727 results about "Relationship extraction" patented technology

The invention relates to a character relationship graph construction method based on integration of an ontology and multiple neural networks. The method comprises the following steps: crawling data related to a character in a certain domain in the Internet; establishing a domain character ontology; extracting data from a structured data table which contains multiple types of entities and has repeated entities to construct a standardized entity table; matching the two class names of the character ontology model with the two entity table names through a semantic mapping algorithm, automaticallyobtaining all entity relationships, and storing the entity relationships in a Neo4j database in a graph structure; for the text data in the structured table, carrying out character entity recognitionand relationship extraction by using a sliding window, entity position characteristics and a bidirectional gating recurrent neural network; and updating the current graph structure of the newly addedrelationship to form a domain character relationship knowledge graph. The character relationship advanced features can be extracted from the original relational data and the text data, manual design is not needed, the recognition effect is improved, and the efficiency of constructing the character relationship graph by the complex webpage text is improved.

The invention discloses a text entity relationship extraction method and a model training method, and can be applied to a natural language processing technology in the field of artificial intelligence; according to the invention, the graph state recurrent neural network and the BERT model are combined, a first vector used for representing semantic features of the text and a second vector used forrepresenting dependency relationship features of the text are extracted from the text; the first vector and the second vector are spliced and then classified; the relationship extraction of the entitypairs is enabled to obtain relatively high accuracy in long sentence and cross-sentence application scenes; the problem of insufficient accuracy in application scenes such as long sentences and crosssentences in the prior art is solved, in addition, in the model training stage, based on the preset rules and the pre-training model, a large amount of annotation data is produced in a remote supervision mode, and a large amount of accurate training data can be obtained at a low cost. Therefore, the method can be widely applied to the natural language processing technology.

The invention discloses a knowledge network-based text indexing system and method. The text indexing system comprises a single text feature extraction unit, a multi-text word relation extraction unit, a knowledge tree generating unit, a knowledge tree application unit and a knowledge base storage unit. The text indexing method comprises the following steps of: partitioning words in a text input to the text indexing system, and acquiring text feature words in the text; deducing a class word TAG corresponding to the text according to node positions of a knowledge tree corresponding to the text feature words; and judging the validity of the TAG through a judgment type model based on the TAG, then extracting a reliable TAG word set, and repositioning a text feature word set through the reliable TAG word set to form a reliable text feature word set. According to the system and the method, content word extraction, class labeling and phrase extraction are integrated, so that the extraction effects can be mutually promoted; and the semantics of the words are expressed through the nodes of the knowledge network, so that different meanings are reduced.

The invention discloses a medical entity relationship extraction method based on feature fusion, and the method comprises the steps: enabling entities in a knowledge base to be aligned to medical corpora through a remote supervision and rule combination method, and constructing an entity pair sentence set; performing word-level vector coding on the sentences based on a convolutional neural networkmodel to obtain overall feature vector representation of the sentences; extracting features in left and right subtree directions on the shortest dependency path of the sentences by using a recurrentneural network respectively, and performing splicing operation; and fusing the sentence overall features and the dependency syntax features which are extracted from the two parts respectively, and performing final relation extraction on the obtained fusion features. According to the method, on the premise that a dependency syntax structure is utilized; entity type characteristics capable of expressing entity relationship types among entities are introduced; the position features and the overall features of the sentences are integrated with the dependency syntactic features, the semantic relationship between the sentences is better learned, the interference of noise data on medical entity relationship extraction is reduced, and the accuracy of medical entity relationship extraction can be improved to a certain extent.

The invention relates to an extraction system of affective characteristic words. The extraction system is characterized by comprising a characteristic selecting module, a characteristic verification module, a relation extraction module, a generalized affective characteristic word list and a narrow affective characteristic word list; the characteristic selecting module utilizes article content in an article set pointed by comment and comment content in a comment set to respectively extract all candidate affective characteristic words and sorted candidate affective characteristic words in the comment content; the relation extraction module constructs a sense relation diagram between words by a template according to the article content, establishes the generalized affective characteristic word list by the all candidate affective characteristic words and the sense relation diagram; and establishes the narrow affective characteristic word list by the sorted candidate affective characteristic words and the sense relation diagram. The method for obtaining affective characteristic words not only is applicable to universal sentiment analysis with larger subject, but also can carry out deeper sentiment analysis in detailed subject. The extraction system of affective characteristic words can be widely applied to sentiment analysis of comment of news, forums, blogs and the like.

The invention relates to a field entity attribute relation extraction method based on distance supervision, belonging to the technical field of natural language processing and depth learning. The method inlcudes constructing a domain knowledge base of Chinese tourist attractions, through the Chinese encyclopedia website and tourism website to obtain a large number of tourism domain text collections, using the constructed tourism domain knowledge base of entity pairs to obtain the relational instance text collections from the tourism domain text collection; using the theme model keyword similarity calculation and keyword pattern matching to denoise; finally, using the training corpus which is composed of positive and negative data under each relationship, the part-of-speech feature, dependency feature and short syntax tree feature of the training corpus are extracted, and the three features are fused into a larger feature with more abundant semantic information, and then the relationship extraction model is trained. Experiments show that the F value of the fusion of the three features extracted from the de-noising training corpus is the highest and the extraction performance is thebest.

The invention discloses an electronic medical record entity relationship extraction method based on a convolutional recurrent neural network. The method comprises the steps of using a data constructorto reconstruct natural statements to obtain a multi-dimensional hierarchical sequence; Mapping the multi-dimensional hierarchical sequence into an input feature vector by adopting a vector representation technology; capturing local and global semantic information of the statement simultaneously by adopting a convolutional recurrent neural network ConvLSTM to obtain an upper sentence vector; adopting a two-stage attention mechanism to capture text content closely associated with the semantic relation, and obtaining a high-stage sentence vector, so as to solve the problem that multiple instances are mistakenly labeled; and performing relation judgment according to the obtained high-level sentence vector to obtain a prediction label. The method does not depend on any external resource characteristics, and the entity relationship extraction performance is improved only by means of data reconstruction and improvement of a network model framework. Meanwhile, the method can be extended to other tasks with the problems of insufficient feature extraction, unbalanced samples and the like.

The invention relates to a domain event graph construction method and device fusing multiple types of facts and entity knowledge. The method comprises the following steps: performing cause relationship extraction and instance cause element extraction on a domain corpus to form an instance cause logic knowledge base; constructing an abstract concept knowledge base with hierarchy; performing entityword abstraction and predicate abstraction on instance events in the instance cause logic knowledge base by utilizing the abstract concept knowledge base to form an abstract cause graph; carrying outentity linking on instance events in the instance cause logic knowledge base by utilizing the entity knowledge graph and adopting an entity linking technology, and fusing event knowledge and entity knowledge to form a cause knowledge graph; and combining the abstract factorial map with the factorial knowledge map to form a domain event map. The advantages of static entity knowledge and action event knowledge can be integrated, the application range of knowledge questions and answers can be widened, and the domain event graph can be used as a common knowledge base to expand domain language resources.

The invention discloses a knowledge graph relational data extraction method based on a semantic syntax interaction network. The method mainly comprises the following steps: collecting a design document of a complex equipment design process, and establishing a design document corpus according to text data of the design document; performing text preprocessing on the design document text data; establishing a relation extraction model based on a semantic syntax multi-round interaction deep neural network; inputting the preprocessed text data and the relationship type label into a relationship extraction model for offline training; and preprocessing the text data of the entity relationship to be predicted, and inputting the preprocessed text data into the trained relationship extraction model to obtain a predicted relationship category. Through multiple rounds of interaction of the semantic information and the syntactic information, the utilization rate of the semantic information and the syntactic information is improved, the semantic information and the syntactic information beneficial to knowledge graph relation data extraction are dynamically and deeply mined, and the flexibility, generalization and accuracy of the model are improved.

The invention provides an entity relationship extraction method and system integrated with dynamic word vector technology. According to the system, an existing knowledge base corresponds to rich unstructured data by using a remote supervision method so as to generate a large amount of training data, so that the problem of insufficient manual annotation corpora is relieved, and the system can reduce the dependence on annotation data, thereby effectively reducing the labor cost. In order to obtain feature information between entities as much as possible, the basic architecture of the model adopts a segmented convolutional neural network; and the semantic information of the example sentences is further extracted by integrating a dynamic word vector technology.

An area extracting unit extracts area information from a page of document information for each area of different types arranged on the page. A relation extracting unit extracts relation information indicating a relation between the area information and the page of the document information that is an extraction source of the area information, from the page of the document information. A registering unit registers the area information and the relation information in area correspondence information stored in a storage unit in association with each other.

The invention relates to a text statement processing method and device, computer equipment and a storage medium. The method comprises the steps of obtaining a sample text statement containing an entity pair and a relationship label of the entity pair; extracting a positive example statement pair and a negative example statement pair from the sample text statement according to the relationship label, and performing positive and negative example sampling processing to obtain a training set; inputting the training set into a to-be-trained relationship extraction model, and generating a loss valuecomprising a comparison loss value; wherein the comparison loss value is used for representing the difference between the similarity of the statements in the positive example statement pair and the similarity of the statements in the negative example statement pair; adjusting parameters of the relation extraction model according to the loss value, and returning to the step of extracting the positive example statement pair and the negative example statement pair from the sample text statement according to the relation label, so as to carry out iterative training until a training stop conditionis met, and obtaining the relation extraction model; wherein the relationship extraction model is used for identifying the entity relationship of the entity pair in the text statement. By adopting the method, the entity relationship extraction accuracy can be effectively improved.