CN111522966A

CN111522966A - Data processing method and device based on knowledge graph, electronic equipment and medium

Info

Publication number: CN111522966A
Application number: CN202010323789.XA
Authority: CN
Inventors: 马工利
Original assignee: Shenzhen Zhuiyi Technology Co Ltd
Current assignee: Shenzhen Zhuiyi Technology Co Ltd
Priority date: 2020-04-22
Filing date: 2020-04-22
Publication date: 2020-08-11

Abstract

The embodiment of the application provides a data processing method and device based on a knowledge graph, electronic equipment and a medium, and relates to the technical field of natural language processing. The method comprises the following steps: acquiring a user input statement; identifying at least one knowledge type corresponding to the user input statement to obtain a knowledge type set; performing fusion processing on the knowledge type set, and updating the knowledge type set; constructing a query graph based on the updated knowledge type set; and inquiring a response corresponding to the query graph from a preset knowledge graph, and feeding back the response. According to the embodiment of the application, the knowledge types corresponding to the sentences input by the user are identified, the identified knowledge types are fused, the knowledge type set is updated, the response of the complex sentences can be covered, the response coverage range based on the knowledge graph is expanded, and therefore the identification accuracy of the complex sentences is improved.

Description

Data processing method and device based on knowledge graph, electronic equipment and medium

Technical Field

The present application relates to the field of natural speech processing technologies, and in particular, to a method, an apparatus, an electronic device, and a medium for question answering based on a knowledge graph.

Background

Knowledge maps have been used to refer broadly to a variety of large-scale knowledge bases. Due to the development of Semantic networks (Semantic nets) and the emergence of large-scale structured Knowledge graphs (Knowledge answers), Knowledge Graph-based questions (Question Answering) have gained wide attention in the fields of Natural Language Processing (NLP) and Knowledge Bases (KB), especially for Question-Answering scenarios of complex questions. However, the recognition effect of the complex question sentence is not good at present, and the knowledge question answering of the complex question sentence is difficult to realize.

Disclosure of Invention

The embodiment of the application provides a data processing method, a data processing device, electronic equipment and a medium based on a knowledge graph, and can solve the problems.

In a first aspect, an embodiment of the present application provides a data processing method based on a knowledge graph, where the method includes: acquiring a user input statement; identifying at least one knowledge type corresponding to the user input statement to obtain a knowledge type set; performing fusion processing on the knowledge type set, and updating the knowledge type set; constructing a query graph based on the updated knowledge type set; and inquiring a response corresponding to the query graph from a preset knowledge graph, and feeding back the response.

Optionally, the fusing the knowledge type set and updating the knowledge type set includes: judging whether the specified knowledge type exists in the knowledge type set or not; if the specified knowledge types exist, judging whether the specified knowledge types exist in plurality or not; and if a plurality of the specified knowledge types exist, fusing the specified knowledge types and updating the knowledge type set.

Optionally, after determining whether there are multiple designated knowledge types if there are fusible knowledge types, the method further includes: and if the plurality of specified knowledge types do not exist, deleting the specified knowledge types.

Optionally, if there are no multiple specified knowledge types, deleting the specified knowledge types, including: if the plurality of the specified knowledge types do not exist, acquiring the number of words corresponding to the specified knowledge types; and when the quantity is determined to be a designated numerical value, deleting the designated knowledge type from the knowledge type set.

Optionally, the constructing a query graph based on the updated knowledge type set includes: and generating a query graph based on the knowledge types in the knowledge type set after the ontology of the preset knowledge graph is connected and updated.

Optionally, the querying a response corresponding to the query graph from a preset knowledge graph and feeding back the response includes: if the number of the query graphs is multiple, carrying out quantization processing on each query graph according to the knowledge type corresponding to each query graph and the word corresponding to the knowledge type to obtain a quantization value of each query graph; sorting the query graphs from high to low according to quantization values, and determining the query graphs with the first specified number as candidate query graphs; and inquiring the answer corresponding to the candidate query graph from a preset knowledge graph, and feeding back the answer.

Optionally, the querying replies corresponding to the candidate query graph from a preset knowledge graph and feeding back the replies include: inquiring the answer corresponding to the candidate query graph from a preset knowledge graph, and taking the candidate query graph inquired to the answer as an optional query graph; and if at least one optional query graph exists, determining the optional query graph with the highest quantization value as a target query graph, and feeding back a reply corresponding to the target query graph.

Optionally, after querying a reply corresponding to the candidate query graph from a preset knowledge graph and taking the candidate query graph queried to the reply as an optional query graph, the method further includes: if the alternative query graph does not exist, new candidate query graphs are determined again from the query graphs after the previously specified number.

Optionally, the knowledge type set includes attributes of words, the fusion processing is performed on the knowledge type set, and before the knowledge type set is updated, the method further includes: acquiring attributes in the user input statement as first attributes; and matching the first attribute with the attribute in the knowledge type set, and adding the unmatched first attribute to the knowledge type set to update the knowledge type set.

Optionally, the obtaining an attribute in the user input sentence as a first attribute includes: and identifying the intention of the user input sentence to obtain a first attribute.

Optionally, the obtaining an attribute in the user input sentence as a first attribute includes: and performing attribute recognition on the user input sentence based on the trained implicit attribute recognition model to obtain a first attribute.

Optionally, after querying a reply corresponding to the candidate query graph from a preset knowledge graph and taking the candidate query graph queried to the reply as an optional query graph, the method further includes: and if the optional query graph does not exist, returning and executing the attribute in the acquired user input statement as a first attribute and subsequent operation.

Optionally, before performing fusion processing on the knowledge type set and updating the knowledge type set, the method further includes: detecting whether an entity exists in the knowledge type set; if not, acquiring a knowledge type set of a historical user input statement as a historical knowledge type set, wherein the historical user input statement is input before the user input statement; adding an entity in the set of historical knowledge types to the set of knowledge types to treat the entity of the historical user input statement as the entity of the user input statement.

In a second aspect, an embodiment of the present application provides a data processing apparatus based on a knowledge-graph, including: the input acquisition module is used for acquiring input sentences of a user; the type identification module is used for identifying at least one knowledge type corresponding to the user input statement to obtain a knowledge type set; the type updating module is used for carrying out fusion processing on the knowledge type set and updating the knowledge type set; the query graph building module is used for building a query graph based on the updated knowledge type set; and the answer query module is used for querying an answer corresponding to the query graph from a preset knowledge graph and feeding back the answer.

Optionally, the type updating module includes: a type judgment submodule, a quantity judgment submodule and a set updating submodule, wherein: the type judgment submodule is used for judging whether the specified knowledge type exists in the knowledge type set or not; the quantity judgment submodule is used for judging whether a plurality of specified knowledge types exist or not if the specified knowledge types exist; and the set updating submodule is used for fusing the plurality of specified knowledge types and updating the knowledge type set if the plurality of specified knowledge types exist.

Optionally, after determining whether there are multiple specified knowledge types if there are fusible knowledge types, the data processing apparatus based on a knowledge graph further includes: a type deletion module, wherein: and the type deleting module is used for deleting the specified knowledge types if a plurality of specified knowledge types do not exist.

Optionally, the type deleting module includes: a quantity acquisition submodule and a type deletion submodule, wherein: the quantity obtaining sub-module is used for obtaining the quantity of the words corresponding to the specified knowledge types if the plurality of specified knowledge types do not exist; and the type deleting submodule is used for deleting the specified knowledge type from the knowledge type set when the quantity is determined to be a specified numerical value.

Optionally, the query graph constructing module includes: a query graph generation sub-module, wherein: and the query graph generation submodule is used for generating a query graph based on the preset knowledge type in the knowledge type set after the ontology of the knowledge graph is connected and updated.

Optionally, the reply query module includes: the device comprises a quantization processing sub-module, a candidate determining sub-module and a reply query sub-module, wherein: the quantization processing sub-module is used for performing quantization processing on each query graph according to the knowledge type corresponding to each query graph and the word corresponding to the knowledge type to obtain a quantization value of each query graph if the number of the query graphs is multiple; the candidate determining submodule is used for sequencing the query graphs from high to low according to quantization values and determining the query graphs with the number specified previously as candidate query graphs; and the answer query submodule is used for querying the answer corresponding to the candidate query graph from a preset knowledge graph and feeding back the answer.

Optionally, the reply query submodule includes: optional confirm unit and optional inquiry unit, wherein: the optional determining unit is used for inquiring the answer corresponding to the candidate query graph from a preset knowledge graph, and taking the candidate query graph inquired to the answer as an optional query graph; and the optional query unit is used for determining the optional query graph with the highest quantization value as a target query graph and feeding back a reply corresponding to the target query graph if at least one optional query graph exists.

Optionally, after querying a reply corresponding to the candidate query graph from a preset knowledge graph and taking the candidate query graph queried to the reply as an optional query graph, the data processing apparatus based on the knowledge graph further includes: a re-determination module, wherein: and a redetermining module, configured to redetermine a new candidate query graph from the query graphs after the previously specified number if the optional query graph does not exist.

Optionally, the set of knowledge types includes attributes of words, and the data processing apparatus based on a knowledge graph further includes: a first attribute obtaining module and a first attribute adding module, wherein: a first attribute obtaining module, configured to obtain an attribute in the user input statement as a first attribute; and the first attribute adding module is used for matching the first attribute with the attribute in the knowledge type set, adding the unmatched first attribute to the knowledge type set so as to update the knowledge type set, returning to execute the fusion processing of the knowledge type set, and updating the knowledge type set and the subsequent steps.

Optionally, the first attribute obtaining module includes: an intent recognition submodule, wherein: and the intention identification submodule is used for carrying out intention identification on the user input sentence to obtain a first attribute.

Optionally, the first attribute obtaining module includes: an implicit attribute identification submodule, wherein: and the hidden attribute recognition submodule is used for carrying out attribute recognition on the user input sentence based on the trained hidden attribute recognition model to obtain a first attribute.

Optionally, after querying a reply corresponding to the candidate query graph from a preset knowledge graph and taking the candidate query graph queried to the reply as an optional query graph, the data processing apparatus based on the knowledge graph further includes: an attribute addition module, wherein: and the attribute increasing module is used for returning and executing the attribute in the acquired user input statement as a first attribute and subsequent operation if the optional query graph does not exist.

Optionally, before the fusing the set of knowledge types and updating the set of knowledge types, the data processing apparatus based on a knowledge graph further includes: entity detection module, history acquisition module and entity increase module, wherein: the entity detection module is used for detecting whether an entity exists in the knowledge type set or not; the history acquisition module is used for acquiring a knowledge type set of a history user input statement as a history knowledge type set if the history acquisition module does not exist, wherein the history user input statement is input before the user input statement; and the entity adding module is used for adding the entities in the historical knowledge type set to the knowledge type set so as to take the entities of the historical user input sentences as the entities of the user input sentences.

In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, where the memory is coupled to the processor, and the memory stores instructions, and when the instructions are executed by the processor, the processor performs the method of the first aspect.

In a fourth aspect, embodiments of the present application provide a computer-readable storage medium, in which program code is stored, and the program code can be called by a processor to execute the method according to the first aspect.

The embodiment of the application provides a data processing method, a data processing device, electronic equipment and a data processing medium based on a knowledge graph. Therefore, the knowledge types corresponding to the sentences input by the user are identified, the identified knowledge types are fused, the knowledge type set is updated, the response of the complex sentences can be covered, the response coverage range based on the knowledge graph is improved, the identification accuracy of the complex sentences is improved, and the response accuracy based on the knowledge graph is improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is a schematic diagram of an application environment suitable for use in embodiments of the present application;

FIG. 2 is a flow diagram illustrating a method for knowledge-graph based data processing according to an embodiment of the present application;

FIG. 3 illustrates a data processing method based on knowledge-graph provided by another embodiment of the present application;

FIG. 4 is a flowchart illustrating step S230 in FIG. 3 according to an exemplary embodiment of the present application;

FIG. 5 is a flowchart illustrating step S250 of FIG. 3 according to an exemplary embodiment of the present application;

fig. 6 is a flowchart illustrating step S253 in fig. 5 according to an exemplary embodiment of the present application;

fig. 7 is a flowchart illustrating step S253 in fig. 5 according to an exemplary embodiment of the present application;

FIG. 8 shows a flow diagram of a method of knowledge-graph based data processing provided by yet another embodiment of the present application;

FIG. 9 is a flow diagram illustrating a method for knowledge-graph based data processing according to yet another embodiment of the present application;

FIG. 10 is a flow diagram illustrating a method for knowledge-graph based data processing according to yet another embodiment of the present application;

FIG. 11 illustrates a diagram of dynamic update of knowledge type sets provided by an exemplary embodiment of the present application;

FIG. 12 is a block diagram of a knowledge-graph based data processing apparatus provided by an embodiment of the present application;

fig. 13 shows a block diagram of an electronic device for executing a data processing method based on a knowledge-graph according to an embodiment of the present application.

Fig. 14 illustrates a storage unit for storing or carrying program code implementing a knowledge-graph based data processing method according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

Definition of terms

Knowledge graph: a knowledge graph is a knowledge base used by Google to enhance its search engine functionality. Essentially, a knowledge graph is a semantic network that exposes relationships between entities (entities), and can formally describe real-world things and their interrelationships.

Currently, due to the development of Semantic Net (Semantic Net) and the emergence of large-scale structured Knowledge graph (KnowledgeGraph), Knowledge graph-based Question Answering (Question Answering) has gained wide attention in the fields of Natural Language Processing (NLP) and Knowledge Base (KB). Especially for complex question and answer scenarios, the conventional processing logic is to use predefined templates or pattern matching. However, the pre-defined template method requires a lot of manual work and even knowledge experts, and when the templates are too many, conflicts are easy to generate and adjustment is not easy.

In addition, there is a method of constructing a query graph using a Finite State Machine (FSM) to answer complex questions based on a knowledge graph. However, when the query graph is built by using the finite state machine, a subject term needs to be found first, and an answer node is found through a core relationship chain, so that a restriction function can only be added to the subject term, and the answer and the subject term cannot be connected through an entity variable, which results in that the complexity of the query is greatly restricted.

Based on the above problems, the inventor proposes a data processing method, device, electronic device and medium based on a knowledge graph in the embodiments of the present application through long-term research, obtains a knowledge type set by obtaining a user input sentence, then identifies at least one knowledge type corresponding to the user input sentence, updates the knowledge type set by performing fusion processing on the knowledge type set, finally constructs a query graph based on the updated knowledge type set, queries a response corresponding to the query graph from a preset knowledge graph, and feeds back the response. Therefore, the knowledge types corresponding to the sentences input by the user are identified, the identified knowledge types are fused, the knowledge type set is updated, the response of the complex sentences can be covered, the limit of the related technology on the query complexity is broken through, the response coverage range based on the knowledge map is improved, the identification accuracy of the complex sentences is improved, and the response accuracy based on the knowledge map is improved.

In order to better understand the method, the apparatus, the electronic device, and the medium for processing data based on a knowledge graph provided in the embodiments of the present application, an application environment suitable for the embodiments of the present application is described below.

Referring to fig. 1, fig. 1 is a schematic diagram illustrating an application environment suitable for the embodiment of the present application. The data processing method based on the knowledge graph provided by the embodiment of the application can be applied to an interactive system 10 shown in fig. 1. The interactive system 10 includes a terminal device 100 and a server 200.

The terminal device 100 may be, but is not limited to, a smart speaker, a smart phone, a tablet computer, a laptop computer, a desktop computer, a wearable electronic device, or other electronic devices that are equipped with a data processing apparatus based on a knowledge graph.

The server 200 and the terminal device 100 are connected through a wireless or wired network to realize data transmission between the terminal device 100 and the server 200 based on the network connection, and the transmitted data includes but is not limited to sentences, replies and the like.

The server 200 may be a conventional server, a cloud server, a server cluster including a plurality of servers, or even a server center including a plurality of servers. Server 200 may be configured to provide a background service for a user, which may include, but is not limited to, a knowledge-graph based response service, and the like, without limitation.

In some embodiments, the terminal device 100 may have a client application installed thereon, and the user may communicate with the server 200 based on the client application (e.g., APP, wechat applet, etc.). Specifically, the server 200 is installed with a corresponding server application, the user may register a user account in the server 200 based on the client application, and communicate with the server 200 based on the user account, for example, the user logs in the user account in the client application, inputs the user account through the client application based on the user account, and may input text information or voice information, and the like, after receiving the information input by the user, the client application may send the information to the server 200, so that the server 200 may receive, process, and store the information, and the server 200 may also receive the information and return a corresponding output information to the terminal device 100 according to the information.

In some embodiments, the means for processing the user input sentence may also be disposed on the terminal device 100, so that the terminal device 100 can realize the interaction with the user without relying on establishing communication with the server 200, and realize the response based on the knowledge graph, in which case the interactive system 10 may only include the terminal device 100.

The above application environments are only examples for facilitating understanding, and it is to be understood that the embodiments of the present application are not limited to the above application environments.

The method, the apparatus, the electronic device and the medium for processing data based on a knowledge graph provided by the embodiments of the present application will be described in detail through specific embodiments.

Referring to fig. 2, fig. 2 is a schematic flowchart illustrating a method for processing data based on a knowledge-graph according to an embodiment of the present application, which can be applied to the terminal device, and the flow shown in fig. 2 will be described in detail below. The above-mentioned data processing method based on knowledge graph may specifically include the following steps:

step S110: and acquiring a user input statement.

The user input sentence is a sentence input by a user, and can be input based on a terminal device, so that the terminal device obtains the user input sentence.

In some examples, such as in a knowledge-graph based question-and-answer scenario, the user input sentence may be a user-input question (query). It should be noted that the question may be a question in the traditional syntax, or may be various types of sentences such as a statement sentence, which is not limited herein.

In some embodiments, the user interface of the terminal device may display an input box as an entry for the user input, and the user may input a sentence in the input box, and the terminal device may detect the input operation, and obtain the user input sentence input by the user, so as to perform a subsequent operation.

Step S120: and identifying at least one knowledge type corresponding to the user input sentence to obtain a knowledge type set.

The knowledge types may include, among others, generic knowledge types and graph knowledge types, such as:

the generic knowledge types may include: logic connection words (and, or), comparison words (larger than or smaller than), number words, quantifier and other cross-domain knowledge types;

the graph knowledge types may include: entity (Entity), Class (Class), Property (DataProperty/Property), relationship (ObjectProperty), Literal volume (Literal), etc. are based on the knowledge type of the ontology Language (WebOntology Language, OWL) developed by W3C. Wherein, in some examples, a category may refer to a category of an entity.

Of course, the knowledge type is not limited to the above type, and the embodiment does not limit this.

In some embodiments, semantic analysis may be performed on a user input sentence to identify at least one knowledge type corresponding to the user input sentence, so as to obtain a knowledge type set including the at least one knowledge type, specifically, an empty natural language understanding item nluistems (natural language understanding items) may be initialized for one user input sentence, and the set is used as the knowledge type set, and then after the identification, the identified knowledge type is placed in the nluistems. For example, if a knowledge type is identified for a user input statement to include the attribute "price," then this knowledge of the attribute "price" may be added to the nlutems corresponding to the user input statement.

Wherein, the semantic analysis can comprise at least one of word segmentation, entity recognition, part of speech tagging, dependency analysis and the like.

Taking entity recognition as an example, an entity is a basic element of the knowledge graph, and the entity can be a key subject word in a user input sentence, such as a name of a person, a place name, an organization and the like, and can be obtained through entity recognition. In some embodiments, the named entity may be obtained by named entity recognition, and the named entity recognition model is specifically implemented based on a named entity recognition model, where the named entity recognition model may be generated based on deep learning model training, in some examples, the deep learning model may be a Recurrent Neural Network (RNN) model, and in an embodiment, a long short Term Memory Network (LSTM) or a Gated Recurrent Unit (GRU) may be further specifically used, and this embodiment does not perform other limitations and details on the named entity recognition model used herein.

In some exemplary embodiments, a comparison term such as "greater than", a quantity term such as "50 ten thousand", a map vehicle type category such as "vehicle", and the like knowledge types may be identified using a reference identification link (Mention Linking). In some examples, a "vehicle" may also be identified as an entity in the knowledge-graph, without limitation. The number words may represent a combination of the number words and the quantity words in this embodiment.

Step S130: and performing fusion processing on the knowledge type set, and updating the knowledge type set.

And performing fusion processing on the knowledge type set, namely performing fusion processing on at least one knowledge type in the knowledge type set, wherein the fusion processing refers to combining part of knowledge types into one knowledge type through fusion (Merge) operation when the knowledge type set comprises a plurality of knowledge types so as to facilitate the generation of a subsequent query graph.

In some embodiments, the knowledge type set may be updated by determining whether the knowledge type set includes a plurality of specified knowledge types, and if so, fusing the plurality of specified knowledge types to update the knowledge type set, so that the knowledge types in the updated knowledge type set are more centralized to extract the expression of the user input sentence, and to better characterize the user input sentence. The specific implementation manner can be seen in the following embodiments, which are not described herein again.

In the present embodiment, "a plurality" means two or more.

Step S140: and constructing a query graph based on the updated knowledge type set.

Based on the updated knowledge type set, that is, based on the knowledge types in the updated knowledge type set, a corresponding query graph can be constructed to query a corresponding response in the knowledge graph, so as to obtain a response corresponding to the question input by the user.

In some embodiments, a connection manner between knowledge types in the knowledge type set can be found based on an ontology (ontology) of the knowledge graph, so that a connection (Connect) operation is performed on knowledge types included in the updated knowledge type set based on the ontology to generate a query graph, so as to implement a query based on the knowledge graph. Specifically, the words corresponding to the knowledge types may be connected to generate a query graph.

The ontology is type information of the knowledge graph and is a graph network. In some examples, in the present embodiment, in the field of computer science and information science, an ontology provides a shared vocabulary, which contains object types or concepts existing in a specific field and their attributes and interrelations, or is used to indicate Formal expressions (Formal Representation) of concepts and their interrelations in a specific field.

In actual operation, the words corresponding to the updated knowledge type sets can be formed into the query graph based on the concepts and the mutual relations in the ontology, so that the relations among all points in the query graph can accord with the mutual relations among the concepts and the attributes in the specific field, the structure of the query graph which does not accord with actual conditions is avoided, the recognition and reply accuracy of the input sentences of the user can be improved, and wrong answers which do not accord with the actual conditions are avoided.

Step S150: and inquiring a response corresponding to the query graph from a preset knowledge graph, and feeding back the response.

In this embodiment, after the query graph is generated, answers may be searched for in the preset knowledge graph in a sub-graph matching manner, or answers corresponding to the query graph may be queried in the preset knowledge graph in other manners.

Wherein the preset knowledge graph may be a knowledge graph set as at least one domain. For example, a knowledge graph related to automobile sales may be preset, and if the user inputs a sentence "car with price greater than 50 ten thousand", a corresponding answer may be queried from the knowledge graph related to automobile sales as a response. The knowledge graph may be stored in a database stored in an RDF format, a Neo4j database, a FlockDB database, a JanusGraph database, or the like, and may be set according to actual needs, which is not limited in this embodiment. The format of the queryable statement is determined by the database.

In some embodiments, the query graph may be converted into a queryable statement to query from a preset knowledge graph with the queryable statement. The format of the queryable statement can be determined by a database stored by a preset knowledge graph. Specifically, the query graph can be converted into a SPARQL statement or a Cypher statement, and then the converted SPARQL statement or Cypher statement is adopted to query and answer in a preset knowledge graph.

Sparql (sparql Protocol and RDF Query language), which is a Query language and data acquisition Protocol developed for RDF (resource description Framework). And converting the query graph into a SPARQL statement, and querying from a preset knowledge graph stored in an RDF format to obtain a result.

Cypher is an official query statement of a Neo4j database, and a result, namely a response corresponding to a query graph, can be obtained by converting the query graph into the Cypher statement and querying the Cypher statement from a preset knowledge graph stored in the Neo4j database.

After the answer corresponding to the query graph is inquired from the preset knowledge graph, the answer can be fed back. The manner of feedback may include, but is not limited to, outputting text of the response (e.g., displayed on an interface of the terminal device), voice, etc., so that the user may receive the response of the feedback. The present embodiment does not limit the form of feedback.

According to the data processing method based on the knowledge graph, the user input sentences are obtained, at least one knowledge type related to the user input sentences is identified, the knowledge types are subjected to fusion processing to update the knowledge type set, the sentence identification coverage can be improved, and the complex sentence identification accuracy is improved. And a query graph is constructed based on the updated knowledge type set to determine a corresponding response, so that the response accuracy can be improved.

Referring to fig. 3, fig. 3 illustrates a method for processing data based on a knowledge-graph according to another embodiment of the present application, which may be applied to the terminal device, and the method may include:

step S210: and acquiring a user input statement.

Step S220: and identifying at least one knowledge type corresponding to the user input sentence to obtain a knowledge type set.

Step S230: and performing fusion processing on the knowledge type set, and updating the knowledge type set.

In some embodiments, step S230 may include steps S231 to S233, please refer to fig. 4, fig. 4 shows a schematic flowchart of step S230 in fig. 3 according to an exemplary embodiment of the present application, and step S230 may include:

step S231: and judging whether the specified knowledge type exists in the knowledge type set.

In some implementations, specifying the type of knowledge may include comparing words, quantifiers.

In addition, because comparables, quantifiers, and the like are often used to modify attributes, a given knowledge type may also include attributes. Specifically, the attribute mainly refers to an attribute, a feature, a characteristic, a feature, a parameter, and the like that the Object (Object) may have, for example, the price, horsepower, and the like of the vehicle; the attribute value mainly refers to a value of an attribute, for example, an attribute value of an attribute "price" of a vehicle may be "50 ten thousand" or the like.

Based on the recognition, at least one knowledge type corresponding to the user input sentence can be determined, a knowledge type set is obtained, and whether the specified knowledge type exists in the knowledge type set or not can be judged.

Step S232: and if the specified knowledge types exist, judging whether the specified knowledge types exist in plurality.

If the specified knowledge type exists, whether the specified knowledge type exists in a plurality of numbers can be judged, namely the specified knowledge type can comprise comparison words, quantifier words and attributes, and if one user input statement comprises at least two specified knowledge types, the specified knowledge type can be judged to exist in a plurality of numbers.

For example, if the comparison word "greater than 50 ten thousand" and the quantity word "50 ten thousand" are recognized for the user input sentence "greater than 50 ten thousand", the knowledge type set includes at least two designated knowledge types, and there are a plurality of designated knowledge types.

Step S233: and if the plurality of specified knowledge types exist, fusing the plurality of specified knowledge types and updating the knowledge type set.

And if a plurality of specified knowledge types exist, fusing the specified knowledge types, and updating the knowledge type set to ensure that at least two specified knowledge types are combined into one knowledge type, namely a restriction function.

For example, for a user input sentence "more than 50 ten thousand cars? If the category "car", the comparison word "greater than", the quantity word "50 ten thousand", and the attribute is "price" are identified, that is, the knowledge type set may include at least the above three, then at this time, the two knowledge types (the comparison word and the quantity word) need to be merged into one knowledge type, that is, a restriction function, and the word corresponding to the restriction function is "greater than 50 ten thousand", and the attribute "price" is restricted together, so as to update the knowledge type set, and at this time, the updated knowledge type includes the restriction function "greater than 50 ten thousand", the category "car", and the attribute "price", so that it is convenient to generate a subsequent query graph.

Since the word "more than 50 ten thousand" corresponding to the merged knowledge type is used to limit other knowledge types, that is, the attribute "price", the merged knowledge type may be recorded as a limiting function in the embodiment of the present application, and other names may be used in other embodiments.

Therefore, the method provided by the embodiment can add the restriction function to the non-subject term, and the restriction function can be added to any position without being limited to the subject term (superior to the existing finite state machine), thereby facilitating the generation of the subsequent semantic query graph. Compared with the existing mode of establishing the query graph through a finite state machine, the method can break through the limitation that the limit function can only be added to the subject term in the mode, namely break through the existing limitation on the query complexity, and improve the question-answer coverage based on the knowledge graph.

In some embodiments, based on the above example, a restriction function may be further fused with the attribute restricted by the restriction function, so that "more than 50 ten thousand" is fused with "price" to obtain "price more than 50 ten thousand" as a word corresponding to a new restriction function, that is, a new restriction function is obtained, and thus the knowledge type set is further updated.

In addition, in some embodiments, if there are no multiple specified knowledge types, the specified knowledge types may be deleted, so as to construct a query graph without fusible knowledge types that cannot be fused, avoid interference with information, and improve recognition accuracy of a sentence, thereby determining content to be expressed by the sentence more accurately without relying on intention recognition.

The deletion processing may be to remove the specified knowledge type from the knowledge type set to other storage locations, to delete the knowledge type completely, or to set the status bit of the knowledge type as unidentified so as to identify the unidentified status bit when subsequently constructing the query graph, that is, to construct the query graph without the knowledge type and the corresponding word thereof, so as to avoid a sentence identification error introduced by the interference information. The present embodiment does not limit the specific manner of the deletion process.

In some embodiments, if there are no multiple specified knowledge types, the number of words corresponding to a specified knowledge type may be obtained first, and when the number is determined to be a specified value, the specified knowledge type may be deleted from the knowledge type set. Since the words specifying the knowledge type in the sentence may have no practical meaning if they appear in isolation, for example, for the user input sentence "recommend a vehicle with low fuel consumption", a "one" of them is easily recognized as the knowledge type of the number word, but since the comparison word and the measure word are not recognized to modify the attribute "fuel consumption", a "one" of this user input sentence has no practical meaning, and may interfere with the final recognition result, so that this knowledge type needs to be deleted from the knowledge type set to prevent the interference with the generation of the query graph.

In some embodiments, if the specified value is 1, then after the fusion process, if there is still an isolated specified knowledge type and the isolated words of the specified knowledge type are also isolated, then the deletion process can be performed on the words of the specified type. If the isolated word of the specified knowledge type is not isolated, that is, if the number of the isolated word of the specified knowledge type is not a specified value, for example, greater than 1, a corresponding query graph can be generated for each isolated word of the specified knowledge type, that is, if the number of the generated query graphs is multiple, the step described later can be seen when determining the response from the multiple query graphs. Therefore, if the number is not a specified value, a corresponding query graph can be generated for each word corresponding to a specified knowledge type respectively so as to query and feed back a response based on the knowledge graph.

Step S240: and generating a query graph based on the ontology of the preset knowledge graph and the knowledge types in the updated knowledge type set.

The specific implementation of step S240 can refer to step S140 in the above embodiments, and is not described herein again.

Step S250: and inquiring a response corresponding to the query graph from a preset knowledge graph, and feeding back the response.

When the number of the generated query graphs is multiple, one or more query graphs with high scores can be selected as candidate query graphs by ranking the query graphs, so as to determine the response and feedback of the user input sentence, specifically, referring to fig. 5, fig. 5 shows a flowchart of step S250 in fig. 3 provided by an exemplary embodiment of the present application, and step S250 may include:

step S251: if the number of the query graphs is multiple, each query graph is quantized according to the knowledge type corresponding to each query graph and the word corresponding to the knowledge type, and the quantized value of each query graph is obtained.

In some embodiments, after the aforementioned identification and fusion processes, the number of the query graphs generated according to the updated knowledge type set may be multiple, for example, if multiple attributes are identified for the user input sentence, or multiple numbers of words are identified and cannot be excluded, that is, if only one word corresponding to each knowledge type cannot be reserved, a query graph is generated by connecting one word from each knowledge type each time, so that multiple query graphs can be obtained.

If the number of the query graphs is multiple, each query graph can be quantized according to the knowledge type corresponding to each query graph and the word corresponding to the knowledge type, so that the quantized value of each query graph is obtained. Specifically, one or more specified features can be constructed, and the feature value of each query graph corresponding to the specified features is calculated as the quantization value corresponding to the query graph. The generated query graph can be scored and sorted through one-dimensional or multi-dimensional features.

The specific feature may be determined according to actual needs, for example, the specific feature may be a coverage rate, and the coverage rate is a coverage rate of a knowledge type appearing in one sentence. In one example, for a user input sentence "car with price greater than 30 ten thousand", if "price", "greater than", "30 ten thousand" and "car" are recognized, all the four words have corresponding knowledge types, and only 1 word ("of") in 9 words has no corresponding knowledge type, then the coverage rate of the knowledge type of the user input sentence can be obtained as 8/9, and the higher the coverage rate is, the higher the score is.

It should be noted that the above examples are only coverage characteristics, and other characteristics (such as distribution of knowledge types, repeated coverage of knowledge types, etc.) can be combined to form a multidimensional characteristic for more comprehensive scoring and sorting.

Step S252: and sequencing the query graphs from high to low according to the quantization values, and determining the query graphs with the first specified number as candidate query graphs.

In some embodiments, the quantized values corresponding to the query graphs obtained through different connection modes may have differences, and by sorting the query graphs from high to low according to the quantized values, the query graphs with the previously specified number can be determined as candidate query graphs, so that the number of queries is effectively reduced, the query efficiency is improved, and excessive resource consumption is avoided.

Step S253: and inquiring the answer corresponding to the candidate query graph from the preset knowledge graph, and feeding back the answer.

In one embodiment, after the candidate query graph is obtained, the candidate query graph may be converted into a queriable sentence, such as a Cypher sentence, and queried in a preset knowledge graph. Specifically, for each candidate query graph, a corresponding answer is queried from a preset knowledge graph and fed back, and answers corresponding to multiple candidate query graphs can be fed back, or an answer corresponding to one of the candidate query graphs can be selected for feedback.

In some embodiments, step S253 may include steps S2531 to S2532 to determine a final target query graph from the candidate query graphs for which answers can be found, and determine a final answer. Specifically, referring to fig. 6, fig. 6 shows a schematic flowchart of step S253 in fig. 5 according to an exemplary embodiment of the present application, where step S253 may include:

step S2531: and querying the answer corresponding to the candidate query graph from the preset knowledge graph, and taking the candidate query graph queried to the answer as an optional query graph.

In practical applications, there may be a case where the candidate query graph cannot query the preset knowledge graph for a response, and the reason why the response cannot be queried may be that the preset knowledge graph does not have a corresponding response, or the candidate query graph is wrong, that is, the user input sentence is not correctly characterized. Candidate query graphs that can be queried for replies may be obtained as alternative query graphs.

Step S2532: and if at least one optional query graph exists, determining the optional query graph with the highest quantization value as a target query graph, and feeding back a reply corresponding to the target query graph.

In some embodiments, if an alternative query graph does not exist, the candidate query graph may be re-determined.

In one embodiment, if no alternative query graph exists, new candidate query graphs may be re-determined from the generated query graphs. Step S253 may include steps S2533 to S2536, so as to determine a final target query graph from the candidate query graphs for which the answer can be found, and determine a final answer, and if none of the candidate query graphs can be queried to answer, re-determine a different candidate query graph for querying. Specifically, referring to fig. 7, fig. 7 is a schematic flowchart illustrating the step S253 in fig. 5 according to an exemplary embodiment of the present application, where the step S253 may include:

step S2533: and querying the answer corresponding to the candidate query graph from the preset knowledge graph, and taking the candidate query graph queried to the answer as an optional query graph.

Step S2534: it is determined whether an alternative query graph exists.

By querying each candidate query graph, if each candidate query graph cannot query a reply, it can be determined that no optional query graph exists, otherwise, it exists.

In this embodiment, after determining whether the optional query graph exists, the method may further include:

if at least one alternative query graph exists, step S2535 may be performed;

if no alternative query graph exists, step S2536 can be performed.

Step S2535: and determining the optional query graph with the highest quantization value as a target query graph, and feeding back a response corresponding to the target query graph.

If at least one optional query graph exists, the optional query graph with the highest quantization value can be determined as the target query graph, and the answer corresponding to the target query graph is fed back. It can be understood that the query graph with a higher quantization value has better characterization capability on the question and better recognition effect, so that more accurate question characterization can be determined through step S2535 to query the response, and a more accurate response is obtained, that is, the response accuracy is improved.

Step S2536: new candidate query graphs are re-determined from the query graphs after the previously specified number.

If there is no alternative query graph, that is, no candidate query graph can be queried from the preset knowledge-graph to the answer, at this time, the candidate query graph may be re-determined from the generated multiple query graphs, so that the new candidate query graph is different from the previously determined candidate query graph, and step S2533 is performed to re-query the answer based on the new candidate query graph.

It should be noted that, for parts not described in detail in this embodiment, reference may be made to the foregoing embodiments, and details are not described herein again.

By the data processing method based on the knowledge graph, the query graph can be dynamically updated to perform knowledge question answering on the premise of no intention of training.

In addition, in an embodiment, if there is no optional query graph, the user input sentence may be re-identified to obtain a new knowledge type, the new knowledge type is added to the knowledge type set to update the knowledge type set, and based on the updated knowledge type set, the step S230 and subsequent steps are returned to be executed, so as to generate a new query graph query reply through fusion processing and subsequent steps, and the like. Specifically, referring to fig. 8, fig. 8 is a flow chart illustrating a method for processing data based on a knowledge-graph according to another embodiment of the present application, where the method may include:

step S301: and acquiring a user input statement.

Step S302: and identifying at least one knowledge type corresponding to the user input sentence to obtain a knowledge type set.

Step S303: and performing fusion processing on the knowledge type set, and updating the knowledge type set.

Step S304: and generating a query graph based on the ontology of the preset knowledge graph and the knowledge types in the updated knowledge type set.

Step S305: if the number of the query graphs is multiple, each query graph is quantized according to the knowledge type corresponding to each query graph and the word corresponding to the knowledge type, and the quantized value of each query graph is obtained.

Step S306: and sequencing the query graphs from high to low according to the quantization values, and determining the query graphs with the first specified number as candidate query graphs.

Step S307: and querying the answer corresponding to the candidate query graph from the preset knowledge graph, and taking the candidate query graph queried to the answer as an optional query graph.

Step S308: it is determined whether an alternative query graph exists.

if at least one optional query graph exists, step S309 may be performed;

if no alternative query graph exists, step S310 may be executed.

Step S309: and determining the optional query graph with the highest quantization value as a target query graph, and feeding back a response corresponding to the target query graph.

Step S310: and acquiring the attribute in the user input statement as a first attribute.

And if the optional query graph does not exist, acquiring the attribute in the user input statement as the first attribute. The algorithm on which the attributes are obtained here is different from the algorithm that previously identified at least one knowledge type in the user input sentence. For example, in the previous algorithm, when the attribute is identified, the attribute may be subjected to string matching, for example, the attribute "price" may be obtained by string matching for the sentence a "car with a price greater than 50 ten thousand", but the attribute "price" may not be determined by string matching for the sentence B "car with a price greater than 50 ten thousand", and there may be missed identification of the attribute.

In some embodiments, the first attribute may be derived by performing intent recognition on a user input sentence. In some embodiments, a large number of questions may be collected and labeled with intentional graph labels to obtain an intention training sample set, where the intention training sample set includes a plurality of questions and an intention label corresponding to each question, and then training is performed based on the intention training sample set by using a deep learning model, such as a Convolutional Neural Network (CNN) model or a sequence-to-sequence (seq2seq) model, to obtain an intention recognition model, so that an intention of a user input sentence may be recognized based on the intention recognition model, an intention corresponding to the user input sentence is obtained, and a first attribute is obtained according to the intention. For example, how much money the user entered the sentence "s 60? The "corresponding intent is" car price query ", and the corresponding first attribute is" price "can be derived.

For example, based on the foregoing example, for a vehicle with a sentence B "greater than 50 ten thousand", it can be judged by the intention recognition that the intention of the sentence B is to ask for the price of the vehicle, so as to acquire the attribute "price" as the first attribute to supplement the recognition, thereby improving the recognition accuracy of the user input sentence.

As another example, how much money is for the question "s 60? "how much money" cannot be identified by character string matching corresponds to the attribute "price" in the knowledge base, and by intention identification, it can be identified that the question is asking for a price, thereby obtaining the attribute "price" as the first attribute.

In other embodiments, attribute recognition may also be performed on the user input sentence based on the trained implicit attribute recognition model to obtain the first attribute. Aiming at the situation that a plurality of attributes are involved in the user input sentence but recognition is not complete, the plurality of attributes involved in the user input sentence can be recognized simultaneously based on the implicit attribute recognition model of deep learning, and the recognition capability of the user input sentence is enhanced.

In some embodiments, a plurality of question sentences may be collected, and each question sentence is labeled with an attribute label, so as to obtain an attribute training sample set, where the attribute training sample set includes the plurality of question sentences and the corresponding attribute labels. The implicit attribute recognition model is then trained based on a training sample set, which in one example may be trained using cosine similarity.

In some embodiments, the implicit attribute model may be constructed based on a bidirectional recurrent neural network (BiRNN), and in particular, a bidirectional long-short term memory (BiLSTM) network or a bidirectional gated cyclic unit (BiGRU) may also be used. Taking the BilSTM as an example, the hidden attribute model may include an Embedding (Embedding) layer and a BilSTM layer, and specifically, after semantic mapping of the vectorized user input sentence and attribute is obtained by using the Embedding (Embedding) layer and the BilSTM layer, the trained hidden attribute recognition model is obtained by using the cosine similarity training model, and can be used for recognizing the hidden attribute in the sentence and enhancing the recognition capability of the user input sentence.

In still other embodiments, the user input sentence may be subjected to the intention recognition based on the intention recognition model and the attribute recognition based on the trained implicit attribute recognition model, and the union of the recognition results of the two models is taken to obtain the first attribute, so that the new attributes recognized by the two models can be added to the knowledge type set of the user input question in the current round. For example, the question "s 60 consumes several oil", the Entity (Entity) "s 60" is obtained by the intention recognition, the attribute (data property) "oil consumption" is obtained by the intention recognition, the intention can be changed by the implicit attribute recognition algorithm, so as to obtain a new attribute "price", thereby not completely inheriting the result of the intention, and improving the recognition capability of the question.

Step S311: and matching the first attribute with the attribute in the knowledge type set, and adding the unmatched first attribute to the knowledge type set to update the knowledge type set.

And matching the first attribute with the attribute in the knowledge type set, and adding the unmatched first attribute to the knowledge type set to update the knowledge type set, so that the previously unidentified attribute can be added to the knowledge type set, and the knowledge type set is dynamically updated, so that the query graph obtained based on the updated knowledge type set can more accurately represent the user input statement.

In an exemplary embodiment, attributes in the user input sentence may be identified based on string matching in step S302. In one example, how much money is being made for the question "s 60? The attribute price in the knowledge graph corresponding to the money cannot be identified through character string matching, the question can be identified to inquire the price through intention identification, the attribute price is obtained as a first attribute, the attribute price is not matched with the original knowledge type set through matching operation, and at the moment, the knowledge type set can be subjected to adding operation to increase the attribute price.

In some embodiments, before the fusion processing is performed on the knowledge types, the missing recognition knowledge types can be added to improve the recognition accuracy of the knowledge types, so as to help improve the recognition accuracy of the user input sentences. Specifically, referring to fig. 9, fig. 9 is a flowchart illustrating a data processing method based on a knowledge graph according to still another embodiment of the present application, which may be applied to the terminal device, where the method may include:

step S410: and acquiring a user input statement.

Step S420: and identifying at least one knowledge type corresponding to the user input sentence to obtain a knowledge type set.

Wherein the set of knowledge types includes attributes of the words.

Step S430: and acquiring the attribute in the user input statement as a first attribute.

Step S440: and matching the first attribute with the attribute in the knowledge type set, and adding the unmatched first attribute to the knowledge type set to update the knowledge type set.

Therefore, the attributes which are not recognized before can be increased to update the knowledge type set, so that the knowledge types contained in the knowledge type set are more accurate, and the recognition accuracy of the input sentences of the user is prevented from being influenced by the missing recognition of the knowledge types.

Step S450: and performing fusion processing on the knowledge type set, and updating the knowledge type set.

Step S460: and constructing a query graph based on the updated knowledge type set.

Step S470: and inquiring a response corresponding to the query graph from a preset knowledge graph, and feeding back the response.

In addition, in some embodiments, when a response corresponding to the query graph is queried from the preset knowledge graph and fed back, if the steps shown in fig. 5 and 6 are implemented, after step S2531, if there is no optional query graph, the user input sentence may be re-identified, and a new knowledge type is obtained and added to the knowledge type set, so as to generate a response corresponding to a new query graph query through fusion processing and subsequent steps. Specifically, referring to fig. 10, fig. 10 is a schematic flow chart illustrating a method for processing data based on a knowledge-graph according to another embodiment of the present application, where the method may include:

step S501: and acquiring a user input statement.

Step S502: and identifying at least one knowledge type corresponding to the user input sentence to obtain a knowledge type set.

Step S503: and acquiring the attribute in the user input statement as a first attribute.

Step S504: and matching the first attribute with the attribute in the knowledge type set, and adding the unmatched first attribute to the knowledge type set to update the knowledge type set.

Step S505: and performing fusion processing on the knowledge type set, and updating the knowledge type set.

Step S506: and constructing a query graph based on the updated knowledge type set.

Step S507: if the number of the query graphs is multiple, each query graph is quantized according to the knowledge type corresponding to each query graph and the word corresponding to the knowledge type, and the quantized value of each query graph is obtained.

Step S508: and sequencing the query graphs from high to low according to the quantization values, and determining the query graphs with the first specified number as candidate query graphs.

Step S509: and querying the answer corresponding to the candidate query graph from the preset knowledge graph, and taking the candidate query graph queried to the answer as an optional query graph.

Step S510: it is determined whether an alternative query graph exists.

In this embodiment, after determining whether the optional query graph exists, the method further includes:

if there is at least one optional query graph, step S511 may be performed;

if the optional query graph does not exist, the step S503 may be executed in return.

In some embodiments, if there is no alternative query graph, the entities involved in the user input statements before the user input statement in the current round may also be added to the knowledge type set of the user input statement in the current round to update the query graph.

In particular, it may be detected whether an entity exists in the set of knowledge types; if the historical knowledge type set does not exist, the knowledge type set of the historical user input statement can be obtained to serve as the historical knowledge type set, and the historical user input statement is input before the user input statement; and adding the entities in the historical knowledge type set to the knowledge type set so as to take the entities of the historical user input sentences as the entities of the user input sentences. Therefore, the knowledge type sets in the current round of question can be integrated with the knowledge type sets in the previous round of question, and multiple rounds of question answering can be realized.

For example, at some point in time is the user entering the first user input sentence "how much money is s 60? "a first query graph is obtained, and then a second user input sentence is input by the user at the next moment as" oil consumption? If there is no entity in the second user input sentence, the entity "s 60" of the previous round of user input question sentence may be inherited, and the entity "s 60" may be added to the knowledge type set of the current round of user input question sentence to obtain a query graph, so that multiple rounds of interaction such as multiple rounds of question and answer may be performed.

In other embodiments, if there is no optional query graph, the knowledge type set may be dynamically updated by using reinforcement learning to dynamically update the knowledge type set, so as to generate a dynamic query graph and answer the knowledge question and answer of the complex question. For example, if the answer obtained by the first recalled candidate query graph is null, the knowledge type set corresponding to the candidate query graph may be modified, and a new candidate query graph is generated for query.

Step S511: and determining the optional query graph with the highest quantization value as a target query graph, and feeding back a response corresponding to the target query graph.

By the data processing method based on the knowledge graph, at least one knowledge type of a user input statement can be identified, after a knowledge type set is obtained for the first time, new attributes which are not contained in an original knowledge type set are added through adding operation, so that the knowledge type set is updated, and therefore the identification capability of the user input statement can be improved.

Furthermore, at least two fusible knowledge types are merged into a new knowledge type by fusing the knowledge types in the knowledge type set and are recorded as a restriction function, so that the characterization capability of a subsequently generated query graph on a user input statement is improved.

And when isolated words of isolated knowledge types exist in the knowledge type set, the isolated words of the specified knowledge types are removed from the knowledge type set, so that the generation of a subsequent query graph is prevented from being interfered.

And then, according to the ontology of the knowledge graph, connecting the isolated knowledge types in the knowledge type set by the connection mode of the isolated knowledge types in the knowledge type set, and generating at least one query graph.

And when the number of the generated query graphs is multiple, performing quantization processing on each query graph to perform score sorting on each query graph, selecting a front appointed number of query graphs with high quantization values as candidate query graphs, querying answers corresponding to the candidate query graphs in a preset knowledge graph, and using the candidate query graphs capable of being queried as optional query graphs, if at least one optional query graph exists, determining the answer corresponding to the optional query graph with the highest quantization value as a final answer for feedback, and if no optional query graph exists, returning to execute an adding operation (namely returning to execute step S503) to continue to select the optimal query graph. Therefore, the response of the complex sentence can be covered, the coverage range of the question and answer based on the knowledge graph is improved, the recognition accuracy of the sentence and the accuracy of the answer are improved, the knowledge response can be carried out on the complex sentence based on the knowledge graph, and in some embodiments, a more accurate query graph can be generated without depending on the intention to carry out the knowledge response even if the intention training is not carried out.

Taking fig. 11 as an example, a process of dynamically updating a set of knowledge types in the data processing method based on a knowledge graph according to the foregoing embodiment is schematically described, and fig. 11 shows a schematic diagram of dynamically updating a set of knowledge types according to an exemplary embodiment of the present application. Specifically, FIG. 11 illustrates add, merge, delete, and join operations on a set of identified knowledge types, where state n (n ∈ [1, 5]) represents the state of the corresponding set of knowledge types after different operations are performed, and the boxes and their internal elements characterize the set of knowledge types and their knowledge type elements, where 1, 2, 3, 4, 5, 6 represent different knowledge type elements.

As shown in fig. 11, if the user inputs a sentence "recommend a car greater than 30 ten thousand? ", a set of knowledge types may be initially derived by identifying the user input sentence, the set of knowledge types including: category 1 "car", comparand 2 "greater than", quantifier 3 "30 ten thousand" and quantifier 4 "one". A new attribute 5 "price" is obtained by implicit attribute recognition and added to the knowledge type set, which now corresponds to state 2. The 'more than' and '30 ten thousand' corresponding to the comparison word 2 and the quantitative word 3 are fused into a knowledge type restriction function 6 'more than 30 ten thousand', and at the moment, the knowledge type set corresponds to the state 3. And for the knowledge type of the number word 4 'one', the corresponding attributes of the comparison word and the limit cannot be found (the 'one' cannot modify the price because the prices of the automobiles are all more than 1 yuan), the number word 4 'one' is deleted, the knowledge type is ignored, and the knowledge type set corresponds to the state 4. And finally, connecting the isolated knowledge types in the knowledge type set based on the ontology, and generating a query graph based on the knowledge type set of the state 5 for query.

It should be understood that the foregoing examples are merely illustrative of the application of the method provided in the embodiments of the present application in a specific scenario, and do not limit the embodiments of the present application. The method provided by the embodiment of the application can also be used for realizing more different applications.

Referring to fig. 12, fig. 12 is a block diagram illustrating a knowledge-graph based data processing apparatus 1200 according to an embodiment of the present application. As will be explained below with respect to the block diagram shown in fig. 12, the knowledge-graph based data processing apparatus 1200 includes: an input acquisition module 1210, a type identification module 1220, a type update module 1230, a query graph construction module 1240, and a reply query module 1250, wherein:

an input obtaining module 1210, configured to obtain a user input sentence;

a type identification module 1220, configured to identify at least one knowledge type corresponding to the user input statement, so as to obtain a knowledge type set;

a type updating module 1230, configured to perform fusion processing on the knowledge type set, and update the knowledge type set;

a query graph construction module 1240, configured to construct a query graph based on the updated knowledge type set;

the answer query module 1250 is configured to query an answer corresponding to the query graph from a preset knowledge graph, and feed back the answer.

Further, the type update module 1230 includes: a type judgment submodule, a quantity judgment submodule and a set updating submodule, wherein:

the type judgment submodule is used for judging whether the specified knowledge type exists in the knowledge type set or not;

the quantity judgment submodule is used for judging whether a plurality of specified knowledge types exist or not if the specified knowledge types exist;

and the set updating submodule is used for fusing the plurality of specified knowledge types and updating the knowledge type set if the plurality of specified knowledge types exist.

Further, after determining whether there are multiple specified knowledge types if there are fusible knowledge types, the apparatus 1200 further includes: a type deletion module, wherein:

and the type deleting module is used for deleting the specified knowledge types if a plurality of specified knowledge types do not exist.

Further, the type deleting module comprises: a quantity acquisition submodule and a type deletion submodule, wherein:

the quantity obtaining sub-module is used for obtaining the quantity of the words corresponding to the specified knowledge types if the plurality of specified knowledge types do not exist;

and the type deleting submodule is used for deleting the specified knowledge type from the knowledge type set when the quantity is determined to be a specified numerical value.

Further, the query graph constructing module 1240 includes: a query graph generation sub-module, wherein:

and the query graph generation submodule is used for generating a query graph based on the preset knowledge type in the knowledge type set after the ontology of the knowledge graph is connected and updated.

Further, the reply query module 1250 includes: the device comprises a quantization processing sub-module, a candidate determining sub-module and a reply query sub-module, wherein:

the quantization processing sub-module is used for performing quantization processing on each query graph according to the knowledge type corresponding to each query graph and the word corresponding to the knowledge type to obtain a quantization value of each query graph if the number of the query graphs is multiple;

the candidate determining submodule is used for sequencing the query graphs from high to low according to quantization values and determining the query graphs with the number specified previously as candidate query graphs;

and the answer query submodule is used for querying the answer corresponding to the candidate query graph from a preset knowledge graph and feeding back the answer.

Further, the reply query submodule includes: optional confirm unit and optional inquiry unit, wherein:

the optional determining unit is used for inquiring the answer corresponding to the candidate query graph from a preset knowledge graph, and taking the candidate query graph inquired to the answer as an optional query graph;

and the optional query unit is used for determining the optional query graph with the highest quantization value as a target query graph and feeding back a reply corresponding to the target query graph if at least one optional query graph exists.

Further, after querying the answer corresponding to the candidate query graph from the preset knowledge graph and taking the candidate query graph queried to the answer as an optional query graph, the knowledge graph-based data processing apparatus 1200 further includes: a re-determination module, wherein:

and a redetermining module, configured to redetermine a new candidate query graph from the query graphs after the previously specified number if the optional query graph does not exist.

Further, the knowledge type set includes attributes of words, and the data processing apparatus 1200 based on knowledge-graph further includes: a first attribute obtaining module and a first attribute adding module, wherein:

a first attribute obtaining module, configured to obtain an attribute in the user input statement as a first attribute;

and the first attribute adding module is used for matching the first attribute with the attribute in the knowledge type set, adding the unmatched first attribute to the knowledge type set so as to update the knowledge type set, returning to execute the fusion processing of the knowledge type set, and updating the knowledge type set and the subsequent steps.

Further, the first attribute obtaining module includes: an intent recognition submodule, wherein:

and the intention identification submodule is used for carrying out intention identification on the user input sentence to obtain a first attribute.

Further, the first attribute obtaining module includes: an implicit attribute identification submodule, wherein:

and the hidden attribute recognition submodule is used for carrying out attribute recognition on the user input sentence based on the trained hidden attribute recognition model to obtain a first attribute.

Further, after querying the answer corresponding to the candidate query graph from the preset knowledge graph and taking the candidate query graph queried to the answer as an optional query graph, the knowledge graph-based data processing apparatus 1200 further includes: an attribute addition module, wherein:

and the attribute increasing module is used for returning and executing the attribute in the acquired user input statement as a first attribute and subsequent operation if the optional query graph does not exist.

Further, before the fusing the set of knowledge types and updating the set of knowledge types, the apparatus 1200 for processing data based on a knowledge graph further includes: entity detection module, history acquisition module and entity increase module, wherein:

the entity detection module is used for detecting whether an entity exists in the knowledge type set or not;

the history acquisition module is used for acquiring a knowledge type set of a history user input statement as a history knowledge type set if the history acquisition module does not exist, wherein the history user input statement is input before the user input statement;

and the entity adding module is used for adding the entities in the historical knowledge type set to the knowledge type set so as to take the entities of the historical user input sentences as the entities of the user input sentences.

The data processing device based on the knowledge graph provided in the embodiment of the application is used for implementing the corresponding data processing method based on the knowledge graph in the foregoing method embodiment, and has the beneficial effects of the corresponding method embodiment, which are not described herein again.

It can be clearly understood by those skilled in the art that the data processing device based on the knowledge graph provided in the embodiment of the present application can implement each process in the foregoing method embodiment, and for convenience and brevity of description, the specific working processes of the foregoing description device and module may refer to the corresponding processes in the foregoing method embodiment, and are not described herein again.

In the embodiments provided in the present application, the coupling or direct coupling or communication connection between the modules shown or discussed may be through some interfaces, and the indirect coupling or communication connection between the devices or modules may be in an electrical, mechanical or other form.

In addition, each functional module in the embodiments of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode.

Referring to fig. 13, a block diagram of an electronic device according to an embodiment of the present application is shown. The electronic device 1300 may be a smart phone, a tablet computer, an electronic book, or other electronic devices capable of running an application. The electronic device 1300 in the present application may include one or more of the following components: a processor 1310, a memory 1320, and one or more applications, wherein the one or more applications may be stored in the memory 1320 and configured to be executed by the one or more processors 1310, the one or more programs configured to perform a method as described in the aforementioned method embodiments.

Processor 1310 may include one or more processing cores. The processor 1310 interfaces with various interfaces and circuitry throughout the electronic device 1300 to perform various functions of the electronic device 1300 and process data by executing or executing instructions, programs, code sets, or instruction sets stored in the memory 1320, as well as invoking data stored in the memory 1320. Alternatively, the processor 1310 may be implemented in hardware using at least one of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor 1310 may integrate one or more of a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), a modem, and the like. Wherein, the CPU mainly processes an operating system, a user interface, an application program and the like; the GPU is used for rendering and drawing display content; the modem is used to handle wireless communications. It is to be understood that the modem may not be integrated into the processor 1310, but may be implemented by a communication chip.

The Memory 1320 may include a Random Access Memory (RAM) or a Read-Only Memory (Read-Only Memory). The memory 1320 may be used to store instructions, programs, code, sets of codes, or sets of instructions. The memory 1320 may include a stored program area and a stored data area, wherein the stored program area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing various method embodiments described below, and the like. The data storage area may also store data created during use by the electronic device 1300, such as phone books, audio-visual data, chat log data, and the like.

Referring to fig. 14, a block diagram of a computer-readable storage medium according to an embodiment of the present disclosure is shown. The computer-readable storage medium 1400 stores program code that can be called by a processor to execute the methods described in the above-described method embodiments.

The computer-readable storage medium 1400 may be an electronic memory such as a flash memory, an electrically-erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a hard disk, or a ROM. Optionally, the computer-readable storage medium 1400 includes a non-volatile computer-readable medium (non-transitory-readable storage medium). The computer readable storage medium 1400 has storage space for program code 1410 for performing any of the method steps described above. The program code can be read from or written to one or more computer program products. Program code 1410 may be compressed, for example, in a suitable form.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not necessarily depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims

1. A method of data processing based on a knowledge graph, the method comprising:

acquiring a user input statement;

identifying at least one knowledge type corresponding to the user input statement to obtain a knowledge type set;

performing fusion processing on the knowledge type set, and updating the knowledge type set;

constructing a query graph based on the updated knowledge type set;

and inquiring a response corresponding to the query graph from a preset knowledge graph, and feeding back the response.

2. The method according to claim 1, wherein said fusing the set of knowledge types and updating the set of knowledge types comprises:

judging whether the specified knowledge type exists in the knowledge type set or not;

if the specified knowledge types exist, judging whether the specified knowledge types exist in plurality or not;

and if a plurality of the specified knowledge types exist, fusing the specified knowledge types and updating the knowledge type set.

3. The method of claim 2, wherein after determining whether there are multiple specified knowledge types if there are fusible knowledge types, the method further comprises:

and if the plurality of specified knowledge types do not exist, deleting the specified knowledge types.

4. The method according to claim 3, wherein the deleting the specified knowledge type if there are no plurality of the specified knowledge types comprises:

if the plurality of the specified knowledge types do not exist, acquiring the number of words corresponding to the specified knowledge types;

and when the quantity is determined to be a designated numerical value, deleting the designated knowledge type from the knowledge type set.

5. The method of any of claims 1-4, wherein constructing a query graph based on the updated set of knowledge types comprises:

and generating a query graph based on the preset ontology of the knowledge graph and the updated knowledge types in the knowledge type set.

6. The method according to claim 1, wherein the querying a preset knowledge graph for a response corresponding to the query graph and feeding back the response comprises:

if the number of the query graphs is multiple, carrying out quantization processing on each query graph according to the knowledge type corresponding to each query graph and the word corresponding to the knowledge type to obtain a quantization value of each query graph;

sorting the query graphs from high to low according to quantization values, and determining the query graphs with the first specified number as candidate query graphs;

and inquiring the answer corresponding to the candidate query graph from a preset knowledge graph, and feeding back the answer.

7. The method according to claim 6, wherein the querying the answer corresponding to the candidate query graph from the preset knowledge graph and feeding back the answer comprises:

inquiring the answer corresponding to the candidate query graph from a preset knowledge graph, and taking the candidate query graph inquired to the answer as an optional query graph;

and if at least one optional query graph exists, determining the optional query graph with the highest quantization value as a target query graph, and feeding back a reply corresponding to the target query graph.

8. The method according to claim 6, wherein the querying the answer corresponding to the candidate query graph from the preset knowledge graph takes the candidate query graph queried to answer as an optional query graph, and the method further comprises:

if the alternative query graph does not exist, new candidate query graphs are determined again from the query graphs after the previously specified number.

9. The method of any of claims 1-8, wherein the set of knowledge types includes attributes of words, the method further comprising:

acquiring attributes in the user input statement as first attributes;

and matching the first attribute with the attribute in the knowledge type set, adding the unmatched first attribute to the knowledge type set to update the knowledge type set, returning to execute the fusion processing of the knowledge type set, updating the knowledge type set and the subsequent steps.

10. The method of claim 9, wherein the obtaining attributes in the user input sentence as first attributes comprises:

and identifying the intention of the user input sentence to obtain a first attribute.

11. The method of claim 9, wherein the obtaining attributes in the user input sentence as first attributes comprises:

and performing attribute recognition on the user input sentence based on the trained implicit attribute recognition model to obtain a first attribute.

12. The method according to claim 9, wherein the querying the answer corresponding to the candidate query graph from the preset knowledge graph takes the candidate query graph queried to answer as an optional query graph, and the method further comprises:

and if the optional query graph does not exist, returning and executing the attribute in the acquired user input statement as a first attribute and subsequent operation.

13. The method according to any one of claims 1-8, wherein before performing the fusion process on the set of knowledge types and updating the set of knowledge types, the method further comprises:

detecting whether an entity exists in the knowledge type set;

if not, acquiring a knowledge type set of a historical user input statement as a historical knowledge type set, wherein the historical user input statement is input before the user input statement;

adding an entity in the set of historical knowledge types to the set of knowledge types to treat the entity of the historical user input statement as the entity of the user input statement.

14. A data processing apparatus based on a knowledge-graph, the apparatus comprising:

the input acquisition module is used for acquiring input sentences of a user;

the type identification module is used for identifying at least one knowledge type corresponding to the user input statement to obtain a knowledge type set;

the type updating module is used for carrying out fusion processing on the knowledge type set and updating the knowledge type set;

the query graph building module is used for building a query graph based on the updated knowledge type set;

and the answer query module is used for querying an answer corresponding to the query graph from a preset knowledge graph and feeding back the answer.

15. An electronic device, comprising:

one or more processors;

a memory;

one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more programs configured to perform the method of any of claims 1-13.

16. A computer-readable storage medium having program code stored therein, the program code being invoked by a processor to perform the method of any of claims 1-13.