52 results about "Sequence annotation" patented technology

The invention relates to a document information extraction method and device based on sequence labeling and a learning model. The method comprises the following steps: training at least one sequence labeling algorithm model to obtain at least one offline sequence labeling algorithm model; Determining the accuracy of the annotation information in each of the offline sequence annotation algorithm models, and converting a to-be-processed document into a text document; Obtaining document structure format property information from the to-be-processed document; And inputting the text document and the structural format property information into the offline sequence labeling algorithm model to obtain labeling information corresponding to the document information in the document. According to the method, the key information of the document can be extracted by using the sequence labeling technology. And by using a multi-model fusion technology, different key information in the document can be extracted by using an optimal model. In addition, business rule reasoning and calculation are carried out on a typeface extraction result, and the application range is wider.

The embodiment of the present application provides a signature information extraction method and device, which can extract the signature information of a rule very quickly and conveniently by extracting the structured information in each statement separately by using a regular expression. Extracting unstructured information uses machine learning classification model and character granularity sequence annotation, which can solve the limitation of traditional way of using mail template alignment to get extracted information. In the implementation process, the TF-IDF word frequency feature and tagged sequence feature are extracted, and the extracted TF-IDF word frequency feature and annotation sequence feature are inputted into address binary classification model and character granularity sequence annotation model respectively, and the name information and address information in each sentence are obtained. Thus, by extracting the TF-IDF word frequency features, the address information canbe identified completely, and the tagged sequence features are used to greatly reduce the negative impact of wrong word segmentation on the identification of name information, so as to accurately extract the mail signature information.

The embodiment of the invention discloses a sequence labeling method, which comprises the following steps: obtaining a training sample set, wherein the training sample set comprises a plurality of pieces of training sample data, and each piece of training sample data comprises an input text sequence and a label corresponding to the input text sequence; preprocessing each piece of training sample data to obtain vector data corresponding to each piece of training sample data; inputting vector data corresponding to each piece of training sample data into a first-order hidden Markov model to construct a feature vector matched with each piece of training sample data; inputting the feature vectors corresponding to the sample data into a neural network model for training to generate a sequence labeling model; and inputting the to-be-labeled sequence into the sequence labeling model to obtain a target label sequence corresponding to the to-be-labeled sequence. The embodiment of the invention further discloses a sequence labeling system, computer equipment and a readable storage medium. The embodiment of the invention has the beneficial effect that the sequence annotation is more accurate.

The invention discloses a data benchmarking method and device and a storage device. The data benchmarking method comprises the steps that original data information is extracted from a data table to bebenchmarked, and the original data information comprises field names and field annotations corresponding to the field names; identifying the field annotation based on a sequence annotation model of deep learning to obtain a characteristic word corresponding to the field name; carrying out first text matching on the characteristic words corresponding to the field names and standard data elements in a standard library; and verifying a result output after the first text is matched. By means of the mode, text matching is conducted on the basis that the feature words are recognized, the credibility of a text matching result is improved, and the mismatching rate in the benchmarking process is reduced.

The invention discloses a bioinformatics method for constructing an antibiotic resistance genbank in the biotechnology field. The method comprises the following steps that: searching the protein sequence of a resistance gene in a GenBank; selecting a sequence with high accuracy as an initial sequence; adopting a Clustalw method for comparison; constructing a hidden Markov model, and searching a GenBank protein database to obtain all sequences which contain protein conserved sites; according to the E values of the sequences and the annotation information of the sequences in the GenBank database, removing the sequences which are highly homologous and do not conform to requirements; and after repeated sequences are removed, adding species annotation information; and integrating all protein sequences to finish database construction. By use of the method, the annotation information and the comparison similarity of a sequence can be comprehensively measured, and sequence collection speed andaccuracy can be improved. By use of the method provided by the invention, the construction of the antibiotic resistance genbank can be finished, and a basic data is provided for researching the primer design, the data analysis and the sequence annotation of the resistance gene.

The invention discloses a medical field entity classification method fusing entity keyword features. The method comprises the steps: text vectorization operation; feature extraction; and sequence annotation. According to the medical field entity classification method fusing entity keyword features provided by the embodiment of the invention, a TFIDF is adopted to assist in constructing a keyword table, the keywords are used as feature input models, a BERT model is adopted to perform text vectorization operation to generate word vectors, the word vectors are input into a BILSTM-CNN hybrid modelto learn features, and then sequence labeling is performed through a CRF layer, so that medical field entity classification can be realized, and the accuracy, recall rate and F1 value of medical field entity classification can be greatly improved.

The invention discloses a Chinese rhythm boundary prediction method based on graph-to-sequence, which specifically comprises the following four parts: (1) word embedding representation characteristics: converting the characteristics into digital representation, so that the technology of mapping words into real number field vectors is called word embedding; (2) a text time sequence feature extraction model, wherein annotation of rhythm boundaries is sequence annotation in the time dimension; (3) text space information: inputting a text sequence, processing the text sequence into a graphic structure, and processing a dependency relationship between rhythm boundaries by adding space information; and (4) combining spatial and temporal features. Time sequence information and space information of the text are combined together to serve as a new feature, and the accuracy of the final rhythm boundary is improved.

The invention relates to a named entity identification method based on a time convolution network. The method comprises the following steps of: firstly, constructing a feature representation layer which mainly consists of a word vector and a character feature layer, wherein the word vector layer and the character vector layer respectively accept words and characters as input, and respectively mapdiscrete One-hot representations to respective continuous dense low-dimensional feature spaces; splicing the word vectors and the character-level vectors to represent features of the words in a particular semantic space; secondly, taking the spliced features as input of a time convolution network, extracting different features through the time convolution network with different fusion convolutionkernel sizes, and obtaining final features h1h2... hn; finally, taking the obtained features as input of a CRF layer; and after the CRF further restrains context annotation, outputting sequence annotation results y1y2... yn. Compared with an existing LSTM network, the TCN network has the advantages that the recognition precision is slightly improved, and the training time is only about 1/3 of thatof the LSTM network.

The invention provides a sequence labeling method and device, a storage medium and computer equipment, and a sequence comprises a to-be-labeled word and a labeled word, and the sequence is used for generating a text. The method comprises the steps: obtaining the sequence; identifying context information of words to be labeled in the sequence; according to the context information, determining a second part-of-speech of the to-be-annotated word in combination with the first part-of-speech of the annotated word adjacent to the to-be-annotated word, wherein the second part-of-speech is used for annotating the to-be-annotated word. According to the method, the part-of-speech of the to-be-annotated word is annotated according to the context information of the to-be-annotated word in the sequence, the part-of-speech annotation accuracy can be improved, the sequence annotation effect is improved, and therefore text generation is effectively assisted.

The invention provides a method for recognizing text segments by using sequence annotation, which comprises the following steps: A, respectively segmenting different text segments of a sample set intoclause sets, and annotating the clause sets by using semantic feature vectors to form semantic feature vector sets; B, performing clustering training on the semantic feature vector set to obtain a clustering model, and performing cluster numbering on each object of the clustering model to form a sequence model; C, establishing mapping between the sequence model and the different text fields, andtraining a sequence labeling model for the mapped cluster sequence; and D, sequentially applying the sequence model and the sequence labeling model, and segmenting the text to be segmented. The methodperforms standardized modeling by taking the sample set as a database template. And during subsequent text segmentation recognition, the method includes standardizing the sentence pattern model in the to-be-segmented text, and mapping the standardized sentence to the sentence features according to the model, so that different expressions representing the same semantics can be expressed to complete text segmentation recognition.

The invention provides an antibody calculation optimization method based on a genetic algorithm. The method covers algorithms such as peptide chain processing, epitope recognition, sequence annotation, CDR H3 sequence design, antibody modeling, molecular docking, antibody property evaluation and the like, and has a full-process automatic antibody design function. Based on known antibody sequence data, a variant antibody sequence formed by combining random sites and random residues is iteratively generated and evaluated by utilizing a genetic algorithm aiming at a heavy-chain highly variable H3section (CDR H3) of the antibody and is subjected to comprehensive scoring comparison with an original antibody, so that an optimized antibody is obtained or a low-quality antibody is removed, and finally, a candidate antibody sequence library is generated, and the biophysical property of the candidate antibody is predicted. According to the invention, basic elements of an antibody calculation optimization process are integrated, and automation of the process is realized on the same platform.

The invention discloses a text entity detection method, and the method comprises the steps: carrying out the matching of each statement instance in a target statement through a seed entity set to obtain a matching result, and generating annotation data corresponding to the target statement according to the matching result; querying a statement instance matched with an unlabeled corpus word frequency table in the target statement, and modifying the labeled data according to a query result to obtain local labeled data; training a sequence annotation neural model by utilizing the local annotationdata; and performing sequence annotation on the unannotated corpus in the target statement by utilizing the trained sequence annotation neural model so as to obtain an entity set of the target statement. According to the method, high-quality entity mining can be realized on the premise of not being limited by the quality and the scale of the unlabeled corpus. The invention further discloses a text entity detection system, a computer readable storage medium and electronic equipment, which have the above beneficial effects.

The invention discloses a sample expansion method and device, a terminal and a readable storage medium. The method comprises the following steps of: selecting sample data from a preset labeled sampledata set as seed data, selecting word data based on the seed data, obtaining seed data of a marked sample data set, obtaining a word type of the word data, determining an expansion mode of the markedsample data set based on the word type, updating the word data in the seed data based on the expansion mode, and taking the updated seed data as expansion sample data to expand the marked sample dataset. Sample data expansion is carried out on the labeled sample data through different expansion modes, according to the method, the cost of obtaining the annotation sample is reduced, the sample expansion efficiency is improved, meanwhile, the generated expansion sample data and the annotated sample data obey the same data distribution, and it can be guaranteed that the model generated by training the sequence annotation model through the expansion sample has very high robustness and accuracy.

The present disclosure provides a method of extracting a relationship from a text, comprising: generating a sequence annotation of the text using a text annotation model, the sequence annotation comprising an annotation for a word in the text, the annotation comprising an entity annotation of the word and a relationship role of the word, the relationship role comprising one of a subject, a predicate, and an object; generating an entity relationship sequence of the text according to the sequence label; and extracting a relationship result set of the text according to the entity relationship sequence. According to the method and device for extracting the relationship from the text, the problems that in the prior art, a relationship extraction method is low in accuracy, poor in universality and low in extraction efficiency can be effectively solved.

The invention discloses a training file generation and evaluation method and device, a computer system and a storage medium, and the method comprises the steps: receiving an original file, obtaining the domain information and training entity of the original file, and processing the original file according to the domain information and training entity, and obtaining a labeled file; identifying semanteme of the annotation file through a preset natural language understanding model, and performing sequence annotation on the annotation file to obtain a training file; and inputting the training fileinto an intelligent search model corresponding to the domain information to obtain a training result, calculating the training result through a hit analysis algorithm to obtain a hit rate, and summarizing the training file and the hit rate to generate a hit analysis report. The technical effect of automatically obtaining the training file is achieved, the generation quality and the generation speed of the training file are guaranteed, and the problem that the labeling quality of the training sample cannot be guaranteed due to the fact that the real hit rate of the training sample cannot be obtained at present is solved.

The invention discloses a sequence labeling method and system, a computer readable storage medium and computer equipment, and the sequence labeling method is based on a joint training mode, and specifically comprises the steps: a data processing step: labeling an original data text in a double-pointer mode, and obtaining a labeled data text; a sequence annotation model construction step: traininga dynamic pre-training model according to the annotation data text to obtain a sequence annotation model; and a sequence labeling step: labeling the real-time data text through the sequence labeling model. According to the method and the system, more accurate information can be obtained, and the model can be deduced according to the context of the entity, so that the problem of multiple meanings of one word can be solved, and the deducing speed is higher.

The present invention relates to a method for automatically extracting a subject of an argumentative article, and belongs to the technical application field of natural language processing. The method disclosed by the present invention comprises: based on the sequence annotation strategy of the random field of the statistical condition, by analyzing semantic features and position characteristics of the subject in the title of the argumentative article and combining with performance of the trained corpus, establishing a commonly used word dictionary and an important word dictionary; using information such as dictionaries and words, locations and the like to carry out sequence feature annotation on the title of the argumentative article; and using the annotated corpus to train and generate the model, so that unknown data can be predicted, the relatively high accuracy can be ensured, and the applicability of the algorithm in different scenarios can be improved. According to the method disclosed by the present invention, automatic extraction of the subject in the argumentative article by the computer can be effectively realized, the main display object of the article can be displayed in an intuitive form, related information of the object can be quickly mastered by the reader in a facilitated manner, related content retrieval and comparison can be facilitated, and automatically extracted phrases can be provided for the computer to carry out various follow-up analysis.

The invention discloses an adversarial interpolation sequence-based annotation data enhancement method and device, equipment and a medium. The method comprises the following steps: acquiring first sample data containing sequence labels; inputting the first sample data into a preset language model, outputting candidate word vectors conforming to context semantic constraints, and forming enhanced second sample data according to the candidate word vectors; and interpolating the first sample data and the second sample data by adopting an adversarial interpolation method to obtain interpolated enhanced sample data. According to the sequence annotation data enhancement method provided by the embodiment of the invention, the language model is utilized to provide the candidate word vector conforming to the context constraint, and the adversarial interpolation is utilized to consider the task characteristics, so that a more difficult sample which enables a machine learning algorithm to easily generate misjudgment is generated, the effect of the sequence model under low resources is improved. The problem that the accuracy of the model is influenced by less annotation data is solved.

The invention discloses a speech recognition result processing method, which comprises the following steps: carrying out semantic feature labeling on a speech recognition result to obtain a semantic labeling result; identifying the semantic annotation result by adopting a sequence annotation model to obtain annotated mark point data; wherein the sequence labeling model is obtained by carrying outdeep learning training according to training data of labeled punctuation; and arranging the labeled punctuation data to obtain a final punctuation adding result. The speech recognition result is subjected to punctuation adding through the sequence labeling model, the punctuation adding efficiency is improved, and good real-time performance is achieved. The invention also discloses a speech recognition result processing system, a computer device and a computer readable storage medium, which have the above beneficial effects.

The invention discloses a language sequence labeling method and device, a storage medium and computer equipment. The method comprises the steps of generating a cross-language vector based on a sourcelanguage vector and a target language vector; generating a language correspondence relationship according to the cross-language vector, the language correspondence relationship comprising a relationship corresponding to the source language and the target language; converting the source language sequence annotation data into conversion data according to the language corresponding relation; trainingthe source language sequence labeling data and the conversion data to obtain a cross-language sequence labeling model; and performing sequence labeling on the target language based on the cross-language sequence labeling model. According to the method and the device, the technical problem of inaccurate and incomplete annotation caused by lack of annotation resources of the target language in a language sequence annotation method in related technologies is solved.

The invention provides a resume information extraction method based on cascading sequence annotation. The method comprises the following steps of 1, analyzing a pdf resume by using a pdfminer, and converting an original pdf into a multi-line text representation; wherein the process mainly solves the problems of disordered sequence and wrong broken lines; step 2, training process data marking, utilizing remotely-supervised data back marking and combining similar items in the marking process; step 3, resume information block division, for sentences obtained through pdfminer, judging the block where each sentence is located according to the classification of each sentence; and step 4, realizing information extraction at a sentence level and a short text fragment level by utilizing the double-layer sequence labeling model. The method is advantaged in that filtering is subsequently realized by utilizing resume block information, so the recall rate is effectively improved, and meanwhile, accuracy is not greatly reduced; through four stages, extraction of the resume information can be effectively realized.

The invention provides a method and a system for extracting dynamic information of the smart home industry, and provides a method for constructing automatic industry dynamic trend capture and automatically generating a report on the basis of industry dynamic data capture and extraction tasks in the field of smart home. According to the method, a smart industry dynamic data extraction mode combining industry priori knowledge and natural language processing sequence annotation can be provided on the basis of the smart home industry background and article structural information extraction, and meanwhile, an industry research report is automatically generated in combination with a text classification model based on deep learning and paragraph abstract extraction of multiple types of indexes. Moreover, the method is a process where a machine learning algorithm is deeply combined with business features of the smart home industry, a natural language analysis business process with a good prediction effect is researched through a large number of practices, the algorithm is efficient and highly targeted, and the process flow highly conforms to a data analysis business, and the success rate of data extraction and report generation is relatively high.

The invention relates to a sequence labeling method in natural language processing, and the method comprises the steps: obtaining a text sequence, inputting the text sequence into a sequence labelingmodel, obtaining a target path, enabling each node in the target path to be a label in a preset label set, and enabling the preset label set to comprise each label corresponding to m coding modes; andarranging the nodes in the target path according to a sequence from first to second in the target path to obtain a labeling sequence corresponding to the text sequence. According to the invention, atext sequence sample coding result is trained according to a plurality of coding modes to obtain a sequence labeling model; the sequence model label is used for inputting the text sequence, the inputtext sequence is processed through the sequence model label, the annotation sequence corresponding to the text sequence is output, annotation of the text sequence is not limited to a single coding mode, and therefore universality of the sequence annotation model to different inputs is improved, and the effect of sequence annotation is improved.