155 results about "Sequence labeling" patented technology

The invention relates to an ancient character image recognition and semantic analysis method, which comprises the following steps of: firstly, acquiring an original image containing irregular deformation and material diversity characteristics from the surface of a cultural relic, and eliminating noise and illumination interference through adaptive filtering; calculating a main inclination angle based on stroke feature distribution, and executing inclination angle correction to obtain a processed image; thirdly, separating independent characters by adopting a region growing algorithm based on stroke features, and extracting feature vectors by utilizing a deep convolutional network for identification; and finally, matching context information in combination with an ancient text corpus, and optimizing sequence labeling by adopting a conditional random field algorithm to obtain semantic output. And if the result is not ideal, the inclination angle correction parameter is adjusted through backtracking and iterative optimization is carried out, so that the overall identification accuracy is improved. According to the method, the problems of irregular deformation and material diversity in the ancient character image can be effectively solved, and the recognition precision and the semantic analysis reliability are improved.

The invention discloses a multi-level cross-modal alignment method based on comparative learning, which is used for improving the accuracy and efficiency of multi-modal sentiment analysis. According to the method, a RoBERTa model and a Vision Transform model are used for coding a text and an image respectively, and text representation and image representation are obtained. The global cross-modal alignment module aligns the representation of the text and the representation of the image by adopting a comparative learning technology to enhance the consistency between the text and the representation of the image. Furthermore, through a local cross-modal alignment module, a cross-attention mechanism is used to perform fine-grained alignment on the text and image representations to identify smaller, more specific semantic units in the associated image and text. According to the method, a multi-task learning framework is adopted to integrate cross-modal information from texts and images, sequence tag prediction is carried out through a conditional random field, and terms and emotions in the aspects are recognized and classified. Experimental results show that the performance of the method on a Twitter-2015 data set and a Twitter-2017 data set is superior to that of an existing single-mode model and an existing multi-mode model, and the performance of multi-mode sentiment analysis is effectively improved.

The invention discloses an unregistered geographical name discovery and spatial position reasoning method and a related device, and relates to the technical field of geographic information, and the method comprises the steps: constructing a geographical name space-time derivation relation knowledge graph based on an open source geographic database; inputting the target place name text data into a sequence labeling model to obtain a labeling result, and extracting candidate derived place names based on the labeling result; performing general name segmentation on the candidate derived geographical names to obtain potential basic geographical names, and judging whether the potential basic geographical names meet preset association conditions or not based on a geographical name space-time derivation relationship knowledge graph; if yes, eliminating potential basic place names; if not, determining the potential basic place name as an unlogged place name; and determining the spatial position of the unregistered place name according to the general name of the unregistered place name and the place name space-time derivation relationship knowledge graph. According to the invention, the automation degree of unregistered place name discovery and spatial position reasoning can be improved.

The invention discloses a relation enhancement type named entity recognition method and device, a medium and a program product, and relates to the field of data processing. The method comprises the following steps: performing feature extraction on text data through a BERT model to obtain a plurality of tokens and semantic feature vector sequences; performing different-scale convolution operations by using a convolutional neural network to obtain local character combination features, and fusing the local character combination features with the semantic feature vector sequence to form a feature enhanced semantic feature vector sequence; the Transform model determines a long-distance dependency relationship of semantic feature vectors in the sequence by using an attention mechanism; generating a long entity structure model based on the long-distance dependency relationship, and then determining an enhanced semantic feature vector sequence; and performing tag sequence optimization processing on the named entity through a conditional random field layer, determining sequence labeling loss, boundary detection loss and relation prediction loss, determining total loss, and adjusting a tag sequence so as to determine the named entity. According to the method, the accuracy and integrity of named entity recognition can be effectively improved.

The invention provides a named entity recognition method based on multi-modal large model fine-grained knowledge generation, and belongs to the technical field of information extraction in natural language processing. According to the method, a multi-modal large model MLLM and a large language model LLM are combined to jointly generate fine-grained auxiliary knowledge, the LLM uses a context learning mode to guide the LLM to generate auxiliary knowledge related to a sample, and the MLLM uses manual annotation data to perform fine tuning on the LLM, so that the MLLM outputs the related auxiliary knowledge according to input text and image information; and combining the obtained auxiliary knowledge with the original text, entering a downstream sequence labeling model for training and reasoning, and completing named entity recognition. According to the method, world knowledge and multi-modal information which are beneficial to information extraction are taken into consideration, the performance of named entity recognition of the downstream sequence marking model is improved, and the entities in the text can be recognized more accurately.

The invention relates to the technical field of natural language processing, in particular to a foreign trade letter intelligent semantic error correction and business term optimization method and device, and the method comprises the steps: obtaining a target foreign trade letter, and constructing a target corpus; performing Chinese word segmentation and part-of-speech tagging on the target foreign trade letter; performing term optimization based on the word segmentation result and the knowledge graph; identifying the letter title by using a conditional random field model, and converting the letter title into structured data; a Bi-LSTM-CRF model is adopted to carry out risk point detection, including Bi-LSTM coding, feature engineering and Max-pooling technologies, a part-of-speech sequence is obtained through a softmax function and a Viterbi path, and sequence labeling is carried out to obtain a risk point detection result; and finally, performing Chinese error correction based on the word segmentation result after part-of-speech tagging and the knowledge graph. The recognition and correction accuracy of foreign trade terminologies is improved, and communication obstacles caused by nonstandard use of the terminologies are effectively reduced.

The application discloses a multi-intent spoken language understanding method based on syntax analysis, which comprises the following steps: firstly, obtaining an intent feature matrix and a slot feature matrix according to a user input sentence, and constructing a multi-level intent feature from the intent feature matrix; secondly, obtaining initial intent labels and initial slot prediction labels by using an intent decoding module and a slot decoding module respectively on the intent feature matrix and the slot feature matrix; then, inputting the multi-level intent feature and the initial slot prediction labels into a slot-intent interaction module to obtain an enhanced intent feature matrix, and inputting the slot feature matrix, the multi-level intent feature and the initial intent labels into an intent-slot interaction module to obtain an enhanced slot feature matrix; finally, inputting the enhanced intent feature matrix and the slot feature matrix into the intent decoding module and the slot decoding module respectively to obtain intent labels and slot sequence labels of the spoken language understanding task. The application improves the accuracy of intent recognition and slot sequence labeling and the accuracy of multi-intent recognition.

The invention provides a multi-mode command entity recognition method based on course comparative learning, and belongs to the technical field of natural language processing and computer vision. The method comprises the following steps: respectively carrying out semantic representation modeling on text and image data to obtain text features and image features; multi-level semantics from coarse granularity to fine granularity between the text features and the image features are aligned through staged comparative learning, and association between the text features and the image features is enhanced; fusing the text features and the image features through a gating multi-interest fusion mechanism to obtain fusion features; and performing sequence labeling on the fusion features by using a dynamically weighted conditional random field model. By combining a learnable gated multi-interest fusion module and a category-sensitive dynamic decoding mechanism, the problems of modal noise interference and category imbalance are effectively relieved, and the accuracy and robustness of multi-modal entity recognition and the applicability and stability of the model in a real social media environment are improved.

The invention discloses a personalized English learning recommendation method and system based on a knowledge graph, and relates to the technical field of intelligent education and natural language processing, and the method comprises the steps: collecting the multi-source learning data of a learner; extracting four types of knowledge entities including vocabularies, grammar, topics and skills by adopting a sequence labeling model, and calculating entity association degree through an attention mechanism to construct a hierarchical knowledge graph; constructing a user ability model based on the learner data, and positioning weak knowledge entities and associated entities thereof to form a target knowledge entity set; screening matched contents from a resource library, analyzing and determining a knowledge logic sequence in combination with a knowledge graph path, and generating a personalized recommendation result; and evaluating the effect according to the learned data and dynamically updating the user capability model and the knowledge graph. According to the invention, structured organization of English knowledge, accurate description of learner ability and coherent recommendation of learning content are realized, and the personalized level and effect of English learning are effectively improved.

The invention relates to the technical field of information retrieval and natural language processing, and particularly discloses a minimum evidence span alignment and accurate reference generation method and system. According to the method, candidate terms are obtained through direct numbering and semantic recall, Span-level evidence alignment is achieved through fusion of sequence labeling and interpretable attribution, the boundary robustness is optimized in combination with anti-fact boundary learning, closed-loop mending is triggered based on the minimum span coverage rate, and finally checkable reference with a standardized RefTag is output. According to the method, the problems of coarse evidence granularity, boundary drift and unreviewable reference of a traditional method are effectively solved, the generation illusion is remarkably reduced, and the auditing traceability is improved.

This invention discloses a method and system for labeling prompt words in a large-scale communication domain model based on knowledge graphs. Belonging to the fields of communication technology and artificial intelligence, it addresses the following technical problems: low efficiency in labeling specialized terms in communication domain prompt word annotation, low accuracy and efficiency of automatic recognition, limited prompt word dimensions, and lagging domain knowledge. The method includes: sequence labeling and entity extraction using a trained combined model to output entities; generating relation triples as candidate relations using a trained generator, and scoring and ranking the selected relation triples using a trained discriminator; semantic parsing of user input to extract key entities and intents; multi-hop reasoning from a dynamic knowledge graph to retrieve relevant entities and relations; eliminating illusory content using a trained GAN network; generating entity prompt words from three levels: basic, extended, and application dimensions; and introducing a triple quality verification mechanism.

The application provides a micro-learning service information extraction method based on a deep sequence labeling model, in the micro-learning service information extraction process, a deep sequence labeling model is used: first, an embedding layer is used to map high-dimensional sparse original data into low-dimensional continuous but dense data representation, and semantic information is extracted; after the embedding layer, two continuous CNN layers are used to further mine and summarize the local features of adjacent inputs in the model; an RNN layer is further arranged after the embedding layer, which is used for extracting and modeling time sequence features; a fusion block is arranged after the CNN and the RNN, which is used for combining different types of potential features; a CRF layer is arranged before a final output layer, which is used for the local constraint of an output sequence.

The application relates to the field of natural language processing, and relates to a campus safety emergency extraction method based on multi-task learning, which comprises the following steps: S1: obtaining original news text of a campus safety emergency; S2: establishing a campus safety emergency extraction model, which comprises a text feature representation module, an event type classification module, a trigger word extraction module and an argument role classification module; S3: executing the text feature representation module to obtain a word vector fused with semantic dependency information; S4: executing the event type classification module to obtain a predicted event type feature vector; S5: executing the trigger word extraction module to obtain a trigger word feature vector; and S6: executing the argument role classification module to obtain an argument role. The method solves the problem that the sequence labeling method leads to labeling conflicts and cannot extract overlapping argument roles; and the method solves the problem that it is difficult to fully extract semantic knowledge and dependency relationships in a sequence in a professional field. The feature expression capability of trigger words is strengthened, and the argument role extraction capability is improved.

The embodiment of the invention relates to a model content auditing method and device, computer equipment and a storage medium, and the method comprises the steps: inputting an original text into a sequence labeling model, and outputting a risk fragment, a risk content confidence coefficient and an attack type in the original text; when the confidence coefficient is greater than a first threshold value, inputting the original text, the risk fragment and the attack type into a trained input auditing model, and outputting a first auditing result; when the first audit result is that the input content is compliant, generating output content; and inputting the output content, the risk fragment and the attack type into the trained output auditing model, so that the output auditing model audits the output content. Therefore, the risk fragments and the attack types are primarily screened through the sequence labeling model, the input content is audited through the input auditing model, the output content is audited through the output auditing model according to different application scenes, and the auditing efficiency, the jail break attack resistance, the accuracy and the safety are improved.

The invention discloses a novel power system standard self-adaptive generation method and system based on a knowledge graph, and belongs to the technical field of power grid data, and the method comprises the steps: carrying out the semantic enhancement and sequence labeling of a standard corpus through an entity extraction model, obtaining an entity, extracting an entity relationship in the standard corpus through a relationship extraction model, and obtaining an entity relationship; constructing an electric power knowledge graph; performing mapping matching on the scene query request and the electric power knowledge graph to obtain a to-be-queried sub-graph, and traversing the electric power knowledge graph by taking the to-be-queried sub-graph as a traversal reference and taking a scene node in the to-be-queried sub-graph as a traversal starting point to obtain a candidate standard set; and constructing a suitability evaluation function by taking the maximum coverage of the scene nodes and the maximum matching degree of parameters corresponding to the technical index nodes in the to-be-queried sub-atlas as targets, and scoring the candidate standard set through the suitability evaluation function to obtain a standard recommendation list. The technical problem that the standard adaptability of a novel power system is difficult to improve in the prior art is solved.

The invention relates to the technical field of computers, in particular to a data sharing method and system based on data desensitization, and the method comprises the steps: carrying out the field category prediction of an initial lexical element sequence through a sequence labeling model, carrying out the analysis of the initial field sequence through combining with a preset field, and achieving the quantitative evaluation of the importance degree of an initial field. The blindness of a traditional regularized desensitization mode is avoided, data security and data availability are both considered, priority setting is carried out with the initial lexical elements as units, the desensitization granularity is dynamically adjusted based on the terminal preset permission level, a high-permission terminal can obtain complete data, and a low-permission terminal can only obtain non-sensitive initial lexical elements, so that the user experience is improved. And the data availability of different terminals is ensured while the data privacy is protected.

The invention discloses a synchronous phase modifier fault diagnosis method based on a knowledge graph and a large language model, and the method comprises the steps: obtaining the operation data of a synchronous phase modifier, and carrying out the preprocessing of the operation data, so as to form a unified corpus and a standardized data set; based on the unified corpus, using a BERT model to encode the corpus to obtain context perception vector representation, and based on the context perception vector representation, performing sequence labeling and relation extraction, and outputting structured knowledge data; based on the structured knowledge data, constructing a synchronous phase modifier fault diagnosis knowledge graph; based on a pre-trained large language model, performing model training and optimization by using the standardized data set, understanding and analyzing a fault phenomenon described by a natural language, and outputting a semantic analysis result; and in combination with the output of the knowledge graph and the semantic analysis result, performing hybrid reasoning to obtain a fault diagnosis result. The method has the advantages of high diagnosis precision, high diagnosis efficiency and the like.

The invention provides a training teaching video intelligent analysis and knowledge point automatic marking method based on multi-modal fusion, and the method comprises the steps: collecting multi-modal data in a training teaching process, and carrying out the time alignment processing; extracting feature information of the modal data, fusing the feature information through a cross-modal fusion architecture, and generating a unified teaching behavior representation vector; based on the teaching behavior representation vector, identifying operation steps in the practical teaching process through a sequence labeling model, and determining the category and the starting and ending time boundary of each operation step; matching the identified operation steps with a preset skill knowledge base, and generating a standardized knowledge point label containing knowledge point content, starting and ending timestamps and confidence information; and storing the standardized knowledge point labels into a database, and constructing a retrieval index. According to the invention, the unstructured teaching video is converted into a searchable, navigable and analyzable knowledge unit, and automatic identification and structured marking of practical teaching operation are realized.

This invention discloses a method for constructing a database for the survey of traditional Chinese medicine (TCM) cultural resources based on knowledge graphs, belonging to the field of database construction technology. Specifically, it includes: collecting classical texts and field survey texts; extracting core entity terms using sequence labeling technology; establishing an ontology model including lineage and medicinal material origin dimensions to form a knowledge architecture for the TCM field; extracting entities and events from the text using neural networks, aligning entities to ontology nodes according to the domain knowledge architecture, and generating a basic TCM resource graph; calculating feature vectors of graph nodes, using inference algorithms to mine implicit connections between origin changes and treatment customs, and completing missing semantic edges in the graph; performing multiple path searches for resource nodes, aggregating their associated nodes in the lineage and regional dimensions, and generating a subgraph containing spatiotemporal evolution trajectories; and converting the generated subgraph into topological data to complete the construction of a network survey database.

This application provides a method, apparatus, and electronic device for generating training data based on psychological process modeling. The method includes: acquiring a set of target text segments; inputting the set of target text segments into a trained binary classification model to obtain a classification result for each text segment in the set; selecting at least one target text segment from the set whose classification result indicates it contains cognitive linkage information; inputting the at least one target text segment into a trained sequence labeling model to obtain at least one linkage data group corresponding to each target text segment; inputting the at least one linkage data group and preset constraint rule information into a trained causal language model to obtain psychological link information corresponding to each linkage data group; and constructing training data based on each linkage data group and its corresponding psychological link information. This solution enables explicit modeling of the psychological motivations behind human interaction behavior, improving the quality of training data.

The application discloses a trigger word recognition and positioning system and method based on BIO sequence labeling, comprising: using an encoder to extract a whole sentence semantic vector, judging whether the input text contains a backdoor attack feature; if the backdoor suspicion is detected, positioning the backdoor trigger word, judging whether each word belongs to the backdoor trigger word through the BIO sequence labeling of each token, and outputting a preliminary trigger word position marking sequence; applying trigger position prior knowledge and multi-strategy rules to correct and optimize the results and filter false positives; when the results have uncertainty or are suspected to have attack avoidance, using a predefined known trigger phrase mode library to scan the input text, capturing hidden or variant trigger word modes, and positioning the missed suspicious trigger words; and outputting the final result of backdoor detection. Through the dual-module cooperation on the architecture and the priori and rule fusion on the strategy, the application can robustly detect the text backdoor trigger and accurately position the trigger content.

The application relates to a video segmentation method, a recommendation method, a system, a device and a medium. The segmentation method comprises the following steps: obtaining a target video; performing voice recognition on the target video to obtain a plurality of subtitle sentences corresponding to the target video and starting and ending moments of the subtitle sentences in the target video; calculating time intervals between each subtitle sentence and a previous subtitle sentence according to the starting and ending moments of each subtitle sentence in the target video, and fusing each subtitle sentence and the corresponding time interval to obtain a plurality of fusion sentences arranged in sequence; identifying and extracting each chapter name from the plurality of fusion sentences based on a sequence labeling algorithm, and determining a corresponding chapter starting moment; and segmenting the target video into a plurality of chapter videos according to each chapter name and the corresponding chapter starting moment. The application improves the accuracy of video segmentation.

The application provides an event information extraction method and device and electronic equipment, and relates to the technical field of natural language processing. The method comprises the following steps: obtaining a sentence to be processed; inputting the sentence into a sequence labeling model to obtain a vector corresponding to the sentence, an entity in the sentence, and a trigger word; for each entity-trigger word pair, concatenating the vector corresponding to the sentence, a position vector of the trigger word in the sentence, and a position vector of the entity in the sentence to obtain a concatenated vector; and inputting a plurality of concatenated vectors into a text classification model to obtain event information in the sentence. Thus, the sequence labeling model and the text classification model are used to automatically extract event information in the sentence, thereby reducing the labor cost, and the event information can be accurately extracted for sentences in different fields and scenarios, and the portability is high.

The invention relates to a text classification method fusing semantic information and structural information, and aims to solve the problems of text graph noise interference, incomplete semantic capture, rigid information fusion and the like in existing graph neural network text classification. The method comprises the steps that firstly, a target text is preprocessed, and phrase blocks with complete semantics are extracted through BERT sequence labeling and B-I-O labeling; constructing an enhanced text graph by taking the phrase blocks as nodes and combining various relationships such as self-loop edges and syntactic dependency edges and cross-sentence anaphora connection; afterwards, redundant edges in the picture are cut through attribute-enhanced personalized PPR, and noise interference is weakened; and finally, inputting the optimized text graph into gradient gating fusion GNN, adaptively integrating BERT context features and global dependency information, outputting node features and completing classification. The method effectively breaks through the limitation of a traditional method, realizes deep fusion of semantic and structural information, has higher classification accuracy, stronger generalization ability and good stability, and is suitable for various short text, long text and professional field text classification scenes.

The invention discloses a radar confrontation domain knowledge extraction method and system based on a BERT model. The method comprises the following steps: firstly, establishing a dynamic and static combined radar confrontation knowledge unified description architecture; performing unstructured data preprocessing on radar confrontation text data from different data sources; then performing entity recognition and relation extraction by using a sequence labeling method based on a pre-training language model; and finally, based on a radar confrontation knowledge unified description architecture, performing knowledge fusion and knowledge representation on knowledge data extracted from different data sources to form a radar confrontation unified description knowledge set. The system is used for realizing the radar confrontation domain knowledge extraction method based on the BERT model. According to the method, knowledge extraction and unified description can be carried out on unstructured text data in the radar countermeasure field, and the method has the advantages of high specialty, high accuracy and high extraction efficiency.

The present invention provides a method, system, storage medium and electronic device for predicting conversation emotions based on emotion dynamics, and relates to the field of emotion prediction technology. The present invention includes obtaining and preprocessing historical conversations to be predicted; obtaining a global inference vector based on the preprocessed historical conversations; weighting attention for each sentence of historical conversation based on the global inference vector; obtaining an emotion dynamic vector based on the weighted historical conversations; obtaining a global optimal emotion sequence based on the emotion dynamic vector, and determining the emotion prediction result. Modeling is performed based on Gross's emotion regulation theory, which has scientific psychological theoretical support, making the prediction results of the model more interpretable; distinguishing the roles in the conversation, modeling the damping (persistence) and contagiousness of emotions; converting the conversation emotion prediction task into a conversation emotion sequence labeling task, analyzing the transfer relationship between emotions, and making the prediction results more reasonable.

The invention provides an entity recognition-oriented aviation intelligence field corpus construction method, which relates to the field of data processing, and comprises the following steps: acquiring aviation intelligence data, and normalizing the original corpus of aviation intelligence; in combination with an aviation intelligence exchange model AIXM, constructing a domain corpus labeling system and forming a labeling criterion, and labeling to obtain an aviation intelligence labeling corpus; and in the construction process, superposition of a sequence labeling layer is carried out based on a regular template, vocabulary feature information is incorporated into character representation, a character representation layer is adjusted to introduce dictionary information, a bidirectional long-short term memory network model is adopted to carry out entity recognition, and construction of the corpus in the aviation intelligence field is completed. The technical problems that no industrial special corpus exists in the current aviation field, and data updating and processing depend on manual work are solved.

The invention discloses a matcha industry innovation and entrepreneurship project tracking system, which belongs to the technical field of project tracking and comprises a project construction drawing design module, a project construction drawing decomposition module, a project construction drawing sequence labeling module, a project construction material drawing coordinate conversion module and a project construction progress planning distribution module. The system comprises a project construction information acquisition module, a project construction material sequence marking module, a project construction material image coordinate conversion module, a project construction progress tracking module and a project construction progress sharing module. The whole process is collected through equipment, tampering cannot be carried out manually, the problem of progress misjudgment is avoided, and meanwhile, videos are collected in real time, and responsibility tracing is facilitated.

The present invention relates to a Chinese named entity recognition method based on local and global character representation enhancement. Existing Chinese named entity recognition methods model it as a character-based sequence labeling problem, but a single Chinese character vector is difficult to represent independent semantics, which brings about entity boundary and type recognition errors. The glyph structure of Chinese characters and their related domain terms contain information specific to domain entities, and effective use of this information is conducive to solving the above problems. To this end, the present invention uses an autoencoding mechanism to fuse the radical structure embedding, radical sequence embedding and contextual semantic embedding of characters to obtain a local character representation; and uses an interactive gating mechanism to combine the global domain term representation corresponding to the character with the local character representation to obtain an enhanced character representation; finally, the enhanced character representation is sent to the Bi-LSTM and CRF layers to obtain a character sequence label. Experiments on a domain Chinese named entity recognition dataset show that the present invention is effective.

The invention discloses a two-channel electric power entity recognition method and system based on BIBASRU and improved Reformer, and relates to the technical field of natural language processing and electric power information extraction, and the method comprises the steps: employing a MacBERT model to carry out pre-training semantic coding, and extracting context semantic features; constructing a BIBASRU structure as a first channel; a Reformer model based on ROPE optimization is constructed to serve as a second channel, and learnable position embedding is replaced with an ROPE coding mechanism; designing a dual-channel fusion mechanism, and dynamically weighting and integrating two paths of feature outputs; and finally, introducing a CRF layer to carry out sequence labeling decoding. According to the method, high-precision and high-efficiency entity recognition is realized in professional texts such as electric power regulations, scheduling logs and equipment reports, the method is remarkably superior to traditional BERT-BiLSTM-CRF models and the like, and the method is suitable for intelligent power grid knowledge graph construction and automatic information extraction scenes.