64 results about "Word recognition" patented technology

The invention discloses a multi-level sensitive word filtering method and system, and relates to the technical field of computers, and the multi-level sensitive word filtering method and system can comprehensively cover antagonism avoidance and semantic fuzzy scenes through a multi-level mechanism of fusion rule matching, expansion word recognition and semantic depth analysis, so that the sensitivity of a user is improved. Therefore, missed judgment and misjudgment are effectively reduced, the system adaptability is improved, missed judgment and misjudgment in the sensitive word filtering process are effectively reduced, and the adaptability and processing accuracy of the system to dynamic changes are improved.

The application belongs to the field of natural language processing in the nuclear power industry, and particularly relates to a professional word segmentation method applied to nuclear power industry corpus, which comprises nuclear power professional vocabulary construction, nuclear power stop word vocabulary construction, nuclear power synonym vocabulary construction, nuclear power same pronoun vocabulary construction, nuclear power field new word recognition, nuclear power field entity automatic recognition, nuclear power field synonym automatic recognition, nuclear power corpus Chinese accurate word segmentation and the like operations. The application has the beneficial effect of completely solving the problem of inaccurate Chinese word segmentation of nuclear power industry corpus, laying a solid foundation for the application of subsequent big data and artificial intelligence in the field of nuclear power natural language processing, and reducing the investment of other practitioners in the field of nuclear power in natural language processing.

This invention discloses an automatic event extraction method for construction quality design specifications, comprising: acquiring and preprocessing the original specification dataset; inputting the data into a pre-trained language model for processing to obtain the event trigger word recognition results of the preprocessed dataset; employing single-trigger word and multi-trigger word classification processing methods, utilizing rule matching methods and deep learning models to jointly process multi-trigger word cases, processing the multi-trigger word recognition results of the preprocessed dataset sequentially according to colon, semicolon, and comma order, and obtaining a complete meta-event dataset using deep learning methods and similarity methods; mining the association between measurement units and measurement terms based on clustering methods, and further supplementing event semantics using similarity methods to improve event expressiveness. This invention, through a combination of grammatical matching and deep learning, completes the generation of specification inspection data, contributing to increasing the number of publicly available Chinese construction datasets.

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

The invention discloses an intelligent management and control method and device for reimbursement applications, equipment and a medium. The method comprises the following steps: according to main table information filled in a reimbursement platform by a user, if it is determined that an attachment verification condition is met, carrying out text information extraction and validity verification on an attachment uploaded by the user; performing sensitive word recognition on the attachment according to the extracted attachment text information and a pre-established sensitive word bank, and determining sensitive words in the attachment and a restricted word bank to which each sensitive word belongs; and according to the restricted word bank to which each sensitive word belongs and a preset attachment verification rule, determining a management and control operation and prompt information for the current reimbursement application, restricting the user to submit the current reimbursement application according to the management and control operation, and displaying the prompt information to the user side. By adopting the technical scheme, the reimbursement application of the user can be comprehensively monitored, and the reimbursement compliance can be accurately identified based on the attachment, so that the risk of illegal reimbursement is reduced.

The application discloses a kind of natural gas marketing strategy execution situation automatic early warning method and system, the method includes: S1: obtaining the natural gas marketing strategy file of a month, and the key word recognition is carried out to natural gas marketing strategy file, extracts natural gas basic data, and it is imported into audit informationization supervision and management platform marketing audit system;S2: obtaining natural gas enterprise financial management system Fmis, marketing system push natural gas system data;S3: according to the natural gas basic data and natural gas system data obtained, using natural gas marketing strategy logic calculation model, real-time calculation natural gas marketing income related quantity;S4: according to natural gas marketing income related quantity, using natural gas sales strategy execution situation suspiciousness multistage screening filtering model is checked gradually, obtains each level suspicious clue, carries out hierarchical classification automatic early warning, fills the blank of domestic about natural gas marketing intelligent audit.

The invention discloses a trigger word recognition and positioning system and method based on BIO sequence tagging, and the method comprises the steps: extracting a whole sentence semantic vector through an encoder, and judging whether an input text contains backdoor attack features or not; if backdoor suspicion is detected, backdoor trigger word positioning is carried out, whether each word belongs to a backdoor trigger word or not is judged through token-by-token BIO sequence labeling, and a preliminary trigger word position labeling sequence is output; performing result correction and optimization by applying trigger position priori knowledge and a multi-strategy rule, and filtering false alarms; when the result is uncertain or suspected to have attack avoidance, scanning the input text by using a predefined known trigger phrase pattern library, capturing a hidden or variant trigger word pattern, and positioning missed suspicious trigger words; and outputting the final fruit of the back door detection. According to the method, through dual-module cooperation on architecture and priori and rule fusion on strategy, text backdoor triggering can be detected in a robust mode, and triggering content can be accurately positioned.

The invention provides an audio wake-up method, intelligent wearable equipment and a storage medium, and the method comprises the steps: detecting the environment sound pressure in real time through a hardware voice activity detection module when an audio processing system is in a low-power-consumption monitoring state; when the environment sound pressure is greater than a preset sound pressure threshold value, the hardware voice activity detection module generates a hardware interrupt signal to trigger the logic processing module to exit a low-power-consumption monitoring state; the logic processing module executes the following steps: controlling the data carrying module to transmit audio data collected by the microphone to the shared memory module, and loading a wake-up word recognition algorithm stored in the external flash memory module to the data processing module; controlling the data processing module to operate a wake-up word recognition algorithm to recognize and process the audio data collected by the microphone in the associated time window; and when the identification processing result satisfies a preset wake-up condition, reporting a wake-up event to the application processing system. According to the technical scheme, the system power consumption is reduced, and the system response speed is increased.

Voice command recognition and natural language recognition are carried out using an accelerometer that senses signals from the vibrations of one or more bones of a user and receives no audio input. Since word recognition is made possible using solely the signal from the accelerometer from a person's bone conduction as they speak, an acoustic microphone is not needed and thus not used to collect data for word recognition. According to one embodiment, a housing contains an accelerometer and a processor, both within the same housing. The accelerometer is preferably a MEMS accelerometer which is capable of sensing the vibrations that are present in the bone of a user as the user is speaking words. A machine learning algorithm is applied to the collected data to correctly recognize words spoken by a person with significant difficulties in creating audible language.

The application discloses a sensitive word recognition method and a sensitive word tree construction method. The recognition method comprises the following steps: in step S11, a character in a text to be detected is read as a first character, and a root node of a sensitive word tree is taken as an initial parent node; in step S12, a character package value of the first character is obtained from a dictionary table, and a child node data table of the parent node is traversed in the sensitive word tree to determine a child node corresponding to the character package value of the first character; in step S13, it is judged whether the child node is a leaf node, if yes, a sensitive word is output, and step S15 is executed; if not, step S14 is executed; in step S14, a next character in the text to be detected is taken as a new first character, and the child node is taken as a new parent node, and step S12 is executed; in step S15, it is judged whether the text to be detected is completely detected, if yes, the method is ended, and if not, the method returns to step S11. The application provides a lightweight solution for quickly recognizing sensitive words.

Embodiments regard generating a requirements statement model. A method can include extracting, from a requirements document, a requirement statement, tagging each word of the requirements statement with a corresponding part of speech tag, identifying dependencies between words and types of dependencies between each word of the requirements statement and associating types of dependencies with each word, generating respective meta-tags for each word of the requirements statement based on the part of speech tag and dependencies, each meta-tag including the parts of speech and type of dependency for each word, identifying, based on the meta-tags and the dependencies, a tree of trees that represent the structure of the requirements statement, and populating the requirements statement model that maps to the identified tree with words of the requirements statement.

This application discloses a multi-model training method, a multi-label qualitative word recognition method, and related equipment. The training method includes: selecting N target characters corresponding to N multi-label qualitative words from sample text containing multi-label qualitative words; encoding the N target characters after masking processing using a pre-trained language model of a word recognition model based on the sample text and the N multi-label qualitative words to obtain a representation vector corresponding to each target character; classifying and recognizing the N multi-label qualitative words based on each representation vector using a classifier of the word recognition model to obtain the predicted word label attributes corresponding to each multi-label qualitative word; adjusting the model parameters of the word recognition model based on the predicted word label attributes corresponding to the N multi-label qualitative words and the N sample word label attributes of the sample text, wherein the sample word label attributes are used to represent the true semantics of the corresponding multi-label qualitative words in the sample text.

The application provides a model evaluation method and device, the method comprising: identifying a preset verification set according to a sensitive word recognition model to obtain the number of true positive examples and the number of true negative examples corresponding to the sensitive word recognition model; determining a target positive example accuracy rate according to the number of true positive examples and the total number of positive examples of the verification set, and determining a target negative example accuracy rate according to the number of true negative examples and the total number of negative examples of the verification set; and performing weighted calculation on the target positive example accuracy rate and the target negative example accuracy rate to obtain evaluation information. The model evaluation method provided by the application can obtain evaluation information representing the recognition accuracy of the model by performing weighted calculation on the target positive example accuracy rate and the target negative example accuracy rate. Since the evaluation information comprehensively considers the accuracy rates of the model in positive example recognition and negative example recognition, the evaluation information can more accurately indicate the recognition accuracy of the model.

The application discloses a WAF rule evaluation method based on a regular analysis tree, relates to the technical field of WAF rule evaluation, and collects regular expression protection rules of an open source WAF as seeds, analyzes and constructs an expression library based on a regular analysis tree, then performs SQL syntax word recognition on a sent test expression, matches all related expressions in the rule expression library through SQL syntax words, traverses the sent expression request, judges whether the target site WAF protection rule is bypassed according to site feedback information, and further evaluates the rule protection capability of the target WAF system. The application can quickly and accurately deduce the target site WAF protection rule, and can reliably and accurately evaluate the protection capability of the target site WAF system.

The invention relates to the technical field of intelligent interaction of pets, and discloses an AI intelligent interaction system for pets, which comprises a pet body, a main control chip, a voice acquisition module, a display module, a sensing module, a feedback execution module, an NFC friend-making module, an offline wake-up word recognition module and an intimacy management module. The induction module is used for identifying shaking actions or touch induction received by the pets, the feedback execution module is used for outputting feedback signals, the NFC friend making module is used for triggering interaction between different pets, and the offline wake-up word identification module is used for waking up the pets without a network. The intimacy management module is used for recording the interaction behavior between the user and the pet body and calculating the intimacy. A social scene is covered through the NFC friend-making module, interaction between toys is achieved, the guidance and use threshold of green hands is lowered, and the whole scene from single-person accompanying to multi-person interaction is covered; due to the integrated design, the AI pet intelligent toy can be developed step by step, the user experience is better, and the activity of the user can be improved.

The invention discloses a wake-up word speech recognition method and device, computer equipment and a storage medium, and belongs to the field of speech recognition. The method comprises the following steps: calculating first probability distribution of each phoneme of input voice through a wake-up word recognition network; judging whether a bank score in the distribution is smaller than a first threshold value or not, if yes, sorting phoneme probabilities, selecting a preset number of target phonemes, and storing the score and the phoneme sequence number of the target phonemes as effective phoneme distribution; the effective phoneme distribution is input into a preset length queue according to a first-in first-out rule, probability values of different time points of the same phoneme sequence number in the queue are subjected to maximum pooling, and an effective phoneme distribution vector is generated; and outputting an identification result based on the vector and a preset registration phoneme distribution vector. The method achieves phoneme-level fine-grained discrimination, reduces the error recognition rate, is light in weight and low in consumption, and is suitable for smart home and other scenes.

The invention relates to the technical field of lexicon adjustment control, and discloses a dynamic difficulty lexicon adjustment control method and system for user cognitive load analysis, and the method comprises the steps: carrying out the cognitive load analysis of multi-modal physiological data, and obtaining the cognitive load characteristics of a user; carrying out capability parameter estimation on the word recognition result sequence, and calculating mastering degree characteristics of the user on the current word bank by utilizing capability parameters; constructing a dynamic difficulty evaluation function to generate the matching degree between the current lexicon difficulty and the user; and generating a difficulty adjustment action of the lexicon by using the reinforcement learning model, and carrying out dynamic difficulty adjustment on the lexicon. According to the method, the vocabulary content most suitable for the cognitive stage of the user can be dynamically pushed according to the current vocabulary mastering degree, the cognitive load bearing capacity and the learning progress of the user, and the cognitive load can be effectively controlled while the learning challenge is guaranteed, so that the active learning motivation of the user is motivated to the greatest extent, and the personalized learning experience and efficiency are enhanced.

The application discloses a method and device for robot wake-up word sensitivity self-adaption to scene noise, which mainly monitors and evaluates the noise level and type in the current environment in real time, and then dynamically adjusts the wake-up word recognition threshold according to the real-time noise evaluation result, so that the self-adaption adjustment mechanism is realized, the sensitivity can be automatically adjusted according to the change of the environmental noise without manual intervention, and when the energy of the environmental sound signal after noise reduction processing is determined to be higher than the wake-up word recognition threshold, it is determined whether to perform the wake-up word recognition, so that the robot can maintain a high wake-up rate in different noise environments, the robot can be self-adapted to various complex and changeable noise environments, and the intelligent level and user experience of the robot are improved.

This invention relates to a method for identifying hot words in customer conversations based on large-scale model technology. The identification method is as follows: model training and construction; users upload batches of historical conversation texts or access conversation streams in real time through the system interface; the system automatically segments the input text into sentences, labels roles, and unifies the encoding format; the trained model processes the preprocessed conversation text; conversation summaries are generated based on hot word graphs; and the results are visualized and output. Compared with existing technologies, this invention achieves accurate identification and classification of the deep meaning of hot words through a multi-dimensional tagging system and semantic understanding; it couples hot word identification and summary generation through multi-task learning and semantic graphs, making the generated summaries both focused and coherent; it supports automated and intelligent analysis of massive conversation data, greatly improving the efficiency and depth of customer sentiment analysis, and providing direct data support for product optimization, service improvement, and risk warning.

The application discloses a speech recognition method, device and electronic equipment, the method sets a hot word database containing multiple hot words in advance, and hot word database is searched and biased decoding is carried out based on the speech feature representation of speech data in the recognition stage, and accurate recognition of personalized and low-frequency hot words is realized. By so setting, the defects that the traditional shallow fusion mode is insufficient to correct the internal bias of the acoustic model and is easy to affect the general recognition performance are overcome, the hot word recognition rate is effectively improved, and the accuracy and stability of the overall recognition system are ensured.

The invention discloses a knowledge-enhanced Chinese multi-mode hatred speech detection method, and belongs to the technical field of natural language processing. The method comprises the following steps: acquiring an image-text pair; extracting visual, scene text, text and hatred knowledge features in parallel through a feature representation module; inputting the text and the prompt template into a large language model to generate a background knowledge text, and splicing the background knowledge text with the text features and hatred knowledge features after BERT coding to form final text features; by means of a knowledge-guided cross-modal attention mechanism, an image-text affinity matrix generated by the CLIP is introduced into an attention map, multi-modal feature fusion is achieved, and the hatred category is predicted. According to the method, large language model background knowledge and a Chinese hatred vocabulary are introduced, the problems of Chinese context understanding, modal difference and hatred word recognition are remarkably relieved, the higher accuracy rate, precision rate, recall rate and F1 score are obtained on a CMMHS data set, and the method is suitable for Chinese multi-modal hatred content auditing on a social platform.

The application relates to the technical field of speech recognition, and provides a speech recognition method and device, electronic equipment and a storage medium, which can greatly improve the accuracy of a speech recognition result, improve the efficiency of speech recognition and reduce the cost of speech recognition through extraction of target entity features in audio data. An audio entity feature extraction unit in a speech recognition model is trained based on a non-text-labeled audio and video training sample, so that not only is manual labeling reduced, labeling cost is reduced, but also the internal relationship between audio data and video data can be fully mined, the pre-training process pays more attention to entities in the audio, the speech recognition model can be applied to a speech recognition downstream task, the speech recognition model has the ability to improve hot word recognition effect. Moreover, the speech recognition model can complete different types of speech recognition tasks, can improve the generalization of the speech recognition model, and expands the application scenarios of the speech recognition model.

The invention discloses a financial electronic document identification method based on artificial intelligence, which comprises the following steps of: acquiring a real object text to be identified to generate an electronic document or intercepting a content part to be identified in the electronic document to obtain an initial processing image; extracting a character part from the initially processed image and finding out a corresponding content coordinate; identifying the text content, and calculating by using the confidence coefficient; table lines in the whole chart are recognized, and intersection points are obtained; obtaining areas divided by the table lines according to the intersection points, traversing the whole image to obtain all the areas, marking the areas, and calculating the coordinates of the central points of the areas; performing verification; after passing, judging the total position error of the whole image; the reliability degree of character recognition is expressed by counting the confidence degree in the whole image; and calculating the recognition effect of the image according to the reliability degree of character recognition and the total position error, and adjusting the confidence coefficient during vocabulary recognition according to the recognition effect. The method is suitable for financial electronic document identification.

A method, device, computer storage medium and terminal for realizing character recognition, the embodiment of the present application acquires a character image containing a preset number of characters based on a PDF document, and performs multi-word recognition on the character image; in the case that the result of multi-word recognition contains a preset number of characters, the final recognition result of the characters contained in the PDF document is determined according to the result of multi-word recognition. The embodiment of the present application can read and render the characters in the PDF file, can determine which character in the PDF the character image corresponds to, in the case that the result of multi-word recognition contains a preset number of characters, the quick confirmation of missed detection or multiple detection is realized, and the accuracy of character recognition is improved according to the result of multi-word recognition; the recognized result is converted into a uniform code (Unicode) and attached to the corresponding character of the PDF, and the accurate recognition of the characters contained in the PDF document is realized.

The embodiment of the invention provides a voice instruction word recognition method and device. The method comprises the following steps: inputting to-be-identified audio into a first encoder for feature coding to obtain audio coding features; inputting the audio coding features into a voice discriminator for voice discrimination to obtain a voice discrimination result; if the voice judgment result indicates that the audio is voice, performing similarity calculation on the audio coding feature and a pre-stored text coding feature of a voice instruction word and an antisense identification word of the voice instruction word; and determining whether the audio is matched with the voice instruction word or not according to a similarity calculation result. According to the embodiment of the invention, the voice instruction word recognition complexity is reduced, and the voice instruction word recognition accuracy is improved.

The application relates to an artificial intelligence-based SMS script sensitive word recognition analysis method, which aims at the technical problem of realizing rich media display and text content synthesis under compliance requirements of marketing SMS, proposes a deep semantic modeling and cross-modal semantic alignment mechanism based on multi-modal data, combines a pre-training language model and a convolutional neural network to respectively extract high-dimensional semantic features of text and images, and optimizes semantic consistency through an attention mechanism and a contrast loss function. Further, the system introduces a dynamic sensitive word recognition and avoidance strategy, effectively filters sensitive expressions, and generates compliance semantic equivalent content through a text reconstruction model when necessary. Finally, the content and the image are fused in format to generate a rich media SMS that can be sent in compliance. The scheme improves the semantic accuracy and content compliance of multi-modal information fusion, ensuring that marketing SMS meets legal compliance requirements and has good user appeal and communication effect.

This invention provides an intelligent negative word construction method and system that combines multi-dimensional dynamic business rules, belonging to the field of data processing technology. Specifically, it includes: a generative language model, through an injected thought chain mechanism and attribute decoupling attention mask, outputs search term attribute vectors and product attribute vectors through inference links and customized masks; constructs a bipartite graph with search term attribute vectors and product attribute vectors as nodes; calculates the semantic distance between attribute nodes using a soft-aligned attention mechanism; establishes synonym alignment connections and conflicting attribute pairs; inputs the conflicting attribute pairs into a discriminative network optimized by adversarial contrastive learning; the discriminative network focuses on the conflicting dimension with the greatest semantic difference through cross-attention gating units; calculates the conflict probability of the conflicting attribute pairs; and makes a comprehensive decision by combining multi-dimensional dynamic business rule thresholds with the conflict probability, thereby improving the reliability of negative word recognition and processing.

This application relates to a method, apparatus, computer device, and storage medium for determining contextual hot words. The method includes: processing the speech to be recognized and candidate hot words in a candidate word list to determine the frame-level speech vector and the text semantic vector of each candidate hot word at each time step; performing temporal accumulation processing of boundary contributions on the frame-level speech vectors to determine an acoustic boundary sequence, wherein the acoustic boundaries in the acoustic boundary sequence are used to indicate the boundary positions of the frame-level speech vectors in the speech to be recognized; performing local alignment of the candidate hot words and the frame-level speech vectors on the acoustic boundary sequence according to the text lexical length of the candidate hot words to determine the acoustic vector group corresponding to the candidate hot words; and filtering each candidate hot word according to the text semantic vector of each candidate hot word and the frame-level speech vector of the acoustic vector group to obtain the contextual hot words of the speech to be recognized, thereby improving the accuracy of contextual hot word recognition.