109 results about "Legal instrument" patented technology

The invention relates to a legal document named entity identification method, in particular to a theft case legal document named entity identification method based on adversarial transfer learning, which comprises the following steps: (1) constructing a legal named entity data set, (2) constructing an adversarial transfer learning neural network model, (3) testing the model, and (4) discussing a result. The method is practicable and clear in structure, has a reference value, can improve the accuracy of legal named entity recognition, can be applied to legal knowledge graph construction, multi-person multi-section case factual decomposition, evidence association analysis, judicial knowledge graph construction and the like based on legal named entity recognition.

The invention discloses a named entity recognition method and system for legal document multi-strategy fusion, and the method comprises the following steps: building a source data corpus, carrying outthe part-of-speech tagging and sequence tagging of the source data corpus, and carrying out the model pre-training; training the labeled data through a BiLSTM-Attention-CRF (Bipolar Long Short Term Memory-Attention-Content Random Field) model to obtain a trained first model; improving the trained first model; establishing a target data corpus, randomly extracting data from the target data of thelegal instrument, and generating a plurality of training sets; carrying out transfer learning on the plurality of training sets, and training the improved first model to obtain models trained by the plurality of training sets; and integrating the models trained by the plurality of training sets by adopting a voting mechanism in ensemble learning to obtain a second model, and performing named entity identification of legal documents by the second model to obtain a final named entity identification result. According to the method, the accuracy and recall rate of named entity recognition are improved under the condition of insufficient annotation corpora.

The invention relates to a knowledge graph-based case information semantic retrieval method and apparatus. The knowledge graph-based case information semantic retrieval method comprises the steps: constructing a legal knowledge graph according to a legal document; carrying out simple identification and intention identification on the questions input by the user; defining a sparql language query template, matching the corresponding sparql language query template according to the intention recognition result to perform first retrieval in the legal knowledge graph, and endowing a first confidencecoefficient to a first retrieval result; building a full-text search engine, carrying out second retrieval on the simple identification result in the full-text search engine, and endowing a second confidence coefficient to a second retrieval result; and outputting a final retrieval result according to the first confidence coefficient and the second confidence coefficient. According to the method,the relationship between entities can be fully mined, complex multi-hop semantic retrieval is completed, a retrieval result can be returned based on a full-text search engine even if semantic retrieval of the legal knowledge graph has output which cannot meet requirements, and the retrieval efficiency and accuracy are further improved.

The invention relates to the field of data processing, and discloses an asset information recognition method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining a legal document, analyzing the legal document, so as to obtain target paragraphs, and enabling the target paragraphs to comprise a litigation participant paragraph and a fact paragraph;performing information extraction on the litigation participant paragraph to obtain litigation participant information; performing anaphora resolution on the fact paragraph according to the litigation participant information to obtain a resolved fact paragraph; and performing information extraction on the resolved fact paragraph by using a pre-trained text classification model and an entity recognition model to obtain asset information. By training the text classification model and the entity recognition model, the asset information in the legal instrument is recognized and extracted, and compared with a traditional rule traversal method, the method is more universal, automatic recognition can be carried out, and the accuracy of information recognition is improved.

The invention provides a multi-task network construction method, which comprises the following steps: processing a text consisting of case description and facts in a known criminal legal document to obtain a preprocessed text and word vectors of a plurality of words; processing and training the preprocessed text and word vectors of a plurality of words by adopting a multi-task network formed by aspecial network of a shared network, charge prediction and legal provision prediction, and finally obtaining a trained multi-task network. The invention also provides a multi-scale charge legal provision joint prediction method, which comprises the following steps: processing a text consisting of case description and facts in the charge legal document to be predicted to obtain a preprocessed textand word vectors of a plurality of words, processing by adopting the trained multi-task network, and judging to obtain a predicted charge and a predicted legal provision. According to the method, thetwo tasks are combined to extract the features together, the prediction accuracy can be effectively improved, and meanwhile, the coverage rate and diversity of prediction on categories are improved.

The invention discloses a legal instrument generation method and device and electronic equipment. The legal document generation method comprises the steps of obtaining structured data and a pre-poseddocument of an in-process case, wherein the structured data comprises first entity data of the in-process case; extracting entity content from the pre-posed document based on an entity recognition engine; extracting second entity data of the in-process case from the entity content based on a legal knowledge graph; performing text definition and logic assembly on the first entity data and the second entity data based on a knowledge base, wherein the knowledge base comprises a label definition base and an antiquing statement assembly rule base; and performing text assembly on the entity data subjected to text definition and logic assembly based on a template model library to generate a legal document. Therefore, the legal instrument generation efficiency is improved, judicial resource wasteis avoided, and the legal instrument generation method has higher adaptability.

The invention discloses a rule and model combination-based legal document information extraction method, and the method comprises the following steps: creating a dictionary and a document entity extraction rule, and performing legal document entity extraction by adopting a rule-based extraction method; taking an extraction result as primary annotation data to perform data annotation operation, andtraining a model and publishing the model; performing legal document entity extraction by adopting an extraction method based on combination of rules and models, and selecting a result with a high evaluation result as an output result; evaluating the result, and if the requirement is met, ending; if not, continuing iteration. The invention further discloses a legal instrument information extraction system which comprises a data acquisition module, an information extraction module, a data labeling module, a data set management module and an evaluation module. According to the method, rules anda model method are effectively combined to form complementation, the legal document information extraction effect is improved, and meanwhile, the expansibility and portability of the system are higher.

The invention relates to the technical field of computers, and provides a case auxiliary information prompting method and device, a storage medium and a server. The method comprises the steps of obtaining a judgment result text of a case; extracting measurement of penalty information and applicable laws of the case from the judgment result text; extracting case condition information of the case from a legal document related to the case; inputting the case information into a pre-constructed first neural network model, and determining the reasonability of the measurement of penalty information according to an output result of the first neural network model; inputting the case information into a pre-constructed second neural network model, and determining whether the applicable law article iscorrect or not according to an output result of the second neural network model; and if the reasonability of the measurement of penalty information is lower than a preset threshold or the applicablelegal provisions are incorrect, outputting auxiliary information used for indicating that the case judgment is wrong. By adoption of the method, a judge can be assisted in supervising and rechecking ajudgment result of the case.

The invention discloses a legal instrument feature extraction method, a related device and a storage medium, and the method comprises the steps: carrying out the pre-recognition of a legal instrument,and determining a paragraph partitioning model and a feature extraction model corresponding to the legal instrument; wherein the feature extraction model comprises a corresponding relationship between a document paragraph and a document element; performing document paragraph division on the legal document through the paragraph division model; and extracting document elements corresponding to thedocument paragraphs from the legal documents after the document paragraphs are divided through the feature extraction model, and outputting an extraction result of the document elements.

The invention relates to the field of artificial intelligence, and provides a legal document generation method, device and equipment and a storage medium. The method comprises the following steps: receiving voice data input by a user; converting the voice data into text data to be processed through a voice conversion technology; creating Graph data, identifying a specific separator through the Graph data, processing a to-be-processed text by clauses, and storing a plurality of candidate key sentences through nodes of the Graph data; inputting the plurality of candidate key sentences into a word segmentation model to obtain words after word segmentation; constructing an edge set E, calculating the sum weight of any two candidate key sentences through the edge set E and the content coveragerate of the text sentences, and linking the candidate key sentences of which the weights are higher than a threshold value; analyzing the importance of the candidate key sentences of which the weightsare higher than a threshold value; sorting the candidate key sentences, and sequentially extracting the candidate key sentences to obtain a plurality of abstract sentences. Repeated labor is reducedto the maximum extent, and the case handling efficiency is improved.

The invention discloses a method, device and system for pushing court trial questions. The method comprises the steps of acquiring case information and a legal knowledge graph of legal cases , wherein the legal knowledge graph at least comprises multiple nodes and associated information and direction information among the multiple nodes, and each node represents one of the following information in the legal knowledge graph: elements for generating legal documents or judgment key points; generating court trial questions based on the case information and the legal knowledge graph; and pushing the court trial problem to the target object. The technical problem that in the prior art, the court trial problem is manually proposed by the target object, and consequently the judgment efficiency islow is solved.

The invention discloses an analysis method for sentences in legal instruments. The method includes reading legal texts of the sentences; analyzing the legal text and segmenting the legal text into text units, wherein the text units comprise beginning, case numbers, parties, origins, facts, evidences, judicial authority or contractor opinions, judgment results and ending information; extracting text units containing sentences; carrying out matching filtering on the text units containing the sentences and preset keywords through a regular expression to obtain a plot set A, carrying out word segmentation on the text units containing the sentences by using vocabulary segmentation, and inputting the text units containing the sentences into a preset sequence labeling model to carry out prediction classification to obtain a plot set B; and performing duplicate removal and combination on the plot set A and the plot set B to obtain a plot set C. According to the method, the work efficiency of judicial workers can be improved, the deviation caused by human subjective factors when the sentences are identified is reduced, and the accuracy and convenience of the method are improved.

The invention discloses a method and system of automatically comparing article similarity to assist legal document error detection. The method comprises the steps of establishing a professional and perfect legal language corpus as a basis material for comparison and error detection; and then detecting the input of a user in real time during the legal document editing process, judging and feeding back the input of the user by comparing the similarity of the input of the user and the corpus, and judging and feeding back the full text by comparing the similarity of the full text structure inputted by the user and the corpus after the full text input is completed. Aiming at the input and editing of the legal documents, the method can detect the input error and nonstandard words of the user inreal time, and can provide modification suggestions for the legal documents inputted by the user overall, thereby effectively solving the above technical problems.

The invention discloses a legal case similarity discrimination and pushing method, and the related discrimination method comprises the steps: carrying out the classification of a target legal case, and extracting historical cases of the same type from a historical case database according to the obtained case types to form a candidate set; performing event sequence representation on the target legal case and each similar historical case in the candidate set; according to the event sequence measurement model, calculating the distance between the event sequence corresponding to the target legal case and the event sequence corresponding to each historical case in the candidate set; and based on the distance of the event sequence and in combination with a scoring function, calculating the similarity between the target legal case and historical cases in the candidate set. According to the method, more comprehensive and accurate class case identification can be realized; meanwhile, the legalinstrument is expressed as a time sequence event sequence, similarity calculation is carried out on the basis of an unsupervised mode, historical cases with high scores are selected for pushing, manpower input is greatly reduced, and pushing intelligence can be better achieved.

The invention discloses a triad similar task-based trial risk early warning method, and the method comprises the following steps: acquiring data from a legal document database, performing keyword matching, and extracting case description and a judgment result; carrying out the text preprocessing, including word segmentation, data enhancement and other operations, and generating multiple triples; generating a case vector by using an LSTM network, and carrying out triple similar task judgment to obtain a representation model of the case; preprocessing the current case and the historical case respectively and inputting the preprocessed cases into a representation model to obtain respective case representations; calculating a topM case with the highest similarity, obtaining a judgment result representation of the topM case, and finally calculating the similarity between the result and the judgment result of the current case to obtain a deviation risk value. According to the method, the current legal document can be analyzed in a triple similarity calculation mode, and the case with relatively high similarity is found out from the historical electronic case. A recommended judgment rangeis given according to the judgment result of the past case, thereby achieving the purpose of trial deviation early warning.

The invention relates to the technical field of big data processing, and provides a legal element processing method and device, electronic equipment and a storage medium. The method includes: obtaining a legal image file; performing optical character recognition (OCR) on the legal image file to obtain a legal document; inputting the legal documents into the intelligent cataloguing model to obtaina plurality of types of documents of the legal documents; inputting each category document into the element extraction model to obtain a first legal element in the category document; extracting eventinformation from the first legal element; judging whether the event type of the event information belongs to a key attention event type or not; if the event type of the event information belongs to the focused event type, querying an associated legal element related to the first legal element through a legal element knowledge graph; and outputting the first legal element and the associated legal element. The invention can be applied to the fields of smart government affairs, smart laws and the like needing legal element processing, so that the development of smart cities is promoted.

The invention provides a quantum clustering system and method for law texts. The quantum clustering system comprises a law document input system, a law text quantum clustering mining system, a database and a law document output system; and related law texts are input by the law document input system, preprocessed and cleaned by the law text quantum clustering mining system and then stored in the database. The law text quantum clustering mining system conducts clustering analysis on the law texts completing preliminary processing and cleaning, wherein the processed law texts or records are described by using quantum state functions, all the texts and records are grouped, the similar texts and records are gathered into one cluster, a large number of texts can be quickly processed by means ofthe superimposition characteristic, the coherence characteristic and the entanglement characteristic of the quantum state functions, text clustering in any shapes and abnormal phenomena and behaviorsof the texts are discovered, and a quantum clustering mining result is output graphically for a decision-making reference of a user.

The invention provides an information extraction method based on legal instruments. The method comprises the following steps: 1, acquiring unsupervised data of the legal instruments, performing data preprocessing and data cleaning on the unsupervised data of the legal instruments, removing noise in the unsupervised data of the legal instruments, and forming a corpus based on the legal instruments; and 2, carrying out new word discovery on the corpus based on the legal instruments by adopting a new word discovery algorithm to obtain a preliminary legal instrument field dictionary candidate set, and carrying out screening processing of denoising, removing general words and combining high-frequency words on the preliminary legal instrument field dictionary candidate set to obtain a final legal instrument field dictionary. The semantic comprehension capability is high, manual annotation is reduced, the learning capability of general vocabularies is enhanced, the precision is high, reference is provided for information extraction of other downstream tasks of legal instruments or other fields, the model can obtain better generalization performance, and the entity classification effect of the model is improved.

Document scanners from different regions of the legal field from 2014 to 2018 are collected, dozens of resolutions are covered, and on the basis of a large amount of legal document data, domain knowledge is fused in combination with legal documents, so that research on Chinese OCR post-processing is carried out. Based on a classic model and a mature OCR (Tesseract, Baidu OCR), an OCR identification model is constructed. A large number of diversified Witnessses are obtained, and manual annotation is omitted. Based on a result of an OCR recognition system, the invention provides an unsupervisedmulti-input OCR correction method, an OCR correction model is constructed, and a large number of artificial marks can be avoided. Experimental results show that the accuracy of an unsupervised learning model proposed on a corpus is improved to a certain extent. It is further indicated that the recognition result of OCR can be well corrected by adopting the multi-input unsupervised learning method.

The invention discloses a Word2Vec-BiLSTM-CRF-based named entity recognition method in the legal field, and the method specifically comprises the following steps: obtaining original data in the legal field, and carrying out the preprocessing of the data, so as to obtain training corpus data; inputting the obtained training corpus data into a Word2Vec algorithm and combining the Word2Vec algorithm with a CBOW model so as to obtain word vectors aiming at the legal field; labeling the training corpus data obtained through preprocessing in combination with template matching and Chinese corpus pause and other modes, obtaining labeled corpus, serving Bi-LSTM as a coding layer of a model, and combining the obtained labeled corpus and obtained word vectors to serve as input and output of the coding layer to obtain text semantic information features; and taking the text semantic information features obtained by the Bi-LSTM layer as input of the CRF, and finally outputting an identification result of the named entity. The method has the advantages that rich entities in the legal document can be identified, fine-grained description of the entities in the legal field and data structuring in the legal field are realized, and the method is of great significance in further mining the relationship among different entities in the legal field.

The invention relates to a person, case and object association relation extraction method based on deep learning and pattern matching. The method comprises the steps: analyzing a legal instrument, and dividing the legal instrument into five parts: basic information of the instrument, command and control content, dialectical opinions, evidence correlation and judgment content; the method comprises the following steps: performing relation definition and entity definition, summarizing all relation sets and entity sets which need to be extracted, performing relation labeling and entity labeling on legal documents according to the defined relation sets and entity sets, performing entity recognition model training on a training set through a Bert-BiLSTM-CRF model, and performing entity recognition on the entity recognition model through a Bert-BiLSTM-CRF model. Calculating the accuracy rate and the recall rate of the test set through the trained model; performing a dependency syntax analysis task, and performing pattern matching definition on required entities, relationships and entity triple relationships after a relationship set and an entity set are defined; entity position information and sentences are input by taking a corresponding relation as a label and a sentence-based attention mechanism is combined with a CNN network, and a deep learning model is trained.

A similar case recommendation method based on network representation learning is mainly used for finding out ten legal documents most similar to legal documents in a huge legal document database underthe condition that the legal documents are given. According to the invention, the method includes taking legal instruments on the Internet as starting points, using a crawler technology for crawlinga large number of legal instruments, and using regular expressions for carrying out preliminary processing on the legal instruments; crawling a large number of legal entities by utilizing a crawler technology, calculating the similarity between the legal entities through bert, constructing a similar entity relationship between the legal entities reaching a threshold, constructing a legal knowledgenetwork by utilizing the legal entities and legal documents, inputting the constructed network into a network representation learning model to obtain vector values of the legal documents, and obtaining ten documents most similar to the legal documents through cosine similarity calculation.

The invention provides a legal instrument generation method and system based on a reading understanding and intention recognition model, and belongs to the technical field of legal consultation. Inputting the long text statement into a long text reading understanding model to obtain initial legal elements; comparing the first-round legal elements with the legal element library to obtain a legal scene that a consultant wants to consult; initiating multiple rounds of conversations with the consultant according to the missing elements to obtain a result of the multiple rounds of conversations; inputting results of the multiple rounds of conversations into an intention recognition model to obtain an intention recognition result; reasoning and decision making are carried out, and legal consultation suggestions or contract documents are automatically generated. The system comprises a data acquisition module, an element acquisition module, a scene acquisition module, a multi-round dialogue module, an intention recognition module and a document generation module, the model convergence time is shortened, the training cost is reduced, and parties can acquire accurate and professional legal consultation related answers.

The invention discloses a legal instrument semantic similarity evaluation method and system based on a deep learning model, and the method comprises the steps: 1, obtaining a target legal instrument and a candidate legal instrument, and obtaining the respective samples, sentence vectors and word vectors of the target legal instrument and the candidate legal instrument; 2, performing data augmentation on samples, sentence vectors and word vectors of the target legal instrument and the candidate legal instrument in a semantic similarity calculation model; 3, based on a semantic similarity calculation model and the augmented data, calculating the semantic similarity of the target legal instrument and the candidate legal instrument; and 4, obtaining a semantic similarity calculation result of the target legal instrument and the candidate legal instrument. When the method is applied, the semantic similarity of the target legal instrument and the candidate legal instrument is calculated conveniently, and the judgment accuracy can be improved.

The invention relates to a criminal legal document criminal name recommendation method and system. The method comprises the steps of obtaining a criminal legal document, and performing word segmentation processing on the criminal legal document to obtain a text set taking entries as units; obtaining a word2vec word vector model and a text separable convolutional neural network model; taking the word2vec word vector model as an embedding layer, and combining the text separable convolutional neural network model to construct a criminal name recommendation model; and obtaining a criminal name label matrix according to the text set by utilizing the criminal name recommendation model, and further determining a criminal name corresponding to the criminal legal document according to the criminalname label matrix. The criminal legal document criminal name recommendation method and system have the advantages of being low in labor cost, high in criminal name obtaining efficiency, high in criminal name recommendation accuracy and the like.

The embodiment of the invention discloses a document processing method and device, electronic equipment and a storage medium, and the method comprises the steps: responding to a legal document processing request of a user, and determining an entity label system of a target legal document; based on the entity label system of the target legal instrument, legal elements are extracted from the targetlegal instrument; and based on the entity label system of the target legal document, obtaining a structured document of the target legal document according to the extracted legal elements. Compared with the prior art, by automatically constructing the structured document of the legal document, the actual information extraction effect is obviously enhanced, the user experience is improved, and a new thought is provided for quickly reading important information from the legal document.

The invention relates to the technical field of document reading, in particular to a legal document reading model and a construction method, and the method comprises the following steps that 1, a Bertlayer codes input chapters and questions; 2, the feature fusion layer fuses part-of-speech tagging and named entity tagging vectors; 3, a modeling layer carries out fragment extraction prediction andnon-type judgment modeling; and 4, an output layer outputs fragment prediction and right and wrong prediction. According to the method, three types of problems of fragment extraction, non-judgment and rejection can be better handled.

The invention provides a long text retrieval model based on comparative learning. The long text retrieval model comprises a legal field pre-training module, a comparative learning module and a retrieval module, wherein the legal field pre-training module is used for constructing a basic long text encoder and performing field pre-training on the long text encoder by using a legal document corpus; the comparative learning module is used for constructing training data from the case annotation data set and performing text vector training on the long text encoder by utilizing the training data, and the training data comprises query statements and positive samples and negative samples of the query statements; and the retrieval module is used for detecting a case corresponding to the long text query statement by adopting the trained long text encoder. By means of the mode, the long text retrieval model can effectively solve the problem that a deep model processes a long text, the document encoder is adjusted by combining the characteristics of class case retrieval and using a domain pre-training and contrast learning method, and the retrieval accuracy and efficiency are improved.

The invention discloses a judicial fact finding generation method and device based on a deep neural network, and a medium. According to the method, features of court trial record text data in a dialogue form are extracted by utilizing a hierarchical sequence model, and vectorized representation of a word level, a sentence level and a dialogue level is obtained; then, the fact element classification task and the missing fact ascertaining task are taken as auxiliary tasks, and a fact ascertaining scene conforming to judicial program logic is constructed under a multi-task learning framework; finally, the feature extraction results are linked and combined, and a judicial fact finding result meeting court trial records is generated by means of an attention mechanism-based Seq2Seq model under amulti-task learning framework. According to the method, the deep sequence learning model is applied to judicial fact finding automatic generation, and compared with a common text generation algorithm, modeling and auxiliary text generation are carried out on factual key information in legal documents, factual elements in original texts are effectively reserved, and controllability and interpretability of the generation model are guaranteed.