CN117390169A - Form data question-answering method, device, equipment and storage medium - Google Patents
Form data question-answering method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN117390169A CN117390169A CN202311691157.9A CN202311691157A CN117390169A CN 117390169 A CN117390169 A CN 117390169A CN 202311691157 A CN202311691157 A CN 202311691157A CN 117390169 A CN117390169 A CN 117390169A
- Authority
- CN
- China
- Prior art keywords
- question
- information
- processed
- language model
- acquiring
- 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 82
- 238000007781 pre-processing Methods 0.000 claims abstract description 35
- 238000000605 extraction Methods 0.000 claims abstract description 17
- 238000005516 engineering process Methods 0.000 claims description 10
- 238000013075 data extraction Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 7
- 238000012163 sequencing technique Methods 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 5
- 238000012216 screening Methods 0.000 claims description 3
- 238000003058 natural language processing Methods 0.000 description 18
- 230000008569 process Effects 0.000 description 6
- 239000013598 vector Substances 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000013519 translation 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
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- 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
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- 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
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Human Computer Interaction (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the technical field of data dialogue questions and answers, in particular to a form data question and answer method, a device, equipment and a storage medium, wherein the method comprises the following steps: acquiring historical table data and extracting historical name information in the historical table data to finely tune a pre-constructed large language model; acquiring to-be-processed form data, extracting to-be-processed name information in the form data, and inputting the to-be-processed name information into the trimmed large language model to obtain a ternary relation prediction result; constructing an undirected graph based on the ternary relation prediction result to obtain a meta concept; constructing a knowledge graph based on the to-be-processed form data, the ternary relation prediction result and the meta concept; acquiring a real-time feedback question and preprocessing to obtain preprocessing information, searching a constructed knowledge graph based on the preprocessing information, and acquiring a preset question-answering large language model feedback answer according to a matching result; the method disclosed by the invention can improve the extraction capability of the large language model to the form data and reduce the illusion of the large language model.
Description
Technical Field
The present invention relates to the field of data dialogue question-answering technologies, and in particular, to a method, an apparatus, a device, and a storage medium for table data question-answering.
Background
When a question-answering system is built based on a large model and an existing knowledge base or database, firstly, the collected documents, forms and other data are required to be vectorized, and the vectorized text data are used as the knowledge base for answering user questions of the large model; when answering the user questions, directly converting the questions input by the user into sentence vectors, and then carrying out matching inquiry in a knowledge base by using a search engine; or extracting keywords from the user input questions by constructing a keyword extraction model, converting the extracted keywords into words, embedding the words into a knowledge base, carrying out matching query, and then taking the result obtained by the matching query and the questions of the user together as the input of a large model, thereby obtaining answers based on knowledge in the existing knowledge base or database.
In fact, in the actual application of the query mode, the accuracy of the output question-answer result is not high, and the answer output by the large model is not matched with the user problem because the knowledge base information obtained by query cannot be well matched with the problem proposed by the user; this problem is more pronounced, especially when the data is of the form type; in the traditional method, table data is usually treated as text data when a knowledge base is constructed, the table knowledge is converted into text data in a text segment form, and a corresponding vector knowledge base is generated; when a knowledge base of form data is constructed, a mode of directly converting the form data into text data is used, so that a large amount of information irrelevant to user questions is returned when relevant contents of the form are searched according to the user questions in actual use, and the accuracy of answering a large model is seriously reduced; in particular, for a table file containing a plurality of workbooks, or a table file with a complex table header, for example, the table header is distributed above and beside a table body, a common data processing mode cannot well describe the content relation among the plurality of workbooks, so that a large model has insufficient processing capability on the complex table file.
It can be seen that there is a need for improvements and improvements in the art.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a table data question-answering method, a device, equipment and a storage medium, which can improve the extraction capability of a large language model to table data and reduce the illusion of the large model data.
The first aspect of the present invention provides a form data question-answering method, comprising: acquiring history table data, extracting history name information in the history table data, and fine-tuning a pre-constructed large language model based on the history name information; acquiring to-be-processed table data, extracting to-be-processed name information in the to-be-processed table data, and inputting the to-be-processed name information into the trimmed large language model to obtain a ternary relation prediction result; constructing an undirected graph based on the ternary relation prediction result, and acquiring a primitive concept according to the constructed undirected graph; constructing a knowledge graph based on the to-be-processed form data, the ternary relation prediction result and the meta concept; the method comprises the steps of obtaining real-time feedback questions, preprocessing the questions to obtain preprocessing information, searching a constructed knowledge graph based on the preprocessing information in a matching mode to obtain auxiliary information, and obtaining answers which are fed back by a preset question-answering large language model and correspond to the questions based on the auxiliary information.
Optionally, in a first implementation manner of the first aspect of the present invention, the obtaining the history table data, extracting history name information in the history table data, and fine-tuning the pre-built large language model based on the history name information specifically includes: acquiring history table data, and extracting history name information in the history table data based on an NLP method; acquiring a data extraction format designed in a few-shot mode, and acquiring a ternary relation extraction problem constructed based on a prompt word; fine-tuning a pre-constructed large language model based on historical name information, a data extraction format and a ternary relation extraction problem; and deploying the trimmed large language model to a server.
Optionally, in a second implementation manner of the first aspect of the present invention, the obtaining the to-be-processed table data, extracting to-be-processed name information in the to-be-processed table data, and inputting the to-be-processed name information into the trimmed large language model to obtain the ternary relationship prediction result, includes: acquiring to-be-processed form data, and extracting to-be-processed name information in the to-be-processed form data based on an NLP method; inputting the name information to be processed into the trimmed large language model to obtain an output result; and carrying out matching screening on the output result by using the regular expression to form a ternary relation prediction result.
Optionally, in a third implementation manner of the first aspect of the present invention, the constructing an undirected graph based on the ternary relationship prediction result, and obtaining the meta concept according to the constructed undirected graph specifically includes: constructing an undirected graph based on the ternary relation prediction result, wherein the constructed undirected graph comprises a plurality of nodes; calculating the centrality of the graph of each node, and sequencing the calculation results of the centrality of the graph; and obtaining the meta concept according to the sequencing result.
Optionally, in a fourth implementation manner of the first aspect of the present invention, the constructing a knowledge graph based on the to-be-processed table data, the ternary relationship prediction result and the meta concept specifically includes: the table data to be processed comprises one or more table files, and each table file comprises one or more workbooks; and connecting the table file into a root node of the knowledge graph, taking the workbook corresponding to the table file and the meta-concepts corresponding to the workbook as nodes of the knowledge graph, and taking the content of the non-meta-concepts in the ternary relation prediction result as a relation to be used for connecting with the corresponding value in the knowledge graph to complete the construction of the knowledge graph.
Optionally, in a fifth implementation manner of the first aspect of the present invention, the obtaining a real-time feedback question, preprocessing the question to obtain preprocessed information, searching a constructed knowledge graph based on the preprocessed information match to obtain auxiliary information, and obtaining an answer corresponding to the question and fed back by a preset question-answer language model based on the auxiliary information, where the method specifically includes: acquiring a real-time feedback problem, and performing word segmentation on the problem to obtain preprocessing information; searching the constructed knowledge graph based on the preprocessing information to obtain auxiliary information; performing similarity matching on the auxiliary information and the preprocessing information by using an NLP method, and acquiring a preset similarity threshold; when the similarity between the problem and the auxiliary information is more than or equal to a preset first similarity threshold value, constructing a first input instruction by adopting a thinking chain technology based on the auxiliary information and the problem; and inputting the constructed first input instruction into a preset question-answering language model to obtain an answer corresponding to the question.
Optionally, in a sixth implementation manner of the first aspect of the present invention, the performing similarity matching on the auxiliary information and the pre-processing information by using an NLP method, and obtaining a preset similarity threshold, further includes: when the second similarity threshold value is less than or equal to the first similarity threshold value, the similarity between the problem and the auxiliary information is less than the preset first similarity threshold value, an inquiry instruction is generated according to the auxiliary information; acquiring selection information fed back by a user according to the inquiry instruction, and constructing a second input instruction by adopting a thinking chain technology based on the auxiliary information, the problems and the selection information; and inputting the constructed second input instruction into a preset question-answering language model to obtain an answer corresponding to the question.
The second aspect of the present invention provides a form data question-answering device, comprising: the fine tuning module is used for acquiring the history table data, extracting history name information in the history table data and fine tuning the pre-constructed large language model based on the history name information; the prediction module is used for acquiring the form data to be processed, extracting the name information to be processed in the form data to be processed, and inputting the name information to be processed into the trimmed large language model to obtain a ternary relation prediction result; the acquisition module is used for constructing an undirected graph based on the ternary relation prediction result and acquiring a primitive concept according to the constructed undirected graph; the construction module is used for constructing a knowledge graph based on the to-be-processed form data, the ternary relation prediction result and the meta concept; and the dialogue module is used for acquiring the real-time feedback questions, preprocessing the questions to obtain preprocessed information, searching the constructed knowledge graph based on the preprocessed information, obtaining auxiliary information, and acquiring answers which are fed back by the preset question-answering large language model and correspond to the questions based on the auxiliary information.
A third aspect of the present invention provides a form data question-answering apparatus, comprising: a memory and at least one processor, the memory having instructions stored therein; at least one of the processors invokes the instructions in the memory to cause the tabular data question-answering device to perform the respective steps of the tabular data question-answering method according to any one of the preceding claims.
A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon which, when executed by a processor, implement the steps of the form data question-answering method according to any one of the above.
According to the technical scheme, the history name information in the history table data is extracted by acquiring the history table data so as to finely adjust the pre-built large language model; acquiring to-be-processed form data, extracting to-be-processed name information in the form data, and inputting the to-be-processed name information into the trimmed large language model to obtain a ternary relation prediction result; constructing an undirected graph based on the ternary relation prediction result to obtain a meta concept; constructing a knowledge graph based on the to-be-processed form data, the ternary relation prediction result and the meta concept; acquiring a real-time feedback question and preprocessing to obtain preprocessing information, searching a constructed knowledge graph based on the preprocessing information, and acquiring a preset answer corresponding to the question fed back by the question-answering large language model according to a matching result; according to the method disclosed by the application, the extraction capability of the large language model on the form data can be improved, the illusion of the large language model is reduced, the accuracy of the question-answer result output by the large language model is improved, and therefore the use experience of a user is improved.
Drawings
FIG. 1 is a first flowchart of a method for asking and answering form data according to an embodiment of the present invention;
FIG. 2 is a second flowchart of a method for asking and answering form data according to an embodiment of the present invention;
FIG. 3 is a third flowchart of a method for asking and answering form data according to an embodiment of the present invention;
FIG. 4 is a fourth flowchart of a method for asking and answering form data according to an embodiment of the present invention;
FIG. 5 is a fifth flowchart of a method for asking and answering form data according to an embodiment of the present invention;
FIG. 6 is a sixth flowchart of a form data question-answering method provided by the present invention;
fig. 7 is a schematic structural diagram of a form data question-answering device according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of a table data question-answering device according to an embodiment of the present invention.
Detailed Description
The present invention provides a method, apparatus, device and storage medium for table data questions and answers, the terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above figures are used for distinguishing similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.
For ease of understanding, a specific flow of an embodiment of the present invention is described below with reference to fig. 1, and one embodiment of a table data question-answering method in the embodiment of the present invention includes:
101. acquiring history table data, extracting history name information in the history table data, and fine-tuning a pre-constructed large language model based on the history name information;
in this embodiment, the Pre-built large language model is an LMM large language model, and is constructed by a deep learning neural network, and adopts a architecture similar to AIG (generating Pre-trained Transformer), and uses a transducer model to process an input text and generate a coherent and logical output; by training a large language model, it can be used to perform various natural language processing tasks to achieve answers to user questions, i.e., to achieve conversations with the user.
LMM (large language model) fine tuning is a technique for tuning a pre-trained language model; this process typically involves adapting a pre-trained large language model, such as AIG-3.5, to a particular application or domain by performing additional training on a particular task; in this embodiment, in order to construct a knowledge graph of the form data, a pre-constructed large language model needs to be adjusted in a targeted manner, specifically, supervised learning is performed on the basis of the pre-constructed large language model, so that the pre-constructed large language model adapts to the target task, even if the pre-constructed large language model has the capability of extracting the ternary relationship for constructing the knowledge graph of the form data.
102. Acquiring to-be-processed table data, extracting to-be-processed name information in the to-be-processed table data, and inputting the to-be-processed name information into the trimmed large language model to obtain a ternary relation prediction result;
in this embodiment, the table data to be processed may include one or more table files, each table file may include one or more workbooks, and each workbook may include one or more tables; the name information to be processed comprises field information of a working book and field information of a table header.
103. Constructing an undirected graph based on the ternary relation prediction result, and acquiring a primitive concept according to the constructed undirected graph;
in this embodiment, since the name information to be processed includes a plurality of field information, in order to improve the accuracy of the constructed knowledge graph, key fields, that is, meta-concepts, in the plurality of field information need to be confirmed.
104. Constructing a knowledge graph based on the to-be-processed form data, the ternary relation prediction result and the meta concept;
in this embodiment, the to-be-processed table data is table data associated with a user question, and auxiliary information associated with the user question can be provided for a preset question-answering language model by constructing a knowledge graph, so that the preset question-answering language model can accurately answer the user question.
105. The method comprises the steps of obtaining real-time feedback questions, preprocessing the questions to obtain preprocessing information, searching a constructed knowledge graph based on the preprocessing information in a matching mode to obtain auxiliary information, and obtaining answers which are fed back by a preset question-answering large language model and correspond to the questions based on the auxiliary information.
In this embodiment, the preset big question-answer language model is an LMM big language model, and the preset big question-answer language model is used for generating an answer corresponding to the question according to the auxiliary information output by the knowledge graph and the question fed back by the user in real time.
The application discloses a form data question-answering method, which is used for finely adjusting a pre-constructed large language model by acquiring historical form data and extracting historical name information in the historical form data; acquiring to-be-processed form data, extracting to-be-processed name information in the form data, and inputting the to-be-processed name information into the trimmed large language model to obtain a ternary relation prediction result; constructing an undirected graph based on the ternary relation prediction result to obtain a meta concept; constructing a knowledge graph based on the to-be-processed form data, the ternary relation prediction result and the meta concept; acquiring a real-time feedback question and preprocessing to obtain preprocessing information, searching a constructed knowledge graph based on the preprocessing information, and acquiring a preset answer corresponding to the question fed back by the question-answering large language model according to a matching result; according to the method disclosed by the application, the extraction capability of the large language model on the form data can be improved, the illusion of the large language model is reduced, the accuracy of the question-answer result output by the large language model is improved, and therefore the use experience of a user is improved.
Referring to fig. 2, a second embodiment of the table data question-answering method in the embodiment of the present invention includes:
201. acquiring history table data, and extracting history name information in the history table data based on an NLP method;
in the embodiment, the history name information in the history table data is extracted based on the NLP method; the NLP method is a natural language processing (Natural Language Processing) method, which is a technique of converting natural language into a form understandable by a computer; the NLP method realizes the functions of man-machine interaction, automatic translation and the like by researching the structure and the semantics of natural language; the NLP method mainly comprises the aspects of grammar analysis, semantic understanding and generation, text classification and the like, wherein the grammar analysis is the basis of the NLP method, and the structure and grammar relation of sentences are extracted mainly by decomposing and analyzing the sentences and are converted into a representation form which can be processed by a computer; semantic understanding and generation are the core of the NLP method, and text representations conforming to semantic rules are generated by analyzing and understanding the semantics of sentences; common NLP methods comprise part-of-speech tagging, syntactic analysis, semantic role tagging, emotion analysis and the like, and are widely applied to the field of natural language processing.
In this embodiment, the history table data may include one or more history table files, each history table file may include one or more history workbooks, each history workbook may include one or more history tables; the history name information comprises field information of a history workbook and field information of a history table header.
202. Acquiring a data extraction format designed in a few-shot mode, and acquiring a ternary relation extraction problem constructed based on a prompt word;
in the embodiment, a ternary relation extraction instruction is designed in a few-shot mode, and the specific method is to correspondingly increase cases of ternary relations related to form categories at the end of the instruction so as to limit a format of the ternary relation output by a pre-built large language model, namely limit a data extraction format; the form of the ternary relationship of the field information of the table is: form field-relationship-form field, e.g., responsibility department-responsible-asset name.
In the embodiment, the problem is extracted based on the ternary relation constructed by the prompt words, wherein the prompt words are a section of characters inserted into an input example, an original task can be expressed as a language model problem, and the correlation among various contents in the header can be obtained by asking a pre-constructed large language model through a construction template; ternary relationship extraction problems can be constructed using methods based on continuity hint words (or soft hints SoftPrompt) or discrete hint words to achieve fine tuning of large language models.
203. Fine-tuning a pre-constructed large language model based on historical name information, a data extraction format and a ternary relation extraction problem;
in the embodiment, an extraction instruction is constructed based on history name information, a data extraction format and a ternary relation extraction problem, and the constructed extraction instruction is input into a pre-constructed large language model as an input feature to further train the pre-constructed large language model, so that a possible ternary relation is obtained; for example, the history table data includes a workbook, and the corresponding table header includes contents including asset name, current leader, responsibility department, use place, warehouse date and use state; the fetch instruction may be: what relationship (ternary relationship extraction problem) may exist between the following entities: asset name, current leader, responsibility, place of use, date of warehouse entry, status of use (history name information), please list in entity-relationship-entity form, such as: current leader-asset name, responsibility department-responsible-asset name (data extraction format); the result output by the large language model is: "the following is a possible relationship: current leader-asset name, responsibility department-responsible-asset name, use place-use-asset name, date of warehouse-in-warehouse-asset name, use status-asset name, current leader-location-use place).
204. Deploying the trimmed large language model to a server;
in this embodiment, the server may be a local server or a cloud server, and the trimmed large language model is deployed to the server, so as to analyze the name, the workbook name and the header field of the table document fed back by the user interface, construct a knowledge graph, and query the content information of the table document in a dialogue manner.
Referring to fig. 3, a third embodiment of a table data question-answering method in an embodiment of the present invention includes:
301. acquiring to-be-processed form data, and extracting to-be-processed name information in the to-be-processed form data based on an NLP method;
302. inputting the name information to be processed into the trimmed large language model to obtain an output result;
303. matching and screening the output result by using a regular expression to form a ternary relation prediction result;
in this embodiment, after the output result is obtained, a result conforming to the ternary relationship, that is, a result conforming to the ternary relationship logic is matched by adopting a regular expression, so that the output result is further screened and removed, and the accuracy of the output ternary relationship prediction result is improved.
Referring to fig. 4, a fourth embodiment of the table data question-answering method in the embodiment of the present invention includes:
401. constructing an undirected graph based on the ternary relation prediction result, wherein the constructed undirected graph comprises a plurality of nodes;
in this embodiment, the ternary relationship prediction result includes a ternary relationship of each content in the header, each content and a relationship between the content and each other are used to construct an undirected graph, each content is used as a node of the undirected graph, and a relationship between the content is used as an edge of the undirected graph to construct the undirected graph.
402. Calculating the centrality of the graph of each node, and sequencing the calculation results of the centrality of the graph;
in this embodiment, the importance degree of each field in the graph is calculated to locate the important field of the complex table data, so as to obtain the meta concept; the importance degree can be judged by utilizing the Centrality (Graph Centrality) of each node Graph in the Graph structure, and the Graph Centrality has a plurality of definitions and calculation methods, including Centrality, feature vector Centrality, intermediate Centrality and connection Centrality; by comprehensively considering centrality and other indexes (such as approximate centrality, graph diameter and the like), whether one field in the table knowledge relationship graph is really positioned at the most center of data contained in one workbook is judged; for example, a graph algorithm-based centrality algorithm may be employed to determine the meta-concept, expressed as:
；
Wherein,represents the centrality of the degree of node x +.>The degree of node x, n is the total number of nodes in the graph structure.
403. Acquiring a meta concept according to the sequencing result;
in this embodiment, the bubble sorting method may be used to sort the calculation results from large to small, where the calculation results may be centrality, and the content with the highest centrality is selected as the meta-concept; for example, the degree center calculation results for the 6 fields "asset name", "current leader", "responsibility department", "use place", "warehouse entry date", "use state" are respectively:
current collar Degree Centrality =0.40
Asset name Degree Centrality =1.00
Responsibility department Degree Centrality =0.20
Degree Centrality =0.40 points of use
Date of warehouse entry Degree Centrality =0.20
State of use Degree Centrality =0.20
The centrality of the asset name is 1 and is highest, so that the asset name is selected as a meta concept, and other fields are used as relations for connection with the value; this example selects only one field as the meta-concept, and in actual use, the meta-concept may be plural.
Referring to fig. 5, a fifth embodiment of a table data question-answering method in an embodiment of the present invention includes:
501. The table data to be processed comprises one or more table files, and each table file comprises one or more workbooks;
502. connecting the table file into a root node of the knowledge graph, taking a workbook corresponding to the table file and a meta concept corresponding to the workbook as nodes of the knowledge graph, and taking the content of a non-meta concept in a ternary relation prediction result as a relation to be used for connecting with a corresponding value in the knowledge graph to complete the construction of the knowledge graph;
in this embodiment, by way of example, the root node of the constructed knowledge graph includes i table files, any table file includes j workbooks, any workbook includes k meta-concepts, and entities in the meta-concepts are respectively connected to values through l fields; further, in the constructed knowledge graph, except for the ternary relation between the entity corresponding to the meta concept and the value, the key meta concept is connected with the original information of the to-be-processed form through the relation with the name of detail.
In the embodiment, specifically, when the ternary relation of the meta concept to the value is constructed, each item corresponding to the meta concept field in the header is selected for ternary relation establishment; illustrating:
Table 1 meta concept and relationship field schematic table
| Meta concept | Field 1 | Field 2 | Field 3 |
| a | 100 | Is that | School |
| b | 200 | Whether or not | Road |
For example, in table one above, the first column is determined to be a meta-concept and the second, third, and fourth column fields are relationship fields; then ternary relationship examples are made for the a and b entities under the meta-concept (table 1): a-field 1- "100", a-field 2- "yes", b-field 3- "road"; two entities in which a and b belong to the meta-concept; the field 1, the field 2 and the field 3 are relation fields; "100", "yes", "road" are values.
Referring to fig. 6, a sixth embodiment of a table data question-answering method in an embodiment of the present invention includes:
601. acquiring a real-time feedback problem, and performing word segmentation on the problem to obtain preprocessing information;
in this embodiment, a chinese word segmentation algorithm may be used to perform word segmentation on the problem that the user feeds back in real time.
602. Searching the constructed knowledge graph based on the preprocessing information to obtain auxiliary information;
in this embodiment, when performing the matching search of the knowledge graph, the meta concept and the field may be modified using the following format, including document name+workbook+meta concept and document name+workbook+field; by matching search, a number of meta-concepts or fields that best match can be obtained, constituting auxiliary information.
603. Performing similarity matching on the auxiliary information and the preprocessing information by using an NLP method, and acquiring a preset first similarity threshold and a preset second similarity threshold;
in this embodiment, similarity matching may be performed using a vector search and similarity matching method represented by FAISS; FAISS is an efficient tool represented by vector search and similarity matching, developed by Facebook AI Research (FAIR); the main characteristics of FAISS include high efficiency performance, support for various similarity metrics, GPU acceleration and scalability; by using FAISS, a fast and accurate similarity search can be performed in a large data set.
In this embodiment, the first similarity threshold and the second similarity threshold are cosine similarity thresholds, and the first similarity threshold may be set to 0.6 and the second similarity threshold may be set to 0.4.
604. When the similarity between the problem and the auxiliary information is more than or equal to a preset first similarity threshold value, constructing a first input instruction by adopting a thinking chain technology based on the auxiliary information and the problem;
in this embodiment, the first input instruction is constructed based on a thought chain technique, and specifically, the end of the first input instruction is added, for example, "please think step by step according to known contents", so as to improve understanding of knowledge by the big-question-answer language model.
605. Inputting the constructed first input instruction into a preset question-answering language model to obtain an answer corresponding to the question;
in this embodiment, for example, assume that the problem of user feedback is: "is there information about an industrial camera? "the best meta concept result of the question matching with the meta concept is" industrial camera ", while the cosine similarity of the" industrial camera "and the user question exceeds 0.6, no further inquiry is needed to the user, thus constructing a first input instruction based on a thinking chain: based on the following known matters, the user's question is briefly answered. If an answer cannot be obtained therefrom, please say "the question cannot be answered based on the known information" or "sufficient relevant information is not provided", it is not allowed to add information other than the known content to the answer. Known is that the use state of the industrial camera is in use, the use place of the industrial camera is A2 302, and the problem is that: is there information about industrial cameras? Please think step by step according to the known content.
After a first input instruction is input into a preset question-answering language model, the answer result is as follows: your good-! From the known information, the following industrial camera related information can be derived:
The industrial camera is in use and the point of use is A2 XXX.
The model and brand of the industrial camera are unknown. The quality of the image taken by the industrial camera is unknown.
-the number of images taken by the industrial camera is unknown.
-the content of the image taken by the industrial camera is unknown.
-the application scenario of the industrial camera is unknown.
-the storage means of the industrial camera is unknown.
The transmission scheme of the industrial camera is unknown.
-the working state of the industrial camera is unknown.
Referring to fig. 6, a seventh embodiment of a table data question-answering method in an embodiment of the present invention includes:
606. when the similarity between the problem and the auxiliary information is smaller than the first similarity threshold, generating an inquiry instruction according to the auxiliary information;
in this embodiment, the query instruction is composed of meta-concepts or fields included in the auxiliary information, and includes several options most similar to the question, that is, when the preset second similarity threshold value is less than or equal to the similarity between the question and the auxiliary information is less than the preset first similarity threshold value, for the accuracy of the answer, the query instruction is generated according to the matched meta-concepts and fields to confirm whether the matched content is required by the user.
607. Acquiring selection information fed back by a user according to the inquiry instruction, and constructing a second input instruction by adopting a thinking chain technology based on the auxiliary information, the problems and the selection information;
in this embodiment, the fed back selection information includes a meta concept or field selected by the user, if the user selects any meta concept or field, the selection information is used as a prefix or suffix of the question, and based on the auxiliary information and the question, a thinking chain technology is adopted to construct a second input instruction, and when the second input instruction is constructed, the auxiliary information part can use a ' field+relationship+field ', and a ' parallel structure representation; if the user does not select any meta concept or field, a preset answer instruction is returned, and the content of the answer instruction may be: "no corresponding results are queried" and possible suggestions are made based on prior knowledge.
608. Inputting the constructed second input instruction into a preset question-answering language model to obtain an answer corresponding to the question;
in this embodiment, for example, assume that the problem of user feedback is: "who has a computer with CPU i 7? The entities of the problem and the meta concept matching the best meta concept are 'CPU operation host', 'notebook', 'programmable controller-other computer', 'embedded development board (Intel)' and 'high performance graphic workstation'.
But because the matching entity is below the threshold of 0.6 with respect to the cosine of the problem, the large language model generates query instructions to further query the user, the contents of the query instructions being:
do you wish to find relevant information of the following items according to the search result that no relevant content is found? CPU operation host, notebook computer, programmable controller-other computer, embedded development board (Intel), high-performance graphic workstation; if the content contains information of your query, please input repeatedly before the question as the question start part, such as a question: CPU calculates the host computer, who has the CPU as i7 computer.
If the user considers the CPU operation host according to the self judgment, the question is continuously asked: "CPU operation host, who has a computer with CPU i 7? The user continues to ask questions as selection information, and a second input instruction is further constructed by adopting a thinking chain technology based on the auxiliary information, the questions and the selection information.
Inputting a second input instruction into the trimmed large language model, wherein the answer result is as follows: from the known content, the following relevant information can be derived for the compliance problem:
1. the asset condition is the details of the CPU operation host, namely the leader is Chen Mou, and the asset condition belongs to a team of departments of a certain research group, and the brand is not detailed and the parameters are not detailed.
2. The current acceptance of the CPU arithmetic host is Chen Mou.
3. The use state of the CPU operation host is in use.
4. The responsible departments and teams of the CPU computing hosts are the a study group.
5. The place of use of the CPU operation host is A2 XXX.
Thus, known contents which meet the problem are:
the leader is Chen Mou belonging to team a.
-branding is not detailed, parameters are not detailed.
-the state of use is in use.
Responsibility departments and teams are the advanced aviation remote sensing technology research group.
The place of use is Ji Hua laboratory A2 zone XXX office.
Since no specific information is provided in the known content about the computer whose CPU is i7, the question cannot be answered.
Further, in this embodiment, when the similarity between the problem and the auxiliary information is less than a preset second similarity threshold, the problem and the prompt fed back by the user are input to a preset big question-answer language model, and a return instruction is obtained, where the return instruction does not include any information in the table data to be processed, and the return instruction may include a preset return document, and the return document may be "no corresponding result is queried".
The table data question-answering method in the embodiment of the present invention is described above, and the table data question-answering device in the embodiment of the present invention is described below, referring to fig. 7, one embodiment of the table data question-answering device in the embodiment of the present invention includes:
The fine tuning module 701 is configured to obtain history table data, extract history name information in the history table data, and fine tune a pre-built large language model based on the history name information;
the prediction module 702 is configured to obtain to-be-processed table data, extract to-be-processed name information in the to-be-processed table data, and input the to-be-processed name information into the trimmed large language model to obtain a ternary relationship prediction result;
an obtaining module 703, configured to construct an undirected graph based on the ternary relationship prediction result, and obtain a meta concept according to the constructed undirected graph;
a construction module 704, configured to construct a knowledge graph based on the to-be-processed table data, the ternary relationship prediction result and the meta concept;
the dialogue module 705 is configured to obtain a question fed back in real time, preprocess the question to obtain preprocessing information, search a constructed knowledge graph based on the preprocessing information, obtain auxiliary information, and obtain an answer corresponding to the question fed back by a preset question-answering language model based on the auxiliary information.
The table data question-answering device in the embodiment of the present invention is described in detail above in fig. 7 from the point of view of the modularized functional entity, and the table data question-answering apparatus in the embodiment of the present invention is described in detail below from the point of view of hardware processing.
Fig. 8 is a schematic structural diagram of a table data questioning and answering apparatus provided in an embodiment of the present invention, where the table data questioning and answering apparatus 800 may generate relatively large differences according to different configurations or performances, and may include one or more processors (central processing units, CPU) 810 (e.g., one or more processors) and a memory 820, and one or more storage media 830 (e.g., one or more mass storage devices) storing application programs 833 or data 832. Wherein memory 820 and storage medium 830 can be transitory or persistent. The program stored in the storage medium 830 may include one or more modules (not shown), each of which may include a series of instruction operations on the tabular data question-answering apparatus 800. Still further, the processor 810 may be configured to communicate with the storage medium 830 and execute a series of instruction operations in the storage medium 830 on the form data question-answering apparatus 800 to implement the steps of the form data question-answering method provided by the above-described method embodiments.
Form data question-answering apparatus 800 can also include one or more power supplies 840, one or more wired or wireless network interfaces 850, one or more input-output interfaces 860, and/or one or more operating systems 831, such as Windows Serve, mac OS X, unix, linux, freeBSD, or the like. It will be appreciated by those skilled in the art that the form data question and answer device structure shown in the present application is not limiting of the form data based question and answer device and may include more or less components than illustrated, or may be combined with certain components, or may be arranged with different components.
The present invention also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, and which may also be a volatile computer readable storage medium, having stored therein instructions that, when executed on a computer, cause the computer to perform the steps of a table data question-answering method.
It will be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process of the system or apparatus and unit described above may refer to the corresponding process in the foregoing method embodiment, which is not repeated herein.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
Finally, it should be noted that: the foregoing is merely a preferred example of the present invention, and the present invention is not limited thereto, but it is to be understood that modifications and equivalents of some of the technical features described in the foregoing embodiments may be made by those skilled in the art, although the present invention has been described in detail with reference to the foregoing embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A form data question-answering method, comprising:
acquiring history table data, extracting history name information in the history table data, and fine-tuning a pre-constructed large language model based on the history name information;
acquiring to-be-processed table data, extracting to-be-processed name information in the to-be-processed table data, and inputting the to-be-processed name information into the trimmed large language model to obtain a ternary relation prediction result;
constructing an undirected graph based on the ternary relation prediction result, and acquiring a primitive concept according to the constructed undirected graph;
constructing a knowledge graph based on the to-be-processed form data, the ternary relation prediction result and the meta concept;
The method comprises the steps of obtaining real-time feedback questions, preprocessing the questions to obtain preprocessing information, searching a constructed knowledge graph based on the preprocessing information in a matching mode to obtain auxiliary information, and obtaining answers which are fed back by a preset question-answering large language model and correspond to the questions based on the auxiliary information.
2. The method for question-answering table data according to claim 1, wherein the steps of obtaining the history table data, extracting history name information from the history table data, and fine-tuning a pre-built large language model based on the history name information include:
acquiring history table data, and extracting history name information in the history table data based on an NLP method;
acquiring a data extraction format designed in a few-shot mode, and acquiring a ternary relation extraction problem constructed based on a prompt word;
fine-tuning a pre-constructed large language model based on historical name information, a data extraction format and a ternary relation extraction problem;
and deploying the trimmed large language model to a server.
3. The method for question-answering table data according to claim 1, wherein the steps of obtaining the table data to be processed, extracting the name information to be processed in the table data to be processed, and inputting the name information to be processed into the trimmed large language model to obtain the ternary relation prediction result include:
Acquiring to-be-processed form data, and extracting to-be-processed name information in the to-be-processed form data based on an NLP method;
inputting the name information to be processed into the trimmed large language model to obtain an output result;
and carrying out matching screening on the output result by using the regular expression to form a ternary relation prediction result.
4. The method for asking and answering table data according to claim 1, wherein the constructing an undirected graph based on the ternary relation prediction result, and obtaining element concepts according to the constructed undirected graph, specifically comprises:
constructing an undirected graph based on the ternary relation prediction result, wherein the constructed undirected graph comprises a plurality of nodes;
calculating the centrality of the graph of each node, and sequencing the calculation results of the centrality of the graph;
and obtaining the meta concept according to the sequencing result.
5. The method for asking and answering table data according to claim 1, wherein the knowledge graph is constructed based on the table data to be processed, the ternary relation prediction result and the meta concept, and specifically comprises:
the table data to be processed comprises one or more table files, and each table file comprises one or more workbooks;
and connecting the table file into a root node of the knowledge graph, taking the workbook corresponding to the table file and the meta-concepts corresponding to the workbook as nodes of the knowledge graph, and taking the content of the non-meta-concepts in the ternary relation prediction result as a relation to be used for connecting with the corresponding value in the knowledge graph to complete the construction of the knowledge graph.
6. The method for obtaining the question and answer of the form data according to claim 1, wherein the obtaining the question fed back in real time, preprocessing the question to obtain the preprocessed information, searching the constructed knowledge graph based on the preprocessed information to obtain the auxiliary information, and obtaining the answer corresponding to the question fed back by the preset question and answer language model based on the auxiliary information, specifically comprises the following steps:
acquiring a real-time feedback problem, and performing word segmentation on the problem to obtain preprocessing information;
searching the constructed knowledge graph based on the preprocessing information to obtain auxiliary information;
performing similarity matching on the auxiliary information and the preprocessing information by using an NLP method, and acquiring a preset first similarity threshold and a preset second similarity threshold;
when the similarity between the problem and the auxiliary information is more than or equal to a preset first similarity threshold value, constructing a first input instruction by adopting a thinking chain technology based on the auxiliary information and the problem;
and inputting the constructed first input instruction into a preset question-answering language model to obtain an answer corresponding to the question.
7. The method for asking and answering table data according to claim 6, wherein the method for matching the similarity between the auxiliary information and the preprocessing information by using the NLP method, and obtaining a preset similarity threshold value, further comprises:
When the second similarity threshold value is less than or equal to the first similarity threshold value, the similarity between the problem and the auxiliary information is less than the preset first similarity threshold value, an inquiry instruction is generated according to the auxiliary information;
acquiring selection information fed back by a user according to the inquiry instruction, and constructing a second input instruction by adopting a thinking chain technology based on the auxiliary information, the problems and the selection information;
and inputting the constructed second input instruction into a preset question-answering language model to obtain an answer corresponding to the question.
8. A form data question-answering apparatus, comprising:
the fine tuning module is used for acquiring the history table data, extracting history name information in the history table data and fine tuning the pre-constructed large language model based on the history name information;
the prediction module is used for acquiring the form data to be processed, extracting the name information to be processed in the form data to be processed, and inputting the name information to be processed into the trimmed large language model to obtain a ternary relation prediction result;
the acquisition module is used for constructing an undirected graph based on the ternary relation prediction result and acquiring a primitive concept according to the constructed undirected graph;
the construction module is used for constructing a knowledge graph based on the to-be-processed form data, the ternary relation prediction result and the meta concept;
And the dialogue module is used for acquiring the real-time feedback questions, preprocessing the questions to obtain preprocessed information, searching the constructed knowledge graph based on the preprocessed information, obtaining auxiliary information, and acquiring answers which are fed back by the preset question-answering large language model and correspond to the questions based on the auxiliary information.
9. A form data question-answering apparatus, characterized in that the form data question-answering apparatus comprises: a memory and at least one processor, the memory having instructions stored therein;
at least one of the processors invokes the instructions in the memory to cause the tabular data question answering device to perform the respective steps of the tabular data question answering method according to any one of claims 1-7.
10. A computer readable storage medium having instructions stored thereon, which when executed by a processor, implement the steps of the form data question-answering method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311691157.9A CN117390169B (en) | 2023-12-11 | 2023-12-11 | Form data question-answering method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311691157.9A CN117390169B (en) | 2023-12-11 | 2023-12-11 | Form data question-answering method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117390169A true CN117390169A (en) | 2024-01-12 |
| CN117390169B CN117390169B (en) | 2024-04-12 |
Family
ID=89468721
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311691157.9A Active CN117390169B (en) | 2023-12-11 | 2023-12-11 | Form data question-answering method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117390169B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118093597A (en) * | 2024-04-23 | 2024-05-28 | 浙江大学 | Table data reconstruction method and device and question-answering method |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021139283A1 (en) * | 2020-06-16 | 2021-07-15 | 平安科技（深圳）有限公司 | Knowledge graph question-answer method and apparatus based on deep learning technology, and device |
| CN113177124A (en) * | 2021-05-11 | 2021-07-27 | 北京邮电大学 | Vertical domain knowledge graph construction method and system |
| CN113254612A (en) * | 2021-05-24 | 2021-08-13 | 中国平安人寿保险股份有限公司 | Knowledge question-answering processing method, device, equipment and storage medium |
| CN113792153A (en) * | 2021-08-25 | 2021-12-14 | 北京百度网讯科技有限公司 | Question and answer recommendation method and device |
| US20230018489A1 (en) * | 2021-07-19 | 2023-01-19 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method for acquiring structured question-answering model, question-answering method and corresponding apparatus |
| CN115658912A (en) * | 2022-09-21 | 2023-01-31 | 安徽科技学院 | Method and system for optimizing knowledge graph and storage medium |
| US20230169101A1 (en) * | 2021-11-30 | 2023-06-01 | International Business Machines Corporation | Conversion of tabular format data to machine readable text for qa operations |
| CN116303971A (en) * | 2023-03-29 | 2023-06-23 | 重庆交通大学 | Few-sample form question-answering method oriented to bridge management and maintenance field |
| CN116821291A (en) * | 2023-06-16 | 2023-09-29 | 成都航天科工大数据研究院有限公司 | Question-answering method and system based on knowledge graph embedding and language model alternate learning |
| CN117033608A (en) * | 2023-09-28 | 2023-11-10 | 中国电子科技集团公司第十研究所 | Knowledge graph generation type question-answering method and system based on large language model |
| CN117033593A (en) * | 2023-08-09 | 2023-11-10 | 深圳阿尔法分子科技有限责任公司 | Method and system for enhancing question answering capability of large language model in specific field |
-
2023
- 2023-12-11 CN CN202311691157.9A patent/CN117390169B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021139283A1 (en) * | 2020-06-16 | 2021-07-15 | 平安科技（深圳）有限公司 | Knowledge graph question-answer method and apparatus based on deep learning technology, and device |
| CN113177124A (en) * | 2021-05-11 | 2021-07-27 | 北京邮电大学 | Vertical domain knowledge graph construction method and system |
| CN113254612A (en) * | 2021-05-24 | 2021-08-13 | 中国平安人寿保险股份有限公司 | Knowledge question-answering processing method, device, equipment and storage medium |
| US20230018489A1 (en) * | 2021-07-19 | 2023-01-19 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method for acquiring structured question-answering model, question-answering method and corresponding apparatus |
| CN113792153A (en) * | 2021-08-25 | 2021-12-14 | 北京百度网讯科技有限公司 | Question and answer recommendation method and device |
| US20230169101A1 (en) * | 2021-11-30 | 2023-06-01 | International Business Machines Corporation | Conversion of tabular format data to machine readable text for qa operations |
| CN115658912A (en) * | 2022-09-21 | 2023-01-31 | 安徽科技学院 | Method and system for optimizing knowledge graph and storage medium |
| CN116303971A (en) * | 2023-03-29 | 2023-06-23 | 重庆交通大学 | Few-sample form question-answering method oriented to bridge management and maintenance field |
| CN116821291A (en) * | 2023-06-16 | 2023-09-29 | 成都航天科工大数据研究院有限公司 | Question-answering method and system based on knowledge graph embedding and language model alternate learning |
| CN117033593A (en) * | 2023-08-09 | 2023-11-10 | 深圳阿尔法分子科技有限责任公司 | Method and system for enhancing question answering capability of large language model in specific field |
| CN117033608A (en) * | 2023-09-28 | 2023-11-10 | 中国电子科技集团公司第十研究所 | Knowledge graph generation type question-answering method and system based on large language model |
Non-Patent Citations (1)
| Title |
|---|
| 徐健;: "基于知识图谱构建5G协议知识库", 移动通信, no. 08, 15 August 2020 (2020-08-15) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118093597A (en) * | 2024-04-23 | 2024-05-28 | 浙江大学 | Table data reconstruction method and device and question-answering method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117390169B (en) | 2024-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111506721B (en) | Question-answering system and construction method for domain knowledge graph | |
| CN107590133A (en) | The method and system that position vacant based on semanteme matches with job seeker resume | |
| US20150074112A1 (en) | Multimedia Question Answering System and Method | |
| WO2019080863A1 (en) | Text sentiment classification method, storage medium and computer | |
| CN109325040B (en) | FAQ question-answer library generalization method, device and equipment | |
| CN111159363A (en) | Knowledge base-based question answer determination method and device | |
| CN110968684A (en) | Information processing method, device, equipment and storage medium | |
| CN117390169B (en) | Form data question-answering method, device, equipment and storage medium | |
| CN108733647B (en) | Word vector generation method based on Gaussian distribution | |
| CN108875065B (en) | Indonesia news webpage recommendation method based on content | |
| CN108038106B (en) | Fine-grained domain term self-learning method based on context semantics | |
| CN112035599A (en) | Query method and device based on vertical search, computer equipment and storage medium | |
| CN115470338B (en) | Multi-scenario intelligent question answering method and system based on multi-path recall | |
| Armouty et al. | Automated keyword extraction using support vector machine from Arabic news documents | |
| CN114329225A (en) | Search method, device, equipment and storage medium based on search statement | |
| CN110795544A (en) | Content search method, device, equipment and storage medium | |
| CN110781285A (en) | Scientific and technical literature retrieval intention construction method | |
| CN116258137A (en) | Text error correction method, device, equipment and storage medium | |
| CN109948154B (en) | Character acquisition and relationship recommendation system and method based on mailbox names | |
| CN114722176A (en) | Intelligent question answering method, device, medium and electronic equipment | |
| CN110728135A (en) | Text theme indexing method and device, electronic equipment and computer storage medium | |
| CN114139537A (en) | Word vector generation method and device | |
| CN114282513A (en) | Text semantic similarity matching method and system, intelligent terminal and storage medium | |
| Alqaraleh | Turkish Sentiment Analysis System via Ensemble Learning | |
| CN116644148A (en) | Keyword recognition method and device, electronic equipment and storage medium |
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 |