CN112966089A

CN112966089A - Problem processing method, device, equipment, medium and product based on knowledge base

Info

Publication number: CN112966089A
Application number: CN202110329818.8A
Authority: CN
Inventors: 张林林; 马文莹; 周颖
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2021-03-26
Filing date: 2021-03-26
Publication date: 2021-06-15

Abstract

The present disclosure provides a problem processing method based on a knowledge base, which is applied to an intelligent question-answering system, wherein the intelligent question-answering system can respond to input information and provide feedback information, and the method comprises the following steps: the method comprises the steps of responding to received input question sentences, obtaining a question classification model, wherein the question classification model is obtained through training based on knowledge data in a knowledge base, determining question types of the input question sentences based on the question classification model, different question types correspond to different question processing methods, and obtaining answer sentences matched with the input question sentences and output answer sentences from the knowledge base based on the question processing methods corresponding to the question types. The present disclosure also provides a problem handling device, apparatus, medium, and product based on a knowledge base. The present disclosure relates to the field of artificial intelligence, and the provided methods and apparatus may be applied, for example, in the field of finance or other fields.

Description

Problem processing method, device, equipment, medium and product based on knowledge base

Technical Field

The present disclosure relates to the field of computer application technologies, and in particular, to a method, an apparatus, a device, a medium, and a product for problem handling based on a knowledge base.

Background

This section is intended to provide a background or context to the embodiments of the disclosure recited in the claims. The description herein is not admitted to be prior art by inclusion in this section.

For enterprises, the service range is widened, the customer requirements are comprehensively met, greater profits are brought to the companies, the method is an important direction of current development, a good enterprise image with thorough service and efficient work is created in the customer center, the core goal pursued by the enterprises is achieved, and the customer service is an effective means for improving the reputation of the enterprises and maintaining the external image and is present in various industries. Because enterprises usually provide services for customers in an artificial seat mode, the mastering degree of enterprise knowledge of customer service staff directly determines the service level and further directly determines the satisfaction degree of the customers for the services. With increasingly burdensome customer care tasks, more and more businesses are becoming aware of the importance of building a large array of professional customer care teams. With the increasing frequency of service function iteration updating and the continuous expansion of service range in the longitudinal and transverse directions, timely and satisfactory customer service is difficult to provide only by means of manual force, and consultation of various services is responded by an intelligent customer service question-answering system in combination with the establishment of a huge question-answering knowledge base.

In the course of implementing the disclosed concept, the inventors found that there are at least the following problems in the prior art: the current intelligent customer service question-answering system is generally divided according to product categories, namely, when identifying customer questions, customer service personnel need to make clear the product names of the questions in a manner of guiding the customers, and then search answers corresponding to the questions in a question-answering knowledge base according to the product names.

Disclosure of Invention

In view of the above, the present disclosure provides a method, an apparatus, a device, a medium, and a program product for processing a question based on a knowledge base, where the method executes a corresponding question processing method based on different question types to obtain an answer sentence matching the input question sentence from the knowledge base, so that the above-mentioned technical problem caused by searching the answer sentence according to a product category can be avoided.

In order to achieve the above object, an aspect of the present disclosure provides a problem processing method based on a knowledge base, which may include: responding to the received input question sentences, and obtaining a question classification model, wherein the question classification model is obtained by training based on knowledge data in the knowledge base; determining the question type of the input question sentence based on the question classification model, wherein different question types correspond to different question processing methods; obtaining answer sentences matched with the input question sentences from the knowledge base based on a question processing method corresponding to the question types; and outputting the answer sentence.

According to an embodiment of the present disclosure, the obtaining of the answer sentence matched with the input question sentence from the knowledge base based on the question processing method corresponding to the question type may include: generating a first feature vector of the input question sentence under the condition that the question type is determined to be a comprehensive type; extracting first entity information and first attribute information of the input question sentence; obtaining a candidate entity set matched with the first entity information from the knowledge base based on the first entity information, wherein the candidate entity set comprises a plurality of second entities; determining a second feature vector of each second entity based on second attribute information of the plurality of second entities; determining a similarity value between the input question sentence and each second entity based on the first feature vector and the second feature vector of each second entity; and obtaining a second entity corresponding to the maximum similarity value from the knowledge base as an answer sentence matched with the input question sentence.

According to an embodiment of the present disclosure, the obtaining, from the knowledge base, a candidate entity set matching the first entity information based on the first entity information may include: obtaining a path length threshold, wherein the path length threshold is used for representing a maximum value of a path length between the first entity and the second entity; obtaining a candidate path from the knowledge base, based on the path length threshold, having a path length with the first entity not exceeding the path length threshold; and using a plurality of second entities covered by the candidate path as a candidate entity set matched with the first entity.

According to an embodiment of the present disclosure, the obtaining of the answer sentence matched with the input question sentence from the knowledge base based on the question processing method corresponding to the question type may include: determining the database type of the knowledge base under the condition that the problem type is determined to be a common type; converting the input question sentence into a query sentence matched with the graph database type according to a preset conversion rule; and obtaining answer sentences matched with the input question sentences from the knowledge base based on the query sentences.

According to an embodiment of the present disclosure, the converting the input question statement into the query statement matching the graph database type according to the preset conversion rule may include: extracting third entity information of the input question sentence under the condition that the common type is a question and answer type or a flow type; extracting third attribute information associated with the third entity information; determining fourth entity information which is the same as or similar to the third entity information in the knowledge base based on the third entity information; determining fourth attribute information that is the same as or similar to the third attribute information in the knowledge base based on the third attribute information; and generating a query sentence matched with the graph database type based on the fourth entity information and the fourth attribute information.

According to an embodiment of the present disclosure, the third entity information and the third attribute information corresponding to the question and answer type may have at least the following correspondence relationship: a piece of third entity information; one piece of third entity information corresponds to one piece of third attribute information; one piece of third entity information corresponds to a plurality of pieces of third attribute information; the plurality of pieces of third entity information correspond to one piece of third attribute information.

According to an embodiment of the present disclosure, the obtaining of the answer sentence matched with the input question sentence from the knowledge base based on the query sentence may include: obtaining an attribute list associated with the third entity information from the knowledge base under the condition that the common type is a flow type, wherein the attribute list comprises a plurality of attribute information which are executed in order; determining an execution logic relationship of the attribute information corresponding to the query statement among the plurality of sequentially executed attribute information based on the attribute list; and obtaining an answer sentence matched with the input question sentence from the knowledge base based on the execution logic relation.

According to an embodiment of the present disclosure, the converting the input question statement into the query statement matching the graph database type according to the preset conversion rule may include: determining a subtype corresponding to the input question sentence when the common type is a comparison type; extracting fifth body information, fifth attribute information and comparison condition information of the input question sentence, wherein the comparison condition information is used for representing the subtype; determining sixth entity information identical or similar to the fifth entity information in the knowledge base based on the fifth entity information; determining sixth attribute information that is the same as or similar to the fifth attribute information in the knowledge base based on the fifth attribute information; and generating a query sentence matched with the graph database type based on the sixth entity information, the sixth attribute information, and the comparison condition information.

According to an embodiment of the present disclosure, in a case where the subtype includes a comparison subtype, the comparison condition information may include comparison condition information corresponding to the comparison subtype, where the comparison subtype is used to indicate that attributes of two types of entities are compared; in a case where the sub-type includes a search sub-type, the comparison condition information may include search condition information corresponding to the search sub-type, where the search sub-type indicates that an entity satisfying the search condition information is searched; in the case where the sub-type includes a generalised sub-type, the comparison condition information may include generalised condition information corresponding to the generalised sub-type, wherein the generalised sub-type indicates a most significant value of the query attribute.

According to an embodiment of the present disclosure, the method may further include: selecting training sample data from the knowledge base according to a preset rule, wherein the training sample data comprises labeled data of the problem type; obtaining an initial problem classification model; and performing deep neural network training on the initial problem classification model based on the training sample data to generate the problem classification model.

In order to achieve the above object, another aspect of the present disclosure provides a knowledge-base-based question processing apparatus applied in an intelligent question-answering system capable of responding to input information and providing feedback information, which may include: a classification model obtaining module, configured to obtain a problem classification model in response to receiving an input problem statement, where the problem classification model is obtained by training based on knowledge data in the knowledge base; a question type determining module, configured to determine a question type of the input question statement based on the question classification model, where different question types correspond to different question processing methods; an answer sentence obtaining module, configured to obtain an answer sentence matched with the input question sentence from the knowledge base based on a question processing method corresponding to the question type; and the answer sentence output module is used for outputting the answer sentences.

According to an embodiment of the present disclosure, the answer sentence obtaining module may include: a generation submodule, configured to generate a first feature vector of the input question statement when the question type is determined to be a comprehensive type; the first extraction submodule is used for extracting first entity information and first attribute information of the input question sentence; a first obtaining sub-module, configured to obtain, from the knowledge base, a candidate entity set that matches the first entity information based on the first entity information, where the candidate entity set includes a plurality of second entities; a first determining submodule, configured to determine a second feature vector of each second entity based on second attribute information of the plurality of second entities; a second determining sub-module, configured to determine a similarity value between the input question statement and each of the second entities based on the first feature vector and the second feature vector of each of the second entities; and a second obtaining submodule, configured to obtain, from the knowledge base, a second entity corresponding to the maximum similarity value as an answer sentence matched with the input question sentence.

According to an embodiment of the present disclosure, the first obtaining sub-module may include: a first obtaining unit, configured to obtain a path length threshold, where the path length threshold is used to characterize a maximum value of a path length between the first entity and the first entity; a second obtaining unit, configured to obtain, from the knowledge base, a candidate path having a path length with the first entity that does not exceed the path length threshold, based on the path length threshold; and a first processing unit, configured to use a plurality of second entities covered by the candidate path as a candidate entity set matching the first entity.

According to an embodiment of the present disclosure, the answer sentence obtaining module may include: a third determining submodule, configured to determine a type of a map database of the knowledge base, if it is determined that the type of the problem is a general type; a first conversion sub-module, configured to convert the input question statement into a query statement matching the graph database type according to a preset conversion rule; and a third obtaining sub-module, configured to obtain, based on the query statement, an answer statement matching the input question statement from the knowledge base.

According to an embodiment of the present disclosure, the first conversion sub-module may include: a first extracting unit, configured to extract third entity information of the input question statement when the general type is a question and answer type or a flow type; a second extracting unit configured to extract third attribute information associated with the third entity information; a first determination unit configured to determine, based on the third entity information, fourth entity information that is the same as or similar to the third entity information in the knowledge base; a second determining unit configured to determine, based on the third attribute information, fourth attribute information that is the same as or similar to the third attribute information in the knowledge base; and a first generating unit configured to generate a query sentence matching the graph database type based on the fourth entity information and the fourth attribute information.

According to an embodiment of the present disclosure, the answer sentence obtaining module may include: a fourth obtaining sub-module, configured to obtain, from the knowledge base, an attribute list associated with the third entity information when the common type is a flow type, where the attribute list includes a plurality of attribute information that are executed in order; a fourth determining submodule, configured to determine, based on the attribute list, an execution logical relationship of attribute information corresponding to the query statement among the plurality of sequentially executed attribute information; and a fifth obtaining submodule, configured to obtain, based on the execution logical relationship, an answer sentence matched with the input question sentence from the knowledge base.

According to an embodiment of the present disclosure, the first conversion sub-module may include: a third determination unit configured to determine a subtype corresponding to the input question sentence when the normal type is a comparison type; a third extracting unit, configured to extract fifth volume information, fifth attribute information, and comparison condition information of the input question sentence, where the comparison condition information is used to represent the subtype; a fourth determination unit configured to determine, based on the fifth volume information, sixth entity information that is the same as or similar to the fifth volume information in the knowledge base; a fifth determining unit configured to determine, based on the fifth attribute information, sixth attribute information that is the same as or similar to the fifth attribute information in the knowledge base; and a second generation unit configured to generate a query expression matching the graph database type based on the sixth entity information, the sixth attribute information, and the comparison condition information.

According to an embodiment of the present disclosure, in a case where the subtype includes a comparison subtype, the comparison condition information may include comparison condition information corresponding to the comparison subtype, where the comparison subtype is used to indicate that attributes of two types of entities are compared; in a case where the sub-type includes a search sub-type, the comparison condition information may include search condition information corresponding to the search sub-type, where the search sub-type indicates that an entity satisfying the search condition information is searched; in the case that the above-mentioned subtype may include an inductive subtype, the above-mentioned comparison condition information includes inductive condition information corresponding to the above-mentioned inductive subtype indicating the most significant value of the query attribute.

According to an embodiment of the present disclosure, the apparatus may further include: the training sample selection module is used for selecting training sample data from the knowledge base according to a preset rule, wherein the training sample data comprises marking data of the problem type; the initial model obtaining module is used for obtaining an initial problem classification model; and the classification model generation module is used for carrying out deep neural network training on the initial problem classification model based on the training sample data to generate the problem classification model.

In order to achieve the above object, another aspect of the present disclosure provides an electronic device including: one or more processing cores, a memory to store one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the repository-based problem handling method as described above.

To achieve the above objects, another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions for implementing the knowledge-base based question processing method as described above when executed.

To achieve the above object, another aspect of the present disclosure provides a computer program comprising computer executable instructions for implementing the knowledge-base based problem processing method as described above when executed.

According to the embodiment of the disclosure, the question type corresponding to the input question sentence is determined based on the question classification model obtained by training knowledge data in the knowledge base, the question processing method corresponding to the question type is determined according to the question processing methods corresponding to different question types, and the answer sentence matched with the input question sentence is obtained from the knowledge base and is output as the feedback information of the input question sentence.

Drawings

The above and other objects, features and advantages of exemplary embodiments of the present disclosure will become readily apparent from the following detailed description read in conjunction with the accompanying drawings. Several embodiments of the present disclosure are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which:

FIG. 1 schematically illustrates a system architecture of a knowledge-base based problem handling method and apparatus suitable for use with embodiments of the present disclosure;

FIG. 2 schematically illustrates a knowledge-graph of a knowledge-base based problem handling method and apparatus suitable for use with embodiments of the present disclosure;

FIG. 3 schematically illustrates a flow diagram of a knowledge-base based problem handling method in accordance with an embodiment of the present disclosure;

FIG. 4 schematically illustrates a flow diagram of a knowledge-base based question processing method according to another embodiment of the present disclosure;

FIG. 5 schematically illustrates a flow diagram of a knowledge-base based question processing method according to another embodiment of the present disclosure;

FIG. 6 schematically illustrates a flow diagram of a knowledge-base based question processing method according to another embodiment of the present disclosure;

FIG. 7 schematically illustrates a flow diagram of a knowledge-base based question processing method according to another embodiment of the present disclosure;

FIG. 8 schematically illustrates a block diagram of a knowledge-base based problem processing apparatus, in accordance with an embodiment of the present disclosure;

FIG. 9 schematically illustrates a block diagram of a knowledge-base based question processing apparatus according to another embodiment of the present disclosure;

FIG. 10 schematically illustrates a schematic diagram of a computer-readable storage medium product suitable for implementing the knowledge-base based question processing method described above, in accordance with an embodiment of the present disclosure; and

FIG. 11 schematically illustrates a block diagram of an electronic device adapted to implement the knowledge-base based question processing method described above, in accordance with an embodiment of the present disclosure.

In the drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

It should be noted that the figures are not drawn to scale and that elements of similar structure or function are generally represented by like reference numerals throughout the figures for illustrative purposes.

Detailed Description

Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood that the description is illustrative only and is not intended to limit the scope of the present disclosure. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the disclosure. It may be evident, however, that one or more embodiments may be practiced without these specific details. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present disclosure.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The terms "comprises," "comprising," and the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components. All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It is noted that the terms used herein should be interpreted as having a meaning that is consistent with the context of this specification and should not be interpreted in an idealized or overly formal sense.

Where a convention analogous to "at least one of A, B and C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B and C" would include but not be limited to systems that have a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.). Where a convention analogous to "A, B or at least one of C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B or C" would include but not be limited to systems that have a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.).

Some block diagrams and/or flow diagrams are shown in the figures. It will be understood that some blocks of the block diagrams and/or flowchart illustrations, or combinations thereof, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable knowledge base-based problem handling device such that the instructions, which execute via the processor, create means for implementing the functions/acts specified in the block diagrams and/or flowchart block or blocks. The techniques of this disclosure may be implemented in hardware and/or software (including firmware, microcode, etc.). In addition, the techniques of this disclosure may take the form of a computer program product on a computer-readable storage medium having instructions stored thereon for use by or in connection with an instruction execution system.

In the related art, the current customer service auxiliary system is generally divided according to product categories, and when identifying problems of customers, customer service personnel need to guide the customers to clearly identify product names and then search according to the product names, so that time and labor are wasted, and a search result is not accurate enough.

To this end, the present disclosure provides a knowledge-base-based question processing method applied in an intelligent question-answering system capable of responding to input information and providing feedback information, the method including a question type determination phase and an answer sentence matching phase. In the problem type determining stage, firstly, in response to receiving input problem sentences, a problem classification model is obtained, the problem classification model is obtained based on pre-training of knowledge data in a knowledge base, then, based on the problem classification model, the problem types of the input problem sentences are determined, and different problem types correspond to different problem processing methods. In the answer sentence matching stage, firstly, an answer sentence matched with the input question sentence is obtained from the knowledge base based on a question processing method corresponding to the question type, and then the answer sentence is output.

It should be noted that the problem processing method and problem processing device based on the knowledge base provided by the present disclosure can be used in a customer service system in the financial field, and can also be used in a customer service system in any field except the financial field. Therefore, the application fields of the problem processing method and the problem processing device based on the knowledge base provided by the present disclosure are not particularly limited.

FIG. 1 schematically illustrates a system architecture of a knowledge-base based problem handling method and apparatus suitable for use with embodiments of the present disclosure. It should be noted that fig. 1 is only an example of a system architecture to which the embodiments of the present disclosure may be applied to help those skilled in the art understand the technical content of the present disclosure, and does not mean that the embodiments of the present disclosure may not be applied to other devices, systems, environments or scenarios.

As shown in fig. 1, the system architecture 100 according to this embodiment may include

terminal devices

101, 102, 103, a network 104 and a server 105. The network 104 serves as a medium for providing communication links between the

terminal devices

101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, to name a few.

The user may use the

terminal devices

101, 102, 103 to interact with the server 105 via the network 104 to receive or send messages or the like. The

terminal devices

101, 102, 103 may have installed thereon various communication client applications, such as a telephone banking application, a mobile banking application, an internet banking application, a short message banking application, a shopping application, a web browser application, a search application, an instant messaging tool, a mailbox client, social platform software, etc. (by way of example only).

The

terminal devices

101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like.

The server 105 may be a server providing various services, such as a background management server (for example only) providing support for websites browsed by users using the

terminal devices

101, 102, 103. The background management server may analyze and perform other processing on the received data such as the user request, and feed back a processing result (e.g., entity information or data in a knowledge base obtained or generated according to the user request) to the terminal device.

It should be noted that the problem processing method based on the knowledge base provided by the embodiment of the present disclosure can be generally executed by the server 105. Accordingly, the knowledge base based problem processing apparatus provided by the embodiments of the present disclosure may be generally disposed in the server 105. The problem handling method based on the knowledge base provided by the embodiment of the present disclosure may also be performed by a server or a server cluster different from the server 105 and capable of communicating with the

terminal devices

101, 102, 103 and/or the server 105. Accordingly, the problem processing apparatus based on the knowledge base provided by the embodiment of the present disclosure may also be disposed in a server or a server cluster different from the server 105 and capable of communicating with the

terminal devices

101, 102, 103 and/or the server 105.

It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

FIG. 2 schematically illustrates a knowledge-graph of a knowledge-base based problem processing method and apparatus suitable for use with embodiments of the present disclosure. The Knowledge Graph (Knowledge Graph) describes entities in the objective world and relationships among the entities in a structured form formed by nodes and edges, so that the information of the internet is expressed into a form closer to the human cognitive world, and the capability of better organizing, managing and understanding mass information of the internet is provided.

As shown in fig. 2, the knowledge-graph 200 describes the bank products and the relationships between the products. Where the entity is demand deposit 210, the entities associated with demand deposit 210 may include banking outlets 220, automated teller machines 230, self-service terminals 240, internet banking 250, personal telephone banking 260, and iPhone cell phone banking 270.

It should be noted that the entity of demand deposit 210 may also include attributes such as deposit interest rate, service channel, service profile, and operating instructions. These several properties may also be understood as specific entities, and the relationships between entities and properties may also be understood as specific types of relationships. The problem processing method based on the knowledge base is provided by the disclosure, and the knowledge base is used for storing entity information and attribute information related to a knowledge graph. The customer accesses the system through a customer service channel, wherein the customer service channel can include but is not limited to a telephone bank, a mobile phone bank, an online bank and a short message bank, the customer service channel can firstly access an intelligent agent, then an intelligent question-answering system (also called an intelligent customer service question-answering system) is connected, the question of the user is intelligently identified, answer sentences matched with input question sentences are searched from a knowledge base, and the searched answer sentence results are returned, so that accurate answers can be quickly given, the quality of service is greatly improved, the pressure of customer service personnel is reduced, and the running cost of enterprises is reduced. Knowledge-graph management techniques may include, but are not limited to Neo4j, tiger graph, Amazon Neptune, as the present disclosure does not limit.

It should be understood that the number of entities and the relationships between the entities in the knowledge-graph of FIG. 2 are merely illustrative. According to the implementation requirement of the customer service system, a corresponding number of entities and the relationship between the entities can be constructed, which is not limited by the disclosure.

FIG. 3 schematically illustrates a flow chart of a knowledge-base based problem handling method according to an embodiment of the present disclosure. As shown in fig. 3, the question processing method 300 may be applied to an intelligent question-answering system capable of responding to input information and providing feedback information, and the question processing method 300 may include operations S310 through S340.

In operation S310, in response to receiving an input question statement, a question classification model is obtained.

According to the embodiment of the disclosure, after receiving a query request of a user, a customer service channel firstly accesses an intelligent agent, and an intelligent question-answering system is connected, wherein an input question sentence can be a character of a query question input by the user through the customer service channel or a directly input question sentence, and for non-character input, unified conversion is required to be performed through the intelligent agent. The focus of the disclosure is on the processing of the problem statement, and the processing of the customer service channel is not described in detail.

As an alternative embodiment, the received input question sentence may be further cleaned, which may include but is not limited to noise cleaning and knowledge fusion.

In operation S320, a question type of the input question sentence is determined based on the question classification model.

According to the embodiment of the disclosure, the problem classification model is obtained by pre-training based on knowledge data in a knowledge base, and different problem types correspond to different problem processing methods.

In operation S330, answer sentences matching the input question sentences are obtained from the knowledge base based on the question processing method corresponding to the question type.

In operation S340, an answer sentence is output.

According to the embodiment of the disclosure, the obtained answer sentences matched with the input question sentences are returned and output through conversion. Corresponding to the input question sentence, the answer sentence can be the character of the searched answer output by the intelligent question-answering system, and for the non-character answer, the unified conversion is needed by the intelligent seat.

By the embodiment of the disclosure, the problem classification model obtained by training based on the knowledge data in the knowledge base is used for determining the problem type corresponding to the input problem statement, according to different question types corresponding to different question processing methods, determining the question processing method corresponding to the question type to obtain answer sentences matched with the input question sentences from a knowledge base to be used as feedback information of the input question sentences for output, the above-mentioned technical problems caused by searching answer sentences by product category can be at least partially avoided, therefore, the answer sentences matched with the input question sentences can be obtained from the knowledge base by executing corresponding question processing methods based on different question types, the technical effects of saving time and labor and improving the accuracy rate of searching the question answers are achieved, the customer service experience of the user is further improved, and the enterprise reputation and the external image are favorably maintained.

The question processing method provided by the present disclosure will be explained in detail below by taking the question type, the sub-type, and the corresponding specific example shown in table 1 as an example, and in particular, several embodiments of obtaining an answer sentence matching an input question sentence from a knowledge base based on the question processing method corresponding to the question type in the present disclosure will be described in detail.

TABLE 1

It is understood that question and answer type, comparison type and flow type question types are more common customer service question categories, and may account for more than 85% of the question and answer knowledge base. The subtypes in table 1 are used to aid in problem understanding and do not substantially affect subsequent problem processing methods.

As shown in table 1, an example corresponding to the generalized sub-type in the generalized problem types "when logging in or operating the bank on the internet, pop up the password box often, ask for the password to be entered, how to solve? ", the type of problem contained therein is complex and not specific to a particular class of problem.

As an alternative embodiment, obtaining answer sentences matching the input question sentences from the knowledge base based on the question processing method corresponding to the question type may include: under the condition that the problem type is determined to be the comprehensive type, generating a first feature vector of an input problem statement; extracting first entity information and first attribute information of an input question sentence; obtaining a candidate entity set matched with the first entity information from a knowledge base based on the first entity information, wherein the candidate entity set comprises a plurality of second entities; determining a second feature vector of each second entity based on second attribute information of the plurality of second entities; determining a similarity value of the input question statement and each second entity based on the first feature vector and the second feature vector of each second entity; and obtaining the second entity corresponding to the maximum similarity value from the knowledge base as an answer sentence matched with the input question sentence.

The invention provides a TF-IDF similarity judgment model for calculating the similarity between texts based on a knowledge graph for similarity calculation, and the model establishes a TF-IDF vector representing the characteristics of each entity. The TF refers to Term Frequency (Term Frequency) and is used for indicating the Frequency of a certain Term appearing in a Document, the TF indicates that the Frequency of the word appearing in the Document is more important to the article, the IDF refers to Inverse file Frequency (Inverse Document Frequency) and is the reciprocal of the Frequency of the Term appearing in all the linguistic data, and the IDF indicates that the Frequency of the word appearing in the Document is more, so that the lower the distinguishing degree of the linguistic data is. In the present disclosure, an entity with the highest similarity value is returned as an answer similar to the input question sentence by calculating the vector similarity of the first feature vector of the input question sentence to the second feature vector of each entity in the candidate entity set. The entities and the relationships between the entities contained in the knowledge graph spectrum formed by the second entities in the candidate entity set are called subgraphs.

As an alternative embodiment, obtaining the candidate entity set matching the first entity information from the knowledge base based on the first entity information may include: acquiring a path length threshold, wherein the path length threshold is used for representing the maximum value of the path length between the first entity and the second entity; based on the path length threshold value, obtaining a candidate path with the path length between the candidate path and the first entity not exceeding the path length threshold value from the knowledge base; and using a plurality of second entities covered by the candidate path as a candidate entity set matched with the first entity.

According to the embodiment of the disclosure, when the second feature vector of the entity in the knowledge base is calculated, not only the attribute of the entity is calculated, but also the attribute of the N-hop (N ≧ 1) entity associated with the entity is simultaneously included in the feature representation range. Taking fig. 2 as an example, if N is 2, the second feature vector of the "demand deposit 210" is calculated by the attribute of the node and the attributes of 6 associated nodes, namely, the bank branch 220, the automatic teller machine 230, the self-service terminal 240, the internet bank 250, the personal telephone bank 260, and the iPhone mobile phone bank 270.

In specific implementation, the TF-IDF similarity determination model obtains a subgraph of a target node corresponding to the second entity as follows: g_i＝{n_j：p_i，j≤N}。

Wherein node n_iSubfigure g of_iIs and node n_iDoes not exceed the set of all nodes of N, where N ≧ 1. If N is 1, it represents g_i＝n_i. So that the corresponding node n_iDocument d of_iCan be represented as a collection of attributes for the nodes in the subgraph gi.

Thus the word t_iThe word frequency (tf) at node j may be expressed as:

idf of a particular word_iCan be expressed as:

the numerator represents the total number of nodes in the corpus, and the denominator represents the inclusion of the word t_iThe number of subgraphs.

The disclosure provides a knowledge graph-based TF-IDF calculation method, which is used for calculating the characteristic vector of each entity in the knowledge graph based on the N-hop relation and the attribute of the entity association. In the calculation result, the entity nodes corresponding to the input question sentences are used as indexes, so that the candidate entity sets can be quickly filtered, and the retrieval efficiency is greatly improved.

It should be noted that the TF-IDF calculation method is a general similarity evaluation method based on a knowledge graph, and the TF-IDF optimization algorithm is also applicable, for example, influence factors of word frequency, part of speech, and sequence are added, and the disclosure does not limit this.

FIG. 4 schematically illustrates a flow diagram of a knowledge-base based problem handling method according to another embodiment of the present disclosure. As shown in fig. 4, the question processing method 400 may be applied to an intelligent question-answering system capable of responding to input information and providing feedback information, and the question processing method 400 may include operations S410 to S480.

In specific implementation, a question sentence is first input (operation S410); next, generating a word vector (operation S420), where the word vector may be a first feature vector of the question statement input in the case that the question type is a comprehensive type; then, the entity is extracted (operation S430); then, a candidate entity set (operation S440), which may be a candidate entity set matching the first entity information obtained from the knowledge base based on the first entity information, and which includes a plurality of second entities; then, the cosine similarity between the word vector and the second feature vector of the candidate entity is calculated (operation S450). For each of the candidate entities, a segmentation is generated (operation S460), and then a segmentation pre-processing is performed (operation S470), and the TF-IDF model is trained (operation S480) to obtain the TF-IDF model.

According to the embodiment of the disclosure, the TF/IDF model is obtained through pre-training, for the input query, except for generating a query word vector, a candidate entity list of a map is obtained through entity query in the query, in the candidate list, the similarity between the query word vector and an entity feature vector is further calculated, and the entity with the highest score of a query result is returned as a candidate answer.

As an alternative embodiment, obtaining answer sentences matching the input question sentences from the knowledge base based on the question processing method corresponding to the question type may include: determining the graph database type of the knowledge base under the condition that the problem type is determined to be a common type; converting the input question sentence into a query sentence matched with the graph database type according to a preset conversion rule; and obtaining an answer sentence matched with the input question sentence from the knowledge base based on the query sentence.

According to the embodiment of the disclosure, the conversion rule is preset to convert the query into the knowledge graph query. Based on the preset conversion rule, the input question sentence can be converted into a query sentence corresponding to any graph database, and for convenience of explanation, the present disclosure will be explained in detail by taking the query sentence Cypher matched with the graph database type Neo4j as an example.

According to the method and the device for processing the problems, the problems are classified through a machine learning method, essential characteristics of various problems are summarized, and natural language of input problem sentences is converted into query language corresponding to the graph database, so that the problem processing method has strong universality and can be applied to various graph databases widely adopted at present.

FIG. 5 schematically illustrates a flow diagram of a knowledge-base based problem handling method according to another embodiment of the present disclosure. As shown in fig. 5, the question processing method 500 may be applied to an intelligent question-answering system capable of responding to input information and providing feedback information, and may include operations S510 to S540.

In a specific implementation, a question sentence is first input (operation S510), then a factor is extracted (operation S520), then a category is extracted (operation S530), and finally a map is retrieved (operation S540) to obtain an answer sentence matching the input question sentence from a knowledge base.

As an alternative embodiment, the element extraction (operation S520) may include entity extraction and relationship (attribute) extraction.

As an alternative embodiment, after the category extraction (operation S530), according to the quantitative relationship between the entities and the relationships (or attributes), the third entity information and the third attribute information corresponding to the question-and-answer type may have at least the following correspondence relationship: a piece of third entity information (i.e., a single entity query, corresponding to the "entity 1: 0 relationship" shown in FIG. 5); one piece of third entity information corresponds to one piece of third attribute information (i.e., entity relationship query, corresponding to the "entity 1: 1 relationship" shown in fig. 5); one piece of third entity information corresponds to a plurality of pieces of third attribute information (i.e., single-entity multi-relationship query, corresponding to the "entity 1: N relationship" shown in fig. 5); the plurality of pieces of third entity information correspond to one piece of third attribute information (i.e., multi-entity single-relationship query, corresponding to the "entity N: 1 relationship" shown in fig. 5). It is understood that "entity 0: 1 relationship "and" entity N: the M relationship "does not fit into the actual scenario, which is not discussed in this disclosure.

Several embodiments of converting an input question sentence into a query sentence matched with a graph database type according to a preset conversion rule will be described below by taking the question type shown in table 1 as a question and answer type, and the sub-type as single entity query, entity relationship query, single entity multi-relationship query and multi-entity single relationship query as an example.

As an alternative embodiment, converting the input question statement into the query statement matching the graph database type according to the preset conversion rule may include: under the condition that the common type is a question and answer type or a flow type, extracting third entity information of the input question sentence; extracting third attribute information associated with the third entity information; determining fourth entity information which is the same as or similar to the third entity information in the knowledge base based on the third entity information; determining fourth attribute information which is the same as or similar to the third attribute information in the knowledge base based on the third attribute information; and generating a query statement matched with the graph database type based on the fourth entity information and the fourth attribute information.

According to the embodiment of the disclosure, besides performing noise cleaning and knowledge fusion on the input question sentences, the key is to extract information of the input question sentences, judge the subcategories of the questions according to the number of entities and relations (attributes) in the questions, convert the subcategories of the questions into a graph query language based on a common preset conversion rule, and extract the obtained information, which may include but is not limited to entity extraction, attribute extraction and relation extraction. The general preset conversion rule may be to extract an entity object in the query condition corresponding to the input question statement, and extract a query attribute or relationship in a return result of the entity object extraction. The entity extraction technology in the related technology is more, and the Python Chinese word segmentation component 'jieba' is adopted in the disclosure. Based on the industry financial word stock, the part of speech and classification can be accurately identified, and the category in the entity, the attribute and the relation can be further determined through the judgment of the part of speech and the classification.

As shown in table 1, an example "what is the demand deposit? ". In specific implementation, the entity can be extracted as 'deposit on schedule' through word segmentation. The query statement generated by converting the query statement to match the graph database type can be described as follows, wherein the name corresponds to the entity name, and can also be represented by other symbols.

MATCH(n)

WHERE n.name ═ deposit on expiry "

RETURN n

It should be noted that for a single entity query, the description attributes of the returned entity may also be optimized, depending on the design structure of the particular knowledge-graph.

As shown in table 1, an example "what is the unit of issuance affiliation of peony-tay card" corresponds to the subtype of entity relationship query in question-and-answer type question type? ". In specific implementation, the entity extracted by word segmentation is "peony-Adita card" and the attribute is "issuing affiliation unit", and the query statement generated by conversion of the entity and matched with the type of the graph database can be described as follows:

MATCH(n)

WHERE n.name ═ peony antitachar "

RETURN, issue affiliation unit "

As shown in table 1, an example of the sub-type correspondence of the multi-entity single-relationship query in the question-and-answer type question type "issue affiliation unit of moto card and peonie card? ". In specific implementation, the entities extracted by word segmentation are "Zhongqing travel card" and "Paeonia suffruticosa", the attribute is "issue joint name unit", and the query sentence generated by conversion of the entities and matched with the type of the graph database can be described as follows:

MATCH(n)

name IN [ "peony inferior taka", "Zhongqing traveling card" ]

RETURN, issue affiliation unit "

As shown in table 1, an example "what is the limit and the handling charge for remittance transfer" corresponds to a subtype of a single-entity multi-relationship query in a question-and-answer type question type? ". In specific implementation, the term extraction entity is "transfer remittance", the attributes are "quota" and "commission", and the query statement generated by converting the term extraction entity and matched with the type of the graph database can be described as follows:

MATCH(n)

WHERE n.name ═ transfer remittance "

RETURN n. "Limited", n. "commission charge"

It can be understood that the question-answering type question type mainly deals with comparing basic condition queries, and in actual matching, methods such as synonym replacement, fuzzy matching and the like are generally required to be used to improve the robustness of result matching.

According to the embodiment of the disclosure, the input question sentences are converted into the query sentences matched with the graph database type based on the preset conversion rules, so that the key characteristics of each question-answering type question can be summarized, the question answering accuracy is greatly improved, the pressure of customer service personnel is reduced, and the operation cost of enterprise customer service is reduced.

An embodiment of obtaining answer sentences matching the input question sentences from the knowledge base will be described below by taking the question types shown in table 1 as flow types, the sub-types as normal flows, and the time-series processing as examples.

FIG. 6 schematically illustrates a flow diagram of a knowledge-base based problem handling method according to another embodiment of the present disclosure. As shown in fig. 6, the question processing method 600 may be applied to an intelligent question-answering system capable of responding to input information and providing feedback information, and may include operations S610 to S640. In specific implementation, a question sentence is first input (operation S610), then an attribute relationship is extracted by using the classifier 610 (operation S620), the extracted attribute relationship may include an order class 620 and a front-back item class 630, then an entity is extracted (operation S630), and finally a graph is retrieved (operation S640) to obtain an answer sentence matched with the input question sentence from a knowledge base.

As an alternative embodiment, obtaining answer sentences matching the input question sentences from the knowledge base based on the query sentences may include: under the condition that the common type is the flow type, obtaining an attribute list associated with the third entity information from the knowledge base, wherein the attribute list comprises a plurality of attribute information which are executed in order; determining an execution logic relation of the attribute information corresponding to the query statement in a plurality of sequentially executed attribute information based on the attribute list; and obtaining answer sentences matched with the input question sentences from the knowledge base based on the execution logic relation.

In specific implementation, the question processing method of the common flow subtype is similar to the question-answer type, and the difference is that the question processing method of the common flow subtype needs to assemble execution results of each flow according to a time sequence. Whether the type falls under the process type depends on the organization of the knowledge graph. The problem processing method of the time sequence subtype mainly aims at judging the logical relation of the extracted predicates. Specifically, a list of entity correspondence or attributes may be preferentially obtained, a current position is determined by similarity of process contents, and a target result is extracted according to the logic predicate.

FIG. 7 schematically illustrates a flow diagram of a knowledge-base based problem handling method according to another embodiment of the present disclosure. As shown in fig. 7, the question processing method 700 may be applied to an intelligent question-answering system capable of responding to input information and providing feedback information, and may include operations S710 to S760.

In particular implementation, a question sentence is first input (operation S710), a first comparison entity is then extracted (operation S720), a comparison relationship (attribute) is extracted (operation S730), which may include, but is not limited to, attribute information corresponding to a comparison subtype, attribute information corresponding to a retrieval subtype, and attribute information corresponding to an induction subtype, a comparison predicate is extracted (operation S740), which may include, but is not limited to, comparison condition information, retrieval condition information, and induction condition information, a second comparison entity is extracted (operation S750), and finally a graph search is performed (operation S760) to obtain an answer sentence matching the input question sentence from a knowledge base.

An example of converting an input question sentence into a query sentence matching a graph database type according to a preset conversion rule will be described below by taking the question type shown in table 1 as a comparative type, and the sub-types as comparative type, retrieval type, and induction type as examples.

As an alternative embodiment, converting the input question statement into the query statement matching the graph database type according to the preset conversion rule includes: determining a subtype corresponding to the input question sentence under the condition that the common type is the comparison type; extracting fifth body information, fifth attribute information and comparison condition information of the input question sentence, wherein the comparison condition information is used for representing a subtype; determining sixth entity information which is the same as or similar to the fifth entity information in the knowledge base based on the fifth entity information; determining sixth attribute information which is the same as or similar to the fifth attribute information in the knowledge base based on the fifth attribute information; and generating a query statement matched with the graph database type based on the sixth entity information, the sixth attribute information and the comparison condition information.

According to embodiments of the present disclosure, comparative types include three categories of sub-problem classifications, namely comparative, search, and induction. The comparative type indicates that two types of entity attributes (relationships) are compared, the retrievable type indicates that the query satisfies a candidate set of comparison conditions, and the inductive type indicates the most valued query, e.g., highest value, lowest value, maximum value, minimum value. The key to comparative problem processing is to extract entities, query conditions, and compare conditions.

As an alternative embodiment, in the case where the subtype includes a comparison subtype, the comparison condition information includes comparison condition information corresponding to the comparison subtype, wherein the comparison subtype is used to indicate that the attributes of the two types of entities are compared; in a case where the subtype includes a search subtype, the comparison condition information includes search condition information corresponding to the search subtype indicating that the search satisfies the search condition information; in the case where the sub-type includes a generalised sub-type, the comparison condition information includes generalised condition information corresponding to the generalised sub-type, where the generalised sub-type is used to indicate the most significant value of the query attribute.

As shown in table 1, an example "how much higher the interest rate of the 2-year deposit period full credit entire is than that of the 1-year deposit period full credit entire? ". In specific implementation, the comparison entity information may be analyzed and extracted as "whole deposit strives for deposit", the attribute information is "2 year" and "1 year", the comparison condition information is interest rate, and the query statement generated by converting the comparison condition information and matching with the graph database type may be described as follows:

MATCH (n { name: "entire deposit entire fetch", "deposit term": 2}),

(m { name: "entire deposit taken", "deposit term": 1})

RETURN n. "interest rate" -m. "interest rate"

As shown in table 1, an example "notify deposit product interest rate lower than 1.25%? ". In specific implementation, the information of the retrieval entity is "notice deposit", the attribute information is "product interest rate", the information of the retrieval condition is 1.25%, and the query statement generated by converting the information and matched with the graph database type can be described as follows:

MATCH(n)

WHERE n.name ═ notify deposit "AND n." product interest rate "<" 1.25% "

RETURN n

As shown in table 1, an example "the highest interest rate of the fixed deposit product? ". In specific implementation, it may be analyzed and extracted that the induction entity information is "fixed deposit", the attribute information is "product interest rate", and the induction condition information is "highest", and the query statement generated by converting the induction entity information and matching the graph database type may be described as follows:

MATCH(n)

WHERE n.name ═ periodic deposits "

RETURN n

ORDER BY

LIMIT1

Or

MATCH(n)

WHERE n.name ═ periodic deposits "

RETURN max (n. "product interest rate")

According to the embodiment of the disclosure, for the comparative question types, the entity information, the attribute information and the condition information corresponding to the sub-types can be extracted based on the input question sentences, the entity information, the attribute information and the condition information are converted into the query sentences matched with the graph database according to the preset conversion rules, the key characteristics of the question sentences of each comparative type are summarized, the accuracy of question answering is greatly improved, the pressure of customer service personnel is reduced, and the running cost of enterprises is reduced.

As an optional embodiment, the problem processing method may further include: selecting training sample data from a knowledge base according to a preset rule, wherein the training sample data comprises marking data of a problem type; obtaining an initial problem classification model; and performing deep neural network training on the initial problem classification model based on the training sample data to generate a problem classification model.

In specific implementation, the question-answering knowledge base has M (for example, 113752) knowledge data records, about a% (for example, 30%) of the knowledge data can be randomly selected for labeling, and a question classification model is trained by a deep neural network method. For the questions of the user, the question categories can be marked through the question classification model, and the questions are transferred to the corresponding question types for processing. The selection method of the training sample data is not limited in the disclosure, and a person skilled in the art can select an appropriate selection method to select any number of sample data from the knowledge base according to actual conditions.

Through the embodiment of the disclosure, the problem classification model is trained by utilizing the deep neural network, the problem classification intelligence can be improved, and the classification efficiency and the classification accuracy are improved.

It should be noted that the problem processing method applicable to the comprehensive problem type (answer sentence search is performed by calculating the similarity between the word vector of the input problem sentence and the feature vector of each entity in the subgraph) is not specific to a certain problem type in nature, and may be actually used to process other problem types.

FIG. 8 schematically illustrates a block diagram of a knowledge-base based problem processing apparatus, in accordance with an embodiment of the present disclosure. As shown in fig. 8, the question processing apparatus 800 may be applied to an intelligent question-answering system capable of responding to input information and providing feedback information, and may include a classification model obtaining module 810, a question type determining module 820, an answer sentence obtaining module 830, and an answer sentence output module 840.

A classification model obtaining module 810, configured to obtain a question classification model in response to receiving an input question statement, where the question classification model is trained based on knowledge data in a knowledge base. Optionally, the classification model obtaining module 810 may be used in the aforementioned operation S310, for example, and is not described herein again.

And a question type determining module 820, configured to determine a question type of the input question statement based on the question classification model, where different question types correspond to different question processing methods. Optionally, the problem type determining module 820 may be used in the aforementioned operation S320, for example, and is not described herein again.

An answer sentence obtaining module 830, configured to obtain an answer sentence matched with the input question sentence from the knowledge base based on the question processing method corresponding to the question type. Optionally, the answer sentence obtaining module 830 may be used in the operation S330, for example, and is not described herein again.

An answer sentence output module 840 for outputting an answer sentence. Optionally, the answer sentence output module 840 may be used in the operation S340, for example, and is not described herein again.

According to the embodiment of the disclosure, the question type corresponding to the input question sentence is determined based on the question classification model obtained by training the knowledge data in the knowledge base, the question processing method corresponding to the question type is determined according to the question processing methods corresponding to different question types, and the answer sentence matched with the input question sentence is obtained from the knowledge base and is output as the feedback information of the input question sentence.

FIG. 9 schematically illustrates a block diagram of a knowledge-base based problem processing apparatus according to another embodiment of the present disclosure. As shown in fig. 9, the question processing apparatus 900 may be applied to an intelligent question-answering system capable of responding to input information and providing feedback information, and may include a question classification module 910, a question-answering type question processing module 920, a comparative type question processing module 930, a flow type question processing module 940, and an integrated type question processing module 950.

According to an embodiment of the present disclosure, the question classification module 910 mainly performs two types of tasks, where the first type of task is to perform cleaning on input question sentences (including entity extraction, attribute extraction, relationship extraction, noise cleaning, and knowledge fusion). The categories in the entities, attributes and relationships are further determined by the judgment of the parts of speech and classification. The second type of task is to transfer the problem to the corresponding sub-module for processing through the problem classification module 910, and after the problem category is determined, transfer to the subsequent processing flow. Specifically, the question-answer input question sentence is transferred to the question-answer question processing module 920 for question processing, the comparative input question sentence is transferred to the comparative question processing module 930 for question processing, the flow-type input question sentence is transferred to the flow-type question processing module 940 for question processing, and the comprehensive input question sentence is transferred to the comprehensive question processing module 950 for question processing.

According to the intelligent customer service question-answering system based on the financial knowledge map, the customer questions are modeled based on the knowledge of the question-answering base, and the user questions are classified into four types through a deep neural network method. According to the analysis of the question types, the customer questions are divided into four types of question answering type, comparison type, flow type and comprehensive type, the method can greatly improve the accuracy of the system for handling the questions, the quality of customer service is improved through an artificial intelligence method, the working pressure of customer service personnel is reduced, and the operation cost of enterprises is reduced.

As an alternative embodiment, the answer sentence obtaining module 830 may include: the generating submodule is used for generating a first feature vector of an input question statement under the condition that the question type is determined to be the comprehensive type; the first extraction submodule is used for extracting first entity information and first attribute information of the input question statement; the first obtaining sub-module is used for obtaining a candidate entity set matched with the first entity information from a knowledge base based on the first entity information, wherein the candidate entity set comprises a plurality of second entities; a first determining submodule, configured to determine a second feature vector of each second entity based on second attribute information of a plurality of second entities; a second determining submodule, configured to determine a similarity value of the input question statement and each second entity based on the first feature vector and the second feature vector of each second entity; and a second obtaining submodule for obtaining a second entity corresponding to the maximum similarity value from the knowledge base as an answer sentence matched with the input question sentence.

As an alternative embodiment, the first obtaining sub-module may include: a first obtaining unit, configured to obtain a path length threshold, where the path length threshold is used to characterize a maximum value of a path length with a first entity; a second obtaining unit, configured to obtain, from the knowledge base, a candidate path having a path length with the first entity that does not exceed the path length threshold based on the path length threshold; and the first processing unit is used for taking a plurality of second entities covered by the candidate path as a candidate entity set matched with the first entity.

As an alternative embodiment, the answer sentence obtaining module 830 may include: the third determining submodule is used for determining the graph database type of the knowledge base under the condition that the problem type is determined to be the common type; the first conversion sub-module is used for converting the input question sentences into query sentences matched with the type of the graph database according to preset conversion rules; and a third obtaining sub-module, which is used for obtaining answer sentences matched with the input question sentences from the knowledge base based on the query sentences.

As an alternative embodiment, the first conversion submodule may include: the first extraction unit is used for extracting third entity information of the input question sentence under the condition that the common type is a question and answer type or a flow type; a second extracting unit configured to extract third attribute information associated with the third entity information; the first determining unit is used for determining fourth entity information which is the same as or similar to the third entity information in the knowledge base based on the third entity information; a second determining unit configured to determine, based on the third attribute information, fourth attribute information that is the same as or similar to the third attribute information in the knowledge base; and a first generating unit configured to generate a query statement matching the graph database type based on the fourth entity information and the fourth attribute information.

As an alternative embodiment, the third entity information and the third attribute information corresponding to the question-answer type may have at least the following correspondence relationship: a piece of third entity information; one piece of third entity information corresponds to one piece of third attribute information; one piece of third entity information corresponds to a plurality of pieces of third attribute information; the plurality of pieces of third entity information correspond to one piece of third attribute information.

As an alternative embodiment, the answer sentence obtaining module 830 may include: a fourth obtaining submodule, configured to obtain, from the knowledge base, an attribute list associated with the third entity information when the common type is the flow type, where the attribute list includes a plurality of attribute information that is executed in order; the fourth determining submodule is used for determining the execution logic relation of the attribute information corresponding to the query statement in the orderly executed attribute information based on the attribute list; and a fifth obtaining submodule, which is used for obtaining answer sentences matched with the input question sentences from the knowledge base based on the execution logic relation.

As an alternative embodiment, the first conversion submodule may include: a third determination unit configured to determine a subtype corresponding to the input question sentence in a case where the normal type is the comparison type; a third extraction unit, configured to extract fifth volume information, fifth attribute information, and comparison condition information of the input question statement, where the comparison condition information is used to characterize a subtype; a fourth determination unit configured to determine, based on the fifth volume information, sixth entity information in the knowledge base that is the same as or similar to the fifth volume information; a fifth determining unit configured to determine, based on the fifth attribute information, sixth attribute information that is the same as or similar to the fifth attribute information in the knowledge base; and a second generating unit configured to generate a query sentence matching the graph database type based on the sixth entity information, the sixth attribute information, and the comparison condition information.

As an alternative embodiment, in the case where the subtype includes a comparison subtype, the comparison condition information may include comparison condition information corresponding to the comparison subtype, wherein the comparison subtype is used to indicate that the attributes of the two types of entities are compared; in the case where the subtype includes a search subtype, the comparison condition information may include search condition information corresponding to the search subtype indicating that an entity satisfying the search condition information is searched for; where the sub-type includes a generalised sub-type, the comparison condition information may include generalised condition information corresponding to a generalised sub-type, where the generalised sub-type is used to indicate the most significant value of the query attribute.

As an alternative embodiment, the apparatus may further comprise: the training sample selection module is used for selecting training sample data from the knowledge base according to a preset rule, wherein the training sample data comprises marking data of the problem type; the initial model obtaining module is used for obtaining an initial problem classification model; and the classification model generation module is used for carrying out deep neural network training on the initial problem classification model based on the training sample data to generate a problem classification model.

It should be noted that the implementation, solved technical problems, implemented functions, and achieved technical effects of each module in the embodiment of the problem processing apparatus part based on the knowledge base are respectively the same as or similar to the implementation, solved technical problems, implemented functions, and achieved technical effects of each corresponding step in the embodiment of the problem processing method part based on the knowledge base, and are not described herein again.

Any number of modules, sub-modules, units, sub-units, or at least part of the functionality of any number thereof according to embodiments of the present disclosure may be implemented in one module. Any one or more of the modules, sub-modules, units, and sub-units according to the embodiments of the present disclosure may be implemented by being split into a plurality of modules. Any one or more of the modules, sub-modules, units, sub-units according to embodiments of the present disclosure may be implemented at least in part as a hardware circuit, such as a field programmable gate array (FNGA), a programmable logic array (NLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or may be implemented in any other reasonable manner of hardware or firmware by integrating or packaging a circuit, or in any one of or a suitable combination of software, hardware, and firmware implementations. Alternatively, one or more of the modules, sub-modules, units, sub-units according to embodiments of the disclosure may be at least partially implemented as a computer program module, which when executed may perform the corresponding functions.

For example, the classification model obtaining module, the question type determining module, the answer sentence obtaining module, the answer sentence outputting module, the generating submodule, the first extracting submodule, the first obtaining submodule, the first determining submodule, the second obtaining submodule, the first obtaining unit, the second obtaining unit, the first processing unit, the third determining submodule, the first converting submodule, the third obtaining submodule, the first extracting unit, the second extracting unit, the first determining unit, the second determining unit, the first generating unit, the fourth obtaining submodule, the fourth determining submodule, the fifth obtaining submodule, the third determining unit, the third extracting unit, the fourth determining unit, the fifth determining unit, the second generating unit, the training sample selecting module, the initial model obtaining module and the classification model generating module may be combined into one module to implement, or any one of them may be split into a plurality of modules. Alternatively, at least part of the functionality of one or more of these modules may be combined with at least part of the functionality of the other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the classification model obtaining module, the question type determining module, the answer sentence obtaining module, the answer sentence output module, the generating sub-module, the first extracting sub-module, the first obtaining sub-module, the first determining sub-module, the second obtaining sub-module, the first obtaining unit, the second obtaining unit, the first processing unit, the third determining sub-module, the first converting sub-module, the third obtaining sub-module, the first extracting unit, the second extracting unit, the first determining unit, the second determining unit, the first generating unit, the fourth obtaining sub-module, the fourth determining sub-module, the fifth obtaining sub-module, the third determining unit, the third extracting unit, the fourth determining unit, the fifth determining unit, the second generating unit, the training sample selecting module, the initial model obtaining module, and the classification model generating module may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FNGA), a programmable logic array (NLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or may be implemented in hardware or firmware in any other reasonable manner of integrating or packaging a circuit, or in any one of or any suitable combination of software, hardware, and software. Or, at least one of the classification model obtaining module, the question type determining module, the answer sentence obtaining module, the answer sentence output module, the generating submodule, the first extracting submodule, the first obtaining submodule, the first determining submodule, the second obtaining submodule, the first obtaining unit, the second obtaining unit, the first processing unit, the third determining submodule, the first converting submodule, the third obtaining submodule, the first extracting unit, the second extracting unit, the first determining unit, the second determining unit, the first generating unit, the fourth obtaining submodule, the fourth determining submodule, the fifth obtaining submodule, the third determining unit, the third extracting unit, the fourth determining unit, the fifth determining unit, the second generating unit, the training sample selecting module, the initial model obtaining module and the classification model generating module may be at least partially implemented as a computer program module, when the computer program modules are run, corresponding functions may be performed.

FIG. 10 schematically illustrates a schematic diagram of a computer-readable storage medium product suitable for implementing the knowledge-base based question processing method described above, according to an embodiment of the present disclosure.

In some possible embodiments, aspects of the present invention may also be implemented in a form of a program product including program code for causing a device to perform the aforementioned operations (or steps) in the method for problem handling based on a knowledge base according to various exemplary embodiments of the present invention described in the above-mentioned "exemplary method" section of this specification when the program product is run on the device, for example, the electronic device may perform operations S310 to S340 shown in fig. 3, operations S410 to S480 shown in fig. 4, operations S510 to S540 shown in fig. 5, operations S610 to S640 shown in fig. 6, and operations S710 to S760 shown in fig. 7.

The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (ENROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

As shown in fig. 10, a program product 1000 that may employ a portable compact disc read only memory (CD-ROM) and include program code and may be run on a device, such as a personal computer, is depicted for a method of knowledge base based problem handling in accordance with an embodiment of the present invention. However, the program product of the present invention is not limited in this respect, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, or device.

A readable signal medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, or device. Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user computing device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a local area network (LAA) or a wide area network (WAA), or may be connected to an external computing device (e.g., through the internet using an internet service provider).

FIG. 11 schematically illustrates a block diagram of an electronic device adapted to implement the knowledge-base based question processing method described above, in accordance with an embodiment of the present disclosure. The electronic device shown in fig. 11 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present disclosure.

As shown in fig. 11, an electronic device 1100 according to an embodiment of the present disclosure includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM)1102 or a program loaded from a storage section 1108 into a Random Access Memory (RAM) 1103. The processor 1101 may comprise, for example, a general purpose microprocessor (e.g., CNU), an instruction set processor and/or associated chipset, and/or a special purpose microprocessor (e.g., Application Specific Integrated Circuit (ASIC)), or the like. The processor 1101 may also include on-board memory for caching purposes. The processor 1101 may comprise a single processing unit or a plurality of processing units for performing the different actions of the method flows according to the embodiments of the present disclosure.

In the RAM 1103, various programs and data necessary for the operation of the electronic device 1100 are stored. The processor 1101, the ROM 1102, and the RAM 1103 are connected to each other by a bus 1104. The processor 1101 performs various operations of the method flow according to the embodiments of the present disclosure by executing programs in the ROM 1102 and/or the RAM 1103. It is noted that the programs may also be stored in one or more memories other than the ROM 1102 and RAM 1103. The processor 1101 may also perform operations S310 to S340 illustrated in fig. 3, operations S410 to S480 as illustrated in fig. 4, operations S510 to S540 as illustrated in fig. 5, operations S610 to S640 as illustrated in fig. 6, and operations S710 to S760 as illustrated in fig. 7 according to an embodiment of the present disclosure by executing the program stored in the one or more memories.

Electronic device 1100 may also include input/output (I/O) interface 1105, input/output (I/O) interface 1105 also connected to bus 1104, according to an embodiment of the disclosure. The system 1100 may also include one or more of the following components connected to the I/O interface 1105: an input portion 1106 including a keyboard, mouse, and the like; an output portion 1107 including a signal output unit such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and a speaker; a storage section 1108 including a hard disk and the like; and a communication section 1109 including a network interface card such as an LAA card, a modem, or the like. The communication section 1109 performs communication processing via a network such as the internet. A driver 1110 is also connected to the I/O interface 1105 as necessary. A removable medium 1111 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 1110 as necessary, so that a computer program read out therefrom is mounted into the storage section 1108 as necessary.

According to embodiments of the present disclosure, method flows according to embodiments of the present disclosure may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable storage medium, the computer program containing program code for performing the method illustrated by the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication portion 1109 and/or installed from the removable medium 1111. The computer program, when executed by the processor 1101, performs the above-described functions defined in the system of the embodiment of the present disclosure. The systems, devices, apparatuses, modules, units, etc. described above may be implemented by computer program modules according to embodiments of the present disclosure.

The present disclosure also provides a computer-readable storage medium, which may be contained in the apparatus/device/system described in the above embodiments; or may exist separately and not be assembled into the device/apparatus/system. The above-mentioned computer-readable storage medium carries one or more programs which, when executed, implement the knowledge-base-based question processing method according to an embodiment of the present disclosure, including operations S310 to S340 shown in fig. 3, operations S410 to S480 shown in fig. 4, operations S510 to S540 shown in fig. 5, operations S610 to S640 shown in fig. 6, and operations S710 to S760 shown in fig. 7.

According to embodiments of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, which may include, for example but is not limited to: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (ENROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. For example, according to embodiments of the present disclosure, a computer-readable storage medium may include the ROM 1102 and/or the RAM 1103 and/or one or more memories other than the ROM 1102 and the RAM 1103 described above.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Those skilled in the art will appreciate that various combinations and/or combinations of features recited in the various embodiments and/or claims of the present disclosure can be made, even if such combinations or combinations are not expressly recited in the present disclosure. In particular, various combinations and/or combinations of the features recited in the various embodiments and/or claims of the present disclosure may be made without departing from the spirit or teaching of the present disclosure. All such combinations and/or associations are within the scope of the present disclosure.

The embodiments of the present disclosure have been described above. However, these examples are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the embodiments are described separately above, this does not mean that the measures in the embodiments cannot be used in advantageous combination. The scope of the disclosure is defined by the appended claims and equivalents thereof. Various alternatives and modifications can be devised by those skilled in the art without departing from the scope of the present disclosure, and such alternatives and modifications are intended to be within the scope of the present disclosure.

Claims

1. A problem processing method based on a knowledge base is applied to an intelligent question-answering system, the intelligent question-answering system can respond to input information and provide feedback information, and the method comprises the following steps:

obtaining a question classification model in response to receiving an input question statement, wherein the question classification model is trained based on knowledge data in the knowledge base;

determining the question type of the input question sentence based on the question classification model, wherein different question types correspond to different question processing methods;

obtaining answer sentences matched with the input question sentences from the knowledge base based on a question processing method corresponding to the question types;

and outputting the answer sentence.

2. The method of claim 1, wherein the obtaining answer sentences matching the input question sentences from the knowledge base based on the question processing method corresponding to the question type comprises:

generating a first feature vector of the input question statement under the condition that the question type is determined to be a comprehensive type;

extracting first entity information and first attribute information of the input question sentence;

obtaining a candidate entity set matched with the first entity information from the knowledge base based on the first entity information, wherein the candidate entity set comprises a plurality of second entities;

determining a second feature vector of each second entity based on second attribute information of the plurality of second entities;

determining a similarity value of the input question statement and each second entity based on the first feature vector and the second feature vector of each second entity;

and taking the second entity corresponding to the maximum similarity value obtained from the knowledge base as an answer sentence matched with the input question sentence.

3. The method of claim 2, wherein the obtaining, from the knowledge base, a set of candidate entities matching the first entity information based on the first entity information comprises:

obtaining a path length threshold, wherein the path length threshold is used for characterizing the maximum value of the path length between the first entity and the second entity;

based on the path length threshold, obtaining candidate paths from the knowledge base, the path length of which between the candidate paths and the first entity does not exceed the path length threshold;

and taking a plurality of second entities covered by the candidate path as a candidate entity set matched with the first entity.

4. The method of claim 1, wherein the obtaining answer sentences matching the input question sentences from the knowledge base based on the question processing method corresponding to the question type comprises:

determining a graph database type of the knowledge base under the condition that the problem type is determined to be a common type;

converting the input question sentence into a query sentence matched with the graph database type according to a preset conversion rule;

and obtaining answer sentences matched with the input question sentences from the knowledge base based on the query sentences.

5. The method of claim 4, wherein said converting said input question statement into a query statement matching said graph database type according to a preset conversion rule comprises:

extracting third entity information of the input question sentence under the condition that the common type is a question and answer type or a flow type;

extracting third attribute information associated with the third entity information;

determining fourth entity information which is the same as or similar to the third entity information in the knowledge base based on the third entity information;

determining fourth attribute information which is the same as or similar to the third attribute information in the knowledge base based on the third attribute information;

and generating a query statement matched with the graph database type based on the fourth entity information and the fourth attribute information.

6. The method according to claim 5, wherein the third entity information and the third attribute information corresponding to the question-answer type have at least the following correspondence relationship:

a piece of third entity information;

one piece of third entity information corresponds to one piece of third attribute information;

one piece of third entity information corresponds to a plurality of pieces of third attribute information;

the plurality of pieces of third entity information correspond to one piece of third attribute information.

7. The method of claim 5, wherein the obtaining answer sentences matching the input question sentences from the knowledge base based on the query sentences comprises:

obtaining an attribute list associated with the third entity information from the knowledge base under the condition that the common type is a flow type, wherein the attribute list comprises a plurality of attribute information which are executed in order;

determining an execution logic relation of the attribute information corresponding to the query statement in the orderly executed attribute information based on the attribute list;

and obtaining answer sentences matched with the input question sentences from the knowledge base based on the execution logic relation.

8. The method of claim 4, wherein said converting said input question statement into a query statement matching said graph database type according to a preset conversion rule comprises:

determining a subtype corresponding to the input question sentence under the condition that the common type is a comparison type;

extracting fifth body information, fifth attribute information and comparison condition information of the input question sentence, wherein the comparison condition information is used for representing the subtype;

determining sixth entity information in the knowledge base, which is the same as or similar to the fifth entity information, based on the fifth entity information;

determining sixth attribute information which is the same as or similar to the fifth attribute information in the knowledge base based on the fifth attribute information;

and generating a query statement matched with the graph database type based on the sixth entity information, the sixth attribute information and the comparison condition information.

9. The method of claim 8, wherein:

in the case that the subtype comprises a comparison subtype, the comparison condition information comprises comparison condition information corresponding to the comparison subtype, wherein the comparison subtype is used for indicating that the attributes of the two types of entities are compared;

in a case where the subtype includes a search subtype, the comparison condition information includes search condition information corresponding to the search subtype indicating that an entity satisfying the search condition information is searched for;

in the case where the sub-type includes an inductive sub-type, the comparison condition information includes inductive condition information corresponding to the inductive sub-type, wherein the inductive sub-type indicates a most significant value of the query attribute.

10. The method of claim 1, wherein the method further comprises:

selecting training sample data from the knowledge base according to a preset rule, wherein the training sample data comprises marking data of a problem type;

obtaining an initial problem classification model;

and performing deep neural network training on the initial problem classification model based on the training sample data to generate the problem classification model.

11. A knowledge-base-based question processing apparatus for use in an intelligent question-answering system capable of responding to input information and providing feedback information, comprising:

a classification model obtaining module, configured to obtain a problem classification model in response to receiving an input question statement, where the problem classification model is obtained by training based on knowledge data in the knowledge base;

the problem type determining module is used for determining the problem type of the input problem statement based on the problem classification model, wherein different problem types correspond to different problem processing methods;

an answer sentence obtaining module, configured to obtain, from the knowledge base, an answer sentence matched with the input question sentence based on a question processing method corresponding to the question type;

and the answer sentence output module is used for outputting the answer sentences.

12. An electronic device, comprising:

one or more processors; and

a memory for storing one or more programs,

wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the method of any of claims 1-10.

13. A computer-readable storage medium storing computer-executable instructions that, when executed, cause a processor to perform the method of any one of claims 1 to 10.

14. A computer program product comprising a computer program which, when executed by a processor, performs the method according to any one of claims 1 to 10.