inference module 120, and an answer generation module 130. The plurality of modules described above are stored in the memory 11 and executed by one or more processors 12 to complete the application. A module as referred to in this application refers to a series of computer program instructions that are capable of performing a particular function. The following description will specifically describe the operations and functions implemented by the parameter extraction module 110, the inference module 120, and the answer generation module 130.

The parameter extraction module 110 is configured to receive and parse a question input by the user through the client, to identify a target parameter representing the user's intention and an attribute parameter associated with the target parameter from the question.

For a given text, the parameter extraction module 110 parses the specific text into parameters of the standard format, and is mainly divided into two parts: the rule template unit 111 and the probability discriminating unit 112.

The rule template unit 111 is mainly configured using a regular expression and a specific syntax structure, wherein a regular expression is used for parameter extraction, and a special syntax structure is used for mapping the extracted parameters in a standard format. Applying a regular expression and a preset grammatical structure, the regular expression is used to extract parameters from the string contained in the natural language question input by the user, and parse the extracted parameters into a preset syntax structure output. A regular expression is a logical formula for string operations. It uses a combination of specific characters defined in advance and a combination of these specific characters to form a "rule string". This "rule string" is used to express a pair of strings. a filtering logic. For example, the rule template for "age" is as follows:

In other words, the client's age is assigned in stages, the value is less than 25 years old, the value is 0, the age of 25-30 is 1, and the age of 31-35 is 2.

In the same way, the customer's academic qualifications are also classified and assigned. The academic qualifications are 0 for primary school and junior high school, 1 for high school, 2 for undergraduate students, and 3 for master's degree.

In the same way, the customer income is also classified and assigned. The annual income is less than 50,000 yuan and the value is 0, 50000-100000 yuan is 1,100,000-2,200,000 yuan is 2, and more than 200,000 yuan is 3.

The probability discriminating unit 112 is mainly trained by the sample and its corresponding classification model, is used to calculate the probability of multiple potential results of a piece of text, and selects a result that best represents the user's intention for parsing. A machine learning model is used to filter out a data structure that best matches the user's intention from all the data structures of the natural language question. For example, the machine learning model can be based on a naive Bayesian classification model, and the naive Bayesian classification model is based on a large number of natural language questions and data structure training corpus training. For example, the user enters the question: "How much is the monthly loan for a 100,000 XXXX bank personal car loan?" The user intention is "interest calculation", and the rule template unit 111 extracts the parameters "XXXX Bank", "personal car loan", "month", "100,000". The resulting data structure may include:

Data Structure 1: (!fb:property.context.LoanAmountRange(argmax(number 1)(number 10)(and(fb:type.loan.loanN fb:loanN.gerengouchedaikuan1)(fb:type.loan.company fb:company .XXXX))(reverse(lambda x(!fb:rank.entity.rank(var x)))))))

Data Structure 2: (!fb:property.context.MonthFeeRate(!fb:property.context.LoanAmountRange fb:company.XXXX))

Data Structure 3: fb:company.XXXX

Data Structure 4: (*(!fb:attribute.attribute.MonthFeeRateD(!fb:property.context.MonthFeeRate(and(and(fb:type.loan.loanNfb:loanN.gerengouchedaikuan1)(fb:type.loan.company fb :company.XXXX))(fb:property.context.LoanAmountRange(fb:attribute.attribute.MaxLoanAmountRange(>=(number 100000))))))))(number100000))

After the four data structures pass the naive Bayesian classification model, the probability discriminating unit 112 filters out a data structure as the data structure most representative of the user's intention.

Further, the parameter extraction module 110 is further configured to convert the extracted target parameters and attribute parameters into parameters in a standard format. For example, in the question “What is the overdue rate of graduate students with an annual income of 300,000 yuan?”, the standard mapping of parameters includes: annual income-recent_income-300000 yuan-3, education-education-Master-3, repayment -debt-overdue-1. Then the question will be parsed as follows:

The inference module 120 is configured to input the target parameter and the attribute parameter into a pre-trained Bayesian network model, and infer the target parameter by using the directed acyclic graph and the conditional probability table set of the Bayesian network model. The value.

The Bayesian network's reasoning is to use the structure of the Bayesian network and its conditional probability table to calculate the probability of taking some other nodes after given the node attribute values. We use a message passing algorithm for precise reasoning. It mainly allocates a processor to each node. Each processor uses the probability passed by the neighboring node and the conditional probability stored in the processor to calculate. Get its own posterior probability and propagate the result to neighboring nodes.

For example, when the question becomes "How much is the overdue rate of the customer with an annual income of 300,000 yuan?", the customer attribute appearing in the question only has annual income -recent_income-300000 -3, repayment -debt-overdue -1 . According to the above-mentioned directed acyclic graph and conditional probability table, when the customer's annual income is determined, the probability of the customer's repayment overdue can be inferred according to the customer's academic qualifications, that is, different academic qualifications will affect the customer's repayment overdue. The probability.

The answer generating module 130 is configured to feed back the value of the target parameter inferred by the Bayesian network model to the user.

When the user enters the question into the Bayesian network model, the value of the target parameter is obtained as follows:

Key:income=3,education=3;debt=1;value:0.01935

In order to make the result more intuitive, the answer generation module 130 converts the output target parameter value of the standard data format into text, and feeds the result in text form as an answer to the user. The results of the above target parameter conversion are as follows:

The overdue rate of graduate students with an annual income of 300,000 yuan is 1.935%.

The Bayesian network-based question and answer system of the present application can understand the user's needs through natural dialogue, and perform deep reasoning according to the user's question to improve the user's human-computer interaction experience.

In addition, the present application also provides a question and answer method based on Bayesian network. Referring to FIG. 3, it is a flowchart of a preferred embodiment of a Bayesian network based question and answer method according to the present application. The method can be performed by a device that can be implemented by software and/or hardware.

In this embodiment, the Bayesian network based question and answer method includes:

Step S10, receiving and parsing a question input by the user through the client, to identify a target parameter representing the user's intention and an attribute parameter associated with the target parameter from the question.

For a given text, the specific text is parsed into the parameters of the standard format, which is mainly divided into two parts: the rule template and the probability discriminant.

Rule templates are primarily configured using regular expressions and specific syntax structures, where regular expressions are used for parameter extraction and special syntax structures are used for standard format mapping of extracted parameters. Applying a regular expression and a preset grammatical structure, the regular expression is used to extract parameters from the string contained in the natural language question input by the user, and parse the extracted parameters into a preset syntax structure output. A regular expression is a logical formula for string operations. It uses a combination of specific characters defined in advance and a combination of these specific characters to form a "rule string". This "rule string" is used to express a pair of strings. a filtering logic. For example, the rule template for "age" is as follows:

That is to say, the customer's age is assigned in stages, the value is less than 25 years old and the value is 0, 25-30 years old is 1, 31-35 years old is 2,...

In the same way, the customer's academic qualifications are also classified and assigned. The academic qualification is 0 for the elementary school-junior high school, 1 for the high school, 2 for the undergraduate, and 3 for the graduate student.

The probabilistic discriminant is mainly trained by the sample and its corresponding classification model, used to calculate the probability of multiple potential results of a piece of text, and select a result that best represents the user's intention for parsing. A machine learning model is used to filter out a data structure that best matches the user's intention from all the data structures of the natural language question. For example, the machine learning model can be based on a naive Bayesian classification model, and the naive Bayesian classification model is based on a large number of natural language questions and data structure training corpus training. For example, the user enters the question: "How much is the monthly loan for a 100,000 XXXX bank personal car loan?" The user intention is "interest calculation", and the rule template unit 111 extracts the parameters "XXXX Bank", "personal car loan", "month", "100,000". The resulting data structure may include:

Data Structure 3: fb:company.XXXX

Data Structure 4: (*(!fb:attribute.attribute.MonthFeeRateD(!fb:property.context.MonthFeeRate(and(and(fb:type.loan.loanNfb:loanN.gerengouchedaikuan1)(fb:type.loan.company fb :company.XXXX)) (fb:property.context.LoanAmountRange(fb:attribute.attribute.MaxLoanAmountRange(>=(number 100000))))))))(number100000))

Further, the step S10 further includes: converting the extracted target parameter and the attribute parameter into a parameter of a standard format. For example, in the question “What is the overdue rate of graduate students with an annual income of 300,000 yuan?”, the standard mapping of parameters includes: annual income-recent_income-300000 yuan-3, education-education-Master-3, repayment -debt-overdue-1. Then the question will be parsed as follows:

Step S20: input the target parameter and the attribute parameter into a pre-trained Bayesian network model, and use the directed acyclic graph and the conditional probability table set of the Bayesian network model to infer the value of the target parameter.

The Bayesian network's reasoning is to use the structure of the Bayesian network and its conditional probability table to calculate the probability of taking some other nodes after given the node attribute values.

We use a message passing algorithm for precise reasoning. It mainly allocates a processor to each node. Each processor uses the probability passed by the neighboring node and the conditional probability stored in the processor to calculate. Get its own posterior probability and propagate the result to neighboring nodes.

In step S30, the value of the target parameter inferred by the Bayesian network model is fed back to the user.

Key:income=3,education=3;debt=1;value:0.01935

In order to make the result more intuitive, the value of the target parameter of the output standard data format is converted into text, and the result in text form is fed back to the user as an answer. The results of the above target parameter conversion are as follows:

The Bayesian network-based question and answer method of the present application can understand the user's needs through natural dialogue, and perform deep reasoning according to the user's question to improve the user's human-computer interaction experience.

A second embodiment of the Bayesian network based question and answer method of the present application is proposed based on the first embodiment. Referring to FIG. 4, in the embodiment, the specific construction steps of the Bayesian network model in FIG. 3 include:

Step S01: extracting, from each historical default data of the historical business data, an attribute associated with the default customer, and calculating conditional mutual information between the attributes;

Step S02: Sort the conditional mutual information values of each attribute in descending order, select an attribute pair with a high conditional mutual information value as a node, and follow a principle of not generating a loop, and construct a maximum weight span tree until n-1 pieces are selected for n nodes. Side, forming an undirected acyclic graph;

Step S03, determining a root node of each node in the undirected acyclic graph, the direction from the root node to the child node is a direction between the nodes, and changing the undirected acyclic graph into a directed acyclic graph;

Step S04: Calculate a conditional probability between random variables represented by each node in the directed acyclic graph according to historical service data, and obtain a conditional probability table set of the Bayesian network model.

The Bayesian network is mainly used to determine the topological relationship between random variables to form a DAG (Directed Acyclic Graph). The method used is to first determine the nodes of the Bayesian network and then use a large amount of training data to learn. The structure of the Bayesian network. Structure learning is performed using TAN (Tree Augmented Naive Bays) algorithm.

Training Bayesian networks, that is, parameter learning, is mainly to determine the conditional probability table, that is, the conditional dependency between random variables. Parameter learning is mainly divided into parameter learning of complete data and parameter learning of incomplete data. Complete data means that each instance has complete observation data, that is, both educational data and income data, and incomplete data refers to certain Some examples are missing or observing anomalies. For example, some people have educational data, others have no educational data and have income data. Usually, it is incomplete data. The parameter learning of the complete observation data adopts the method of maximum likelihood estimation. For the parameter learning of incomplete data, the EM algorithm (Expectation-maximization) is adopted.

Calculating a conditional probability between random variables represented by each node in the DAG according to historical business data, and obtaining a conditional probability table set of the Bayesian network model.

The Bayesian network in this embodiment includes a DAG and a set of probability tables, as shown in Figures 2c, 2d.

In Figure 2c, three nodes in the DAG represent three random variables, and the directed edges represent conditional dependencies between random variables.

In other embodiments, the random variable represented by each node may be a directly observable variable or a hidden variable, which refers to a variable that cannot be directly observed or can be observed but still needs to be integrated by other methods. Variables, such as intelligence levels.

In Figure 2d, each element in the conditional probability table corresponds to a unique node in the DAG, storing the joint conditional probability of this node for all its immediate precursor nodes:

Among them, E is the academic qualification of the defaulting customer, I is the annual income, P is the probability, T is the overdue condition of repayment, and F is the normal situation of repayment.

For example, extract historically related attributes from a financial service institution's historical default data, such as: default customer age, education, annual income, gender, nationality, work experience, assets (whether there is a car or room), whether Have insurance and marital status, etc., and calculate conditional mutual information between different attributes.

In the TAN, there will be the addition of class variable attributes, because the premise of the association between attributes is to recalculate under certain classification attributes. Different class attribute values will have different attribute associations, so the calculation formula is as follows :

Where P(x, y|c) is the joint distribution of two random variables x and y, P(x|c), P(y|c) are the marginal distribution of random variables X and Y, respectively, and C is the class variable. X, Y respectively represent the attribute variables associated with the default customer, and I(X, Y|C) represents the conditional mutual information between the attributes X and Y.

If the conditional mutual information between the above attributes is calculated as: mutual information value of education and annual income (0.8)> mutual information value of annual income and overdue (0.7)> mutual information value of age and annual income (0.4)>sex Mutual information value with overdue (0.2). Then, the attribute pairs with higher mutual information values are sequentially selected as nodes.

Further, the step S02 further includes: presetting a mutual information threshold as a criterion for retaining a plurality of attribute pairs or edges. The reason why the value of mutual information is selected from high to low is to preserve the edge of the associated dependency with higher relevance. Assuming that the preset mutual information threshold is 0.5, then the attribute pair whose mutual information value is higher than 0.5 is selected as the node, that is, the academic qualification, the annual income, and the overdue as nodes, forming an undirected acyclic graph as shown in FIG. 2a.

The "overdue" node, the "educational" node, and the "annual income" node are connected to form a directed acyclic graph as shown in Figure 2b.

The Bayesian network-based question and answer method of the present application, by constructing a Bayesian network model, enables the question and answer method to understand the user's needs through natural dialogue, and performs deep reasoning according to the user's question to improve the user's human-computer interaction experience.

In addition, the embodiment of the present application further provides a computer readable storage medium, where the Bayesian network-based question and answer program is stored on the computer readable storage medium, and the Bayesian network-based question and answer program is executed by the processor. Implement the following operations:

The specific implementation manner of the computer readable storage medium of the present application is substantially the same as the specific implementation method of the above-mentioned Bayesian network based question and answer method, and therefore will not be described again.

It is to be understood that the term "comprises", "comprising", or any other variants thereof, is intended to encompass a non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a series of elements includes those elements. It also includes other elements not explicitly listed, or elements that are inherent to such a process, device, item, or method. An element that is defined by the phrase "comprising a ..." does not exclude the presence of additional equivalent elements in the process, the device, the item, or the method that comprises the element.

The serial numbers of the embodiments of the present application are merely for the description, and do not represent the advantages and disadvantages of the embodiments. Through the description of the above embodiments, those skilled in the art can clearly understand that the foregoing embodiment method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be through hardware, but in many cases, the former is better. Implementation. Based on such understanding, the technical solution of the present application, which is essential or contributes to the prior art, may be embodied in the form of a software product stored in a storage medium (such as ROM/RAM as described above). , a disk, an optical disk, including a number of instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the methods described in the various embodiments of the present application.

The above is only a preferred embodiment of the present application, and is not intended to limit the scope of the patent application, and the equivalent structure or equivalent process transformations made by the specification and the drawings of the present application, or directly or indirectly applied to other related technical fields. The same is included in the scope of patent protection of this application.

Claims

A Bayesian network-based question answering device, comprising: a memory, a processor, wherein the memory stores a Bayesian network-based question answering program, and the Bayesian network-based question answering program is The processor implements the following steps when executed:

Parameter extraction step: receiving and parsing a question input by the user through the client, to identify a target parameter representing the user's intention and an attribute parameter associated with the target parameter from the question;

Inference step: inputting the target parameter and the attribute parameter into a pre-trained Bayesian network model, and inferring the value of the target parameter by using the directed acyclic graph and the conditional probability table set of the Bayesian network model;

The answer generation step: feeding back the value of the target parameter inferred by the Bayesian network model to the user.
The Bayesian network-based question answering device according to claim 1, wherein the modeling step of the Bayesian network specifically comprises:

Extracting the attributes associated with the default customer from each historical default data of the historical business data, and calculating the conditional mutual information value between the attributes;

The conditional mutual information values of each attribute are sorted in descending order, and the attribute pairs with high conditional mutual information values are selected as nodes, and the principle of not generating loops is constructed, and the maximum weight span tree is constructed until n-1 edges are selected for n nodes. An undirected acyclic graph;

Determining the root node of each node in the undirected acyclic graph, the direction from the root node to the child node is the direction between the nodes, and changing the undirected acyclic graph into a directed acyclic graph;

Calculating a conditional probability between random variables represented by each node in the directed acyclic graph according to historical service data, and obtaining a conditional probability table set of the Bayesian network model.
The Bayesian network-based question answering apparatus according to claim 2, wherein the calculation formula of the conditional mutual information value between the attributes is as follows:

Where P(x, y|c) is the joint distribution of two random variables x and y, P(x|c)P, P(y|c)P are the marginal distribution of random variables X and Y, respectively, C is The class variable, X and Y respectively represent the attribute variables associated with the default customer, and I(X, Y|C) represents the conditional mutual information between the attributes X and Y.
The Bayesian network-based question answering device according to claim 1, wherein the parameter extraction step comprises:

The extracted target parameters and attribute parameters are converted into parameters in a standard format.
The Bayesian network-based question answering device according to claim 2, wherein the parameter extraction step comprises:

The extracted target parameters and attribute parameters are converted into parameters in a standard format.
The Bayesian network-based question answering device according to claim 3, wherein the parameter extraction step comprises:

The extracted target parameters and attribute parameters are converted into parameters in a standard format.
The Bayesian network-based question answering device according to any one of claims 1 to 6, wherein the answer generating step comprises:

The value of the target parameter inferred by the Bayesian network model is converted into text, and the result of the text format is fed back to the user as an answer.
A question and answer method based on Bayesian network, characterized in that the method comprises:

Parameter extraction step: receiving and parsing a question input by the user through the client, to identify a target parameter representing the user's intention and an attribute parameter associated with the target parameter from the question;

Inference step: inputting the target parameter and the attribute parameter into a pre-trained Bayesian network model, and inferring the value of the target parameter by using the directed acyclic graph and the conditional probability table set of the Bayesian network model;

The answer generation step: feeding back the value of the target parameter inferred by the Bayesian network model to the user.
The Bayesian network-based question and answer method according to claim 8, wherein the modeling step of the Bayesian network specifically comprises:

Extracting the attributes associated with the default customer from each historical default data of the historical business data, and calculating the conditional mutual information value between the attributes;

The conditional mutual information values of each attribute are sorted in descending order, and the attribute pairs with high conditional mutual information values are selected as nodes, and the principle of not generating loops is constructed, and the maximum weight span tree is constructed until n-1 edges are selected for n nodes. An undirected acyclic graph;

Determining the root node of each node in the undirected acyclic graph, the direction from the root node to the child node is the direction between the nodes, and changing the undirected acyclic graph into a directed acyclic graph;

Calculating a conditional probability between random variables represented by each node in the directed acyclic graph according to historical service data, and obtaining a conditional probability table set of the Bayesian network model.
The Bayesian network-based question and answer method according to claim 9, wherein the calculation formula of the conditional mutual information value between the attributes is as follows:

Where P(x, y|c) is the joint distribution of two random variables x and y, P(x|c)P, P(y|c)P are the marginal distribution of random variables X and Y, respectively, C is The class variable, X and Y respectively represent the attribute variables associated with the default customer, and I(X, Y|C) represents the conditional mutual information between the attributes X and Y.
The Bayesian network-based question and answer method according to claim 8, wherein the parameter extraction step comprises:

The extracted target parameters and attribute parameters are converted into parameters in a standard format.
The Bayesian network-based question and answer method according to claim 9, wherein the parameter extraction step comprises:

The extracted target parameters and attribute parameters are converted into parameters in a standard format.
The Bayesian network-based question and answer method according to claim 10, wherein the parameter extraction step comprises:

The extracted target parameters and attribute parameters are converted into parameters in a standard format.
The Bayesian network-based question answering method according to any one of claims 8 to 13, wherein the answer generating step comprises:

The value of the target parameter inferred by the Bayesian network model is converted into text, and the result of the text format is fed back to the user as an answer.
A computer readable storage medium, characterized in that the computer readable storage medium stores a Bayesian network-based question and answer program, and the Bayesian network-based question and answer program is executed by the processor to implement the following steps:

Parameter extraction step: receiving and parsing a question input by the user through the client, to identify a target parameter representing the user's intention and an attribute parameter associated with the target parameter from the question;

Inference step: inputting the target parameter and the attribute parameter into a pre-trained Bayesian network model, and inferring the value of the target parameter by using the directed acyclic graph and the conditional probability table set of the Bayesian network model;

The answer generation step: feeding back the value of the target parameter inferred by the Bayesian network model to the user.
The computer readable storage medium according to claim 15, wherein the step of constructing the Bayesian network specifically comprises:

Extracting the attributes associated with the default customer from each historical default data of the historical business data, and calculating the conditional mutual information value between the attributes;

The conditional mutual information values of each attribute are sorted in descending order, and the attribute pairs with high conditional mutual information values are selected as nodes, and the principle of not generating loops is constructed, and the maximum weight span tree is constructed until n-1 edges are selected for n nodes. An undirected acyclic graph;

Determining the root node of each node in the undirected acyclic graph, the direction from the root node to the child node is the direction between the nodes, and changing the undirected acyclic graph into a directed acyclic graph;

Calculating a conditional probability between random variables represented by each node in the directed acyclic graph according to historical service data, and obtaining a conditional probability table set of the Bayesian network model.
The computer readable storage medium according to claim 16, wherein the conditional mutual information value between the attributes is calculated as follows:

Where P(x, y|c) is the joint distribution of two random variables x and y, P(x|c)P, P(y|c)P are the marginal distribution of random variables X and Y, respectively, C is The class variable, X and Y respectively represent the attribute variables associated with the default customer, and I(X, Y|C) represents the conditional mutual information between the attributes X and Y.
The computer readable storage medium of claim 15, wherein the parameter extraction step comprises:

The extracted target parameters and attribute parameters are converted into parameters in a standard format.
The computer readable storage medium according to claim 16 or 17, wherein the parameter extraction step comprises:

The extracted target parameters and attribute parameters are converted into parameters in a standard format.
The computer readable storage medium of claim 19, wherein the answer generating step comprises:

The value of the target parameter inferred by the Bayesian network model is converted into text, and the result of the text format is fed back to the user as an answer.