86 results about "Semantic dependency" patented technology

The invention relates to a multi-granularity semantic chunk based entity attribute and attribute value extracting method, and belongs to the technical field of Web mining and information extraction. The method comprises the following steps that a corpus set is constructed and free text extraction is performed; a corpus is subjected to word segmentation, part-of-speech tagging and phrase recognition; the corpus is subjected to semantic role labeling; the corpus is subjected to dependency grammar analysis; the corpus is subjected to semantic dependency analysis; candidate entities, attributes and attribute value triads based on three granularities of words, phrases and semantic roles are extracted; the candidate entities, attributes and attribute value triads are corrected and subjected to error classification by means of a trained classifier. Compared with the prior art, the entities, attributes and attribute value triads based on three granularities of words, phrases and semantic roles are automatically extracted from a free text, the entity attribute and attribute value extraction accuracy and efficiency are improved, and the wide application prospect is achieved in the fields of theme detection, information retrieval, automatic abstracting, question and answer systems and the like.

The invention discloses an event argument role extraction method based on a multi-head attention mechanism. The method comprises the following implementation steps: step (1), preprocessing a data settext, and outputting the preprocessed text and a corresponding tag; step (2), training a bidirectional GRU network integrated with a multi-head attention supervision mechanism; (3) performing semanticdependency analysis on the text, and outputting a semantic dependency path between the trigger word and the candidate argument; (4) inputting the preprocessed text into the network in the step (2) for training, outputting the code of each word, and fusing the semantic dependency path outputs (trigger word codes, candidate argument codes and semantic dependency paths) argument classification structures in the step (3); and (5) inputting the argument classification structure into a classification network for training and classifying. According to the method, the text is analyzed by using a neural network method fused with a multi-head attention supervision mechanism, and the method has a good argument role extraction capability for a plurality of event situations in event sentences.

The invention relates to a relation extraction method and system based on an attention cycle gated graph convolutional network, and the method comprises the steps of carrying out the semantic dependency analysis of a statement, enabling word embedding to be connected with a position feature, and obtaining a final word embedding representation; constructing a BLSTM network layer, and extracting a word context feature vector; applying an attention mechanism to the dependency tree to obtain a soft adjacency matrix of a fully connected graph with weight information; transmitting the word context feature vector and the soft adjacency matrix into a gated graph convolutional network, and extracting a high-order semantic dependence feature to obtain vector representation of a statement; and extracting vector representations of the two marked entities, splicing the extracted vector representations of the two marked entities with the vector representation of the statement, transmitting the spliced vector representation of the statement into a full connection layer of the gated graph convolutional network, calculating the probability of each relationship type and predicting the relationship type, and finally obtaining the relationship type of the statement. According to the invention, key information loss is avoided, and the relationship extraction performance is improved.

The invention provides a natural language emotion analysis method based on a depth network. On the basis of a memory network, semantic dependent information is introduced to guide the execution of theattention mechanism, and the context moment information including the whole emotion information of a sentence is also used to provide background information for the current analysis object word. Thewhole model includes an embedding module, a memory sequence building module, a semantic dependency mask attention module, a context moment affective learning module and an output module. In the model,the semantic dependency information of object words and context obtained from dependency syntax tree is introduced into the memory network, so that the memory sequences of each layer are generated dynamically, which leads to the execution of attention mechanism in the multi-layer module of the memory network. In addition, in order to introduce the whole affective information of a sentence, that is, the relationship information between all the object words in the same sentence, we propose a context-based learning task, which assists the affective analysis of specific object words by multitasking learning.

The invention discloses a method and a device for positioning a voice control service, which belong to the technical field of smart television. The method includes the following steps: segmenting the words of a voice control text to get a word segmentation result; carrying out resource annotation on the word segmentation result to get a resource annotation result; searching for a service classification template matching the voice control text in preset service classification templates based on the word segmentation result and the resource annotation result; and when no matching service classification template is found, analyzing the syntactic structure based on the word segmentation result and the resource annotation result and according to a custom semantic dependency relation to determine a voice control service corresponding to the voice control text. By analyzing the syntactic structure of the voice control text, the syntactic structure information is retained, voice control text comprehension error is avoided, and the accuracy of voice control service positioning is improved.

The invention relates to an Internet instant message communication system, which aims to ensure that an intelligent function of an instant message (IM) chat tool is available by introducing a novel function of automatically dispatching message to a common IM chat tool. To establish an automatic message dispatcher in the IM chat tool, the invention adopts processing methods such as information retrieval of natural language counting technology, machine translation and the like and realizes automatic message dispatch to a corresponding receiver in an environment with a single window and a plurality of message receiving objects of an Internet IM system by establishing a semantic association model, calculating the semantic dependency between a new message and the messages of different conversations and selecting the most related conversation.

The invention discloses a self-service query statistics analysis method based on natural language processing. The method comprises the steps that a power communication data query statistics analysis rule base, a knowledge base and a corpus are constructed; a multi-dimensional data query statistic analysis model for power communication standardization and programming is established; a query statistics analysis statement or a paragraph is input, a query statistics analysis natural language processing task is formed, and a semantic dependency tree is generated; each element in the semantic dependency tree corresponds to a query statistic analysis model, a structured query statement is generated through matching of the rule base and the knowledge base, and finally a query statistics analysis result set is generated. According to the method, self-defined, fast and flexible data configuration, processing, analysis and management can be supported, automatic and intelligent planning and designof a communication network can be achieved, and the operation and maintenance level is improved.

The invention provides a voice inquiry method and device and an electronic device, wherein, the method is applied to a server. The method comprises the following steps: acquiring a user voice and converting the voice into a Chinese text; processing Chinese text to obtain a text vector; Inputting the text vector into the preset transformation model to get the output result, wherein The transformation model is obtained by training the attention model; inputting The output result into a preset SQL statement template, and obtaining a formalized SQL statement according to the preset semantic dependency relationship; Getting domain information; Based on the domain information and formal SQL statements, obtaining the executable SQL statements. As that speech is convert into a text vector, The text vector is converted into executable SQL statement by using pre-established conversion model, SQL statement template and domain information, and the database is queried by using executable SQL statement, which simplifies the query mode of the database and can be adapted to the database queries of different service systems, and improves the portability of the database queries.

The invention discloses an intelligent conference summary generation method and system, and the method comprises the following steps: obtaining conference audio and video signals; extracting voiceprint features of the real-time voice signals; performing feature matching on the voiceprint features and a preset voiceprint database, identifying speaker identities corresponding to different voice signals, and separating speaking voices of different speakers; respectively transliterating the separated voice signals into characters, and generating a preliminary text corresponding to the speaker; performing error correction and optimization processing on the preliminary text to obtain a target text; and extracting important sentences in the target text to form an abstract, and arranging the abstract into the conference summary according to a preset template. The beneficial effects of the method are that the method can improve the detection time, improves the execution efficiency of an algorithm, and reduces the operation time, and responsible text processing flows are reduced, and the reusability of codes is improved. The workload of algorithm researchers is reduced, and bidirectional semantic dependence can be better captured.

The invention discloses a text inference method based on limited semantic dependency analysis. The method comprises the following steps of: 1, according to a semantic dependency representation mechanism, establishing a Chinese text inference marking resource, wherein a text inference marking instance comprises a text T, a hypothesis H, a semantic dependency graph of the hypothesis and an inference type; 2, on the basis of the Chinese text inference marking resource, performing semantic dependency analysis on the hypothesis H which is newly input under the limitation of the text T which is newly input so as to judge whether the text T can infer the hypothesis H; and 3, evaluating, analyzing and summarizing an analysis process and a judgment result of the step 2, and improving the performance of the analysis process with feedback. By adoption of the method, the judgment of text inference is formalized into the problem of the limited semantic dependency analysis, and the text inference extends from entailment to preset and implication in type. The method contributes to processing of the inference of discourse and dialogue corpora which are relatively complicated.

The invention provides a remote supervision relation extraction method with entity perception based on a PCNN model. The method specifically comprises the following steps: combining word embedding with head entity and tail entity embedding and relative position embedding by using a multi-head self-attention mechanism to generate enhanced word semantic representation of a perceptible entity, whichcan capture semantic dependence between each word and an entity pair; introducing a global door, and combining the enhanced word representation perceived by each entity in the input sentence with theaverage value of the enhanced word representations to form a final word representation input by the PCNN, and in addition, in order to determine a key sentence segment in which the most important relationship classification information appears. According to the method, another gate mechanism is introduced, and different weights are allocated to each sentence segment, so the effect of key sentencesegments in the PCNN is highlighted. Experiments show that the remote supervision relationship extraction method provided by the invention can improve the prediction capability of the remote supervision relationship in the sentence.

The invention discloses a text matching method using a semantic parsing structure. The method comprises the following steps: defining an initial corpus set Cqa and a supplementary corpus set Cq; defining Semantic structure DP-tree corresponding to text by using a semantic dependency analysis method; Defining a kernel function of the text and a metric function of text similarity based on the semantic structure; Carrying out kernel clustering on the text; obtaining an aggregated text class function(shown in the specification), wherein i = 1, 2, ..., M, q'ij is ni sample points which are selectedfrom each cluster and are closest to the cluster; And through manual audit, approving the Ci class and marking the Ci class with a specific tag Ti. According to the invention, syntactic analysis structures such as a syntactic structure are used as a comparison basis; A convolution kernel function theory and tree kernels (tree kernel, TK) are combined to define a kernel function representing the distance between two tree syntactic structures, and internal and external knowledge of syntactic similarity, word vectors, word sense networks and the like is introduced, so that the similarity betweentexts can be accurately judged.

The present invention provides a method and system directed to predicting implicit rhetorical relations between two spans of text, e.g., in a large annotated corpus, such as the Penn Discourse Treebank (“PDTB”), Rhetorical Structure Theory corpus, and the Discourse Graph Bank, and particularly directed to determining a rhetorical relation in the absence of an explicit discourse marker. Surface level features may be used to capture pragmatic information encoded in the absent marker. In one manner a simplified feature set based only on raw text and semantic dependencies is used to improve performance for all relations. By using surface level features to predict implicit rhetorical relations for the large annotated corpus the invention approaches a theoretical maximum performance, suggesting that more data will not necessarily improve performance based on these and similarly situated features.

Embodiments of the invention provide a sensitive word identification method, device and equipment. The method comprises the steps of determining context information corresponding to each character ina segmented word; a word vector sequence of the segmented words is generated according to the semantic relation in the context information, and the word vector sequence comprises word vectors of the segmented words and word vectors of the segmented words in the context information; inputting the word vector sequence of the segmented word into a pre-trained recognition model to obtain a recognitionresult whether the segmented word is a sensitive word or not; for sensitive word variants, even if glyph or character pronunciation is changed, the context semantic dependency relationship is not changed, so that the sensitive words are recognized based on the context semantic dependency relationship in the scheme, the variants of the sensitive words can be recognized, and the recognition effectis improved.

The invention relates to a cross-domain emotion classification system based on attention mechanism fusion. The system comprises a comment text preprocessing module used for obtaining vector forms of texts in a source domain and a target domain; a text semantic learning module which is used for learning a semantic dependency relationship between words; an attention mechanism fusion module which isused for fusing different attention modes to obtain comprehensive weights of words for text classification; a hierarchical attention module which is used for calculating attention weights of the textfrom a word level and a sentence level respectively and judging weights of words for sentence representation and sentences for document representation; and an emotion category output module which is used for obtaining a final emotion classification result by utilizing the classification function. According to the method, the potential universal features of the target domain and the source domain can be automatically extracted, the features are abstracted and combined, and finally the emotion category of the text of the target domain is recognized.

The invention relates to a sentence semantic distance measurement method. The method comprises the following steps: firstly, carrying out word segmentation and stop word removal preprocessing on a sentence data set; selecting a word meaning similarity scheme, and setting a threshold value to execute normalization of synonymous words and synonymous words; then, calculating the vector space distanceof the two statements by combining smooth inverse frequency weighting and common component removal; measuring the word order distance of the two statements according to the out-of-order degree; calculating the semantic dependency distance of the two statements by combining the semantic dependency quintuple features; and finally, carrying out hybrid weighting calculation on the vector space distance, the word order distance and the semantic dependency distance. Measurement is from three dimensions of sentence vector representation, sentence word sequence and sentence component dependency, andfinally a final semantic distance is obtained in a weighted summation manner. A word level calculation means is utilized, and a sentence level operation idea is absorbed, and through introduction andcreative combination of a vector space distance, a word order distance and a semantic dependency distance, the semantic distance of the sentences is more comprehensively and reasonably measured.

The invention discloses an SNOMED-CT-based medical noun dictionary construction and expansion method. The method comprises the steps of performing Chinese word segmentation on a target text to obtainwords in each sentence; secondly, analyzing sentences of the segmented words by utilizing a semantic dependency algorithm so as to obtain noun phrases; finally, taking an existing normalized medical dictionary SNOMED-CT as a basis to be compared with the medical phrases extracted from the text, screening the candidate noun phrases through predefined similarity, wherein the selected noun phrases are regarded as new medical noun phrases, and therefore the dictionary is formed. According to the method, the problems of multi-expression, abbreviation, abbreviation and the like of the same concept are solved, and a basic medical term dictionary with wider coverage is constructed. The method serves as the first part of natural language processing, and can improve the speed and precision of subsequent named entity recognition, relation extraction, entity alignment and other tasks.

The invention discloses a recurrent neural network event sequential relationship recognition method based on semantic attention, and the method mainly comprises the following steps: firstly carrying out syntactic dependency analysis on an inputted event sentence, intercepting a related trigger word semantic dependency branch, and then obtaining a corresponding hidden state vector through a recurrent neural network; then, calculating attention weight vectors except the trigger word, fusing different segmented words according to different weights, then splicing the fused segmented words with thetrigger word, and obtaining event sentence state vectors; and finally, putting the event sentence state vector into a softmax function to predict a time sequence relationship. According to the method, semantic information implied in event sentences can be effectively captured, and different segmented words can be effectively associated and fused, so that the event sequential relationship recognition precision is improved.

The invention discloses an answer extraction method based on a semantic dependency tree. The method comprises the step of selecting an optimal answer from the candidate answer sentences by comprehensive consideration of a vector similarity, a word form similarity and a dependency path length similarity on the basis of considering a semantic relation between an interrogative sentences and candidateanswer sentences. According to the method, compared with other algorithm, the accurate rate for selection of the optimal answer sentence is greatly improved.

The invention discloses a news text oriented social emotion sorting method based on a hierarchical state neural network, which pays attention to semantic hierarchical structure information of documents and solves related emotion sorting problems in social emotion detection. The method comprises the steps of preprocessing a news text; encoding the words through a sentence state recurrent neural network to obtain sentence representation; encoding sentences through a document state recurrent neural network to obtain document representation; and on the basis of document representation, mapping byusing a multi-layer perceptron, and normalizing softmax to obtain a sorting result of related emotions. Compared with a previous related emotion sorting method, the method has the advantages that thehidden states of all words or sentences are coded at the same time in each time step, and long-distance semantic dependence can be better captured. Besides, a hierarchical structure mechanism is adopted to capture a key hierarchical semantic structure in the document, an important part for evoking emotions in the document is dynamically highlighted, and the performance of related emotion sorting can be improved.

The invention relates to a visual angle level text sentiment classification method and system based on external knowledge, and the method comprises the steps: carrying out the characterization of a visual angle level text, and capturing a bidirectional semantic dependence relation of the text; combining the current word with an external knowledge synonym through a dynamic attention mechanism, andintroducing a sentinel vector to avoid misleading of external knowledge to the model; judging the contribution degree of each word to the view word through a position attention mechanism; calculatingan attention score of each memory content, nonlinearly combining the attention score of each layer with an output result of an upper layer by utilizing a threshold circulation unit, and taking the last layer as emotion feature representation of the text; and obtaining a final sentiment classification result by utilizing the classification function. According to the method, the view-level text sentiment classification performance can be improved, and the resource consumption can be reduced.

The invention provides a graph convolution-based relationship extraction method, which comprises the following steps of: language analysis preprocessing: performing word segmentation and dependency syntactic analysis on an original sentence in a data set by means of a natural language analysis tool to obtain a word segmentation result of the original sentence, and constructing a dependency syntactic tree for representing a semantic dependency relationship between words in the original sentence, generating an adjacent matrix according to a topological relation between nodes in the dependency syntax tree; querying word vectors: converting each word of the original sentence into a corresponding word vector by querying a word vector table to obtain a vectorized representation of the original sentence; feature extraction through the graph convolutional neural network: inputting the adjacent matrix and the vectorized representation of each word into the graph convolutional neural network, and performing learning to obtain feature representation; and relation classification: splicing the feature representations and then sending the spliced feature representations into a learning neural network to obtain a final representation, then obtaining probability distribution of the entity pairs on each relation according to the feature representations, and predicting the relation with the maximum probability, namely the relation type of the subject entity and the object entity in the sentence predicted by the model.

The invention discloses a data restoration method and apparatus, and belongs to the technical field of information. The method comprises: according to data in a resource description framework (RDF) knowledge base, obtaining RDF graph data and a neighbor step number; according to the RDF graph data and the neighbor step number, determining at least one frequent graph structure; mining a semantic dependency relationship in each frequent graph structure; and according to the semantic dependency relationship, detecting abnormal data in the RFD graph data and restoring the abnormal data. According to the data restoration method and apparatus, the at least one frequent graph structure is determined according to the RDF graph data and the neighbor step number, and the abnormal data in the RDF graph data is detected according to the mined semantic dependency relationship in each frequent graph structure, so that the number of error types of the detected abnormal data is relatively large and the data restoration range is expanded by restoring the detected abnormal data.

The invention relates to a Chinese question semantic comprehension method by utilizing semantic dependency analysis, which comprises the following steps of: step 1, collecting and sorting a question data set in an application field; step 2, preprocessing the obtained question data set; step 3, constructing a semantic dependency analysis model based on BiLSTM (Bidirectional Long Short Term Memory); step 4, constructing a semantic dependency structure analysis rule; and step 5, mapping the query triad list obtained by conversion in the step 4 into a basic diagram mode in the SPARQL statement to form a triad mode relationship. According to the method, through the semantic dependency analysis technology, in combination with the Chinese word segmentation technology, a new mode of efficient semantic understanding can be provided for users, and the method plays an important role in intelligent questioning and answering, search recommendation and other systems in the specific application field.

The invention relates to a Chinese structured event extraction method, and belongs to the technical field of information extraction. The Chinese structured event extraction method comprises the following steps of: performing syntactic dependency structure analysis on an unstructured text statement by using a syntactic dependency analysis tool to obtain a syntactic dependency analysis tree; analyzing the characteristics of the Chinese linguistic and syntactic dependency analysis tree, constructing a core verb chain, and identifying all events existing in the statement; adding a semantic dependency relationship to the syntactic dependency analysis tree by means of a semantic dependency analysis tool, and constructing a syntactic semantic dependency analysis tree; adjusting dependency structures of event core verbs, prepositions and passive language states in the syntactic semantic dependency analysis tree, and constructing a syntactic semantic dependency analysis event graph. According to the method, data does not need to be manually annotated, the structured event can be well extracted, and the extraction accuracy and recall rate are high.

The invention discloses a query statistic-based guidance searching method for a P2P system. The method comprises the following steps: (1) establishing a statistic guidance table SGT; establishing a 'statistic guidance table' with a double layer structure based on history query statistics and semantic analysis results, wherein only a history queried currently is saved in an upstream table entry, and is ranked according to query time; and node information and advantage factors which respond to the current query or similar query are saved in an downstream table entry; (2) querying files by usingthe statistic guidance table, and performing querying by using information in the SGT according to the semantic dependency of the current query and SGT history query after the statistic guidance tableis established by nodes; automatically starting an underlying query strategy if the history saved in the SGT has little dependency with the current query; and (3) adaptively updating the statistic guidance table, including two methods, namely active updating and passive updating, wherein the passive mode takes place after the node forwards a query request.

The invention discloses a spatial relationship extraction method and device based on semantic dependency, and the method comprises the steps: constructing a semantic dependency recognition model, extracting a spatial relationship tuple from an unstructured text, firstly, defining semantic dependency types, namely a trigger word-containing role type and a trigger word-free role type; then inputting a to-be-recognized text and pre-labeled space elements into a semantic dependency recognition model based on the deep self-attention network, and obtaining all semantic dependency recognition results among the space elements in combination with the defined semantic dependency type; and finally, combining the obtained semantic dependency recognition results, and outputting a complete spatial relationship tuple. According to the method, a spatial relationship extraction problem is converted into a semantic dependency recognition problem, and various spatial relationships including a spatial relationship containing a trigger word and a spatial relationship not containing the trigger word can be processed at the same time; according to the method, the spatial semantic information in the text can be effectively expressed, the semantic dependence among the spatial elements is extracted, manual feature design is not needed, the generalization performance is high, and the accuracy rate is high.

The invention belongs to the technical field of intelligent services, and relates to a semantic recognition method for inscriptions on ancient bronze objects, which comprises the following steps of S1, performing pre-training on a BERT model by adopting an inscriptions-on-ancient-bronze-objects training set to obtain an inscriptions-on-ancient-bronze-objects enhanced context vector; s2, substituting the inscriptions-on-ancient-bronze-objects enhanced context vector into a BiLSTM model to obtain a inscriptions-on-ancient-bronze-objects implicit vector matrix H representing inscriptions-on-ancient-bronze-objects sentence context information; s3, performing semantic role identification and semantic dependency relationship identification on the inscriptions on ancient bronze objects accordingto the inscriptions-on-ancient-bronze-objects implicit vector matrix H; s4, establishing an inscriptions-on-ancient-bronze-objects knowledge graph according to the inscriptions-on-ancient-bronze-objects semantic role label and the semantic dependency relationship; and S5, substituting the to-be-identified inscriptions-on-ancient-bronze-objects into the inscriptions-on-ancient-bronze-objects knowledge graph for identification. The causal association of the inscriptions-on-ancient-bronze-objects description content and the semantic dependency relationship between semantic elements are comprehensively considered and fused, and the inscriptions-on-ancient-bronze-objects meaning can be understood through the context information, so that the recognition result is more accurate.

The invention discloses an English text concept understanding method. The method is an understanding model composed of an English text understanding preprocessing module, an English text keyword concept semantic feature extraction module, an English text keyword and concept semantic dependency relationship extraction module and a candidate answer selection module which are connected in sequence. After an English text and questions related to the English text are processed by the understanding method, related concept answers of the questions can be finally obtained. According to the method of the invention, the problem of English text concept understanding is solved, and the answer result is more accurate than that of a traditional English text understanding method.