CN117435749A - Method, device and storage medium for generating knowledge graph - Google Patents
Method, device and storage medium for generating knowledge graph Download PDFInfo
- Publication number
- CN117435749A CN117435749A CN202311767192.4A CN202311767192A CN117435749A CN 117435749 A CN117435749 A CN 117435749A CN 202311767192 A CN202311767192 A CN 202311767192A CN 117435749 A CN117435749 A CN 117435749A
- Authority
- CN
- China
- Prior art keywords
- ontology
- level
- knowledge graph
- knowledge
- graph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000003860 storage Methods 0.000 title claims abstract description 7
- 238000010276 construction Methods 0.000 claims abstract description 86
- 239000013598 vector Substances 0.000 claims abstract description 59
- 238000012545 processing Methods 0.000 claims description 8
- 238000012217 deletion Methods 0.000 claims description 6
- 230000037430 deletion Effects 0.000 claims description 6
- 238000012163 sequencing technique Methods 0.000 claims description 6
- 238000000638 solvent extraction Methods 0.000 claims description 2
- 238000003745 diagnosis Methods 0.000 description 16
- 230000008569 process Effects 0.000 description 16
- 239000003795 chemical substances by application Substances 0.000 description 5
- 230000009471 action Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 239000000446 fuel Substances 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 230000002452 interceptive effect Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 238000009960 carding Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012552 review Methods 0.000 description 2
- 108020001568 subdomains Proteins 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000004378 air conditioning Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 239000000725 suspension Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
- G06N5/022—Knowledge engineering; Knowledge acquisition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Human Computer Interaction (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a method, a device and a storage medium for generating a knowledge graph. The method comprises the following steps: receiving a target field keyword, and generating first-level ontology construction prompt information; transmitting the knowledge graph to an ontology construction model to generate a first-level knowledge graph ontology; expanding the previous level ontology according to the expandability sequence to generate a current level knowledge graph ontology; repeatedly executing the body expansion until the number of predefined levels is reached; creating a question generation prompt message according to the knowledge graph body, transmitting the question generation prompt message to a question generation model, and generating an inquiry statement; vectorizing the query statement and the domain document text; retrieving field document text vectors similar to the query sentence vectors, and combining the field document text vectors into a retrieval context; creating answer generation prompt information according to the query statement and the search context, transmitting the answer generation prompt information to an answer generation model, and generating an answer; and generating a knowledge graph according to the entity and the attribute of the answer and the ontology. The invention improves the efficiency and quality of knowledge graph construction.
Description
Technical Field
Embodiments of the present invention relate generally to knowledge-graph construction in a specific field, and more particularly, to a method, apparatus, and storage medium for generating a knowledge-graph.
Background
In the current artificial intelligence field, a knowledge graph is widely applied to various complex problem solving and decision support scenes as an important knowledge organization tool. The knowledge graph provides an intuitive and efficient knowledge acquisition and reasoning mode by graphically representing the entities and their relationships. However, the construction process of the knowledge graph generally requires a great deal of manpower investment, including links of knowledge extraction, labeling of entities and relations, verification of knowledge and the like in the professional field, and has high cost.
Although large language models exhibit strong capabilities in terms of natural language understanding and generation, the large language models have difficulty learning deep expertise during training because their knowledge cannot be updated and the expertise and vertical domain knowledge is very small in large corpora.
Therefore, in a specific application field, such as construction of knowledge graphs of user manuals of different automobile types, the knowledge graphs still need to be relied on.
On the other hand, neural networks face challenges in addressing the facts accuracy problem. In order to provide deterministic factual information, an accurate knowledge base must be relied upon. Therefore, in this case, the presence of the knowledge graph is particularly important.
Disclosure of Invention
In order to solve the above-mentioned problems in the prior art, in a first aspect, an embodiment of the present invention provides a method for generating a knowledge-graph, the method including the steps of:
s101, receiving target domain keywords input by a user as first-level keywords;
s102, generating first-level ontology construction prompt information according to the first-level keywords, wherein the first-level ontology construction prompt information comprises an instruction for constructing a knowledge graph ontology based on the first-level keywords, pre-specified source definition information and a first-level knowledge graph ontology format, and the first-level knowledge graph ontology format limits knowledge graph ontology entries in the first-level knowledge graph ontology to include entities and attributes;
s103, transmitting the first-level ontology construction prompt information to an ontology construction model so that a first-level knowledge graph ontology is generated by the ontology construction model according to the first-level ontology construction prompt information, and the first-level knowledge graph ontology is stored in an ontology library;
s104, generating entity expansion allocation prompt information according to a previous-level knowledge graph body stored in the ontology base, wherein the entity expansion allocation prompt information comprises the previous-level knowledge graph body and an instruction for ordering the expandability of the attributes of the knowledge graph body items in the previous-level knowledge graph body;
S105, transmitting the entity expansion allocation prompt information to an entity expansion allocation model so that the entity expansion allocation model generates the expandability sequencing of the attribute of the knowledge graph body of the previous level according to the entity expansion allocation prompt information;
s106, sequentially executing the following processing on the knowledge graph body items in the knowledge graph body of the previous level according to the order of attribute scalability from high to low in the scalability sequencing: generating a current level keyword corresponding to the knowledge graph body item according to the entity and attribute of the knowledge graph body item and the previous level keyword;
s107, generating current level ontology construction prompt information according to the current level keyword, wherein the current level ontology construction prompt information comprises an instruction for constructing a current level knowledge graph ontology based on the current level keyword, pre-designated source definition information and a current level knowledge graph ontology format, wherein the current level knowledge graph ontology format limits knowledge graph ontology entries in the current level knowledge graph ontology to include entities and attributes, and designates that the entities of the current level knowledge graph ontology entries are attributes of corresponding previous level knowledge graph ontology entries;
S108, transmitting the current-level ontology construction prompt information to the ontology construction model so as to generate a current-level knowledge graph ontology according to the current-level ontology construction prompt information by the ontology construction model, and storing the current-level knowledge graph ontology in the ontology library, wherein the entity of the current-level knowledge graph ontology entry is stored in association with the attribute of the corresponding previous-level knowledge graph ontology entry;
s109, judging whether the number of the levels of the generated knowledge graph body reaches a predefined level number threshold value, and if so, turning to a step S110; if not, go to step S104;
s110, creating problem generation prompt information according to the knowledge graph ontology stored in the ontology library, and transmitting the problem generation prompt information to a problem generation model so as to generate query sentences by the problem generation model for the entity and the attribute of each knowledge graph ontology item in the stored knowledge graph ontology;
s111, vectorizing the query statement to generate a query statement vector;
s112, partitioning a pre-stored domain document text to generate a domain document text block, vectorizing the domain document text block to generate a domain document text vector, and storing the domain document text vector in a vector library;
S113, retrieving a domain document text vector similar to the query sentence vector in the vector library, and combining domain document text blocks corresponding to the similar domain document text vector into a retrieval context;
s114, creating answer generation prompt information according to the query statement and the search context, wherein the answer generation prompt information comprises an instruction for searching the answer of the query statement in the search context;
s115, transmitting the answer generation prompt information to an answer generation model so that the answer generation model generates an answer of the query statement according to the answer generation prompt information;
s116, generating a knowledge graph item of a knowledge graph according to the answer and the entity and the attribute of the knowledge graph body item corresponding to the query statement, wherein the knowledge graph item comprises an entity, an attribute and an attribute value, and the entity and the attribute of the knowledge graph item are the entity and the attribute of the corresponding knowledge graph body item respectively, and the attribute value of the knowledge graph item is the answer.
In some embodiments, the method comprises: after all the knowledge-graph ontology entries to be expanded in the knowledge-graph ontology of the previous level are executed in steps S106-S108, the knowledge-graph entries to be expanded in the knowledge-graph ontology of the current level are executed again.
In some embodiments, the method comprises: after steps S106-S108 are performed on a particular knowledge-graph ontology entry in a previous-level knowledge-graph ontology, steps S106-S108 are performed on a current-level knowledge-graph ontology corresponding to the particular knowledge-graph ontology entry until a predefined number of levels threshold is reached, and steps S106-S108 are performed on a next knowledge-graph ontology entry of the particular knowledge-graph ontology entry in the previous-level knowledge-graph ontology.
In some embodiments, step S106 further comprises: taking a preset expansion number of knowledge graph ontology entries which are in front in the order as knowledge graph ontology entries to be expanded according to the order of attribute expandability from high to low, and executing the following processing on the knowledge graph ontology entries to be expanded: and generating a current level keyword corresponding to the knowledge graph body entry according to the entity and attribute of the knowledge graph body entry and the previous level keyword.
In some embodiments, the predetermined number of extensions is determined according to the number of layers of the current hierarchy.
In some embodiments, the predefined number of levels threshold is fixed or determined according to an order of attribute extensibility ordering of corresponding knowledge-graph entries in a specified level.
In some embodiments, the method further comprises: displaying the body library; receiving an operation instruction of a user for selecting a knowledge graph ontology entry in the ontology library, wherein the operation instruction comprises deletion, editing and addition; executing corresponding operation on the selected knowledge-graph body entry according to the operation instruction; and/or displaying the knowledge graph; receiving an operation instruction of a user for selecting a knowledge graph item in the knowledge graph, wherein the operation instruction comprises deletion, editing and addition; and executing corresponding operation on the selected knowledge graph entry according to the operation instruction.
In some embodiments, the pre-specified source definition information includes one or more of a database of the ontology-build model itself, a specified database address, a specified search engine.
In a second aspect, an embodiment of the present invention proposes an apparatus for generating a knowledge-graph, the apparatus comprising: the target domain keyword receiving module is configured to receive target domain keywords input by a user and serve as first-level keywords; a first-level ontology construction hint information generation module configured to generate first-level ontology construction hint information according to the first-level keyword, wherein the first-level ontology construction hint information includes an instruction to construct a knowledge-graph ontology based on the first-level keyword, pre-specified source definition information, and a first-level knowledge-graph ontology format, wherein the first-level knowledge-graph ontology format defines knowledge-graph ontology entries in a first-level knowledge-graph ontology as including entities and attributes; a first-level knowledge graph body generating module configured to transmit the first-level ontology construction prompt information to an ontology construction model so as to generate a first-level knowledge graph body by the ontology construction model according to the first-level ontology construction prompt information, and store the first-level knowledge graph body in a ontology library; the entity expansion allocation prompt information generation module is configured to generate entity expansion allocation prompt information according to a previous-level knowledge graph body stored in the ontology base, wherein the entity expansion allocation prompt information comprises the previous-level knowledge graph body and an instruction for ordering expandability of attributes of knowledge graph body items in the previous-level knowledge graph body; the expandability ranking generation module is configured to transmit the entity expansion allocation prompt information to an entity expansion allocation model so that the entity expansion allocation model generates expandability ranking of the attribute of the previous-level knowledge graph body according to the entity expansion allocation prompt information; the current-level keyword generation module is configured to sequentially execute the following processing on the knowledge graph body items in the knowledge graph body of the previous level according to the order of attribute scalability from high to low in the scalability sequencing: generating a current level keyword corresponding to the knowledge graph body item according to the entity and attribute of the knowledge graph body item and the previous level keyword; a current level ontology construction hint information generation module configured to generate a current level ontology construction hint information according to the current level keyword, wherein the current level ontology construction hint information includes an instruction to construct a current level knowledge graph ontology based on the current level keyword, pre-specified source definition information, and a current level knowledge graph ontology format, wherein the current level knowledge graph ontology format defines a knowledge graph ontology entry in the current level knowledge graph ontology as including an entity and an attribute, and specifies that the entity of the current level knowledge graph ontology entry is an attribute of a corresponding previous level knowledge graph ontology entry; a current-level knowledge-graph ontology generating module configured to transmit the current-level ontology construction prompt information to the ontology construction model, so that a current-level knowledge-graph ontology is generated by the ontology construction model according to the current-level ontology construction prompt information, and the current-level knowledge-graph ontology is stored in the ontology library, wherein an entity of the current-level knowledge-graph ontology entry is stored in association with an attribute of a corresponding previous-level knowledge-graph ontology entry; the hierarchy number judging module is configured to judge whether the hierarchy number of the generated knowledge graph body reaches a predefined hierarchy number threshold value, and if so, the query sentence generating module is switched to; if not, turning to an entity expansion allocation prompt information generation module; an inquiry sentence generation module configured to create a question generation hint information from the knowledge graph ontology stored in the ontology base and transmit the question generation hint information to a question generation model so that an inquiry sentence is generated by the question generation model for the entity and attribute of each of the knowledge graph ontology entries in the stored knowledge graph ontology; the query sentence vectorization module is configured to vectorize the query sentence to generate a query sentence vector; the field document text vectorization module is configured to block a pre-stored field document text, generate a field document text block, vectorize the field document text block, generate a field document text vector, and store the field document text vector in a vector library; a search context generation module configured to search the vector library for a domain document text vector similar to the query sentence vector, and combine domain document text blocks corresponding to the similar domain document text vector into a search context; an answer generation prompt creation module configured to create answer generation prompt according to the query sentence and the search context, wherein the answer generation prompt includes an instruction to search an answer of the query sentence in the search context; the answer generation module is configured to transmit the answer generation prompt information to an answer generation model so that the answer generation model generates an answer of the query statement according to the answer generation prompt information; and a knowledge graph generation module configured to generate a knowledge graph entry of a knowledge graph according to the answer and the entity and attribute of the knowledge graph ontology entry corresponding to the query statement, wherein the knowledge graph entry comprises an entity, an attribute and an attribute value, and wherein the entity and the attribute of the knowledge graph entry are the entity and the attribute of the corresponding knowledge graph ontology entry, respectively, and the attribute value of the knowledge graph entry is the answer.
In a third aspect, embodiments of the invention provide a storage medium storing computer readable instructions which, when executed by a processor, perform a method according to any of the embodiments described above.
The embodiment of the invention provides a scheme for constructing a knowledge graph by multi-agent cooperation based on a large language model, and aims to automate and automate the knowledge graph construction process, reduce the participation of manpower as much as possible and reduce the labor cost. Through AI agent based on big language model, can be with environment interactive execution long-term and complicated task to playing different roles and cooperate each other, thereby realize the automated construction of knowledge graph.
The knowledge graph construction process is automated and a plurality of models are utilized to play different roles, so that remarkable efficiency improvement is brought. This improvement is mainly manifested in the following two aspects:
1. automation of the knowledge graph body construction process: the different roles of the various models and the mutual cooperation make up for the problem of insufficient knowledge coverage caused by relying on experts in a single field. This combination makes the ontology-build process more efficient and accurate.
2. Automating the completion process of the knowledge graph entity: the traditional knowledge graph construction process generally needs to manually review a large amount of document data to carry out complicated manual carding, so that a large amount of manpower and material resources are consumed, and the efficiency is low. Valuable information can be extracted from a large amount of unstructured data through automatic construction, and a knowledge graph can be quickly and accurately complemented. The construction speed is greatly improved, and the quality and the integrity of the knowledge graph are improved.
Drawings
The above, as well as additional purposes, features, and advantages of embodiments of the present invention will become apparent in the following detailed written description and claims upon reference to the accompanying drawings. Several embodiments of the present invention are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which:
FIG. 1 shows a flow chart of a method of generating a knowledge-graph, in accordance with an embodiment of the invention;
FIG. 2 shows a data flow diagram of a method of generating a knowledge-graph, in accordance with an embodiment of the invention;
fig. 3 shows a block diagram of an apparatus for generating a knowledge-graph, according to an embodiment of the invention.
In the drawings, the same or corresponding reference numerals indicate the same or corresponding parts.
Detailed Description
The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are presented merely to enable those skilled in the art to better understand and practice the invention and are not intended to limit the scope of the invention in any way.
In one aspect, embodiments of the present invention provide a method of generating a knowledge-graph. Referring to fig. 1-2, fig. 1 shows a flowchart of a method 100 of generating a knowledge-graph according to an embodiment of the present invention, and fig. 2 shows a data flow diagram of a method of generating a knowledge-graph according to an embodiment of the present invention.
As shown in fig. 1, the method 100 includes steps S101-S116.
In step S101, a target domain keyword input by a user is received as a first-level keyword.
In step S102, first-level ontology construction hint information is generated from the first-level keywords. Wherein the first-level ontology construction hint information includes instructions to construct a knowledge-graph ontology based on the first-level keywords, the pre-specified source definition information, and the first-level knowledge-graph ontology format. Wherein the first level knowledge-graph ontology format defines knowledge-graph ontology entries in the first level knowledge-graph ontology as comprising entities and attributes, i.e. each knowledge-graph ontology entry is a binary group comprising an entity and an attribute.
As one embodiment of the present invention, the pre-specified source definition information includes one or more of a database of the ontology-build model itself, a specified database address, and a specified search engine.
In step S103, the first-level ontology-build hint information is transmitted to the ontology-build model so that a first-level knowledge-graph ontology is generated by the ontology-build model from the first-level ontology-build hint information, and the first-level knowledge-graph ontology is stored in the ontology library.
In step S104, entity expansion allocation hint information is generated according to the previous-level knowledge graph body stored in the ontology base, wherein the entity expansion allocation hint information includes the previous-level knowledge graph body and an instruction for ordering the expandability of the attributes of the knowledge graph body items in the previous-level knowledge graph body.
In step S105, the entity extension allocation hint information is transmitted to the entity extension allocation model, so that the entity extension allocation model generates the scalability rank of the attribute of the previous-level knowledge graph body according to the entity extension allocation hint information.
In step S106, the following processes are sequentially performed on the knowledge-graph ontology entries in the knowledge-graph ontology of the previous hierarchy in the order of high-to-low attribute scalability in the scalability ranking: and generating a current level keyword corresponding to the knowledge graph body entry according to the entity and attribute of the knowledge graph body entry and the previous level keyword. In other words, the attribute of the knowledge-graph ontology to be expanded may be added to the keywords as a new target field.
As an embodiment of the present invention, all the knowledge-graph body entries in the previous hierarchy may be used as the knowledge-graph body entries to be expanded, and the expansion may be performed in order of high scalability.
As another embodiment of the present invention, attribute scalability may be quantized into a numeric value, and a knowledge-graph ontology entry whose numeric value exceeds a predetermined scalability threshold value is used as a knowledge-graph ontology entry to be expanded.
As still another embodiment of the present invention, a predetermined expansion number of knowledge-graph body entries preceding in the order may be fetched as the knowledge-graph body entries to be expanded in the order in which the attribute scalability is from high to low.
In the above embodiment, for example, the predetermined expansion number may be determined according to the number of layers of the current hierarchy.
In practical application, any implementation mode for determining the ontology entries of the knowledge graph to be expanded can be adopted according to requirements.
In step S107, current-level ontology construction hint information is generated according to the current-level keywords, wherein the current-level ontology construction hint information includes an instruction to construct a current-level knowledge-graph ontology based on the current-level keywords, pre-specified source definition information, and a current-level knowledge-graph ontology format, wherein the current-level knowledge-graph ontology format defines knowledge-graph ontology entries in the current-level knowledge-graph ontology as including entities and attributes, and specifies that the entities of the current-level knowledge-graph ontology entries are attributes of corresponding previous-level knowledge-graph ontology entries.
In step S108, the current-level ontology-build hint information is transmitted to the ontology-build model, so that a current-level knowledge-graph ontology is generated by the ontology-build model according to the current-level ontology-build hint information, and the current-level knowledge-graph ontology is stored in an ontology library, wherein entities of the current-level knowledge-graph ontology entries are stored in association with attributes of the corresponding previous-level knowledge-graph ontology entries.
As an embodiment of the present invention, after steps S106 to S108 are performed on all the knowledge-graph body entries to be expanded in the knowledge-graph body of the previous level, steps S106 to S108 may be performed on the knowledge-graph entries to be expanded in the knowledge-graph body of the current level.
As another embodiment of the present invention, after steps S106 to S108 are performed on a specific knowledge-graph body entry in a previous-level knowledge-graph body, steps S106 to S108 may be performed on a current-level knowledge-graph body corresponding to the specific knowledge-graph body entry until a predefined number of levels threshold is reached, and then steps S106 to S108 may be performed on a next-knowledge-graph body entry of the specific knowledge-graph body entry in the previous-level knowledge-graph body.
In other words, the above two embodiments describe the use of breadth-first and depth-first policies, respectively, in an iterative process. In practical application, any one or combination of the two can be adopted according to the requirements. For example, a second-level knowledge-graph ontology may be first generated from a first-level knowledge-graph ontology based on a breadth-first policy, and then expanded for the second-level knowledge-graph ontology based on a depth-first policy until a pre-specified level is reached. For another example, ontology expansion may be performed first from a first level knowledge graph ontology to a specified level based on depth-first policies, and then from the specified level on an ontology expansion based on breadth-first policies.
In step S109, it is determined whether the number of levels of the generated knowledge-graph body all reach a predefined level number threshold, and if so, the process goes to step S110; if not, go to step S104.
As an embodiment of the present invention, the predefined number of levels threshold may be a fixed value, i.e. the same number of levels is extended for all knowledge-graph ontologies.
As another embodiment of the present invention, the predefined number of levels threshold may be determined according to an order of attribute extensibility ordering of corresponding knowledge-graph entries in the specified level. For example, in a first-level knowledge graph ontology, the scalability rank is in the first N bits, its level number threshold is set to a first threshold, the scalability rank is in the next N bits, its level number threshold is set to a second threshold, where the first threshold is greater than the second threshold. Thus, more levels can be extended for more highly extensible ontology entries, and less levels can be extended for less extensible ontology entries.
In step S110, a question generation hint is created from the knowledge graph ontology stored in the ontology library, and is transmitted to the question generation model, so that an inquiry sentence (query) is generated by the question generation model for the entity and attribute of each of the stored knowledge graph ontology entries. The query sentence is, for example, a question in a natural language form.
In step S111, the query term vector is generated by vectorizing the query term.
In step S112, the pre-stored domain document text is segmented to generate domain document text blocks, and the domain document text blocks are vectorized to generate domain document text vectors, which are stored in a vector library.
The pre-stored domain document text is, for example, a document inside an enterprise, an enterprise product related document, private data, etc., and is generally knowledge not possessed by a general language model.
In step S113, a domain document text vector similar to the query sentence vector is retrieved in the vector library, and domain document text blocks corresponding to the similar domain document text vector are combined as a retrieval context.
For example, the domain document text vector similar to the query term vector may be a domain document text vector having a similarity at top k, or a domain document text vector having a similarity exceeding a predetermined similarity threshold.
In step S114, answer generation hint information is created from the query sentence and the search context, wherein the answer generation hint information includes an instruction to search for an answer to the query sentence in the search context.
In step S115, the answer generation hint information is transmitted to the answer generation model so that an answer of the query sentence is generated by the answer generation model according to the answer generation hint information.
In step S116, a knowledge-graph entry of the knowledge-graph is generated according to the entity and the attribute of the knowledge-graph ontology entry corresponding to the query sentence, wherein the knowledge-graph entry includes an entity, an attribute, and an attribute value, and wherein the entity and the attribute of the knowledge-graph entry are the entity and the attribute of the corresponding knowledge-graph ontology entry, respectively, and the attribute value of the knowledge-graph entry is the answer. That is, each knowledge graph entry is a triplet including entities, attributes, attribute values, combined from previously generated entities, attributes, and currently generated answers to the ontology entries.
As shown in fig. 2, the model according to the embodiment of the present invention includes: the ontology construction model, the entity extension allocation model, the difficult problem generation model and the answer generation model. The four models can all adopt a large language model as an original model, play different roles in the process of generating a knowledge graph after training, and realize different tasks.
For ease of understanding, the functions that can be implemented by these four models are described below as examples.
The ontology-build model may generate task-related language reasoning tracks and actions based on the ReAct framework by prompting the language model, e.g., invoking a search engine api, or invoking an original model's own database, which enables the model to dynamically reason to create, maintain, and adjust a high-level plan of actions, while also interacting with external environments (e.g., search engines), incorporating additional information into the reasoning, building a domain-specific underlying concept, attribute, and relationship.
The entity extension allocation model may be based on understanding the concepts of entities, attributes and relationships in the ontology base, deciding and allocating the most extensible entities and sub-domains.
The problem generation model may convert entity relationships or attributes in the ontology library into query sentences, such as in natural language.
The answer generation model can search and extract answers of summarized questions based on document data in the professional field, and fill a knowledge graph to form a final triplet.
Referring to fig. 2, a user may issue an operation instruction to the ontology library and the knowledge graph, respectively.
As an embodiment of the present invention, the method may further include: displaying a body library; receiving an operation instruction of a user for selecting a knowledge graph ontology entry in an ontology library, wherein the operation instruction comprises deletion, editing and addition; and executing corresponding operation on the selected knowledge-graph body entry according to the operation instruction.
As an embodiment of the present invention, the method may further include: displaying the knowledge graph; receiving an operation instruction of a user for selecting a knowledge graph item in the knowledge graph, wherein the operation instruction comprises deletion, editing and addition; and executing corresponding operation on the selected knowledge graph entry according to the operation instruction.
The user who sends the operation instruction can be, for example, a domain knowledge expert, and the generated ontology library and the generated knowledge graph are manually interfered through the operation instruction to perform quality inspection and optimization on data in the ontology library and the knowledge graph.
The method for generating the knowledge graph provided by the embodiment of the invention automatically completes the construction method of the knowledge graph in a multi-model cooperation mode. The strong capability of a large language model is utilized, and the intelligent agents of multiple roles are used for interactive collaboration, so that the automatic construction of the knowledge graph is realized. The method can automatically construct basic concepts, attributes and relations in a specific field. By means of the reasoning capability of the large language model, the professional domain knowledge is automatically acquired by calling the search engine, and therefore accurate definition of the ontology concept relationship is assisted. The limitation of pure manual knowledge acquisition is overcome, and the efficiency and quality of ontology construction are improved. The decision and allocation can be automated for iterating over the entities and sub-domains with the highest scalability. The ontology library can be rapidly and efficiently expanded, and the rapid growth of the knowledge graph is realized. The intelligent question-answering system fills in triples of the knowledge graph. Based on a vector retrieval mode, the knowledge graph can be automatically filled and complemented by combining reading understanding and answer summarizing capabilities of the large model, and automatic updating and maintenance of the knowledge graph are realized.
A specific example of a method of generating a knowledge-graph according to an embodiment of the present invention is described below.
Two key information input by a user are received into a React framework: task goals and action types. For example, the body of an automobile user manual knowledge graph needs to be constructed, and the following configuration can be performed:
action type: search engine API
Task goal: building a knowledge graph body of a knowledge graph of an automobile user manual, and outputting (< entity >, < attribute >) in the form of a binary group
The ontology-build model produces an ontology result in the form of the following two tuples: (< car >, < manufacturer >); (< car >, < model >); (< car >, < year of production >); (< car >, < fuel type >); (< car >, < horsepower >); (< car >, < torque >); (< car >, < fault diagnosis >); … ….
And the entity with highest expandability is decided by the entity expansion distribution model.
1) And configuring rules for expandability judgment in the service field comprising the entity and the attribute. For example:
prompting of priority decision rules: (< car >, < manufacturer >); (< car >, < model >); (< car >, < year of production >); (< car >, < fuel type >); (< car >, < horsepower >); (< car >, < torque >); (< car >, < fault diagnosis >).
By way of example only, where only a certain vehicle model is considered, the automobile user manual is regarded as a domain document for which the number of extensible subclasses is ordered from large to small.
Model output: (< car >, < fault diagnosis >); (< car >, < horsepower >); (< car >, < torque >); (< car >, < fuel type >); (< car >, < year of production >); (< car >, < model >); (< car >, < manufacturer >); … ….
The entity with the highest priority is transferred to the ontology building model according to a greedy strategy (namely, the entity with the highest priority is selected and ordered at each time or called the entity with the highest priority), and is removed from a priority list, and the ontology building steps are repeatedly executed, so that the extension of the sub-class ontology can be completed: (< failure diagnosis >, < engine failure >); (< failure diagnosis >, < brake system failure >); (< fault diagnosis >, < electrical system fault >); (< failure diagnosis >, < exhaust system failure >); (< failure diagnosis >, < cooling system failure >); (< failure diagnosis >, < fuel system failure >); (< failure diagnosis >, < driveline failure >); (< failure diagnosis >, < suspension system failure >); (< failure diagnosis >, < air conditioning and heating system failure >); (< failure diagnosis >, < body/appearance failure >); (< failure diagnosis >, < airbag system failure >); (< trouble diagnosis >, < navigation/entertainment system trouble >); (< failure diagnosis >, < tire failure >); … ….
After multiple iterations, the ontology can be expanded to a certain scale. Optionally, the quality inspection and optimization of the data in the ontology library may be performed by manual intervention by domain knowledge specialists.
By means of the question generation model, the entity relationship or attribute is converted into query sentences, such as question sentences in natural language form, for example:
prompt "(< automobile >, < manufacturer >) converts question" in knowledge-graph questions and answers "
And then outputting a model: "who is the manufacturer of the car? "; "who made the car? "; "what company the car is made? "; "which manufacturer's hands this car comes from? "; "which manufacturer produced the car? "; … ….
The automatic completion and filling of the knowledge graph are completed through a question-answering system based on retrieval, wherein the following key implementation steps are involved:
(1) The method comprises the steps of setting the maximum length of each block for ensuring the semantic integrity of a segmented text, and performing recursive segmentation according to the following segmentation symbol priority as a cutting point to ensure that the maximum segmentation is maximized in a maximum length limit class.
[";", "；", "。", "！", "!", "?", "\n\n", "\n", " ", ""]
(2) And carrying out the vector quantization processing of the ebedding on each block, and storing the vector into a vector library. The emmbedding algorithm may be, for example, model calculation emmbedding trained based on the BERT architecture.
(3) The problem generated in step three was vectorized using the same emmbedding model as in (2).
(4) And (3) calculating the vector similarity of the step (3) and the step (2), and selecting a block of the similarity top k as the context of question-answering retrieval.
(5) And packaging the questions and the contexts as prompts and transmitting the prompts and the contexts to an answer generation model, wherein the prompt templates are as follows:
{context}
-----------------------------------
please give the answer of "{ query }". The answer content can only come from the above text, and the answer is not to be built. Please output strictly in the following format:
-----------------------------------
if it is unable to answer, output "N| is unaware of"
If answer is possible, output "Y| < answer >".
Where { context } is context and { query } is question.
(6) Automatically filling the output result of the model, for example, an original document of an automobile user manual describes a model of an automobile of the masses, and generating the following triples: < car >, < manufacturer >, < masses for a car >.
In another aspect, an embodiment of the present invention provides an apparatus for generating a knowledge-graph. Referring to fig. 3, a block diagram of an apparatus for generating a knowledge-graph according to an embodiment of the present invention is shown. The apparatus comprises modules 301-316.
The target domain keyword receiving module 301 may be configured to receive a target domain keyword input by a user as a first-level keyword.
The first-level ontology construction hint information generation module 302 may be configured to generate first-level ontology construction hint information from the first-level keywords, wherein the first-level ontology construction hint information includes instructions to construct a knowledge-graph ontology based on the first-level keywords, pre-specified source definition information, and a first-level knowledge-graph ontology format, wherein the first-level knowledge-graph ontology format defines knowledge-graph ontology entries in the first-level knowledge-graph ontology as including entities and attributes.
The first-level knowledge-graph ontology generating module 303 may be configured to transmit first-level ontology-build hints to the ontology-build model, so as to generate a first-level knowledge-graph ontology from the first-level ontology-build hints by the ontology-build model, and store the first-level knowledge-graph ontology in the ontology library.
The entity extension allocation hint information generation module 304 may be configured to generate entity extension allocation hint information according to a previous-level knowledge-graph ontology stored in an ontology base, where the entity extension allocation hint information includes the previous-level knowledge-graph ontology and an instruction to rank expandability of attributes of knowledge-graph ontology entries in the previous-level knowledge-graph ontology.
The extensibility rank generation module 305 may be configured to transmit the entity extension assignment hint information to the entity extension assignment model so that an extensibility rank of the attribute of the previous-level knowledge-graph body is generated by the entity extension assignment model from the entity extension assignment hint information.
The current-level keyword generation module 306 may be configured to sequentially perform the following processing on the knowledge-graph ontology entries in the knowledge-graph ontology of the previous level in order of high-to-low attribute scalability in the scalability ranking: and generating a current level keyword corresponding to the knowledge graph body entry according to the entity and attribute of the knowledge graph body entry and the previous level keyword.
The current-level ontology construction hint information generation module 307 may be configured to generate a current-level ontology construction hint information from the current-level keywords, wherein the current-level ontology construction hint information includes instructions to construct a current-level knowledge-graph ontology based on the current-level keywords, pre-specified source definition information, and a current-level knowledge-graph ontology format, wherein the current-level knowledge-graph ontology format defines knowledge-graph ontology entries in the current-level knowledge-graph ontology as including entities and attributes, and specifies that the entities of the current-level knowledge-graph ontology entries are attributes of corresponding previous-level knowledge-graph ontology entries.
The current-level knowledge-graph ontology generating module 308 may be configured to transmit current-level ontology-build hints to the ontology-build model to generate a current-level knowledge-graph ontology from the current-level ontology-build hints by the ontology-build model, and store the current-level knowledge-graph ontology in an ontology library, wherein entities of the current-level knowledge-graph ontology entries are stored in association with attributes of corresponding previous-level knowledge-graph ontology entries.
The number of levels determination module 309 may be configured to determine whether the number of levels of the generated knowledge-graph ontology all reach a predefined number of levels threshold, and if so, go to the query statement generation module; if not, turning to an entity expansion allocation prompt information generation module.
The query sentence generation module 310 may be configured to create a question generation hint information from the knowledge-graph ontology stored in the ontology library and transmit the question generation hint information to the question generation model to generate a query sentence by the question generation model for the entity and attribute of each of the knowledge-graph ontology entries in the stored knowledge-graph ontology.
The query term vectorization module 311 may be configured to vectorize query terms to generate query term vectors.
The domain document text vectorization module 312 may be configured to block pre-stored domain document text, generate domain document text blocks, and vectorize the domain document text blocks, generate domain document text vectors, and store the domain document text vectors in a vector library.
The search context generation module 313 may be configured to search a vector library for a domain document text vector that is similar to the query term vector, and combine domain document text blocks corresponding to the similar domain document text vector into a search context.
The answer generation hint information creation module 314 may be configured to create answer generation hint information from the query statement and the search context, wherein the answer generation hint information includes instructions to search for an answer to the query statement in the search context.
The answer generation module 315 may be configured to transmit the answer generation hint information to the answer generation model so that an answer to the query statement is generated by the answer generation model based on the answer generation hint information.
The knowledge-graph generation module 316 may be configured to generate a knowledge-graph entry of the knowledge-graph from the answer and the entity and attribute of the knowledge-graph ontology entry corresponding to the query statement, wherein the knowledge-graph entry includes an entity, an attribute, and an attribute value, and wherein the entity and the attribute of the knowledge-graph entry are the entity and the attribute of the corresponding knowledge-graph ontology entry, respectively, and the attribute value of the knowledge-graph entry is the answer.
It should be noted that, the functions implemented by each module in the device for generating a knowledge graph according to the embodiment of the present invention correspond to each step of the method for generating a knowledge graph described above, and specific embodiments, examples and beneficial effects thereof refer to the description of the method above.
In yet another aspect, embodiments of the present invention provide a storage medium storing computer readable instructions that, when executed by a processor, perform a method of generating a knowledge-graph according to any of the embodiments described above.
The invention provides a method for constructing a knowledge graph by multi-agent cooperation based on a large language model, which aims to automate and automate the knowledge graph construction process, reduce the participation of manpower as much as possible and lower the labor cost. Through AI agent based on big language model, can be with environment interactive execution long-term and complicated task to playing different roles and cooperate each other, thereby realize the automated construction of knowledge graph.
The knowledge graph construction process is automated and a plurality of models are utilized to play different roles, so that remarkable efficiency improvement is brought. This improvement is mainly manifested in the following two aspects:
1. Automation of the knowledge graph body construction process: the different roles of the various models and the mutual cooperation make up for the problem of insufficient knowledge coverage caused by relying on experts in a single field. This combination makes the ontology-build process more efficient and accurate.
2. Automating the completion process of the knowledge graph entity: the traditional knowledge graph construction process generally needs to manually review a large amount of document data to carry out complicated manual carding, so that a large amount of manpower and material resources are consumed, and the efficiency is low. Valuable information can be extracted from a large amount of unstructured data through automatic construction, and a knowledge graph can be quickly and accurately complemented. The construction speed is greatly improved, and the quality and the integrity of the knowledge graph are improved.
The foregoing description of embodiments of the invention have been presented for the purpose of illustration and is not intended to be exhaustive or to limit the invention to the precise form disclosed. It will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for elements thereof without departing from the scope of the invention. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the invention without departing from the essential scope thereof. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed as the best mode contemplated for carrying out this invention, but that the invention will include all embodiments falling within the scope of the appended claims.
Claims (10)
1. A method of generating a knowledge-graph, the method comprising the steps of:
s101, receiving target domain keywords input by a user as first-level keywords;
s102, generating first-level ontology construction prompt information according to the first-level keywords, wherein the first-level ontology construction prompt information comprises an instruction for constructing a knowledge graph ontology based on the first-level keywords, pre-specified source definition information and a first-level knowledge graph ontology format, and the first-level knowledge graph ontology format limits knowledge graph ontology entries in the first-level knowledge graph ontology to include entities and attributes;
s103, transmitting the first-level ontology construction prompt information to an ontology construction model so that a first-level knowledge graph ontology is generated by the ontology construction model according to the first-level ontology construction prompt information, and the first-level knowledge graph ontology is stored in an ontology library;
s104, generating entity expansion allocation prompt information according to a previous-level knowledge graph body stored in the ontology base, wherein the entity expansion allocation prompt information comprises the previous-level knowledge graph body and an instruction for ordering the expandability of the attributes of the knowledge graph body items in the previous-level knowledge graph body;
S105, transmitting the entity expansion allocation prompt information to an entity expansion allocation model so that the entity expansion allocation model generates the expandability sequencing of the attribute of the knowledge graph body of the previous level according to the entity expansion allocation prompt information;
s106, sequentially executing the following processing on the knowledge graph body items in the knowledge graph body of the previous level according to the order of attribute scalability from high to low in the scalability sequencing: generating a current level keyword corresponding to the knowledge graph body item according to the entity and attribute of the knowledge graph body item and the previous level keyword;
s107, generating current level ontology construction prompt information according to the current level keyword, wherein the current level ontology construction prompt information comprises an instruction for constructing a current level knowledge graph ontology based on the current level keyword, pre-designated source definition information and a current level knowledge graph ontology format, wherein the current level knowledge graph ontology format limits knowledge graph ontology entries in the current level knowledge graph ontology to include entities and attributes, and designates that the entities of the current level knowledge graph ontology entries are attributes of corresponding previous level knowledge graph ontology entries;
S108, transmitting the current-level ontology construction prompt information to the ontology construction model so as to generate a current-level knowledge graph ontology according to the current-level ontology construction prompt information by the ontology construction model, and storing the current-level knowledge graph ontology in the ontology library, wherein the entity of the current-level knowledge graph ontology entry is stored in association with the attribute of the corresponding previous-level knowledge graph ontology entry;
s109, judging whether the number of the levels of the generated knowledge graph body reaches a predefined level number threshold value, and if so, turning to a step S110; if not, go to step S104;
s110, creating problem generation prompt information according to the knowledge graph ontology stored in the ontology library, and transmitting the problem generation prompt information to a problem generation model so as to generate query sentences by the problem generation model for the entity and the attribute of each knowledge graph ontology item in the stored knowledge graph ontology;
s111, vectorizing the query statement to generate a query statement vector;
s112, partitioning a pre-stored domain document text to generate a domain document text block, vectorizing the domain document text block to generate a domain document text vector, and storing the domain document text vector in a vector library;
S113, retrieving a domain document text vector similar to the query sentence vector in the vector library, and combining domain document text blocks corresponding to the similar domain document text vector into a retrieval context;
s114, creating answer generation prompt information according to the query statement and the search context, wherein the answer generation prompt information comprises an instruction for searching the answer of the query statement in the search context;
s115, transmitting the answer generation prompt information to an answer generation model so that the answer generation model generates an answer of the query statement according to the answer generation prompt information;
s116, generating a knowledge graph item of a knowledge graph according to the answer and the entity and the attribute of the knowledge graph body item corresponding to the query statement, wherein the knowledge graph item comprises an entity, an attribute and an attribute value, and the entity and the attribute of the knowledge graph item are the entity and the attribute of the corresponding knowledge graph body item respectively, and the attribute value of the knowledge graph item is the answer.
2. The method according to claim 1, characterized in that the method comprises:
After all the knowledge-graph ontology entries to be expanded in the knowledge-graph ontology of the previous level are executed in steps S106-S108, the knowledge-graph entries to be expanded in the knowledge-graph ontology of the current level are executed again.
3. The method according to claim 1, characterized in that the method comprises:
after steps S106-S108 are performed on a particular knowledge-graph ontology entry in a previous-level knowledge-graph ontology, steps S106-S108 are performed on a current-level knowledge-graph ontology corresponding to the particular knowledge-graph ontology entry until a predefined number of levels threshold is reached, and steps S106-S108 are performed on a next knowledge-graph ontology entry of the particular knowledge-graph ontology entry in the previous-level knowledge-graph ontology.
4. The method of claim 1, wherein step S106 further comprises:
taking a preset expansion number of knowledge graph ontology entries which are in front in the order as knowledge graph ontology entries to be expanded according to the order of attribute expandability from high to low, and executing the following processing on the knowledge graph ontology entries to be expanded: and generating a current level keyword corresponding to the knowledge graph body entry according to the entity and attribute of the knowledge graph body entry and the previous level keyword.
5. The method of claim 4, wherein the predetermined number of extensions is determined based on the number of layers of the current hierarchy.
6. The method of claim 1, wherein the predefined number of levels threshold is fixed or determined according to an order of attribute extensibility ordering of corresponding knowledge-graph entries in a specified level.
7. The method according to claim 1, wherein the method further comprises:
displaying the body library;
receiving an operation instruction of a user for selecting a knowledge graph ontology entry in the ontology library, wherein the operation instruction comprises deletion, editing and addition;
executing corresponding operation on the selected knowledge-graph body entry according to the operation instruction;
and/or the number of the groups of groups,
displaying the knowledge graph;
receiving an operation instruction of a user for selecting a knowledge graph item in the knowledge graph, wherein the operation instruction comprises deletion, editing and addition;
and executing corresponding operation on the selected knowledge graph entry according to the operation instruction.
8. The method of claim 1, wherein the pre-specified source definition information includes one or more of a database of the ontology-build model itself, a specified database address, a specified search engine.
9. An apparatus for generating a knowledge-graph for implementing the method of any one of claims 1-8, the apparatus comprising:
the target domain keyword receiving module is configured to receive target domain keywords input by a user and serve as first-level keywords;
a first-level ontology construction hint information generation module configured to generate first-level ontology construction hint information according to the first-level keyword, wherein the first-level ontology construction hint information includes an instruction to construct a knowledge-graph ontology based on the first-level keyword, pre-specified source definition information, and a first-level knowledge-graph ontology format, wherein the first-level knowledge-graph ontology format defines knowledge-graph ontology entries in a first-level knowledge-graph ontology as including entities and attributes;
a first-level knowledge graph body generating module configured to transmit the first-level ontology construction prompt information to an ontology construction model so as to generate a first-level knowledge graph body by the ontology construction model according to the first-level ontology construction prompt information, and store the first-level knowledge graph body in a ontology library;
The entity expansion allocation prompt information generation module is configured to generate entity expansion allocation prompt information according to a previous-level knowledge graph body stored in the ontology base, wherein the entity expansion allocation prompt information comprises the previous-level knowledge graph body and an instruction for ordering expandability of attributes of knowledge graph body items in the previous-level knowledge graph body;
the expandability ranking generation module is configured to transmit the entity expansion allocation prompt information to an entity expansion allocation model so that the entity expansion allocation model generates expandability ranking of the attribute of the previous-level knowledge graph body according to the entity expansion allocation prompt information;
the current-level keyword generation module is configured to sequentially execute the following processing on the knowledge graph body items in the knowledge graph body of the previous level according to the order of attribute scalability from high to low in the scalability sequencing: generating a current level keyword corresponding to the knowledge graph body item according to the entity and attribute of the knowledge graph body item and the previous level keyword;
a current level ontology construction hint information generation module configured to generate a current level ontology construction hint information according to the current level keyword, wherein the current level ontology construction hint information includes an instruction to construct a current level knowledge graph ontology based on the current level keyword, pre-specified source definition information, and a current level knowledge graph ontology format, wherein the current level knowledge graph ontology format defines a knowledge graph ontology entry in the current level knowledge graph ontology as including an entity and an attribute, and specifies that the entity of the current level knowledge graph ontology entry is an attribute of a corresponding previous level knowledge graph ontology entry;
A current-level knowledge-graph ontology generating module configured to transmit the current-level ontology construction prompt information to the ontology construction model, so that a current-level knowledge-graph ontology is generated by the ontology construction model according to the current-level ontology construction prompt information, and the current-level knowledge-graph ontology is stored in the ontology library, wherein an entity of the current-level knowledge-graph ontology entry is stored in association with an attribute of a corresponding previous-level knowledge-graph ontology entry;
the hierarchy number judging module is configured to judge whether the hierarchy number of the generated knowledge graph body reaches a predefined hierarchy number threshold value, and if so, the query sentence generating module is switched to; if not, turning to an entity expansion allocation prompt information generation module;
an inquiry sentence generation module configured to create a question generation hint information from the knowledge graph ontology stored in the ontology base and transmit the question generation hint information to a question generation model so that an inquiry sentence is generated by the question generation model for the entity and attribute of each of the knowledge graph ontology entries in the stored knowledge graph ontology;
The query sentence vectorization module is configured to vectorize the query sentence to generate a query sentence vector;
the field document text vectorization module is configured to block a pre-stored field document text, generate a field document text block, vectorize the field document text block, generate a field document text vector, and store the field document text vector in a vector library;
a search context generation module configured to search the vector library for a domain document text vector similar to the query sentence vector, and combine domain document text blocks corresponding to the similar domain document text vector into a search context;
an answer generation prompt creation module configured to create answer generation prompt according to the query sentence and the search context, wherein the answer generation prompt includes an instruction to search an answer of the query sentence in the search context;
the answer generation module is configured to transmit the answer generation prompt information to an answer generation model so that the answer generation model generates an answer of the query statement according to the answer generation prompt information;
And a knowledge graph generation module configured to generate a knowledge graph entry of a knowledge graph according to the answer and the entity and attribute of the knowledge graph ontology entry corresponding to the query statement, wherein the knowledge graph entry comprises an entity, an attribute and an attribute value, and wherein the entity and the attribute of the knowledge graph entry are the entity and the attribute of the corresponding knowledge graph ontology entry, respectively, and the attribute value of the knowledge graph entry is the answer.
10. A storage medium storing computer readable instructions which, when executed by a processor, perform the method of any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311767192.4A CN117435749B (en) | 2023-12-21 | 2023-12-21 | Method, device and storage medium for generating knowledge graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311767192.4A CN117435749B (en) | 2023-12-21 | 2023-12-21 | Method, device and storage medium for generating knowledge graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117435749A true CN117435749A (en) | 2024-01-23 |
| CN117435749B CN117435749B (en) | 2024-03-15 |
Family
ID=89550199
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311767192.4A Active CN117435749B (en) | 2023-12-21 | 2023-12-21 | Method, device and storage medium for generating knowledge graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117435749B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120010867A1 (en) * | 2002-12-10 | 2012-01-12 | Jeffrey Scott Eder | Personalized Medicine System |
| US8429179B1 (en) * | 2009-12-16 | 2013-04-23 | Board Of Regents, The University Of Texas System | Method and system for ontology driven data collection and processing |
| CN111966793A (en) * | 2019-05-20 | 2020-11-20 | 云号（北京）科技有限公司 | Intelligent question-answering method and system based on knowledge graph and knowledge graph updating system |
| US20210027175A1 (en) * | 2019-07-26 | 2021-01-28 | Bae Systems Information And Electronic Systems Integration Inc. | Context-infused ontology for knowledge models |
| WO2021040727A1 (en) * | 2019-08-30 | 2021-03-04 | Siemens Aktiengesellschaft | Plc code generation with a knowledge graph |
| CN113239210A (en) * | 2021-05-25 | 2021-08-10 | 河海大学 | Water conservancy literature recommendation method and system based on automatic completion knowledge graph |
| CN114064918A (en) * | 2021-11-06 | 2022-02-18 | 中国电子科技集团公司第五十四研究所 | Multi-modal event knowledge graph construction method |
| CN114417004A (en) * | 2021-11-10 | 2022-04-29 | 南京邮电大学 | Method, device and system for fusing knowledge graph and case graph |
| CN114610846A (en) * | 2022-03-09 | 2022-06-10 | 西安科技大学 | Knowledge graph expanding and complementing method for heuristic bionic knowledge grafting strategy |
| CN114880493A (en) * | 2022-04-22 | 2022-08-09 | 镇江智栎高科技有限公司 | Cross-modal retrieval algorithm based on text concept expansion |
| CN116450833A (en) * | 2022-12-31 | 2023-07-18 | 西南交通大学 | Knowledge graph construction system for complex equipment |
| CN117033721A (en) * | 2023-06-25 | 2023-11-10 | 长春市把手科技有限公司 | Legal consultation report generation system and method based on legal knowledge graph |
-
2023
- 2023-12-21 CN CN202311767192.4A patent/CN117435749B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120010867A1 (en) * | 2002-12-10 | 2012-01-12 | Jeffrey Scott Eder | Personalized Medicine System |
| US8429179B1 (en) * | 2009-12-16 | 2013-04-23 | Board Of Regents, The University Of Texas System | Method and system for ontology driven data collection and processing |
| CN111966793A (en) * | 2019-05-20 | 2020-11-20 | 云号（北京）科技有限公司 | Intelligent question-answering method and system based on knowledge graph and knowledge graph updating system |
| US20210027175A1 (en) * | 2019-07-26 | 2021-01-28 | Bae Systems Information And Electronic Systems Integration Inc. | Context-infused ontology for knowledge models |
| WO2021040727A1 (en) * | 2019-08-30 | 2021-03-04 | Siemens Aktiengesellschaft | Plc code generation with a knowledge graph |
| CN113239210A (en) * | 2021-05-25 | 2021-08-10 | 河海大学 | Water conservancy literature recommendation method and system based on automatic completion knowledge graph |
| CN114064918A (en) * | 2021-11-06 | 2022-02-18 | 中国电子科技集团公司第五十四研究所 | Multi-modal event knowledge graph construction method |
| CN114417004A (en) * | 2021-11-10 | 2022-04-29 | 南京邮电大学 | Method, device and system for fusing knowledge graph and case graph |
| CN114610846A (en) * | 2022-03-09 | 2022-06-10 | 西安科技大学 | Knowledge graph expanding and complementing method for heuristic bionic knowledge grafting strategy |
| CN114880493A (en) * | 2022-04-22 | 2022-08-09 | 镇江智栎高科技有限公司 | Cross-modal retrieval algorithm based on text concept expansion |
| CN116450833A (en) * | 2022-12-31 | 2023-07-18 | 西南交通大学 | Knowledge graph construction system for complex equipment |
| CN117033721A (en) * | 2023-06-25 | 2023-11-10 | 长春市把手科技有限公司 | Legal consultation report generation system and method based on legal knowledge graph |
Non-Patent Citations (3)
| Title |
|---|
| 吴微: "小规模知识库指导下的细分领域知识图谱构建研究", 中国优秀硕士学位论文全文数据库信息科技辑, no. 3, 15 March 2022 (2022-03-15), pages 143 - 111 * |
| 官赛萍;靳小龙;贾岩涛;王元卓;程学旗;: "面向知识图谱的知识推理研究进展", 软件学报, no. 10, 8 February 2018 (2018-02-08), pages 74 - 102 * |
| 马超;刘亚淑;骆功宁;王宽全;: "基于级联随机森林与活动轮廓的3D MR图像分割", 自动化学报, no. 05, 11 October 2018 (2018-10-11), pages 178 - 188 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117435749B (en) | 2024-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108959613B (en) | RDF knowledge graph-oriented semantic approximate query method | |
| CN105975531B (en) | Robot dialog control method and system based on dialogue knowledge base | |
| CN109284363A (en) | A kind of answering method, device, electronic equipment and storage medium | |
| CN105808590B (en) | Search engine implementation method, searching method and device | |
| Acid et al. | An information retrieval model based on simple Bayesian networks | |
| CN111026886B (en) | Multi-round dialogue processing method for professional scene | |
| CN112800170A (en) | Question matching method and device and question reply method and device | |
| CN110866093A (en) | Machine question-answering method and device | |
| CN112749266B (en) | Industrial question and answer method, device, system, equipment and storage medium | |
| CN111078837A (en) | Intelligent question and answer information processing method, electronic equipment and computer readable storage medium | |
| US8175997B2 (en) | Method of applying user-defined inference rule using function of searching knowledge base and knowledge base management system therefor | |
| CN112818092B (en) | Knowledge graph query statement generation method, device, equipment and storage medium | |
| CN111143539A (en) | Knowledge graph-based question-answering method in teaching field | |
| CN112632239A (en) | Brain-like question-answering system based on artificial intelligence technology | |
| CN114511085A (en) | Entity attribute value identification method, apparatus, device, medium, and program product | |
| KR20120071966A (en) | Method and apparatus for retrieving software components using case based reasoning system, and method for providing explanation | |
| CN112559760B (en) | CPS (cyber physical system) resource capacity knowledge graph construction method for text description | |
| CN116821307B (en) | Content interaction method, device, electronic equipment and storage medium | |
| CN117435749B (en) | Method, device and storage medium for generating knowledge graph | |
| CN115469860B (en) | Method and system for automatically generating demand-to-software field model based on instruction set | |
| CN114881019A (en) | Data hybrid storage method and device for multi-modal network | |
| CN116737964B (en) | Artificial intelligence brain system | |
| CN117829242B (en) | Model processing method and related equipment | |
| CN118035409A (en) | Question answering method and device, storage medium and computing equipment | |
| CN117094392A (en) | Expert system based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |