202 results about "Label" patented technology

The invention relates to the technical field of cross-language information processing, and provides a cross-language text fusion intelligent alignment method and system.The cross-language text fusion intelligent alignment method comprises the steps that a hierarchical alignment model is constructed through preprocessing and label recognition of multi-coding-type texts, deep semantic feature extraction and labeling of a multi-language pre-training model and text semantic and format information analysis; transform is taken as a core, cross-language semantic association is enhanced through a multi-head attention mechanism, character-level and paragraph-level format collaboration is realized through label weight allocation and condition constraint, the format alignment accuracy is obviously improved in a multi-language mixed typesetting scene, and the multi-language mixed typesetting efficiency is improved. The analysis capability of the model on the structured information can be extended to layout relation processing of texts, images and tables, so that the comprehensive alignment efficiency in a multi-modal fusion scene is remarkably improved. The accuracy of text and label fusion is guaranteed, the problem of confusion of messy codes and labels is avoided, and high-precision cross-language text alignment from semantics to formats, from single mode to multiple modes and from semantics to formats is achieved.

The invention provides a large-model multi-label classification method, system and device based on ReAct and a vector library, and belongs to the technical field of artificial intelligence. The method comprises the following steps: constructing a multi-level label mapping tool according to each label data table corresponding to a preset multi-level label system; and constructing a label retrieval tool based on a pre-trained text similarity model and a preset label vector database. Constructing a ReAct tool chain based on a preset label dynamic adjustment tool, a multi-level label mapping tool and a label retrieval tool; the preset label dynamic adjustment tool is constructed on the basis of cue words and a large language model and is used for updating candidate labels according to the to-be-labeled text corpus, the candidate labels output by the label retrieval tool and label rules; and based on a preset ReAct cue word project, the large language model and the ReAct tool chain, constructing a large model multi-label classification module so as to input the to-be-labeled text corpus from the user terminal into the large model multi-label classification module for multi-label classification.

The invention discloses a commercial customer service system based on large model emotion recognition labeling and correction, and belongs to the technical field of natural language processing and deep learning. In order to solve the problems that an existing emotion recognition system depends on large-scale manual labeling, label quality is unstable and small sample performance is poor, a multi-model collaborative labeling and iterative optimization mechanism is adopted, and fusion labels are generated through automatic basic model selection, small sample LoRA fine adjustment, double-model divergence detection and large model arbitration. And a refining training set is constructed for iterative fine tuning to form a closed-loop optimization system. The method can effectively reduce the labeling cost, improves the label consistency and the emotion recognition precision under complex semantics, and is suitable for business information, financial public opinions and intelligent customer service scenes.

The invention discloses a multi-round cross validation and weak supervision noise cleaning method based on a large language model. According to the method, weak supervised learning and large language model reasoning capability are fused, and multi-round cleaning and label optimization are performed on a low-confidence sample by introducing a small amount of high-quality label seeds and combining automatic rule construction and cross validation processes. According to the method, a verification and feedback system with a large language model as a core is constructed, efficient purification and enhancement of weak tags in large-scale text data are achieved, and high-quality training data and intelligent tag optimization support are provided for natural language processing tasks such as relation extraction, text classification and entity recognition.

The invention relates to a cross-language entity alignment method based on large language model knowledge enhancement, and belongs to the field of knowledge graph alignment. According to the method, firstly, entity and relation embedding is enhanced by applying a large language model, and the distinction degree of positive and negative samples in the training process is improved; then, adaptive fusion weighting is introduced to weaken various embedded noises, and bidirectional flexible voting is introduced to generate more reliable pseudo labels on label-free data; the CLEA-LLM uses a teacher-student structure, a teacher encoder and a student encoder generate three types of embedding according to two knowledge maps and knowledge features, features obtained by the teacher encoder are subjected to adaptive fusion weighting to generate joint features, and a pseudo-mapping probability matrix is obtained through bidirectional flexible voting and diversity correction to supervise learning of a learning encoder. Therefore, the entity alignment precision is improved.

The invention provides an open domain label system construction method and device based on a large language model, and relates to the technical field of data processing. The method comprises the following steps: constructing a label association pool according to different to-be-labeled contents in a first time period and initial labels output by labeling the to-be-labeled contents by using a first large language model; determining an initial tag system and a tag cluster of any initial tag based on the tag occurrence frequency of each initial tag; based on a preset semantic understanding task description, determining a normalized tag of each initial tag by using the zero sample capability of the second large language model; constructing a label directed graph set; and if it is detected that the first tag in the first tag directed graph is a normalized tag of the second tag in the directed graph and the normalized tag of the first tag in the second tag directed graph is a third tag, constructing a target tag system. According to the method, the time from emerging topic appearance to label system response is shortened, and the semantic accuracy of the label system is improved.

Disclosure are a backdoor attack method and system for a classification task in a code model, the method includes: S1. collecting and preprocessing clean samples to obtain importance variable names; S2. classifying the variable names of the clean samples according to label categories to obtain a plurality of trigger sets; and selecting target labels from the clean samples; S3. performing score calculation on the variable names in the trigger sets corresponding to the target labels; replacing one importance variable name with the variable name having a maximum C score in the clean samples to obtain poisoned samples, and repeating the above process until the labels are changed into the target labels; and S4. randomly inserting the triggers in the poisoned samples into the clean samples to form negative samples; and performing an attack by using an attack model obtained based on the negative, poisoned and clean samples.

Disclosed in the present application are a model training method, apparatus and device, and a storage medium and a computer program product. The model training method comprises: acquiring target training data, and on the basis of the target training data, training a classification model, so as to obtain simulated data related to the classification model; acquiring a label type sequence; on the basis of the label type sequence and the simulated data related to the classification model, determining a first loss function; and using the first loss function to update parameters of a first encoder network and parameters of a second encoder network until a loss value of the first loss function converges, and generating a noisy label detection model.

The invention relates to a multi-label text classification method based on positive and negative label learning and label correlation, and belongs to the field of multi-label classification. Comprising the following steps: constructing a feedforward neural network model with double hidden layers; initializing a model component; reading features and label information of samples in the training set, and generating a feature matrix and a label matrix; randomly initializing a weight matrix and an offset parameter of the model; inputting the feature matrix into an input layer of the model, and calculating neuron output layer by layer; calculating gradients of weight matrixes and bias parameters among layers in the model by adopting a composite error function, dynamically adjusting the gradients by utilizing an Adam optimization algorithm, and updating the weight matrixes and the bias parameters; when the error change amplitude is lower than a threshold value or reaches a preset number of iterations, stopping training; and after model convergence, predicting the test set to form a final multi-label classification result. According to the method, the limitation of traditional text classification is broken through through a deep learning technology, and high-precision and high-efficiency classification of complex text data is realized.

A method and system of UI automation includes receiving a demonstration of an automation to be performed on an application. One or more objects and one or more corresponding labels associated with the demonstration are detected on the application. The demonstration is transformed into one or more natural language instructions. An object is semantically selected during a runtime action based on a large language model (LLM). The semantic selection is reflective of an intention captured in the natural language instruction despite any change in a corresponding label of the object.

The invention relates to the technical field of intelligent label recommendation, and discloses a standardized label framework collaborative recommendation system based on artificial intelligence. The system comprises a multi-level professional label knowledge base, a large language model reasoning unit, a semantic consistency verification module, a path verification module, a collaborative decision module and a label recommendation engine. The multi-level professional label knowledge base stores a standardized label system containing a multi-level structure; after receiving the input content, the large language model reasoning unit generates an initial label set according to label system level constraint reasoning; a semantic consistency verification module compares the label and content semantic matching degree and outputs a result, and a path verification module verifies whether a label generation path accords with a hierarchical structure and gives a compliance indication; the collaborative decision-making module corrects the initial label set according to a preset rule when the semantic verification result conflicts or the confidence coefficient is insufficient; and integrating the corrected label set by the label recommendation engine, and outputting a final standardized label recommendation result to adapt to multi-field labeling requirements.

Aspects of the present disclosure provide techniques for training an item extraction machine learning model. Embodiments include extracting text and bounding box coordinates from a structured document and creating structured text by adjusting formatting of the extracted text based on the extracted bounding box coordinates and adding table delimiter tags to the extracted text based on detecting one or more tables in the structured document. Embodiments include providing the structured text to a language processing machine learning model along with a prompt instructing the language processing machine learning model to generate a label indicating variables present in the structured text and values for the variables. Embodiments include receiving the label from the language processing machine learning model in response to the structured text and the prompt and training the item extraction machine learning model through a supervised learning process based on training data comprising the structured text and the label.

The invention provides a marking task-oriented training course generation method and device, a medium and a product. The method comprises the steps of obtaining label sets corresponding to a plurality of service platforms respectively, wherein each label set comprises a label field and a corresponding label rule; based on entities pointed by the semantic information of the annotation fields in each annotation set, constructing an annotation knowledge graph; the annotation knowledge graph comprises a first node for representing an entity and a second node for representing an annotation field, a connection relationship for indicating semantic association is established between the first node and the second node, and each second node is associated with a corresponding annotation rule; if the annotation rules corresponding to different second nodes connected with the same first node in the annotation knowledge graph have differences, generating a cross-platform contrast training course based on the differences; and if the labeling rules corresponding to different second nodes connected with the same first node in the labeling knowledge graph are the same, generating a cross-platform unified training course.

An example apparatus disclosed includes interface circuitry, machine readable instructions, and programmable circuitry to at least one of execute or instantiate the machine readable instructions to identify a first source of object label representation and a second source of object label representation, the first source or the second source including an estimated label prototype vector associated with an input text-based object query, determine a first contextualized embedding for the first source and a second contextualized embedding for the second source, and combine the first contextualized embedding and the second contextualized embedding to generate a candidate object representation associated with the input text-based object query.

Proposed are a data classification method and apparatus. The data classification method that is performed by the data classification apparatus includes extracting features from input data through a learning network model and outputting prediction results based on the features, and the learning network model compares local features derived through an individual layer other than the final layer of the learning network model with label embedding vectors corresponding to a classification label.

The embodiment of the invention provides a data preprocessing and optimizing method and device for a generative model in a question and answer scene. In an intelligent question and answer service scene, on the basis of answering the user question, another tag question can be provided for the user to select. Tag questions may be determined based on questions generated by the generative model. The generative model herein may be a natural language processing model such as a large language model. In order to optimize the generative model, sample data for training can be determined according to the operation performed by the user on the pushed label problem in the online use process. A plurality of preference data pairs are automatically extracted from a single operation of a user through data screening and processing to serve as training samples. Therefore, the generated model is optimized by using the extracted training sample, and the generation result of the generated model can be guided, so that the generated model generates a tag problem which better conforms to human preference.

Techniques for evaluating the efficacy of large language models on classification tasks are disclosed. A prompt that includes an instruction and a content item to be classified is submitted multiple times to a large language model. For each submission of the prompt, a corresponding classification label from a set of two or more classification labels is returned. Each classification label is compared to the expected classification label for the content item using a label distance value metric. Using the label distance value metric, a confidence score is generated.

The small sample intention recognition method based on label semantics and gating attention is high in recognition precision, the generalization performance is remarkably enhanced, and the data utilization efficiency is extremely high. The invention discloses a small sample intention recognition method based on tag semantics and gating attention. The method comprises the following steps: (1) establishing a triple contrast learning task; (2) establishing a mask language modeling task; (3) establishing a self-attention double-layer gating prototype network; and (4) establishing an intention detection metric learning and loss function.

The embodiment of the invention provides a model training method and device, equipment and a computer storage medium, and relates to the technical field of artificial intelligence. The method comprises the steps that when a multi-label text classification model is trained, the relevance between category labels marked by training data is analyzed, exclusiveness and relevance between category pairs are determined, the multi-label text classification model learns the relevance relation between the exclusiveness and the relevance between the category labels, and the classification efficiency of the multi-label text classification model is improved. And the text feature vector is fused with the text feature vector of the training data, so that the predicted multi-class result output by the multi-label text classification model is improved. Besides, since the prediction of the multi-class result learns exclusiveness between class labels, the multi-label text classification model can be prevented from outputting class pairs with exclusiveness at the same time, so that the accuracy of the model output result is further improved.

Techniques for compiling switch blocks are disclosed. One or more embodiments analyze and rewrite a set of pattern labels in a switch block in a compile-time process for determining whether the switch block is exhaustive. At compile-time, a system populates a set with case labels from a switch block. The system applies a set of rules to iteratively re-write pattern labels in the set and checks whether the re-written set, and hence the original switch block, is exhaustive. If the compiler determines that (a) the set does not appear to be exhaustive, and (b) the set includes patterns labels, then the compiler determines whether the set may be rewritten before further analysis. The compiler iteratively re-writes and re-analyzes the case label set for exhaustivity until the case label set is determined to be exhaustive or cannot be rewritten further.

The invention relates to the technical field of natural language processing, and provides a large language model event extraction method fusing label reconstruction and a multi-dimensional instruction set, which comprises the following steps of: 1, reconstructing a label; 2, constructing a multi-dimensional instruction set; constructing two different multi-dimensional instruction sets for the multi-dimensional instruction library by adopting a multi-dimensional layered alternate combination strategy through a second-level instruction architecture and a third-level instruction architecture; 3, fine adjustment of the model; taking the reconstructed document-level event text and the event record sequences in two different formats as input and output of fine tuning respectively, and performing fine tuning on an LLaMA-3. 2-1B model by using a multi-dimensional instruction set and adopting a LoRA technology; 4, event extraction; and extracting events by using a large model for fusing label reconstruction and multi-dimensional instruction set fine tuning. According to the method, the data labeling cost is reduced through label reconstruction, the semantic analysis capability of a large language model on financial field events is enhanced by utilizing a multi-dimensional hierarchical instruction set, and the financial field event extraction performance is improved.

Systems, apparatus, articles of manufacture, and methods are disclosed to train models with a dynamic margin. An example apparatus includes interface circuitry, machine-readable instructions, and at least one processor circuit to be programmed by the machine-readable instructions to determine a positive label similarity value and a negative label similarity value, the positive and negative similarity values based on a query embedding, determine a negative-to-positive difference value and a positive-to-negative difference value associated with the positive label similarity value and the negative label similarity value, determine a positive-to-negative difference value associated with the positive label similarity value and the negative label similarity value, and cause training of a label classifier with a loss function having a dynamic margin when the positive-to-negative difference value satisfies a threshold.

An example may train a cross encoder embedding model using a ranking instruction, a combined input, a pseudo label, and a combined loss. The combined loss includes a ranking loss and a first retrieval loss. A first entity embedding of an entity and a first item embedding of an item may be obtained from the trained cross encoder embedding model. A first input including the first entity embedding obtained from the trained cross encoder embedding model, a second input including the first item embedding obtained from the trained cross encoder embedding model, and a second retrieval loss, may be used to train a dual encoder retrieval model to produce a trained dual encoder retrieval model. A system may use output of the trained dual encoder retrieval model to include or exclude items from a presentation of digital content items to the entity via a device.

The present application relates to a kind of based on the method of abandoning anti-label noise dialogue state tracking, belong to natural language processing technical field.This method is encoded to each data in data set, judges the prediction mode of slot by the slot classifier based on abandonment, for the slot value generator decoding using for the slot needing to generate, slot value replicator decoding is used to the slot needing to reason.A special category "abandonment" is added in slot classifier, it indicates that the model considers this prediction too complex or there is label noise, and gives up the prediction and learning of this sample, and trains the dialogue state tracking model of anti-label noise through the loss function after revision.This method improves the coding efficiency and decoding speed, effectively reduces the influence of labeling error, so that the model can also maintain good performance in the data set with larger label noise.

The invention provides a data processing method and device and a storage medium. The method comprises the steps that keywords corresponding to intention tags of a preset category can be obtained; the keywords corresponding to the intention labels of the preset category are preprocessed, preprocessed keywords are obtained, and preprocessing comprises at least one of normalization processing and data augmentation processing; according to categories of intention tags corresponding to the preprocessed keywords, the preprocessed keywords are combined, at least one intention tag sequence is generated, and the intention tag sequence comprises at least one category of intention tags and keywords corresponding to the intention tags; and utilizing the intention prediction template corresponding to the at least one intention category and the at least one group of intention label sequences to generate an intention prediction text corresponding to the at least one intention category, the intention prediction text being used for training a large language model, and the large language model being used for intention prediction. By means of the intention prediction text obtained through the method, the training effect on the large language model can be improved.

The application provides an intent recognition model training method and device based on pseudo labels, and an intent recognition method and device, which comprises the following steps: inputting sample text into a language model to extract a feature vector; clustering the sample text based on the feature vector, taking the clustering result as a pseudo label, and calculating the normalized mutual information between the real label of the labeled sample text and the pseudo label; determining the confidence score corresponding to each sample; the confidence score is used to quantify the noise in the pseudo label, filter high-confidence samples, and take the corresponding pseudo label as a self-supervised signal to iteratively optimize the language model until convergence; after iteration, reinitialize the clustering, update the clustering result, the normalized mutual information, and the iteration number; when the iteration number reaches an upper limit or the normalized mutual information increment is less than a threshold, terminate the training and determine the language model as an intent recognition model; the problem that noise pseudo labels continuously spread and accumulate, leading to a decline in the ability of the model to recognize new intents, can be solved; and the ability of the model to recognize new intents is improved.

The application discloses an index construction method, a data query method and related equipment, and relates to the technical field of computers. The method comprises the following steps: acquiring a timeline composed of labels of time-series data, and assigning a timeline identifier to the timeline, wherein the labels comprise label keys and label values, and the timeline is composed of label values of different label keys; determining a set of timeline identifiers corresponding to the label values, so as to obtain an identifier set of the label values; and constructing an index structure for retrieving the timeline based on the label values, the timeline identifiers and the identifier set, wherein the index structure comprises a first index layer and a second index layer, so as to create a first mapping relationship between the label values and the identifier set through the first index layer, and create a second mapping relationship between the timeline identifiers and the timelines through the second index layer. The application solves the problem in the prior art that an index cannot efficiently support multi-dimensional retrieval of timelines.

The invention discloses a software demand quality evaluation data processing method and device, equipment and a medium, and the method comprises the steps: obtaining an initial software demand text, and recognizing a candidate demand text containing a predefined defect indication word through semantic analysis; for each candidate text, matching or generating a high-quality demand text corresponding to semantics from a historical demand library or a domain knowledge library, and constructing a difference sample pair; and labeling the sample pair input into the arbitration system based on the large language model, and automatically generating a quality label corresponding to the demand text. And generating a high-quality evaluation data set based on all the labeled difference sample pairs. According to the technical scheme, automation of data set construction can be achieved, the structural design can effectively reveal context ambiguity of the demand text, and a reliable reference is provided for training and evaluating a high-precision demand quality detection model. The method can be widely applied to the technical field of software.

The invention discloses a label determination method and device, equipment and a storage medium, and relates to the technical field of computers.The method comprises the steps that an identification text corresponding to comment information is generated according to the comment information of a current object and an identification instruction, and the identification text is input into a pre-trained large language model, the big language model outputs identification features of the comment information; inputting the comment information and the recognition features of the comment information into a pre-trained discrimination model, and determining a discrimination result of the recognition features of the comment information according to the output of the discrimination model; and determining the label of the current object for the identification feature of the correctly identified comment information according to the judgment result. According to the technical scheme, the large language model determines the more accurate recognition features of the comment information according to the recognition text containing the comment information and the recognition instruction for the comment information, and determines the tag which is more accurate and better matched with the current object according to the recognition features of the comment information with the higher matching degree with the recognition features of the comment information.

The invention provides a reference processing method and system, and the method comprises the steps: responding to a label selection operation triggered by a cursor in a writing region in a target reference process, and displaying a reference region; acquiring the current position of the cursor in the writing area; in response to a label adding operation for the reference area, content corresponding to the selected label is added to the current position of the cursor, and the content can be a graph, a table, a block formula, a chapter or a reference in an article. According to the method, manual annotation is not needed, and different contents can be quickly and accurately quoted in the form of the graphical interface according to the reference area, so that the writing efficiency and the use experience of a user are improved.