48 results about "Knowledge base question answering" patented technology

The invention discloses a dialogue generation model training method, which comprises the following steps: obtaining a dialogue data set, the dialogue data in the dialogue data set comprising questionsand annotation replies corresponding to the questions; based on the questions in the dialogue data set, obtaining the coded representation of the question through an encoder in a constructed dialoguegeneration model; fusing the coded representation of the question and knowledge information of transfer learning of the question from a knowledge base question and answer model through a decoder in the dialogue generation model to obtain a prediction reply corresponding to the question output by the dialogue generation model; and determining a loss function based on the prediction reply and the annotation reply corresponding to the question, and adjusting parameters of the dialogue generation model through the loss function until the loss function of the dialogue generation model is convergent. According to the method, the knowledge information can be better fused by the model, so that the dialogue generation accuracy and reasonability are improved. The invention further discloses a dialogue generation method and device, equipment and a medium.

The invention discloses an end-to-end context-based knowledge base question and answer method and device, and the method and device consider the connection relationship between entities and relationships in a knowledge base, enable the two tasks including entity links and relationship prediction contained in the knowledge base to be mutually promoted, and jointly improve the two tasks. The methodmainly comprises the steps of prreprocessing a problem and removing special symbols; constructing a candidate entity set related to the problem based on the knowledge base, and constructing a candidate relationship set according to the relationship of the candidate entities in the knowledge base; for each entity in each candidate set, extracting the context of the entity in the question; dividingthe candidate relationships into different granularities; predicting a subject entity and a predicate relationship based on the CERM model; and finding the object entity in the knowledge base as an answer to return by using a predicted subject entity and relationship. Entity links and relationship prediction in knowledge base questions and answers are integrated into a unified prediction model. Joint prediction of the subjective entities and relationships is achieved, and accuracy of questions and answers is improved.

The invention belongs to the technical field of data processing, and discloses a knowledge base question and answer entity linking method and system based on similarity, and the method comprises the steps that entities in a question are recognized through a deep learning method, and the end-to-end entity linking is carried out; in the candidate entity generation stage, named entity identificationis carried out by using a Bert feature extraction network and a BiLSTM-CRF sequence labeling model to generate candidate entities; in the disambiguation stage of the candidate entities, relational words in the questions are extracted through a certain rule and sorted according to the similarity between the relational words and the candidate relations, and the question and answer time of the knowledge base is shortened. According to the method, an end-to-end thought is applied to knowledge base questioning and answering, knowledge base questioning and answering questions are combined with an advanced computer technology, and named entity recognition is performed by using a Bert feature extraction network and a BiLSTM-CRF sequence labeling model to generate candidate entities. According to the method, the problem that the candidate entities are polysemous in one word is relieved, and the entity linking accuracy is improved.

The invention discloses a questioning and answering method and system based on BERT and a knowledge base, is applied to the field of information retrieval. A named entity recognition model based on BERT-CRF and a language model and a text similarity dichotomy model based on BERT and the language model in order to overcome at the defects of an existing knowledge base questioning and answering system. the two models are trained, the two trained models are adopted to process the linguistic data of the to-be-answered questions, correct answers to the questions can be obtained, and answers are automatically rewritten.

The invention provides a method and a system for improving question and answer accuracy of a knowledge base. The method comprises the following steps: obtaining a user question to be answered, extracting a theme entity in the user question, retrieving the knowledge base by using the theme entity, taking path information of each obtained candidate answer as a candidate path, and preprocessing the user question to obtain vector representation of the user question; scoring each step relationship on the candidate path by utilizing the vector representation by utilizing an attention mechanism to obtain a relationship confidence coefficient of each step relationship on the candidate path, and summing all relationship confidence coefficients on the candidate path to obtain a path confidence coefficient of the relationship path; and sorting all the candidate paths according to the set path confidence coefficients, and outputting the candidate path with the highest path confidence coefficient as an answer result of the user question. According to the invention, the effect of the intermediate node in the whole relationship inference is enhanced, and the accuracy of the relationship inferenceis improved.

The invention provides an automatic template generation method for knowledge base questions and answers. The automatic template generation method comprises a relational dictionary construction processand a process of automatically generating a question template and a query template according to question answer pairs. The relational dictionary construction process is used for constructing a relational dictionary for enabling natural language phrases to correspond to knowledge base relations on the basis of a large number of corpora. The process of automatically generating the question templateand the query template according to the question-answer pair comprises the following steps: firstly, obtaining a query graph from a knowledge base according to the question-answer pair, aligning a natural language question with the query graph, and automatically generating the question template and the query template on the basis of alignment. The method does not depend on manual template construction, can solve the problem of automatic generation of templates of knowledge base questions and answers, is automatic and efficient in template generation process, solves the problems that a traditional manual template construction method is high in template generation cost and limited in template number, and facilitates development of a series of subsequent applications (such as natural language knowledge base questions and answers).

The invention discloses a semantic similarity calculation method based on FSM multi-round questions and answers. According to the semantic similarity calculation method based on FSM multi-round questions and answers, according to user input questions, the user input questions and knowledge base question and answer pair data into a Transformer DSSM semantic similarity calculation model to carry outmultiple rounds of matching, and a candidate answer is returned t the user; the problem that a traditional customer service system often consumes a large number of human resources under the conditions of insufficient service peak people and insufficient valley people is solved, and the efficiency of obtaining answers of common questions in related fields by the user is improved.

The invention belongs to the technical field of network natural language processing. The invention discloses a Chinese intelligent question and answer system method based on word similarity of a network platform. In a knowledge base question and answer system, each question and each answer are regarded as two word sets, each word in the question set is matched with each word in the answer set, theword similarity is calculated, then the maximum similarity value is obtained, and then the average value of the maximum value is obtained; The method is simple and high in efficiency; According to the method, the problem of data sparsity of an existing vector included angle cosine method is solved; Meanwhile, the situation of inaccurate answer extraction caused by inconsistent question and answerlanguage structures in the existing mode matching method is also overcome; According to the word similarity algorithm, answers can be found in the knowledge base question and answer system more reasonably and efficiently.

The invention provides an entity linking method in a knowledge base question answering system, which comprises the following steps: obtaining a subject term set from a question sentence; searching inthe knowledge base according to the obtained set of subject terms to obtain a preliminary set of candidate entities; For each entity in the preliminary candidate entity set, extracting a correspondingfeature from the entity, the question and the knowledge base; and obtaining a score of the entity according to the extracted feature of each entity in the preliminary candidate entity set, and obtaining a candidate entity set according to the score. The invention improves the accuracy and efficiency of the entity link.

The invention discloses a knowledge base question answering device and a method for establishing the knowledge base question answering device. The device comprises an index module, which is used for offline establishing a knowledge base with sentence vector index according to semantic vectors; A receiving module, configured to receive a user question and answer request; And a question and answer retrieval module, configured to retrieve answers to the questions and answers on line according to the user question and answer request in the knowledge base. The present application solves the technical problem of low efficiency in indexing a question answering knowledge base. As that index of the question answer knowledge base is established by the semantic vector, the efficiency is improved andthe index establishment time is reduced. In addition, the knowledge base question answering device of the present application can be applied to a question answering system for answering user questions.

The invention discloses a knowledge base question-answering method fusing multiple loss functions and an attention mechanism. According to the method, questions and candidate answers are used as input, Bi-LSTM and Bi-GRU are used as main feature extractors, an attention mechanism is integrated, two loss functions are used for optimizing a model, model parameters are updated by calculating back propagation of loss values, and a network model is trained until the model converges; and finally, the questions and the candidate answers are mapped to a feature space with the same dimension through network training, the semantic similarity of the questions and the candidate answers is calculated by using the inner product between the feature vectors of the questions and the candidate answers, andthe difference between different answers is expanded by using the cosine similarity between the candidate answers. Through testing on a SimpleQuestions data set, the model has relatively strong feature mapping capability and relatively high accuracy, so that the superiority of the method is proved.

The invention discloses an entity linking method based on RoBERTa and a heuristic algorithm. The method comprises the following steps that a sequence labeling model based on a pre-trained language model RoBERTa is used for labeling a problem, and the RoBERTa model stacks 12 layers of transcriber structures to obtain multi-level grammatical semantic information in the problem; dynamic representation of each word in the question based on context is obtained through a multi-head attention mechanism in transformator, and then the entity mention range in the question is obtained; and after the entity mention range is obtained, directly matching the entity mention with the knowledge base entity by using a heuristic algorithm to complete entity linking. The method can be applied to various knowledge base questioning and answering scenes, and provides underlying basic services for many advanced applications.

The invention provides a knowledge base question-answering method in the electric power field, which comprises the steps of performing part-of-speech tagging and syntactic analysis on acquired questions in the electric power field to obtain question expressions of the questions in the electric power field; identifying the question expression of the power field question based on a pre-constructed key phrase identification model to obtain a key entity and a key attribute phrase of the power field question; based on the key entities and the key attribute phrases, carrying out retrieval in a powerdomain knowledge base, and obtaining the question answers, wherein the key phrase recognition model comprises training open domain questions to obtain key entities and key attribute phrases corresponding to the open domain questions. According to the method and the system, high dependence of electric power domain model training on terms in the prior art is avoided, and knowledge base questions and answers in the electric power domain can be achieved only by utilizing easily obtained open domain resources.

The invention discloses a query relaxation method for question and answer of a knowledge base, which is characterized by comprising the following steps: segmenting a result-free SPARQL statement, analyzing query conditions, and extracting predicates of an inference rule to be learned to form a predicate set; for the predicate set obtained in the previous step, obtaining an inference rule set and aconfidence calculation model of each predicate; supplementing corresponding predicates based on the reasoning rule set obtained in the above steps, recombining query conditions, and querying candidate results; and scoring and sorting the candidate results based on the confidence calculation model obtained in the above steps, reserving a part of high-confidence results as final results, and outputting an inference rule enabling each result to be established. According to the method, efficient and accurate result prediction of the SPARQL statement without the query result is realized.

The invention discloses a knowledge base questioning and answering method based on deep learning, and the method comprises the following steps: performing theme entity identification on natural language questions of a user to identify a plurality of theme entities; performing weight assignment according to the plurality of theme entities to obtain a plurality of central entities with different weights; selecting candidate answer paths according to the plurality of central entities with different weights, and calculating a similarity total score; sorting and weighting the candidate answer pathsaccording to the similarity total score to obtain a plurality of candidate answer paths with different weights; and performing function matching calculation on the plurality of central entities withdifferent weights and the plurality of candidate answer paths with different weights to obtain a final answer, and feeding back the final answer to the user. According to the method, the problem thata traditional question-answering method identifies wrong theme entities or cannot identify theme entities is solved, the error rate of a theme entity identification model is reduced, the accuracy of an attribute relationship detection model is improved, and the question-answering accuracy of a whole knowledge base is improved.

The invention is suitable for the technical field of knowledge bases, and provides a knowledge base question and answer extraction method and system, a mobile terminal and a storage medium, and the method comprises the steps: obtaining manual question and answer data, and carrying out the question recognition of the manual question and answer data, so as to obtain a client questioning sentence; carrying out response query in the manual question and answer data according to the client question to acquire customer service answer information, wherein a plurality of customer service answering sentences are stored in the customer service answer information; calculating question-answer matching degrees between the customer questioning sentences and the customer service answering sentences respectively, and setting the customer service answering sentence corresponding to the maximum value in the question-answer matching degrees and the customer questioning sentences as question-answer pairs;and extracting and storing the question-answer pairs in a question-answer knowledge base. According to the method, based on automatic acquisition of the client questioning sentences and the customer service answering sentences and calculation design of the question-answer matching degree between the client questioning sentences and the corresponding customer service answering sentences, high-quality question-answer pairs can be automatically extracted from real manual question and answer data, and the extraction efficiency of the question-answer pairs is improved.

The invention relates to a knowledge base question-answering method fusing fact texts, and belongs to the field of natural language processing. According to the invention, natural language questions and candidate answer triples are analyzed respectively, entities, entity types and relationships in the triples are converted into fact texts; the natural language questions and the fact texts are mapped into numerical vectors in a low-dimensional semantic space through a pre-training language model BERT, and then cosine similarity is adopted for calculation and sorting, so that the knowledge basequestion-answering method model fusing the fact text is established; the model can learn the score relation between the natural language questions and the candidate answer triples so as to find the answer most similar to the semantics of the natural language questions in the knowledge base, and the question-answering method achieves a good effect.

The invention relates to a knowledge base question-answering method fusing a multi-head-attention mechanism and relative position encoding, belonging to the field of natural language processing. According to the invention, a Transformer encoder is introduced to replace BiLSTM to encode questions; meanwhile, due to the structure problem of Transformer, the Transformer encoder is insufficient in capability of acquiring information of relative position words in sentences; and the relative position encoding thought in Transform-XL is adopted in the invention, an absolute position encoding formula used in Transformer is rewritten, relative position encoding is used for replacing absolute position encoding, and thus, the situation that the capacity of acquiring the information of relative position words is insufficient is made up for.

An attribute matching method in knowledge base questions and answers based on a neural network comprises the following steps of replacing entities in questions, generating candidate attributes according to the entities, and sending the segmented words of the candidate attributes to a word embedding layer in the neural network; learning semantic representation of the upper and lower questions by using bidirectional LSTM; calculating according to the word vector representation of the questions and attributes to obtain a word meaning similarity matrix, and similarly, obtaining a semantic similarity matrix according to semantic representation; taking the maximum values of the two similarity matrixes from the question direction and the attribute direction respectively to obtain four vectors, and then obtaining the similarity of the question and the attribute through a full connection layer; and selecting the highest similarity and the attribute corresponding to the highest similarity, if the similarity is greater than a threshold, adding the attribute, and replacing the text in the question to perform the next round of attribute matching. According to the method, the final similarity iscalculated by considering the context semantic representation and the word meaning representation of the questions and the attributes, so that the attribute matching accuracy is improved; corresponding predicate texts in the questions can be positioned, and the multi-hop problem can be iteratively processed.

The invention relates to a question decomposition type semantic analysis method fusing a fact text, and belongs to the technical field of natural language processing. The method comprises the following steps of: dividing into three stages of decomposition, extraction and analysis, firstly decomposing a complex problem into a simple sub-problem sequence, then extracting key information in an original problem and the sub-problems, and finally generating a structured query statement by integrating the information. Meanwhile, in order to avoid the situation of entity judgment errors or subject entity missing in the decomposition process, triples in the knowledge base are converted into fact text information described by a natural language, a fact text base is constructed, an attention mechanism is adopted to obtain richer knowledge, and the purposes of enhancing entity representation information and reducing error propagation are achieved. According to the method, the fact text information is fused, semantic analysis is carried out on the complex problem by adopting a question decomposition mode, the understanding ability of the question and answer model on the complex problem is improved, and therefore the problem that the processing effect of a knowledge base question and answer technology on the complex problem is poor is solved.

The invention discloses a question answering method and device, an electronic device and a storage medium, and relates to the field of knowledge base question answering. The specific implementation scheme is as follows: determining a natural language question template matched with a natural language question, wherein the natural language question template can be matched with at least one type of natural language question; searching a first language question template corresponding to the natural language question template according to a corresponding relationship between the natural language question template and the first language question template; and generating a first language question corresponding to the natural language question by adopting the first language question template. Thenatural language question template adopted by the embodiment of the invention can be matched with at least one type of natural question template, the business retrieval requirement can be used withoutcompiling a large number of templates, and the labor and time cost is saved.

The invention discloses a dialogue knowledge base question and answer implementation method. The method comprises the steps that 1, a fuzzy reasoning grammar system is defined in advance; 2, obtaining a current question to be answered and historical dialogue information, and obtaining vector representation through a GloVe model; step 3, obtaining a hidden state vector of the encoder; step 4, identifying the named entity and the type thereof from the feature representation; step 5, obtaining an entity ID in the knowledge base; step 6, obtaining a logic form of the problem under the current grammar framework; 7, predicting a relation ID and an entity type ID in the problem; 8, obtaining a query statement which can be executed on the knowledge base; and 9, executing the query statement in the knowledge base to obtain an answer corresponding to the current question. Compared with the prior art, the method has the advantages that the intention of the user question in the dialogue is automatically identified, the response accuracy in the dialogue is improved, the method is suitable for open domain knowledge graph question answering, and the user experience satisfaction can be improved.

The invention discloses a complex question knowledge base question answering method based on embedded and candidate subgraph pruning, and belongs to the technical field of data processing. The complexity of a relation is distinguished based on dependency syntactic analysis, and a candidate subgraph range is preliminarily screened out; the candidate sub-graphs are pruned through a pruning method based on tail entities and relation types to reduce interference brought by error paths in the candidate sub-graphs during model training; a short text matching model based on a neural network is trained, so that the matching score of a question and a correct question and answer path context is relatively high. When a new question and answer data set is constructed by SPE-QA, the complexity degree of a relationship in the question is analyzed based on a dependency syntax, and a candidate subgraph range is preliminarily screened out; a relation path type selector is trained, and the candidate sub-graphs is further pruned; and a short text matching model based on the neural network is constructed, so that the matching score of the question and the correct question and answer path context is relatively high.

The embodiment of the invention provides a text query method and device integrating retrieval and machine reading understanding, a readable storage medium and computing equipment, and aims to realizehigh-precision search and directly extract answers from search results and return the answers to a user. The method comprises the steps of receiving a query request of a user, wherein the query request comprises a query text; conducting searching according to the query text to obtain a preset first number of candidate documents; inputting the preset first number of candidate documents and the query text into a preset binary classification model, and selecting a preset second number of candidate documents from the preset first number of candidate documents; inputting the preset second number ofcandidate documents and the query text into a preset paragraph extraction reading understanding model, and selecting a preset third number of paragraphs or sentences from the preset second number ofcandidate documents; and returning the preset third number of paragraphs or sentences to the user.

The invention discloses a question and answer method and system, electronic equipment and a storage medium, and relates to the technical field of knowledge maps. The specific implementation scheme isas follows: receiving input problem information; analyzing the problem information to obtain a plurality of keywords in the problem information; obtaining a keyword attribute list according to the plurality of keywords, wherein the keyword attribute list has a target attribute corresponding to each keyword; obtaining a candidate list of the question information, the candidate list having a plurality of answer options; and according to the target attribute corresponding to each keyword in the keyword attribute list, screening the plurality of answer options in the candidate list to obtain a correct answer. According to the knowledge base question-answering method based on decision screening, decision screening can be conducted on the answer options according to the keyword attributes by referring to the thought of the exclusion method, so that question answering of choice questions of complex knowledge reasoning types can be simplified through the knowledge base question-answering method based on decision screening, the accuracy is guaranteed, and the recall rate of correct answers is increased.

The invention discloses a knowledge base question and answer method, electronic equipment and a readable storage medium. The method comprises the following steps: step 1, constructing a corpus for a specific field; 2, training an MA-B model to classify questions to be answered; step 3, extracting user questions with the same field identifiers as the to-be-answered questions, and calculating similarity scores of the to-be-answered questions and the user questions; 4, determining similar questions of the to-be-answered questions by using the similarity scores, and taking answers corresponding to the similar questions as candidate answers; step 5, calculating a correlation score between each candidate answer and the question to be answered, and determining a final answer based on the correlation score and the similarity score; the invention is small in calculation amount, and the obtained final answer is highly matched with the to-be-answered question.

The invention discloses a knowledge base question-answering system construction method based on template matching and deep learning, and the method comprises the following steps: S1, designing and constructing a question-answering template, wherein when the question-answering template is designed, question-answering completeness must be achieved, each question possibly asked by a user must contain a corresponding question-answering template, and this part requires a designer to fully investigate business problems; s2, designing and constructing an ontology map, and designing the ontology map according to the entity data, the relational data, the scene business and the intention template; s3, constructing a marking layer; s4, constructing a trigger layer; s5, constructing a matching layer; s6, constructing an alignment layer; and S7, constructing a query layer. According to the invention, a mode of combining template matching and model prediction and combining ES search and model prediction is used, so that the coverage rate and the accuracy rate of the question-answering system are higher, the robustness of the question-answering system is enhanced, the diversity of questions is considered, the range and the form of questions and answers are expanded, and the question-answering system becomes richer.

The invention discloses a knowledge base question-answering system and device based on question generation, and relates to an automatic question-answering system. The invention aims to solve the problems of high labeling cost, large workload and long consumed time due to the fact that a knowledge graph-based question-answering method needs professional knowledge personnel to label a special data set. A template database of the system is used for storing templates; the triple extension module reads the triple and analyzes the triple, and all templates under the relationship are selected from the template library; symbols corresponding to the triples in the template are replaced with entities to generate sentences; the full-text retrieval module is used for segmenting a query text queried bya user, converting query statements segmented into words into Luene internal representation Query objects, and retrieving a group of sentences related to user query as a candidate set; and the semantic matching module is used for sorting the candidate sets by adopting a semantic matching network based on a pre-training model Bert, taking a triple corresponding to the highest score as an answer and returning the answer to the user. The invention is mainly used for realizing automatic questioning and answering.

The invention discloses a data processing method and device, equipment and a medium, which are used for solving the problem of low accuracy of response information determined by existing knowledge base questions and answers. According to the embodiment of the invention, after the target attribute information matched with the inquiry information is determined, the response information corresponding to the inquiry information is determined by comprehensively considering the target attribute information and the question type of the inquiry information, so that the determined response information is closer to the content to be inquired by the inquiry information, the accuracy of the response information is improved, and the user experience is improved. And the user experience is improved.