38 results about "Language technology" patented technology

The invention provides an audit report automatic generation method based on natural language processing, and relates to the technical field of natural languages, and the method comprises the steps: obtaining audit document data passing the examination and approval, the audit document data comprising an audit manuscript, an audit problem, an audit evidence obtaining sheet and an audit verification sheet; performing semantic analysis on the audit document data through a natural language processing model, and extracting key information of an audit subject, a risk point and a compliance conclusion; inputting the key information into a structured processing algorithm to generate structured data matched with a preset auditing template; and inputting the structured data into a pre-training text generation model for language reconstruction, generating an audit report text, automatically filling the audit report text into a corresponding field of an audit template, and outputting a complete audit report.

For semantic data leak prevention, one or more semantic interpretations of an input text are obtained with a generative pre-trained transformer, GPT, configured for interpreting the input text with reference to a target context, and it is determined whether the input text comprises a hidden data leakage by checking the semantic interpretations of the input text with a data leak prevention, DLP, system configured for detecting a data leakage based on a collection of words and phrases of interest. As a result, enabled is a prevention of hidden data leaks when a sensitive information is hidden within a text, for example, by using advanced language techniques.

The invention relates to the technical field of computer programming languages, and particularly discloses a computer language programming method based on business logic. Aiming at the defects of deep coupling of business logic and technology implementation, high cross-domain adaptation cost, hardware operation code redundancy, high asynchronous programming complexity and the like in the prior art, the invention provides the following core schemes: a modularized development framework, a cross-platform adaptive mechanism, an event hierarchical model, a hardware intention analysis engine and a code, namely a document system. According to the method, the development complexity of a multi-field system is remarkably reduced, the business logic expression efficiency and maintainability are improved, the method is suitable for cloud micro-services, embedded equipment and mixed language scenes, and a technical basis is provided for intelligent programming.

This disclosure relates to method and system for mixed language text understanding for Generative Artificial Intelligence (GenAI) models. The method may include receiving a raw parallel corpus of two languages. The method may further include generating a cross-domain codemix parallel corpus and a first set of linguistic features from the raw parallel corpus using statistical and linguistic techniques. The method may further include determining a complexity of each of the plurality of samples of the cross-domain codemix parallel corpus based on a set of complexity parameters. The method may further include sequentially fine-tuning a pre-trained multilingual translation model using each of the plurality of samples in the curriculum learning dataset to obtain a generic pre-trained codemix understanding model.

The invention discloses an intelligent office assistant system based on natural language processing, and relates to the technical field of natural language processing. When the system runs, a user intention vector I is generated by acquiring query behavior data of a user, generating a user behavior feature vector B and analyzing the real demand of the user by using a natural language technology; the method comprises the following steps of: performing similarity calculation with an entry vector R of a rule and regulation inventory, generating a matching result sim (I, R), generating a personalized recommendation list TC, comparing the personalized recommendation list TC with a pushing execution threshold TZ, accurately controlling the pushing content, adaptively adjusting the pushing execution threshold TZ by using an operation record after a user receives the pushing content, and realizing dynamic response to user demand change. According to the technical scheme, a comprehensive and efficient solution is provided, the problem of inaccurate matching caused by dependence on keyword search is solved, the phenomenon that the pushed content is redundant or does not accord with requirements is effectively avoided, and the information matching efficiency and the user satisfaction degree are greatly improved.

The invention relates to a speech enhancement method based on bispectrum nonlinear feature coupling, and belongs to the technical field of natural languages. The method comprises the following steps: performing feature extraction on time-frequency domain representation of input noisy voice through an encoder to generate encoding features; inputting the coding features into a bispectrum feature extractor, extracting high-order nonlinear coupling features through complex frequency spectrum conversion and bispectrum analysis, and fusing the coding features and the bispectrum features through jump connection to form complementary feature expression; an amplitude mask decoder and a phase decoder are adopted to decode the fusion features, and an enhanced amplitude spectrum and an enhanced phase spectrum are predicted respectively; and combining a multi-task loss function with a generative adversarial training mechanism to jointly optimize the performance of the speech enhancement model. According to the method, the speech enhancement quality in a noise environment is remarkably improved, and particularly, higher robustness is shown in a non-stationary noise scene.

The present invention provides a quantization method, device, electronic device, storage medium and computer program product, which relate to the field of natural language technology. The quantization method comprises: for each weight to be quantized in a dialogue model, Gaussian fitting is performed on the weight to be quantized to obtain a Gaussian distribution function; based on the Gaussian distribution function, a quantization parameter value range of linear quantization is determined; maximum likelihood estimation is performed on the estimated parameter variables within the quantization parameter value range of linear quantization to obtain a target threshold; and the target threshold is used to perform a quantization operation on the weight to be quantized. The present invention can significantly reduce the precision loss in the quantization process without using dialogue data set reasoning, thereby improving the accuracy of the dialogue.

The present invention provides a character recognition method and system based on language technology, which relates to the field of financial bill recognition. The method includes: obtaining a bank bill image, dividing the bank bill image into grids, and screening according to the fractal dimension of each grid area to obtain a first character candidate area; taking the pixels of the first character candidate area as nodes and calculating topological invariants to eliminate abnormal areas according to the topological invariants to obtain a second character candidate area; generating a topographic map by brightness mapping according to the fractal dimension of the second character candidate area and performing dynamic segmentation path planning on the topographic map to obtain a third character candidate area; performing adhesion detection on the third character candidate area and elastically separating the adhesion points to obtain a target character area. The present invention can accurately distinguish adhesive characters, solve the problem of adhesion of microtext; and can retain the stroke continuity while separating the adhesive area.

The invention provides a Chinese word segmentation method based on deep learning, and relates to the technical field of natural languages, and the method comprises the steps: dividing a Chinese word segmentation public data set according to a proportion, and carrying out the pre-training, and obtaining the multi-dimensional features of the word granularity; and constructing a Chinese word segmentation model TC-CRF based on deep learning, and performing Chinese word segmentation processing on the multi-dimensional features by using the Chinese word segmentation model TC-CRF. According to the method, the problem that long dependence cannot be obtained by an existing convolution feature extraction method is solved, the problem that the global transformer obtains too large noise is solved, and Chinese word segmentation is effectively realized by using deep learning.

The invention discloses a target detection method, electronic equipment, a storage medium and a program product, belongs to the technical field of computer languages, and is used for reducing the time delay of target detection of a large model. The method comprises the following steps: preprocessing a to-be-detected image to obtain a mapping feature sequence of the to-be-detected image in a text embedding space; preprocessing category text information of a target object in the to-be-detected image to obtain a text embedding sequence of the category text information in the text embedding space; and determining detection features of the target object according to a preset query vector, the mapping feature sequence and the text embedding sequence.

In order to prevent semantic data leakage, one or more semantic interpretations of an input text are obtained using a generative pre-trained translator (GPT), where the GPT is to interpret the input text with reference to a target context; and determining whether the input text includes hidden data leaks by checking the semantic interpretation of the input text using a data leak prevention (DLP) system, where the DLP system is used to detect data leaks based on a set of words and phrases of interest. Accordingly, it is possible to prevent hidden data leakage when sensitive information is hidden in text (e.g., by a high-level language technology).

The invention discloses an information generation method and device, electronic equipment and a readable storage medium, and belongs to the technical field of natural languages. The method comprises the steps that first question information is input into a first translation layer of a generative question and answer model, the first translation layer comprises at least two first memory parameters, and the at least two first memory parameters are obtained by conducting feature extraction on first user historical behavior data; based on the at least two first memory parameters, processing the first question information through a first translation layer to obtain first response information output by the generative question and answer model, the first response information being response information corresponding to the first question information; wherein the first user historical behavior data comprises at least one of the following items: historical browsing records of the user, historical preference data of the user and historical problem information of the user.

The invention discloses an emotion perception feedback action type non-linguistic interaction robot, which comprises an emotion perception module, a feedback action module and a non-linguistic interaction module, and is characterized in that an emotion perception result is firstly used for judging a target needing to take action; whether the target negative emotion is relieved before and after the action and the relieving degree are evaluated; the feedback action is to increase or decrease the weight of the action in the action library and increase or decrease the occurrence probability of the action based on the evaluation result of the emotion perception on whether the negative emotion is relieved and the relieving degree after the action is adopted; according to the non-language interaction, the robot does not adopt the voice recognition technology and the artificial intelligence language technology, and the robot interacts with the target only through non-language actions. The method has the prominent advantages of prominent effect, excellent environmental adaptability and easiness in popularization.

The invention provides an industrial chain intelligent description method and system based on a thinking map. The method comprises the following steps: determining entity attributes corresponding to different industrial entities in an industrial chain according to multi-source industrial data and an entity relationship between the different industrial entities to construct an industrial chain ontology model, in the process of operating the industrial chain ontology model, using a natural language technology and a graph algorithm extraction technology to obtain operation knowledge corresponding to each industrial entity, generating a knowledge database, using the knowledge database to derive an industrial chain structure and an industrial chain risk of an industrial chain, using the industrial chain ontology model to simulate the industrial chain risk, and using the industrial chain ontology model to simulate the industrial chain risk. According to the method, a plurality of risk features included in an industrial chain structure are determined, a visualization technology is utilized to simulate the industrial chain structure and a scene corresponding to the risk features, a real-time knowledge system of an industrial chain is obtained and displayed, and work such as supply chain optimization and industrial planning can be better served by constructing the dynamically updated industrial chain knowledge system.

The invention discloses an RAG enhanced small-scale language model question and answer accuracy improving method and system. The method comprises the following steps: adjusting a text form of an answer basis in a knowledge base according to a text fragmentation processing mode; obtaining and utilizing a first verification set to perform retrieval optimization on the RAG model; taking a question expanded by using a natural language technology as input, and performing retrieval in a knowledge base associated with a bid invitation scene by using the RAG model after retrieval optimization to obtain a plurality of initial answer bases; respectively outputting the plurality of initial answer bases to an initial answer base scoring model, screening and sorting the plurality of initial answer bases according to a scoring result, and completing de-duplication filtering processing; performing content simplification summarization on each processed initial answer basis by using a natural language technology to obtain a final answer basis; and according to the question and a final answer basis, generating an answer by using an RAG model. According to the method, the question and answer accuracy of the RAG enhanced small-scale language model can be improved.

The invention relates to an agent workflow-based big language technology practice learning platform and a construction method, and the platform comprises an account management module which is used for creating at least one learning account, and / or modifying the account information of a to-be-corrected learning account, and / or deleting a to-be-deleted learning account; the construction module is used for constructing a knowledge database of at least one knowledge field based on an enhanced retrieval generation technology; the interaction module is used for performing learning interaction with the learner; the application development module is used for carrying out application development based on the development requirements; and the control module is used for determining a to-be-interacted module from the construction module, the interaction module and the application development module when a learning request is received, so that the to-be-interacted module executes a dialogue action and a data processing action based on the learning request. Therefore, the problem of relatively high use cost caused by large construction resource demand and relatively weak expandability of a learning platform in related technologies is solved, the applicability of a practical learning platform is improved, and the use cost is reduced.

The invention discloses a harmful text content detection method and device, electronic equipment and a storage medium, relates to the technical field of natural languages, and mainly aims to solve the problem of poor text detection accuracy caused by missing detection and false detection of harmful text content. The method comprises the steps of performing sensitive word matching on a to-be-detected text statement based on a first data set, wherein the first data set comprises a plurality of sensitive words; performing first detection on the text statement matched with the sensitive word based on a semantic detection model to obtain a first detection result, the semantic detection model being obtained by performing model training on a lightweight natural language processing model based on a second data set, statement samples in the second data set are statements including sensitive words and having harmless meanings; and performing second detection on the harmful statements in the first detection result based on a large language model to obtain a second detection result, and generating a harmful text content detection result based on verification information of the second detection result.

The invention discloses a news text generation method based on a text style, and relates to the technical field of false news text research. Comprising the following steps: S1, replacing sentences highlighted by real news with credible but false information through a language loading technology by using a seq2seq model; and S2, performing feature analysis and detection on the generated text by using a pre-training language model. According to the method, a new training corpus is generated through a seq2seq technology based on obviously different text styles and potential intentions, sentences of highlighted text styles and potential intentions of real news are replaced with credible but false information by the training corpus, and the false news detection method based on the text styles is provided accordingly. According to the method, the style characteristics of the false news are obtained by utilizing the pre-training language model, so that the training efficiency and the generalization ability of the model are improved, and the reliability of a false news detection system is guaranteed on the basis.

The invention relates to the technical field of natural languages, and provides a code version verification method, device and equipment based on a code hook and a computer readable storage medium. Before a submitted code is submitted to a code library through the code hook, the submitted target code is intercepted; according to the method, the target code is intercepted, version or branch inspection is performed on the intercepted target code, and the characteristics of the code library and the collaboration platform are combined, so that the code library automatically intercepts and submits the code for inspection when the code library is in butt joint with various different collaboration platforms, and the risk that developers submit the code to other versions or other branches by mistake is reduced. Under the background of multi-team and multi-version parallel development in the current software development industry, the management cost is reduced while the cooperation efficiency and the development efficiency are improved, so that the development efficiency is improved while the error submission is effectively reduced.

The application discloses a method and system for mining PQL based on a large language model analysis process, and relates to the technical field of process query language, and the method comprises the following steps: S1, establishing a large language model; S2, knowledge base and explanation preparation; S3, PQL analysis prompt; S4, PQL statement explanation; S5, local highlight display; and S6, PQL statement fine tuning. The method and system can parse the PQL statement input by the user by using the large language model, and convert the PQL statement into natural language text that is easy to understand, so as to help the user quickly understand the specific meaning of the PQL, thereby reducing the understanding requirement of the user for the PQL statement while facilitating the reading and understanding of the user by feeding back the natural language analysis result with high readability without manual intervention.

The invention belongs to the technical field of languages, and discloses a cross-language platform communication system which is composed of a Flutter environment module, an OC message center module, an RN environment module and an OC environment module. And the OC message center module sends subscription messages to the Flutter environment module, the RN environment module and the OC environment module. Other cross-platform language processing schemes are introduced, so that the development process is remarkably simplified, the framework design is simple, the implementation difficulty is low, maintenance is easy, efficient and stable operation is ensured, more importantly, pages on different language platforms can be seamlessly connected, free circulation of information is achieved, language and technical barriers are broken, and the development efficiency is improved. The working efficiency of business developers is greatly improved, the communication cost, the labor cost and the time cost among teams are greatly reduced, and through the innovative scheme, enterprises are assisted to quickly respond to market requirements and accelerate product iteration, so that a precedent is occupied in fierce market competition.

The invention provides an intelligent question and answer processing method, system and device and a medium, and the method comprises the steps: obtaining a question text of a user; the question text is input into an intention recognition model for intention recognition, intention label information output by the intention recognition model is obtained, and the intention label information is used for indicating a plurality of different target execution strategies corresponding to the question text, the target execution strategy is a rule matching strategy, a graph query strategy, a deep reasoning strategy or a legal document task strategy; according to all the target execution strategies, performing strategy processing on the question text to obtain an intermediate answer corresponding to each target execution strategy; and performing content optimization on all the intermediate answers to obtain a target answer of the question text. The method can effectively improve the practicability of knowledge questions and answers, and is beneficial to improving the efficiency of legal document processing. The invention relates to the technical field of natural languages.

This invention provides a dialogue processing method, apparatus, and computer-readable storage medium, belonging to the field of natural language processing technology. The dialogue processing method includes acquiring multiple sets of dialogue streams; determining at least one topic boundary based on the semantics of the multiple sets of dialogue streams; segmenting all dialogue streams into at least two topic segments based on each topic boundary; assigning dynamic weights to memory items in each topic segment, wherein each memory item stores at least one piece of memory information; the dynamic weights are determined at least based on the timeliness of the memory item, its relevance to the current query, and its coherence with the current topic; and, in response to the current dialogue, acquiring multiple candidate memory items from all topic segments, and, based on the dynamic weight of each candidate memory item, filtering and outputting a set of memory items that meets a target length from the multiple candidate memory items.

The invention provides a territorial space domain data analysis method and system based on a DSL (Digital Subscriber Language) technology, and the method comprises the steps: defining a data namespace of a territorial space domain by adopting a domain-specific language, the types of the data namespace at least comprising a source data configuration class, a basic entity configuration class, a professional entity configuration class and a comprehensive entity configuration class; collecting a source pasting data set; constructing a basic entity data set; calculating the basic entity to obtain a professional entity; performing statistics and clustering analysis on all professional entities to complete data analysis; according to the method, the domain-specific language technology is applied to the territorial space field for the first time, natural resource business knowledge and related GIS spatial data can be subjected to standard extraction, the processes of data analysis and the like are standardized, a rapid and automatic means is provided for natural resource data collection, processing, business analysis and the like, and the method is suitable for popularization and application. The digital transformation of the territorial space field is facilitated, and the scientificity and accuracy of comprehensive decision and implementation management of natural resources are improved.

The application provides a kind of electric power equipment asset management data belonging to class and group recommendation method, comprising: using natural language technology to process the text data of electric power asset management system, obtain effective text data;Effective text data is handled to entity, and knowledge graph is constructed based on entity result;According to the topological structure information of knowledge graph, the influence of knowledge graph node is calculated;Extract the standard entity of user input asset information, and extract the subgraph from knowledge graph that matches the standard entity;According to the theme probability distribution and node influence of matched subgraph, the recommended asset class and asset group are obtained.The application solves the problem that the existing recommendation method for electric power asset data management platform has low recommendation accuracy.

The invention discloses a special medical consultation reply generation method and device, terminal equipment and a storage medium, and belongs to the technical field of natural languages, and the method comprises the steps: obtaining an image text splicing feature according to image data and first text data inputted by a user; a plurality of expert feed-forward networks are selected through a routing network, image text splicing features are processed in parallel through all the expert feed-forward networks, feature processing results of all the expert feed-forward networks are fused in a gating weighting mode, a reasoning result is obtained, evidence retrieval is conducted on the reasoning result through a retrieval enhancement technology, and consistency checking is conducted. When the consistency check is passed, generating a text reply according to a reasoning result; and extracting a text condition constraint from second text data input by the user, and injecting the text condition constraint into the improved condition diffusion model through an attention layer to obtain an image reply. The problem that large model generation reply accuracy is low under the specialized background can be solved.

The present invention discloses an intelligent office assistant system based on natural language processing, which relates to the field of natural language processing technology. When the system is in operation, it obtains user query behavior data to generate a user behavior feature vector B, and uses natural language technology to analyze the user's actual needs to generate a user intention vector I. Subsequently, a similarity calculation is performed with the entry vector R of the rules and regulations inventory to generate a matching result sim(I, R). A personalized recommendation list TC is then generated and compared with the push execution threshold TZ to accurately control the pushed content. The push execution threshold TZ is adaptively adjusted based on the user's operation record after receiving the pushed content, achieving dynamic response to changes in user needs. This provides a comprehensive and efficient solution, overcomes the problem of inaccurate matching caused by relying on keyword searches, effectively avoids the phenomenon of redundant or inconsistent pushed content, and significantly improves information matching efficiency and user satisfaction.

The invention discloses an electric power field hidden danger analysis method, system and device based on a large model and a medium, and relates to the technical field of natural languages, and the method comprises the steps: constructing an electric power hidden danger map according to an accident hidden danger case of a target electric power system; initializing learnable prompt vectors for each type of nodes and edges of the electric power hidden danger map, inputting the learnable prompt vectors into the pre-training large model, and generating a target semantic embedding representation; according to the target semantic embedding representation, performing similarity weighted analysis on the semantic embedding representation between each node and the corresponding neighbor node to generate an enhanced node representation; inputting the enhanced node representation into a preset expert routing model for training, wherein the expert routing model is designed to introduce a task decomposition mechanism and a multi-target joint loss function for dynamic tuning; and inputting target user query information into the trained expert routing model to carry out efficient query, and outputting to obtain an accurate and reliable power field hidden danger coping scheme.

The invention relates to the technical field of natural languages, in particular to a conference summary generation method and device, computer equipment, a storage medium and a product. The method comprises the following steps: acquiring a key frame image under at least one reference moment displayed in a video conference and audio data of the video conference; performing data splitting on the audio data to obtain audio sub-data at each reference moment; performing data fusion on the key frame image and the audio sub-data at each reference moment to obtain a conference summary of the video conference; according to the method and the device, the multimodal information of each reference moment in the video conference is subjected to summary generation, so that the conference summary contains the visual information in the video conference, and a user can conveniently search specific conference content from the conference summary in the later period.