CN110727779A - Question-answering method and system based on multi-model fusion - Google Patents
Question-answering method and system based on multi-model fusion Download PDFInfo
- Publication number
- CN110727779A CN110727779A CN201910984294.9A CN201910984294A CN110727779A CN 110727779 A CN110727779 A CN 110727779A CN 201910984294 A CN201910984294 A CN 201910984294A CN 110727779 A CN110727779 A CN 110727779A
- Authority
- CN
- China
- Prior art keywords
- question
- original
- knowledge
- model
- candidate
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 70
- 230000004927 fusion Effects 0.000 title claims abstract description 54
- 238000004458 analytical method Methods 0.000 claims abstract description 40
- 239000013598 vector Substances 0.000 claims description 12
- 238000013507 mapping Methods 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 8
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 238000001914 filtration Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 4
- 238000013136 deep learning model Methods 0.000 abstract description 5
- 238000002347 injection Methods 0.000 abstract description 5
- 239000007924 injection Substances 0.000 abstract description 5
- 238000004891 communication Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
Images
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
-
- 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
- G06F16/355—Creation or modification of classes or clusters
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention discloses a question-answering method and a question-answering system based on multi-model fusion, wherein the method comprises the following steps: the method comprises the steps of constructing a knowledge base and a knowledge map, carrying out problem analysis on an input original problem by combining the knowledge base and the knowledge map to obtain problem analysis data, and retrieving the problem analysis data based on a matching method indicated by a fusion model to obtain a problem answer corresponding to the original problem. In the application, the process of searching the answer corresponding to the original problem by adopting the fusion model can realize higher efficiency on the premise of ensuring the accuracy, and the deep learning model is increasingly robust along with the increase of data volume, so that the whole model can more accurately understand the semantics, and the real 'intelligence injection' of the robot can be realized.
Description
Technical Field
The invention relates to the technical field of artificial intelligence question-answering systems, in particular to a question-answering method and system based on multi-model fusion.
Background
With the development of big data and deep learning techniques, it would no longer be a fantasy to create an automated man-machine dialog system as our personal assistant or chat partner.
Currently, people pay more and more attention to the dialog system in various fields, and the development of the dialog system is greatly promoted by the continuous progress of deep learning technology. For conversational systems, deep learning techniques may utilize large amounts of data to learn feature representation and reply generation strategies, where only a small amount of manual work is required. The existing robot based on the intention identification and the dialogue management needs a large amount of labeled corpora to perform the feature learning, the labeling process consumes a large amount of manpower and material resources, the execution efficiency of the whole system is influenced, the system maintenance is not facilitated, and the robustness and the growing performance of the system are influenced.
Disclosure of Invention
The embodiment of the invention provides a question-answering method and a question-answering system based on multi-model fusion, which can solve the problems.
The first aspect of the embodiments of the present invention provides a question-answering method based on multi-model fusion, which may include:
constructing a knowledge base and a knowledge map, wherein the knowledge base comprises a source knowledge base for storing source knowledge, a grammar base for storing synonyms, keyword sets and question templates and an FAQ question base for storing frequently asked question sets and answers, and the knowledge map is based on knowledge data sets of industry-related laws, regulations and management systems with label data;
performing problem analysis on an input original problem by combining a knowledge base and a knowledge graph to obtain problem analysis data, wherein the problem analysis data comprises key words in the original problem, entity names of questions, problem classification and possible problem templates to be matched;
and retrieving question analysis data based on a matching method indicated by a fusion model to obtain a question answer corresponding to the original question, wherein the fusion model comprises a knowledge graph reasoning model, a TF-IDF model, a Siamese model and a Bert model.
A second aspect of the embodiments of the present invention provides a question-answering system based on multi-model fusion, which may include:
the system comprises a basic data construction module, a knowledge base and a knowledge map, wherein the knowledge base comprises a source knowledge base for storing source knowledge, a grammar base for storing synonyms, keyword sets and question templates and an FAQ question base for storing frequently asked question sets and answers, and the knowledge map is based on the knowledge data set of industry-related laws, regulations and management systems with label data;
the system comprises an original question analysis module, a question analysis module and a question matching module, wherein the original question analysis module is used for carrying out question analysis on an input original question by combining a knowledge base and a knowledge graph to obtain question analysis data, and the question analysis data comprises key words in the original question, entity names of questions, question classifications and possible question templates to be matched;
and the answer searching module is used for retrieving question analysis data based on a matching method indicated by the fusion model to obtain the answer of the question corresponding to the original question, wherein the fusion model comprises a knowledge graph reasoning model, a TF-IDF model, a Siamese model and a Bert model.
A third aspect of embodiments of the present invention provides a computer device, which includes a processor and a memory, where the memory stores at least one instruction, at least one program, a code set, or a set of instructions, and the at least one instruction, the at least one program, the code set, or the set of instructions is loaded and executed by the processor to implement the multi-model fusion-based question-answering method according to the above aspect.
A fourth aspect of the embodiments of the present invention provides a computer storage medium, where at least one instruction, at least one program, a code set, or a set of instructions is stored in the computer storage medium, and the at least one instruction, the at least one program, the code set, or the set of instructions is loaded and executed by a processor to implement the multi-model fusion-based question-answering method according to the above aspect.
In the embodiment of the invention, the four models are fused, and a specific method for fusion is provided, so that the model with strong interpretability and large difference is realized. The process of searching the answer corresponding to the original problem by adopting the fusion model has higher efficiency on the premise of ensuring the accuracy, and the deep learning model becomes more and more robust along with the increase of data volume, so that the whole model has more accurate understanding on the semantics, and the real 'intelligence injection' on the robot is realized.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flow chart of a question-answering method based on multi-model fusion according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a predicted answer using a fusion model according to an embodiment of the present invention;
FIG. 3 is a flowchart illustrating a predicted answer of the fusion model according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of similarity calculation provided by an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a question-answering system based on multi-model fusion according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a computer device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The terms "including" and "having," and any variations thereof, in the description and claims of this invention and the above-described drawings are intended to cover a non-exclusive inclusion, and the terms "first" and "second" are used for distinguishing designations only and do not denote any order or magnitude of a number. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those steps or elements listed, but may alternatively include other steps or elements not listed, or inherent to such process, method, article, or apparatus.
It should be noted that the question-answering method based on multi-model fusion provided by the application can be applied to the application scene of the knowledge training question-answering in the company.
In the embodiment of the present invention, the question-answering method based on multi-model fusion may be applied to a Computer device, where the Computer device may be a terminal such as a smart phone, a tablet Computer, a PC (Personal Computer), or other electronic devices with computing processing capability.
As shown in fig. 1, the question-answering method based on multi-model fusion at least includes the following steps:
and S101, constructing a knowledge base and a knowledge graph.
Specifically, the computer device needs to construct a knowledge base and a knowledge map before searching answers, wherein the knowledge base can comprise a source knowledge base for storing source knowledge, a grammar base for storing synonyms, keyword sets and question templates, and an FAQ question base for storing frequently asked question sets and answers, and the knowledge map can be a knowledge map of a knowledge data set based on industry-related laws, regulations and management systems with tagged data.
In one implementation, the computer device may collect relevant questions of training routines to construct question examples, classify the question examples according to the relations and attributes related to the question examples, recognize the participles using the named entities, extract simplified question examples only containing keywords after removing stop words, and store the simplified question examples in a file. Preferably, the device may extract a unique word as a professional word in the domain and add the word to the segmentation dictionary.
In one implementation, the computer device may attribute the simplified problem instances relating to the same relationship or attribute as the same class, and may determine the keywords in the simplified problem template of the same class as sets of synonyms, each set of synonyms being understood to correspond to a relationship or attribute in the knowledge-graph.
In one implementation, the computer device may configure each question template with a path to extract an answer to the question template from the knowledge graph and constraints to extract the answer from the original question. Optionally, the format of the constraint may be field name + parameter, where the parameter is a synonym, and the value of the parameter may be a word or a word in the original question replaced by the word set name.
In one implementation, a computer device may perform named entity recognition on a simplified problem sample, build a syntax tree using recognized participles and corresponding part-of-speech tags, classify the problem sample using the structure of the syntax tree, divide the problem into a plurality of major classes, continue dividing each major class into a plurality of minor classes, build an index for each minor class problem, build a mapping between a keyword and a problem minor class index, and assign the keyword to a problem minor class by mapping; the mapping relation is a problem template, and a problem template library is formed. It will be appreciated that entities in the named entity identification may be replaced with wildcards.
In an alternative embodiment, the device may further count answers and answers of high-frequency questions to form a frequently asked question set, and store frequently asked questions and related answers into a question library, where each question corresponds to a standard answer, i.e., a mapping between the questions and the answers, so as to construct an FAQ question library.
And S102, carrying out problem analysis on the input original problem by combining the knowledge base and the knowledge graph to obtain problem analysis data.
It will be appreciated that the question resolution data may include keywords in the original question, the entity name of the question, the question category and possibly the question template to be matched. In specific implementation, the computer device may pre-process the input original question based on the knowledge base and the knowledge graph, determine a keyword corresponding to the original question, further determine a question classification corresponding to the keyword of the original question based on the question template base, then judge an actual number of questions asked corresponding to the original question, and allocate a question template to be matched to the original question according to the actual number of questions asked.
In a preferred implementation, the device may first determine whether the original question sentence has a misspelling, and if so, correct the original question sentence by combining spelling rules with the algorithm of the language model, and output the correct sentence, otherwise, directly output the user sentence. Further, the output user sentence can be subjected to filtering of symbols such as punctuation and/or filtering of stop words, conversion of synonyms and conversion of numbers into Chinese data.
In one implementation, the device may further perform named entity recognition on the original problem to obtain a corresponding entity name, and then link the entity name to a map node of the knowledge map based on the entity link, and if the entity name cannot be linked to the map node, use all the problems in the FAQ question library as a template of the problem to be matched.
In an optional implementation manner, the device may perform word segmentation on the remaining part of the original problem except the entity name by using a dictionary containing professional words, match words obtained by the word segmentation in a grammar library, and determine corresponding keywords.
Further, the device may match the keyword with a question template in a question template library, search for a possible question classification corresponding to the keyword, and if the question classification can be located, all questions in the question templates that meet the keyword in the question classification are taken as possible question templates to be matched; if the specific classification of the question cannot be located, all questions of the FAQ question bank are used as the question template to be matched. It can be understood that, in the process of matching the keywords with the question templates, each keyword may be respectively matched with all question templates, all keywords may be simultaneously matched with all questions, or a part of keywords may be selected to be matched with all text question templates.
Furthermore, the device can record and judge the current question for the second time, and for the first question, the question template corresponding to the specific classification of the question asked for the first time can be used as the question template to be matched, so as to retrieve the corresponding answer; the method comprises the steps of reading K preceding sentence questioning data aiming at question not to be questioned for the first time, judging whether the question is a question or a plurality of questions, increasing constraint conditions to ask the questions one by one in a reverse mode if the questions are a plurality of questions, extracting questioning keywords after the questions are clarified after positive answers are obtained, repositioning question classification, and generating a plurality of question templates to be matched.
S103, retrieving the question analysis data based on a matching method indicated by the fusion model to obtain a question answer corresponding to the original question.
In a specific implementation, the fusion model can be a model after a knowledge graph inference model, a TF-IDF model, a Siamese model and a Bert model are fused.
In this application, a method for generating a candidate answer set may be determined according to the number of the to-be-matched question templates, for example, when there are only 1 to-be-matched question templates and corresponding triples are retrieved from the knowledge graph, the device may use the triples as one of the candidate questions, and determine the graph candidate answer set corresponding to the candidate question, and have a higher weight value.
In one implementation, when the number of the problem templates to be matched is more than one, or the map nodes cannot be located from the knowledge map, the specific method for extracting the answers includes: calculating the similarity of the texts of the original problem and the problem template to be matched by adopting a fusion model, and generating a plurality of candidate answers; further, selecting the frequently asked questions with the maximum similarity as candidate questions, matching the original questions to the candidate questions if the similarity between the original questions and the question templates to be matched is greater than a similarity threshold, otherwise, outputting prompt information which cannot answer the questions if the answers of the questions cannot be matched; furthermore, according to the mapping relation between the matched candidate question and the question and answer indicated by the FAQ question bank, the corresponding answer is selected as the answer of the original question.
In a preferred implementation, a specific process of calculating the similarity between the original question and the short text of the question template to be matched by the device using the fusion model may be as shown in fig. 2, and includes: generating a sparse matrix corresponding to a TF-IDF vector of an original question based on a TF-IDF model, calculating the similarity between the sparse matrix and a question template to be matched, ordering the questions ranked from high to low by K according to the similarity to serve as a first candidate answer set, and recording the similarity weight Ai; further, calculating the similarity between the original problem and a problem template to be matched by adopting a Simese model, returning the problem with the confidence degree from high to low ranked K as a candidate answer set, and recording the similarity weight Bi; further, calculating semantic similarity between the original question and the question template to be matched by adopting a Bert model, sorting according to the semantic similarity to obtain the most similar K questions as a third candidate answer set, and recording the similarity weight Ci; further, fusing the map candidate answer set, the first candidate answer set, the second candidate answer set and the third candidate answer set based on the similarity weight, and training corresponding two-classification neural networks according to the word vectors and the word vectors of the result original question and the candidate question; and finally, the probability of the positive class obtained by SoftMax is used as the probability of the candidate answer by adopting the two-class neural network, the final candidate answer is output in a probability descending order, and the highest confidence level is taken as the final answer to be output.
It should be noted that, in the process of model fusion, the matching text itself is added as one of the inputs, and in the process of considering the weight, the sentence itself is considered at the same time, so that the number of features in the network structure is increased, and the network of the fusion model has better robustness.
In a specific implementation manner of the present application, the training recognition process of the Siamese model may be as shown in fig. 3 and fig. 4:
the siernese Network has two sub-networks which are identical in structure and share weights. Two inputs X1 and X2 are received, respectively, and converted into vectors Gw (X1) and Gw (X2), and then a distance Ew between the two output vectors is calculated by some distance metric. The training sample used for training the Siamese Network is a tuple (X1, X2, y), and the label y ═ 0 indicates that X1 and X2 are of different types (dissimilar, not repeated, depending on the application scenario). And y is 1, which means that X1 is of the same type as X2 (similar). Compared with the traditional embedding of word-level or char-level, the method adopts word + char merging input; the BiGPU is used as a main body of the twin neural network, and compared with the BilSTM, the BiGPU has better performance on a data set and higher training efficiency; the similarity of the vector is calculated by adopting the Manhattan space distance, and compared with cosine, the similarity has stronger interpretability on the text and better performance.
In the embodiment of the invention, the four models are fused, and a specific method for fusion is provided, so that the model with strong interpretability and large difference is realized. The process of searching the answer corresponding to the original problem by adopting the fusion model has higher efficiency on the premise of ensuring the accuracy, and the deep learning model becomes more and more robust along with the increase of data volume, so that the whole model has more accurate understanding on the semantics, and the real 'intelligence injection' on the robot is realized.
The question-answering system based on multi-model fusion provided by the embodiment of the invention will be described with reference to fig. 5. It should be noted that the question-answering system shown in fig. 5 is used for executing the method according to the embodiment of the present invention shown in fig. 1, and for convenience of description, only the portion related to the embodiment of the present invention is shown, and details of the technology are not disclosed, please refer to the embodiment of the present invention shown in fig. 1.
Referring to fig. 5, a schematic structural diagram of a question-answering system based on multi-model fusion is provided for an embodiment of the present invention. As shown in fig. 5, the question-answering system 10 of the embodiment of the present invention may include: a basic data construction module 101, an original question analysis module 102 and an answer search module 103.
The basic data construction module 101 is used for constructing a knowledge base and a knowledge graph, wherein the knowledge base comprises a source knowledge base for storing source knowledge, a grammar base for storing synonyms, keyword sets and question templates, and an FAQ question base for storing frequently asked question sets and answers, and the knowledge graph is based on knowledge data sets of industry-related laws, regulations and management systems with label data.
The original question analysis module 102 is configured to perform question analysis on an input original question by combining a knowledge base and a knowledge graph to obtain question analysis data, where the question analysis data includes keywords in the original question, an entity name of a question, a question classification, and a possible question template to be matched.
And the answer searching module 103 is used for retrieving question analysis data based on a matching method indicated by a fusion model to obtain a question answer corresponding to the original question, wherein the fusion model comprises a knowledge graph reasoning model, a TF-IDF model, a Siamese model and a Bert model.
It should be noted that, in the system embodiment corresponding to the fourth above-mentioned method embodiment of this embodiment, a process of cooperatively implementing answer search between different modules in this embodiment is the same as that in the above-mentioned embodiment, and the implementation process may refer to the detailed description of the above-mentioned method embodiment, and is not described herein again.
In the embodiment of the invention, the four models are fused, and a specific method for fusion is provided, so that the model with strong interpretability and large difference is realized. The process of searching the answer corresponding to the original problem by adopting the fusion model has higher efficiency on the premise of ensuring the accuracy, and the deep learning model becomes more and more robust along with the increase of data volume, so that the whole model has more accurate understanding on the semantics, and the real 'intelligence injection' on the robot is realized.
An embodiment of the present invention further provides a computer storage medium, where the computer storage medium may store a plurality of instructions, where the instructions are suitable for being loaded by a processor and executing the method steps in the embodiments shown in fig. 1 to 4, and a specific execution process may refer to specific descriptions of the embodiments shown in fig. 1 to 4, which are not described herein again.
The embodiment of the application also provides computer equipment. As shown in fig. 6, the computer device 20 may include: the at least one processor 201, e.g., CPU, the at least one network interface 204, the user interface 203, the memory 205, the at least one communication bus 202, and optionally, a display 206. Wherein a communication bus 202 is used to enable the connection communication between these components. The user interface 203 may include a touch screen, a keyboard or a mouse, among others. The network interface 204 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface), and a communication connection may be established with the server via the network interface 204. The memory 205 may be a high-speed RAM memory or a non-volatile memory (non-volatile memory), such as at least one disk memory, and the memory 205 includes a flash in the embodiment of the present invention. The memory 205 may optionally be at least one memory system located remotely from the processor 201. As shown in fig. 6, memory 205, which is a type of computer storage medium, may include an operating system, a network communication module, a user interface module, and program instructions.
It should be noted that the network interface 204 may be connected to a receiver, a transmitter or other communication module, and the other communication module may include, but is not limited to, a WiFi module, a bluetooth module, etc., and it is understood that the computer device in the embodiment of the present invention may also include a receiver, a transmitter, other communication module, etc.
Processor 201 may be used to call program instructions stored in memory 205 and cause computer device 20 to perform the following operations:
constructing a knowledge base and a knowledge map, wherein the knowledge base comprises a source knowledge base for storing source knowledge, a grammar base for storing synonyms, keyword sets and question templates and an FAQ question base for storing frequently asked question sets and answers, and the knowledge map is based on knowledge data sets of industry-related laws, regulations and management systems with label data;
performing problem analysis on an input original problem by combining a knowledge base and a knowledge graph to obtain problem analysis data, wherein the problem analysis data comprises key words in the original problem, entity names of questions, problem classification and possible problem templates to be matched;
and retrieving question analysis data based on a matching method indicated by a fusion model to obtain a question answer corresponding to the original question, wherein the fusion model comprises a knowledge graph reasoning model, a TF-IDF model, a Siamese model and a Bert model.
In some embodiments, apparatus 20, when constructing the grammar library and the knowledge graph, is specifically configured to:
constructing a knowledge graph of a knowledge data set based on industry-related laws, regulations and management systems with labeled data, wherein texts and categories of each record in the knowledge data set are in one-to-one correspondence;
and preprocessing the collected problem samples to obtain simplified problem samples, and establishing a grammar tree corresponding to a grammar library by using word segmentation and part-of-speech tagging.
In some embodiments, the device 20 is specifically configured to, when preprocessing the collected question samples to obtain simplified question samples and building a syntax tree corresponding to the syntax library by using the word segmentation and the part-of-speech tagging,:
classifying the collected problem instances according to the corresponding relation and attribute of the problem instances, identifying word segmentation by using a named entity, removing stop words and extracting simplified problem instances only containing keywords;
classifying simplified problem examples relating to the same relationship or attribute into a problem template of the same class;
configuring a path for extracting an answer of the question template from the knowledge graph and a constraint condition for extracting the answer from the original question for each question template;
and carrying out named entity recognition on the simplified problem sample, establishing a grammar tree by adopting the recognized participles and corresponding part-of-speech labels, and forming a problem template library by the mapping relation between the keywords indicated by the grammar tree and the problem subclass indexes.
In some embodiments, the apparatus 20, when performing problem resolution on the input original problem by combining the knowledge base and the knowledge graph to obtain problem resolution data, is specifically configured to:
preprocessing the input original problem based on a knowledge base and a knowledge graph, and determining a keyword corresponding to the original problem;
determining a problem classification corresponding to a keyword of an original problem based on a problem template library;
and judging the actual question asking times corresponding to the original question, and distributing a question template to be matched for the original question according to the actual question asking times.
In some embodiments, the apparatus 20, when preprocessing the input original question based on the knowledge base and the knowledge graph and determining the keyword corresponding to the original question, is specifically configured to:
when the input original problem is determined to be free of spelling errors, punctuation filtering and stop word filtering are carried out on the original problem, and synonym conversion is adopted to convert numbers in the original problem into Chinese data;
carrying out named entity recognition on an original problem to obtain a corresponding entity index, and linking the entity index to a map node of a knowledge map based on entity link;
and (3) segmenting the rest parts except the entity names in the original problem by adopting a dictionary containing professional words, matching the words obtained by segmenting in a grammar library, and determining corresponding keywords.
In some embodiments, apparatus 20 is further configured to:
and when the entity designation cannot be linked to the map node, determining all the problems in the FAQ problem library as the problem template to be matched.
In some embodiments, when the device 20 retrieves the question analysis data based on the matching method indicated by the fusion model to obtain the question answer corresponding to the original question, it is specifically configured to:
when the number of the problem templates to be matched is more than 1 or the problem templates to be matched cannot be linked to the map nodes, calculating the similarity of the original problem and the short texts of the problem templates to be matched by adopting a fusion model, and generating a plurality of candidate answers;
selecting a frequently asked question with the largest similarity as a candidate question, and matching the original question with the candidate question if the similarity between the original question and a question template to be matched is greater than a similarity threshold value;
and selecting the corresponding answer as the answer of the original question according to the mapping relation between the matched candidate question and the question and answer indicated by the FAQ question library.
In some embodiments, apparatus 20 is further configured to:
and when the number of the problem templates to be matched is only 1 and the corresponding triple is retrieved from the knowledge graph spectrum, taking the triple problem as a candidate problem, and determining the graph candidate answer set corresponding to the candidate problem.
In some embodiments, the device 20 is specifically configured to, when calculating the similarity between the original question and the short text of the question template to be matched by using the fusion model to generate a plurality of candidate answers:
generating a sparse matrix corresponding to a TF-IDF vector of an original question based on a TF-IDF model, calculating the similarity between the sparse matrix and a question template to be matched, ordering the questions ranked from high to low by K according to the similarity to serve as a first candidate answer set, and recording the similarity weight Ai;
calculating the similarity between the original question and the first candidate answer set by adopting a Simese model, returning the question with the confidence coefficient from high to low and K before ranking as a candidate answer set, and recording the similarity weight Bi;
calculating semantic similarity between every two original problems and the problem template to be matched by adopting a Bert model, obtaining the most similar K problems as a third candidate answer set after sorting according to the semantic similarity, and recording similarity weight Ci;
fusing the chart candidate answer set, the first candidate answer set, the second candidate answer set and the third candidate answer set based on the similarity weight, and training corresponding two-classification neural networks according to the character vectors and the word vectors of the result original question and the candidate question;
and (3) adopting a binary neural network to take the probability of the positive class obtained by SoftMax as the probability of the candidate answer, outputting the final candidate answer in a probability descending order, and taking the highest confidence level as the final answer to output.
In the embodiment of the invention, the four models are fused, and a specific method for fusion is provided, so that the model with strong interpretability and large difference is realized. The process of searching the answer corresponding to the original problem by adopting the fusion model has higher efficiency on the premise of ensuring the accuracy, and the deep learning model becomes more and more robust along with the increase of data volume, so that the whole model has more accurate understanding on the semantics, and the real 'intelligence injection' on the robot is realized.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.
The above disclosure is only for the purpose of illustrating the preferred embodiments of the present invention, and it is therefore to be understood that the invention is not limited by the scope of the appended claims.
Claims (10)
1. A question-answering method based on multi-model fusion is characterized by comprising the following steps:
constructing a knowledge base and a knowledge graph, wherein the knowledge base comprises a source knowledge base for storing source knowledge, a grammar base for storing synonyms, keyword sets and question templates and an FAQ question base for storing frequently asked question sets and answers, and the knowledge graph is based on knowledge data sets of industry-related laws, regulations and management systems with label data;
performing problem analysis on the input original problem by combining the knowledge base and the knowledge graph to obtain problem analysis data, wherein the problem analysis data comprises keywords in the original problem, entity names of questions, problem classification and possible problem templates to be matched;
and retrieving the question analysis data based on a matching method indicated by a fusion model to obtain a question answer corresponding to the original question, wherein the fusion model comprises a knowledge graph reasoning model, a TF-IDF model, a Siamese model and a Bert model.
2. The method of claim 1, wherein constructing the grammar library and the knowledge graph comprises:
constructing a knowledge graph of a knowledge data set based on industry-related laws, regulations and management systems with labeled data, wherein texts and categories of each record in the knowledge data set are in one-to-one correspondence;
and preprocessing the collected problem samples to obtain simplified problem samples, and establishing a grammar tree corresponding to a grammar library by using word segmentation and part-of-speech tagging.
3. The method of claim 2, wherein preprocessing the collected question samples to obtain simplified question samples, and building a syntax tree corresponding to a syntax library using word segmentation and part-of-speech tagging comprises:
classifying the collected problem instances according to the corresponding relation and attribute of the problem instances, identifying word segmentation by using a named entity, removing stop words and extracting simplified problem instances only containing keywords;
classifying simplified problem examples relating to the same relationship or attribute into a problem template of the same class;
configuring a path for extracting an answer of each question template from the knowledge graph and a constraint condition for extracting the answer from the original question for each question template;
and carrying out named entity recognition on the simplified problem sample, establishing a grammar tree by adopting the recognized participles and corresponding part-of-speech labels, and forming a problem template library by the mapping relation between the keywords indicated by the grammar tree and the problem subclass indexes.
4. The method of claim 3, wherein performing problem resolution on the input original problem in combination with the knowledge base and the knowledge graph to obtain problem resolution data comprises:
preprocessing the input original problem based on the knowledge base and the knowledge graph, and determining a keyword corresponding to the original problem;
determining a question classification corresponding to the keywords of the original question based on the question template library;
and judging the actual question asking times corresponding to the original question, and distributing a question template to be matched for the original question according to the actual question asking times.
5. The method according to claim 4, wherein the preprocessing the input original question based on the knowledge base and the knowledge graph, and determining the keyword corresponding to the original question comprises:
when it is determined that the input original problem is not misspelled, punctuation filtering, stop word filtering, synonym conversion and digit conversion are carried out on the original problem to Chinese data; (ii) a
Conducting named entity recognition on the original problem to obtain a corresponding entity designation, and linking the entity designation to a map node of the knowledge map based on entity link;
and segmenting the rest parts except the entity names in the original problem by adopting a dictionary containing professional words, matching the words obtained by segmenting in the grammar library and determining corresponding keywords.
6. The method of claim 5, further comprising:
and when the entity designation cannot be linked to the graph node, determining all the questions in the FAQ question bank as question templates to be matched.
7. The method according to claim 5, wherein the retrieving the question analysis data based on the matching method indicated by the fusion model to obtain the answer to the question corresponding to the original question comprises:
when the number of the problem templates to be matched is more than 1 or the problem templates to be matched cannot be linked to the map nodes, calculating the similarity of the texts of the original problem and the problem templates to be matched by adopting a fusion model, and generating a plurality of candidate answers;
selecting a frequently asked question with the largest similarity as a candidate question, and matching the original question with the candidate question if the similarity between the original question and the question template to be matched is greater than a similarity threshold value;
and selecting corresponding answers as answers of the original questions according to the mapping relation between the matched candidate questions and the questions and answers indicated by the FAQ question bank.
8. The method of claim 7, further comprising:
and when the number of the problem templates to be matched is only 1 and the corresponding triples are retrieved from the knowledge graph, taking the triples as candidate problems and determining a graph candidate answer set corresponding to the candidate problems.
9. The method according to claim 7, wherein the calculating the similarity of the texts between the original question and the question template to be matched by using the fusion model to generate a plurality of candidate answers comprises:
generating a sparse matrix corresponding to a TF-IDF vector of an original question based on a TF-IDF model, calculating the similarity between the sparse matrix and the question template to be matched, ranking the questions K before from high to low according to the similarity as a first candidate answer set, and recording the similarity weight Ai;
calculating the similarity between the original question and the first candidate answer set by adopting a Simese model, returning the question with the confidence coefficient from high to low and K before ranking as a candidate answer set, and recording the similarity weight Bi;
calculating semantic similarity between every two original questions and the question templates to be matched by adopting a Bert model, obtaining the most similar K questions as a third candidate answer set after sorting according to the semantic similarity, and recording similarity weight Ci;
fusing the map candidate answer set, the first candidate answer set, the second candidate answer set and the third candidate answer set based on similarity weight, and training corresponding two-classification neural networks according to word vectors and word vectors of the original question and the candidate question;
and adopting the two-classification neural network to take the probability of the positive class obtained by SoftMax as the probability of the candidate answer, outputting the final candidate answer in a probability descending order, and taking the highest confidence level as the final answer to output.
10. A question-answering system based on multi-model fusion is characterized by comprising:
the system comprises a basic data construction module, a knowledge base and a knowledge map, wherein the knowledge base comprises a source knowledge base for storing source knowledge, a grammar base for storing synonyms, keyword sets and question templates, and an FAQ question base for storing frequently asked question sets and answers, and the knowledge map is based on knowledge data sets of industry-related laws, regulations and management systems with label data;
the system comprises an original question analysis module, a question analysis module and a question matching module, wherein the original question analysis module is used for carrying out question analysis on an input original question by combining the knowledge base and the knowledge graph to obtain question analysis data, and the question analysis data comprises keywords in the original question, entity names of questions, question classification and possible question templates to be matched;
and the answer searching module is used for retrieving the question analysis data based on a matching method indicated by a fusion model to obtain the question answer corresponding to the original question, wherein the fusion model comprises a knowledge graph reasoning model, a TF-IDF model, a Siamese model and a Bert model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910984294.9A CN110727779A (en) | 2019-10-16 | 2019-10-16 | Question-answering method and system based on multi-model fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910984294.9A CN110727779A (en) | 2019-10-16 | 2019-10-16 | Question-answering method and system based on multi-model fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110727779A true CN110727779A (en) | 2020-01-24 |
Family
ID=69221335
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910984294.9A Pending CN110727779A (en) | 2019-10-16 | 2019-10-16 | Question-answering method and system based on multi-model fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110727779A (en) |
Cited By (40)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111259987A (en) * | 2020-02-20 | 2020-06-09 | 民生科技有限责任公司 | Method for extracting event main body based on BERT (belief-based regression analysis) multi-model fusion |
| CN111339277A (en) * | 2020-02-28 | 2020-06-26 | 中国工商银行股份有限公司 | Question-answer interaction method and device based on machine learning |
| CN111444724A (en) * | 2020-03-23 | 2020-07-24 | 腾讯科技（深圳）有限公司 | Medical question-answer quality testing method and device, computer equipment and storage medium |
| CN111460119A (en) * | 2020-03-27 | 2020-07-28 | 海信集团有限公司 | Intelligent question and answer method and system for economic knowledge and intelligent equipment |
| CN111552819A (en) * | 2020-04-28 | 2020-08-18 | 腾讯科技（深圳）有限公司 | Entity extraction method and device and readable storage medium |
| CN111625658A (en) * | 2020-07-28 | 2020-09-04 | 杭州翔毅科技有限公司 | Voice interaction method, device and equipment based on knowledge graph and storage medium |
| CN111681765A (en) * | 2020-04-29 | 2020-09-18 | 华南师范大学 | A multi-model fusion method for medical question answering system |
| CN111708703A (en) * | 2020-06-18 | 2020-09-25 | 深圳前海微众银行股份有限公司 | Test case set generation method, apparatus, device, and computer-readable storage medium |
| CN111782786A (en) * | 2020-07-01 | 2020-10-16 | 中关村科学城城市大脑股份有限公司 | Multi-model fusion question-answering method, system and medium for urban brain |
| CN111858850A (en) * | 2020-06-29 | 2020-10-30 | 银盛支付服务股份有限公司 | A Method for Accurate and Fast Scoring of Questions and Answers on Intelligent Customer Service |
| CN111949783A (en) * | 2020-08-12 | 2020-11-17 | 中国银行股份有限公司 | Question and answer result generation method and device in knowledge base |
| CN112101423A (en) * | 2020-08-22 | 2020-12-18 | 上海昌投网络科技有限公司 | Multi-model fused FAQ matching method and device |
| CN112131353A (en) * | 2020-10-12 | 2020-12-25 | 吉林大学 | Method for solving full text interpretation analysis problem |
| CN112182180A (en) * | 2020-09-27 | 2021-01-05 | 京东方科技集团股份有限公司 | Question and answer processing method, electronic device, and computer-readable medium |
| CN112380343A (en) * | 2020-11-19 | 2021-02-19 | 平安科技（深圳）有限公司 | Problem analysis method, problem analysis device, electronic device and storage medium |
| CN112463944A (en) * | 2020-12-22 | 2021-03-09 | 安徽商信政通信息技术股份有限公司 | Retrieval type intelligent question-answering method and device based on multi-model fusion |
| CN112614024A (en) * | 2020-12-30 | 2021-04-06 | 成都数之联科技有限公司 | Case fact based intelligent law strip recommendation method, system, device and medium |
| CN112632225A (en) * | 2020-12-29 | 2021-04-09 | 天津汇智星源信息技术有限公司 | Semantic searching method and device based on case and event knowledge graph and electronic equipment |
| CN112632265A (en) * | 2021-03-10 | 2021-04-09 | 北京沃丰时代数据科技有限公司 | Intelligent machine reading understanding method and device, electronic equipment and storage medium |
| CN112651224A (en) * | 2020-12-24 | 2021-04-13 | 天津大学 | Intelligent search method and device for engineering construction safety management document text |
| CN112685434A (en) * | 2020-12-21 | 2021-04-20 | 福建新大陆软件工程有限公司 | Operation and maintenance question-answering method based on knowledge graph |
| CN112700016A (en) * | 2020-12-28 | 2021-04-23 | 济南浪潮高新科技投资发展有限公司 | Predictive maintenance method and apparatus |
| CN112749265A (en) * | 2021-01-08 | 2021-05-04 | 哈尔滨工业大学 | Intelligent question-answering system based on multiple information sources |
| CN112949674A (en) * | 2020-08-22 | 2021-06-11 | 上海昌投网络科技有限公司 | Multi-model fused corpus generation method and device |
| CN113139615A (en) * | 2021-05-08 | 2021-07-20 | 北京联合大学 | Unmanned environment target detection method based on embedded equipment |
| CN113515643A (en) * | 2021-05-19 | 2021-10-19 | 思必驰科技股份有限公司 | User identity verification method and device |
| CN113971403A (en) * | 2021-09-23 | 2022-01-25 | 西安电子科技大学 | Entity identification method and system considering text semantic information |
| CN114238598A (en) * | 2021-12-07 | 2022-03-25 | 北京妙医佳健康科技集团有限公司 | Question-answering system and labeling, auditing and model training method thereof |
| CN114860913A (en) * | 2022-05-24 | 2022-08-05 | 北京百度网讯科技有限公司 | Intelligent question-answering system construction method, question-answering processing method and device |
| CN114996419A (en) * | 2022-05-09 | 2022-09-02 | 成都数之联科技股份有限公司 | Intelligent question answering method and device for weapon equipment, electronic equipment and storage medium |
| CN115470338A (en) * | 2022-10-27 | 2022-12-13 | 之江实验室 | A multi-scene intelligent question answering method and system based on multi-channel recall |
| CN115858760A (en) * | 2023-03-02 | 2023-03-28 | 广州嘉为科技有限公司 | Knowledge graph question-answer reasoning method, device and storage medium |
| US11783131B2 (en) | 2020-09-10 | 2023-10-10 | International Business Machines Corporation | Knowledge graph fusion |
| CN117194602A (en) * | 2023-09-06 | 2023-12-08 | 书音(上海)文化科技有限公司 | Local knowledge base updating method and system based on large language model and BERT model |
| CN117407515A (en) * | 2023-12-15 | 2024-01-16 | 湖南三湘银行股份有限公司 | Answer system based on artificial intelligence |
| CN117688424A (en) * | 2023-11-24 | 2024-03-12 | 华南师范大学 | Method, system, device and medium for classifying teaching data generated by retrieval enhancement |
| CN117891929A (en) * | 2024-03-18 | 2024-04-16 | 南京华飞数据技术有限公司 | Knowledge graph intelligent question-answer information identification method of improved deep learning algorithm |
| CN117972108A (en) * | 2024-01-26 | 2024-05-03 | 建龙西林钢铁有限公司 | Knowledge graph construction method and system based on steel mill safety production |
| CN118170892A (en) * | 2024-05-13 | 2024-06-11 | 山东第一医科大学附属省立医院(山东省立医院) | Method and device for assisting interpretation of imaging examination in physical examination report |
| CN119088945A (en) * | 2024-01-25 | 2024-12-06 | 浙江孚宝智能科技有限公司 | A large-scale language model question answering system based on health care knowledge |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778256A (en) * | 2015-04-20 | 2015-07-15 | 江苏科技大学 | Rapid incremental clustering method for domain question-answering system consultations |
| CN108804521A (en) * | 2018-04-27 | 2018-11-13 | 南京柯基数据科技有限公司 | A kind of answering method and agricultural encyclopaedia question answering system of knowledge based collection of illustrative plates |
| CN109308321A (en) * | 2018-11-27 | 2019-02-05 | 烟台中科网络技术研究所 | A knowledge question answering method, knowledge question answering system and computer readable storage medium |
| CN109829042A (en) * | 2018-12-28 | 2019-05-31 | 北京科技大学 | A kind of knowledge Q-A system and method based on Biomedical literature |
-
2019
- 2019-10-16 CN CN201910984294.9A patent/CN110727779A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778256A (en) * | 2015-04-20 | 2015-07-15 | 江苏科技大学 | Rapid incremental clustering method for domain question-answering system consultations |
| CN108804521A (en) * | 2018-04-27 | 2018-11-13 | 南京柯基数据科技有限公司 | A kind of answering method and agricultural encyclopaedia question answering system of knowledge based collection of illustrative plates |
| CN109308321A (en) * | 2018-11-27 | 2019-02-05 | 烟台中科网络技术研究所 | A knowledge question answering method, knowledge question answering system and computer readable storage medium |
| CN109829042A (en) * | 2018-12-28 | 2019-05-31 | 北京科技大学 | A kind of knowledge Q-A system and method based on Biomedical literature |
Non-Patent Citations (2)
| Title |
|---|
| WATARU SAKATA，等: "FAQ Retrieval using Query-Question Similarity and BERT-Based Query-Answer Relevance", 《HTTPS://WWW.RESEARCHGATE.NET/PUBLICATION.332960569》 * |
| 陈柯锦等: "基于多尺度相似度特征的答案选择算法", 《系统工程与电子技术》 * |
Cited By (54)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111259987A (en) * | 2020-02-20 | 2020-06-09 | 民生科技有限责任公司 | Method for extracting event main body based on BERT (belief-based regression analysis) multi-model fusion |
| CN111259987B (en) * | 2020-02-20 | 2023-12-29 | 民生科技有限责任公司 | Method for extracting event main body by multi-model fusion based on BERT |
| CN111339277A (en) * | 2020-02-28 | 2020-06-26 | 中国工商银行股份有限公司 | Question-answer interaction method and device based on machine learning |
| CN111444724A (en) * | 2020-03-23 | 2020-07-24 | 腾讯科技（深圳）有限公司 | Medical question-answer quality testing method and device, computer equipment and storage medium |
| CN111460119B (en) * | 2020-03-27 | 2024-04-12 | 海信集团有限公司 | Intelligent question-answering method and system for economic knowledge and intelligent equipment |
| CN111460119A (en) * | 2020-03-27 | 2020-07-28 | 海信集团有限公司 | Intelligent question and answer method and system for economic knowledge and intelligent equipment |
| CN111552819A (en) * | 2020-04-28 | 2020-08-18 | 腾讯科技（深圳）有限公司 | Entity extraction method and device and readable storage medium |
| CN111552819B (en) * | 2020-04-28 | 2022-07-15 | 腾讯科技（深圳）有限公司 | Entity extraction method and device and readable storage medium |
| CN111681765A (en) * | 2020-04-29 | 2020-09-18 | 华南师范大学 | A multi-model fusion method for medical question answering system |
| CN111681765B (en) * | 2020-04-29 | 2023-08-11 | 华南师范大学 | Multi-model fusion method of medical question-answering system |
| WO2021253904A1 (en) * | 2020-06-18 | 2021-12-23 | 深圳前海微众银行股份有限公司 | Test case set generation method, apparatus and device, and computer readable storage medium |
| CN111708703A (en) * | 2020-06-18 | 2020-09-25 | 深圳前海微众银行股份有限公司 | Test case set generation method, apparatus, device, and computer-readable storage medium |
| CN111858850A (en) * | 2020-06-29 | 2020-10-30 | 银盛支付服务股份有限公司 | A Method for Accurate and Fast Scoring of Questions and Answers on Intelligent Customer Service |
| CN111782786B (en) * | 2020-07-01 | 2022-02-15 | 中关村科学城城市大脑股份有限公司 | Multi-model fusion question-answering method, system and medium for urban brain |
| CN111782786A (en) * | 2020-07-01 | 2020-10-16 | 中关村科学城城市大脑股份有限公司 | Multi-model fusion question-answering method, system and medium for urban brain |
| CN111625658A (en) * | 2020-07-28 | 2020-09-04 | 杭州翔毅科技有限公司 | Voice interaction method, device and equipment based on knowledge graph and storage medium |
| CN111949783A (en) * | 2020-08-12 | 2020-11-17 | 中国银行股份有限公司 | Question and answer result generation method and device in knowledge base |
| CN112101423A (en) * | 2020-08-22 | 2020-12-18 | 上海昌投网络科技有限公司 | Multi-model fused FAQ matching method and device |
| CN112949674A (en) * | 2020-08-22 | 2021-06-11 | 上海昌投网络科技有限公司 | Multi-model fused corpus generation method and device |
| US11783131B2 (en) | 2020-09-10 | 2023-10-10 | International Business Machines Corporation | Knowledge graph fusion |
| CN112182180A (en) * | 2020-09-27 | 2021-01-05 | 京东方科技集团股份有限公司 | Question and answer processing method, electronic device, and computer-readable medium |
| CN112131353A (en) * | 2020-10-12 | 2020-12-25 | 吉林大学 | Method for solving full text interpretation analysis problem |
| CN112380343B (en) * | 2020-11-19 | 2024-06-25 | 平安科技（深圳）有限公司 | Problem analysis method, device, electronic equipment and storage medium |
| CN112380343A (en) * | 2020-11-19 | 2021-02-19 | 平安科技（深圳）有限公司 | Problem analysis method, problem analysis device, electronic device and storage medium |
| CN112685434A (en) * | 2020-12-21 | 2021-04-20 | 福建新大陆软件工程有限公司 | Operation and maintenance question-answering method based on knowledge graph |
| CN112463944A (en) * | 2020-12-22 | 2021-03-09 | 安徽商信政通信息技术股份有限公司 | Retrieval type intelligent question-answering method and device based on multi-model fusion |
| CN112463944B (en) * | 2020-12-22 | 2023-10-24 | 安徽商信政通信息技术股份有限公司 | Search type intelligent question-answering method and device based on multi-model fusion |
| CN112651224A (en) * | 2020-12-24 | 2021-04-13 | 天津大学 | Intelligent search method and device for engineering construction safety management document text |
| CN112700016A (en) * | 2020-12-28 | 2021-04-23 | 济南浪潮高新科技投资发展有限公司 | Predictive maintenance method and apparatus |
| CN112632225A (en) * | 2020-12-29 | 2021-04-09 | 天津汇智星源信息技术有限公司 | Semantic searching method and device based on case and event knowledge graph and electronic equipment |
| CN112632225B (en) * | 2020-12-29 | 2022-08-30 | 天津汇智星源信息技术有限公司 | Semantic searching method and device based on case and event knowledge graph and electronic equipment |
| CN112614024A (en) * | 2020-12-30 | 2021-04-06 | 成都数之联科技有限公司 | Case fact based intelligent law strip recommendation method, system, device and medium |
| CN112614024B (en) * | 2020-12-30 | 2024-03-08 | 成都数之联科技股份有限公司 | Legal intelligent recommendation method, system, device and medium based on case facts |
| CN112749265A (en) * | 2021-01-08 | 2021-05-04 | 哈尔滨工业大学 | Intelligent question-answering system based on multiple information sources |
| CN112632265A (en) * | 2021-03-10 | 2021-04-09 | 北京沃丰时代数据科技有限公司 | Intelligent machine reading understanding method and device, electronic equipment and storage medium |
| CN113139615A (en) * | 2021-05-08 | 2021-07-20 | 北京联合大学 | Unmanned environment target detection method based on embedded equipment |
| CN113515643A (en) * | 2021-05-19 | 2021-10-19 | 思必驰科技股份有限公司 | User identity verification method and device |
| CN113971403A (en) * | 2021-09-23 | 2022-01-25 | 西安电子科技大学 | Entity identification method and system considering text semantic information |
| CN114238598A (en) * | 2021-12-07 | 2022-03-25 | 北京妙医佳健康科技集团有限公司 | Question-answering system and labeling, auditing and model training method thereof |
| CN114996419A (en) * | 2022-05-09 | 2022-09-02 | 成都数之联科技股份有限公司 | Intelligent question answering method and device for weapon equipment, electronic equipment and storage medium |
| CN114860913A (en) * | 2022-05-24 | 2022-08-05 | 北京百度网讯科技有限公司 | Intelligent question-answering system construction method, question-answering processing method and device |
| CN114860913B (en) * | 2022-05-24 | 2023-12-12 | 北京百度网讯科技有限公司 | Intelligent question and answer system construction method, question and answer processing method and device |
| CN115470338A (en) * | 2022-10-27 | 2022-12-13 | 之江实验室 | A multi-scene intelligent question answering method and system based on multi-channel recall |
| CN115858760A (en) * | 2023-03-02 | 2023-03-28 | 广州嘉为科技有限公司 | Knowledge graph question-answer reasoning method, device and storage medium |
| CN117194602A (en) * | 2023-09-06 | 2023-12-08 | 书音(上海)文化科技有限公司 | Local knowledge base updating method and system based on large language model and BERT model |
| CN117194602B (en) * | 2023-09-06 | 2024-04-19 | 书音(上海)文化科技有限公司 | Local knowledge base updating method and system based on large language model and BERT model |
| CN117688424A (en) * | 2023-11-24 | 2024-03-12 | 华南师范大学 | Method, system, device and medium for classifying teaching data generated by retrieval enhancement |
| CN117407515A (en) * | 2023-12-15 | 2024-01-16 | 湖南三湘银行股份有限公司 | Answer system based on artificial intelligence |
| CN117407515B (en) * | 2023-12-15 | 2024-04-30 | 湖南三湘银行股份有限公司 | Answer system based on artificial intelligence |
| CN119088945A (en) * | 2024-01-25 | 2024-12-06 | 浙江孚宝智能科技有限公司 | A large-scale language model question answering system based on health care knowledge |
| CN117972108A (en) * | 2024-01-26 | 2024-05-03 | 建龙西林钢铁有限公司 | Knowledge graph construction method and system based on steel mill safety production |
| CN117891929B (en) * | 2024-03-18 | 2024-05-17 | 南京华飞数据技术有限公司 | Knowledge graph intelligent question-answer information identification method of improved deep learning algorithm |
| CN117891929A (en) * | 2024-03-18 | 2024-04-16 | 南京华飞数据技术有限公司 | Knowledge graph intelligent question-answer information identification method of improved deep learning algorithm |
| CN118170892A (en) * | 2024-05-13 | 2024-06-11 | 山东第一医科大学附属省立医院(山东省立医院) | Method and device for assisting interpretation of imaging examination in physical examination report |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110727779A (en) | Question-answering method and system based on multi-model fusion | |
| CN116628172B (en) | Dialogue method for multi-strategy fusion in government service field based on knowledge graph | |
| CN111708703B (en) | Test case set generation method, device, equipment and computer readable storage medium | |
| CN111353310B (en) | Named entity identification method and device based on artificial intelligence and electronic equipment | |
| CN111026842B (en) | Natural language processing method, natural language processing device and intelligent question-answering system | |
| CN109960728B (en) | Method and system for identifying named entities of open domain conference information | |
| CN111639171A (en) | Knowledge graph question-answering method and device | |
| CN116795973B (en) | Text processing method and device based on artificial intelligence, electronic equipment and medium | |
| WO2022048194A1 (en) | Method, apparatus and device for optimizing event subject identification model, and readable storage medium | |
| CN112395421B (en) | Course label generation method and device, computer equipment and medium | |
| CN116991875B (en) | SQL sentence generation and alias mapping method and device based on big model | |
| CN113742733A (en) | Reading comprehension vulnerability event trigger word extraction and vulnerability type identification method and device | |
| CN113468311B (en) | Knowledge graph-based complex question and answer method, device and storage medium | |
| CN118132669A (en) | Intelligent indexing method based on large language model | |
| CN113505786A (en) | Test question photographing and judging method and device and electronic equipment | |
| CN112035668A (en) | Event subject recognition model optimization method, device and equipment and readable storage medium | |
| CN112036169A (en) | Event recognition model optimization method, device and equipment and readable storage medium | |
| CN111708870A (en) | Deep neural network-based question answering method and device and storage medium | |
| AU2019290658B2 (en) | Systems and methods for identifying and linking events in structured proceedings | |
| CN116186220A (en) | Information retrieval method, question and answer processing method, information retrieval device and system | |
| CN113505889B (en) | Processing method and device of mapping knowledge base, computer equipment and storage medium | |
| CN111178080A (en) | Named entity identification method and system based on structured information | |
| CN117291192B (en) | Government affair text semantic understanding analysis method and system | |
| CN118395987A (en) | BERT-based landslide hazard assessment named entity identification method of multi-neural network | |
| CN117828024A (en) | Plug-in retrieval method, device, storage medium and equipment |
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 | ||
| CB02 | Change of applicant information |
Address after: Xinyada technology building, 3888 Jiangnan Avenue, Binjiang District, Hangzhou City, Zhejiang Province 310000 Applicant after: Sinyada Technology Co.,Ltd. Address before: Xinyada technology building, 3888 Jiangnan Avenue, Binjiang District, Hangzhou City, Zhejiang Province 310000 Applicant before: SUNYARD SYSTEM ENGINEERING Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200124 |
|
| RJ01 | Rejection of invention patent application after publication |