53 results about "Concept extraction" patented technology

Information retrieval systems face challenging problems with delivering highly relevant and highly inclusive search results in response to a user's query. Contextual personalized information retrieval uses a set of integrated methodologies that can combine automatic concept extraction / matching from text, a powerful fuzzy search engine, and a collaborative user preference learning engine to provide accurate and personalized search results. The system can include constructing a search query to execute a search of a database parsing an input query from a user conducting the search of the database into sub-strings, and matching the sub-strings to concepts in a semantic concept network of a knowledge base. The system can further map the matched concepts to criteria and criteria values that specify a set of constraints on and scoring parameters for the matched concepts. Furthermore, the system can learn user preferences to construct one or more profiles for producing personalized search results.

Information retrieval systems face challenging problems with delivering highly relevant and highly inclusive search results in response to a user's query. Contextual personalized information retrieval uses a set of integrated methodologies that can combine automatic concept extraction / matching from text, a powerful fuzzy search engine, and a collaborative user preference learning engine to provide accurate and personalized search results. The system can include constructing a search query to execute a search of a database. The system can parse an input query from a user conducting the search of the database into sub-strings, and can match the sub-strings to concepts in a semantic concept network of a knowledge base. The system can further map the matched concepts to criteria and criteria values that specify a set of constraints on and scoring parameters for the matched concepts.

An online video search system, including a tag discoverer including a web encyclopedia crawler for (i) accessing a web encyclopedia to find web pages related to at least one designated reference topic, and (ii) retrieving a plurality of web pages by performing an n-level depth recursive traversal of the web pages found, and web pages that are hyper-linked thereto, a concept extractor for extracting important concepts founds in the retrieved plurality of web pages, and a user interface for providing at least of the important concepts extracted by the web page processor to an online video search engine. A method and a computer-readable storage medium are also described and claimed.

A system for extracting ontological information from a body of text includes an input module configured to receive a verb phrase. The system also includes a parsing module configured to parse one or more sentences from the body of text into parse tree format to generate a set of parsed sentences. The system further includes a named-entity-recognition module configured to identify a subset of parsed sentences from the set of parsed sentences, identify a subset of noun phrases from the subset of parsed sentences, classify a first noun phrase in subset of noun phrases as an entity, and classify a second noun phrase in subset of noun phrases as a property. The system also includes a concept-extraction module configured to identify and output a conceptual relationship between the first entity and the first property based at least partially on grammatical relationship of the first entity and the first property.

Automatic extraction of objects in a presentation-oriented document comprises receiving the presentation-oriented document (POD) in which content elements are spatially arranged in a given layout organization for presenting contents to human users; receiving a set of descriptors that semantically define the objects to extract from the POD based on attributes comprising the objects; using the set of descriptors to identify content elements in the POD that match the attributes in the set of descriptors defining the objects, and assigning semantic annotations to the identified elements based on the descriptors; creating a semantic and spatial document model (SSDM) containing spatial structures of the identified content elements in the POD and the semantic annotations assigned to the identified contents elements; extracting the identified content elements from the POD based on the set of descriptors and the SSDM to create a set of object instances; and performing at least one of: i) using the object instances to generate semantic and spatial wrappers that can be reused on a different POD, and ii) storing the object instances in a data repository.

The invention provides a concept-based unstructured text big data analysis mining method and a device thereof. The method comprises the following steps of: (1) performing pre-processing, including word segmentation and named entity recognition; (2) performing concept extraction and concept expression identification on an input text; (3) performing analysis mining on the concept expression of the input text according to mining rules; (4) calculating confidence levels of the mining results; (5) outputting the mining results according to the confidence levels; and (5) visually showing the mining results. A mining model of the method comprises three trees, including an ontology tree, an element tree and a concept tree. The device comprises a modelling unit (1), a pre-processing unit (2), a concept extraction and expression identifying unit (3), an analysis mining unit (4) and a visual show unit (5). The method and the device have the following advantages that: diversity of services and natural language expressions is separated in the modelling process, and investment in service maintenance is lowered; and the mining method can greatly improve accuracy of analysis mining.

The embodiment of the invention discloses a construction method and device of a human object attribute classification knowledge graph, a system and a storage medium. The construction method includes the steps of obtaining human object data, conducting concept extraction on the human object data, and determining at least one concept of human objects; according to preset principles, determining attributive classifications of the concepts and incidence relations among the concepts; according to the attributive classifications of the concepts and the incidence relations among the concepts, constructing the human object attribute classification knowledge graph. By the adoption of the construction method and device of the human object attribute classification knowledge graph, the system and thestorage medium, the problems are solved that human object attributes are disordered in classification and not uniform in information, the purpose of automatically constructing the human object attribute classification knowledge graph is achieved, and the comprehensiveness and uniformity of human object attribute generalization are improved.

The invention provides a domain concept extraction method for open texts. The method includes the steps of firstly, traversing an open text set, and extracting candidate domain concepts from all the open texts; secondly, obtaining the word vector associated with the corresponding candidate domain concept for each candidate domain concept through the phrase resolution result, contextual information and encyclopedia classification information of the candidate domain concept, and using all words in the word vector as domain labels associated with the candidate domain concept; thirdly, establishing a candidate domain concept set A through all the candidate domain concepts obtained the first step, establishing a domain label set B through the domain labels obtained in the second step, and conducting iterative computation through the HITS algorithm to obtain the domain relevancy of all the candidate domain concepts; fourthly, judging the domain concepts through the domain relevancy of all the candidate domain concepts. By means of the method, accuracy and the recall rate can be increased, and the important low-frequency concepts can be better identified.

The invention discloses a domain concept extraction method based on Deep Learning. The method includes extracting samples in a training corpus, adopting word frequency, document frequency, inverse document frequency, word length, word frequency variance and domain consensus as feature vectors, training and acquiring a deep network model, which is capable of representing the complex mapping correspondence between the word-type filed concept multi-dimensional feature vectors and class labels, on the basis of the Deep Learning technology, and finally comparing the deep network model established on the basis of the Deep Learning technology, an optimized BP neural network model and mainstream KNN and SVM models in the testing step. According to the tests, the optimal test effect is acquired through the deep network model established on the basis of the Deep Learning technology.

Techniques for generating automated tags for a video file are described. The method includes receiving one or more manually generated tags associated with a video file, based at least in part on the one or more manually entered tags, determining a preliminary category for the video file, and based on the preliminary category, generating a targeted transcript of the video file, wherein the targeted transcript includes a plurality of words. The method further includes generating an ontology of the plurality of words based on the targeted transcript, ranking the plurality of words in the ontology based on a plurality of scoring factors, and based on the ranking of the plurality of words, generating one or more automated tags associated with the video file.

The invention discloses an autonomous data lake construction system and method based on associated data. The system comprises a data source input module, a heterogeneous data preprocessing module, a metadata discovery and extraction module, a metadata fusion and association module, a meta-model optimization and construction module, an instance knowledge extraction module, a knowledge packaging module, a knowledge correction and fusion module, an instance concept extraction module and a meta-model verification and evolution module. Based on the associated data, the directory index updated in real time and the instance knowledge graph capable of being quickly positioned through the directory are generated while the data lake is constructed, and the data lake with the autonomous ability is finally obtained through the internal structure and semantic association of the directory index and the instance knowledge graph, so that the data lake is easily managed and retrieved by external usersand more requirements are met.

The invention belongs to the technical field of big data. The invention relates to a knowledge extraction and fusion method based on big data. Aiming at the problem that big data integration and knowledge extraction are greatly inconvenient due to the characteristics of timeliness, multi-source heterogeneity, weak relevance, isolation dispersity and the like of big data, the following scheme is provided: the method comprises the following steps of concept extraction, concept classification relation extraction, concept non-classification relation extraction, entity alignment and entity linking.According to the invention, for the obtained big data, an entity, relationship and attribute category system of each facet is constructed, syntax meaning analysis is carried out, candidate knowledgepoints are discovered, and then feature selection is conducted, so the knowledge of entity-relation-entity and entity-attribute-attribute value is automatically extracted from mass data, the completeness of big data acquisition is improved, the credibility and validity of the acquired data are improved, high availability, dynamic expansion and updating of a knowledge graph are supported, and effective fusion of big data is realized.

The invention provides a BERT-based government affair official document ontology concept extraction method. The method comprises the following steps of (1) obtaining government affair official document data; (2) performing text data preprocessing on the public government affair official document data; (3) establishing linguistic rules of terms; (4) performing official document ontology term extraction; (5) estimating the category number of the official document ontology terms; (6) carrying out word vector representation for the official document ontology terms; (7) completing term clustering;(8) extracting an official document ontology concept; and (9) realizing evaluation and verification of an ontology concept extraction effect. Effective technical means of government affair work are overall planned, powerful support and guarantee are provided for application of government affair official affairs such as sharing exchange, information retrieval, information extraction and governmentaffair atlas construction, the clustering effect of official document terms is improved, and solid guarantee and support are provided for precision of official document ontology concept extraction.

The invention discloses a method for generating a character test questionnaire based on an image and a surveying interactive method. The method for generating the character test questionnaire comprises the following steps: acquiring an image set which the user likes and a character nature truth set of the user and establishing a first relationship between the image and the user; extracting a concept and establishing the relationship between the image and the concept; extracting image features of each image in the image set and establishing a second relationship of the image and the user; confirming a concept set shared by a user set under a given user set according to the relationship between the image and the concept and the relationship of the image and the user; screening out the concept set at specific character distinguishing degree; and screening out the image with representative character from the image set under each concept to be taken as an image option of a visualization problem; generating the questionnaire through the screened-out concept and the image option under the concept. According to the embodiment of the invention, the accurate user character can be acquired within a shorter time period and the cross-language property is better.

In a method and system for reviewing, searching and analyzing raw data in a data corpus a corpus optimization module converts the raw data to an optimized corpus. A search composition module operates on the optimized corpus to derive a set of search parameters and a concept extraction module extracts a set of initial concept clusters using the set of search parameters. A hybrid review module receives the set of initial concept clusters from the concept extraction module and allows a user to review the optimized corpus using a user interface until the user declares the review complete. A visualization module visualizes the results of the review, search and analysis of the raw data in the data corpus after the user declares the review complete.

The invention discloses a text-oriented domain classification relationship automatic learning method. The method comprises the steps of adopting MEDLINE as a corpus library; performing term extractionand concept extraction; performing syntax similarity and semantic similarity-based five dimension similarity calculation for extracted concepts; performing weighting on the similarity of each dimension to obtain a final similarity matrix; based on this, performing hierarchical clustering to obtain an initial tree diagram; and performing corresponding pruning and cluster marking on the tree diagram to finally obtain a tree diagram reflecting a classification relationship among the concepts. According to the method, a large amount of manual marking is not required, so that the manpower and timeoverhead is saved; extracted terms and a UMLS metathesaurus of an authoritative knowledge base are mapped to obtain accurate domain concepts; by adopting a distributed method of the hierarchical clustering, and in combination with domain background knowledge, the five dimension similarity calculation is provided; and an extremal distance estimation-based unsupervised hierarchical clustering dynamic pruning method is proposed, so that the domain-related classification relationship can be better obtained.

Disclosed is an intelligent tea tree insect pest diagnosis prototype system based on cloud ontology. The system comprises a concept extraction module, a tea tree insect pest field ontology clouding module, a concept-to-concept taxonomic relation clouding module, a concept-to-concept non-taxonomic relation clouding module, an intelligent checking module, a diagnosis module and a result output module. The system aims to solve the expression problem of uncertain knowledge of the ontology of the tea tree insect pest field, the process of combining cloud model concepts with the ontology to establish the cloud ontology is taken into consideration, a clouding method for tea tree insect pest field concepts and concept-to-concept relations is researched, and uncertain knowledge in the field is confirmed and applied to an intelligent diagnosis expert system.

The invention belongs to the technical fields of electrical power systems and energy storage systems, and particularly relates to a cloud model-based energy storage system typical-curve mining method. A determined application scenario of an energy storage system and a proper control policy are selected, and an energy storage system longitudinal timing sequence power is counted; frequency distribution of an energy storage system longitudinal timing sequence power value is decomposed into a plurality of cloud models different in granularity, expectations of a normality cloud model group are weighted and summed, and an energy storage system typical-curve is given. The calculated amount can be greatly reduced, and clear perceiving of the overall charging and discharging power of the energy storage system is facilitated; a backward cloud generator in each cloud model can convert precise data into a cloud model represented by numerical characteristics, thus realizing numerical value-to-concept extraction and conversion, namely realizing the effect that the energy storage system typical-curve is mined by energy storage system charging and discharging power data.

The invention provides an entity relation joint extraction method and system based on active deep learning, and relates to the technical field of computer natural language processing. The method comprises the following steps: firstly, acquiring a to-be-labeled sample data set as a corpus, performing concept extraction on the corpus, and defining an entity category set and a relationship category set; carrying out sample sampling by using a to-be-labeled sampling method based on active learning to obtain a to-be-labeled sample data set; performing data enhancement on the to-be-labeled sample data set by using an improved EDA method; then, according to the defined entity and relationship category set, labeling data of the to-be-labeled sample data set by adopting a BIO-OVE / R-HT labeling strategy; and finally, inputting the labeled data into an entity relationship joint extraction model for training; and when the model is used for prediction, decoding the predicted label by using a decoding rule corresponding to the labeling strategy to obtain a triple. According to the system, the entity relationship is extracted, and meanwhile, the extracted entity relationship is used for quickly constructing a knowledge graph and managing the knowledge graph.

The invention provides a method and a device for extracting an entity relationship in a text, electronic equipment and a storage medium, and the method comprises the steps of inputting a to-be-extracted text into a pre-trained concept extraction model to obtain a concept sequence; determining a plurality of tuples to be judged corresponding to the concept sequence according to a preset tuple generation rule; according to at least one feature judgment rule, after generating relationship feature vectors corresponding to the to-be-judged relationship tuples, combining the relationship feature vectors into a relationship feature matrix of the to-be-extracted text; inputting the relation feature matrix into a pre-trained tuple judgment model to obtain tuple judgment result values corresponding to the to-be-judged relation tuples, and further determining a target entity relation of the to-be-extracted text. Thus, the step of obtaining tuples of different dimensions is reduced, and meanwhile, based on judgment of the relation feature matrix, a more reliable basis is provided for judgment of each relation tuple to be judged, and the efficiency and accuracy of extracting the entity relation in the text according to the relation tuple are improved.

The invention provides a character-level clinical concept extraction named entity recognition method and system. The method comprises the steps of: introducing character-level representation through apre-trained word vector, then marking a target entity according to the context, and finally, restraining a marking result through a rule, thereby effectively improving the recognition accuracy of a clinical named entity. In addition, a Nadam optimization mode is adopted, so that the recognition speed is greatly increased.

A method for extracting concept of coal mine safety accident ontology is provided, This method combines a word vector and a conditional random field to extract the concept of coal mine safety accidentontology, fully considers the semantic characteristics and domain characteristics of domain words, and solves the problems of lack of traditional research methods in semantic relationship analysis and inconsistent data management in the field of coal mine safety, and improves the reusability of knowledge. Experiments show that the proposed method improves the precision of concept extraction in the field of coal mine safety compared with the traditional concept extraction method based on CRFs. At the same time, it also proves that the word vector model provided in this invention is better thana traditional CBOW model and a skip-gram model in performance.

An online video search system, including a tag discoverer including a web encyclopedia crawler for (i) accessing a web encyclopedia to find web pages related to at least one designated reference topic, and (ii) retrieving a plurality of web pages by performing an n-level depth recursive traversal of the web pages found, and web pages that are hyper-linked thereto, a concept extractor for extracting important concepts founds in the retrieved plurality of web pages, and a user interface for providing at least of the important concepts extracted by the web page processor to an online video search engine. A method and a computer-readable storage medium are also described and claimed.

The invention discloses an automatic subjective question marking neural network model with concept enhanced representation and unidirectional attention implication. A concept series in questions is automatically identified by combining a bidirectional long-short term memory neural network BiLSTM and a conditional random field CRF of machine learning; through a multi-head attention mechanism, enhanced representation modeling of a concept word embedding vector sequence on a answer word embedding vector sequence is realized; answer context information is coded through the BiLSTM; through a one-way attention implication matching mode, semantic inclusion of student answers to reference answers is estimated, information is gathered on the basis of one-way implication matching vectors, and probability distribution prediction of student answer score intervals is carried out. The model comprises a concept extraction layer, an answer presentation layer, a concept enhancement presentation layer, a context presentation layer, a one-way implication attention layer, an aggregation layer and a prediction layer. The model has the advantages that extra semantic analysis and artificial rules are not needed; the matching precision of paper marking is improved; and the adaptability and practicability of a paper marking system are expanded.

The invention claims a robot data interoperation domain ontology construction method based on depth learning, which comprises the pretreatment of data source, the robot domain term extraction and concept extraction based on depth learning, and the construction of a relationship model between robot data and concept. This method solves the key problem of data interoperability in robot heterogeneoussystems, that is, how to solve the problem of semantic heterogeneity of heterogeneous data sources. This method is mainly applied to data interoperability in manufacturing heterogeneous systems, completes the semi-automatic construction of robot ontology, is the perfection of the existing ontology theory and application research in China, and fills the blind spot of the application research of ontology theory in the field of industrial robots. Compared with the traditional manual method, the ontology construction process is more convenient and quick, and it is suitable for the ontology construction of a large number of data sources.

The embodiment of the invention provides a concept extraction method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the term extraction of a to-be-extracted text according to a preset word list, obtaining a first candidate concept list, carrying out the entity linking of the to-be-extracted text according to a preset knowledge graph, and obtaining a second candidate concept list; reordering the candidate concepts in the first candidate concept list and the second candidate concept list, and obtaining a concept extraction result of the to-be-extracted text according to a reordering result, wherein the text to be extracted is an unstructured text. According to the concept extraction method and device, the electronic equipment and the storage medium provided by the embodiment of the invention, the candidate concepts obtained by performing term extraction and entity link acquisition on the to-be-extracted text are reordered, and theconcept extraction result is obtained according to the reordering result, so that under the condition of less annotation data or even no annotation data, concepts are extracted from the unstructured text more efficiently and accurately.

The invention discloses a data clustering method and device. According to the invention, a local density clustering method is adopted, and a foundation is laid for clustering of sensor data through determining the distance between nodes, the local density of each node and the shortest distance between each node and a node with higher local density; and then the category to which each clustering center node and the nodes except for the clustering center nodes belong is determined according to the determined local density of each node and the shortest distance between each node and a node with higher local density, thereby realizing automatic clustering the nodes, accomplishing automatic extraction of data concepts, not only breaking through defects of the traditional k-means clustering method, but also realizing clustering for data of any shapes. In addition, the data clustering method and device lay a foundation for realizing collaborative analysis of heterogeneous equipment, interoperations of the equipment and the like, the reliability and complementarity of the information are ensured, and the accuracy of data concept extraction is improved.

An apparatus is configured to structure contents of a meeting. The apparatus includes a voice recognizer configured to recognize a voice to generate text corresponding to the recognized voice, and a clustering element configured to cluster the generated text into subjects to generate one or more clusters. The apparatus further includes a concept extractor configured to extract concepts of each of the generated clusters, and a level analyzer configured to analyze a level of each of the extracted concepts. The apparatus further includes a structuring element configured to structure each of the extracted concepts based on the analysis.