113 results about "Text stream" patented technology

The invention relates to the technical field of natural language processing, in particular to an AIGC content safety monitoring system and method based on dynamic reasoning and context awareness, and the system comprises a semantic graph construction module which is used for extracting entity nouns and predicate verbs in a text according to a received AIGC interaction text flow, generating a semantic concept node set, and sending the semantic concept node set to a database; and performing directed connection and hierarchical nesting on the concepts in the semantic concept node set according to a logic direction according to a subject-predicate-object dependency relationship rule. According to the method, the curvature value of the semantic track is calculated, and the similarity between the direction vector and the center of the sensitive semantic cluster is combined for double verification, so that sudden turning of an intention in a dialogue process or progressive induction to a sensitive field can be perceived, and abnormal mutation can be recognized through a curvature pulse form; therefore, hostile attack behaviors are accurately captured in real time in dynamic interaction, and the defense capability for context dependent attacks and implicit induction behaviors is improved.

The invention relates to the field of drawing recognition, in particular to a PDF drawing data extraction method and system based on intelligent recognition. Comprising the following steps: reading an internal structure of a PDF engineering drawing to obtain a native text stream, a vector path and a grating image; identifying the native text flow through a shunt preprocessing framework to form structured text data; rendering the vector path and the grating image to obtain a background image; analyzing the structured text data by utilizing the intelligent recognition model through the character recognition and extraction sub-model, obtaining drawing metadata and recording the position, and obtaining a character recognition result; analyzing the background image through a graphic element recognition and classification sub-model, recognizing and classifying component elements, and obtaining a graphic recognition result; and performing fusion according to the visual space corresponding relation to form a drawing analysis result. According to the method, the adaptive capacity of engineering drawings with various sources and different qualities is improved through the shunting preprocessing framework and the intelligent identification model.

The invention discloses a multi-modal fusion short message compliance and security dual auditing method and system, and the method comprises the steps: encrypting account information through employing an irreversible algorithm based on SHA256 and a random salt value, and decomposing the content of a short message into a text stream, a link stream and a symbol stream through employing a regular expression; traversing each character in the short message content by adopting a prefix mode based on a Trie tree in combination with an AC automaton algorithm to detect sensitive words; performing symbol semantic classification mapping and analysis on the special symbol feature data extracted from the symbol stream; carrying out sending behavior analysis on the text feature data extracted from the text stream; performing special detection on link feature data extracted from the link stream; based on the factors corresponding to the sensitive words, the semantics, the behaviors and the links and the weights of the factors, a multi-modal feature fusion decision risk assessment algorithm is adopted to output a risk score value and a risk decision rule of the risk score value so as to execute short message interception operation or short message release operation. According to the invention, full-dimension perception and dynamic defense of risks can be realized.

The invention discloses a lightweight member marketing method and device for real-time conversation, a terminal and a storage medium, and the method comprises the steps: responding to a received user voice stream, continuously generating an incomplete intermediate text stream through streaming voice recognition, and inputting the intermediate text stream into a lightweight marketing response model; the lightweight marketing response model predicts the potential intention of the user in real time based on the intermediate text stream, and pre-assembles one or more candidate response frameworks from the fragmented marketing corpus; when it is detected that the utterance of the user ends, the real intention of the user is determined based on a complete final text stream generated by streaming speech recognition; and arbitrating the candidate reply framework according to the weighted decision factor set to obtain an arbitration result, determining final reply content according to the arbitration result, converting the final reply content into an instant response call, and outputting the instant response call. According to the method, the time delay of the real-time call marketing process can be reduced, and the concurrent processing capability is improved.

The invention discloses a simultaneous interpretation method and system based on a large model and electronic equipment, and the method comprises the steps: extracting bilingual parallel corpora related to terms from professional resources based on a standardized professional dictionary, obtaining qualified corpora through data enhancement processing and manual screening, and constructing a multi-level corpus according to the levels of words, sentences and paragraphs; the method comprises the following steps: receiving an input audio stream in real time, extracting acoustic features through preprocessing, inputting a pre-established large-scale speech recognition model, and carrying out incremental decoding on the acoustic features in a sliding window mode; and calling a sentence boundary prediction network to judge a pause point, and outputting a text stream with a timestamp. Performing fine tuning on the large-scale speech recognition model by using a multi-level corpus, translating a text stream based on the fine-tuned large-scale speech recognition model, and constraining term translation according to a standardized professional dictionary; and synchronously displaying the audio output in the translation result and the subtitles. According to the scheme, the terminology recognition and translation accuracy is improved, and simultaneous interpretation delay is reduced.

The invention discloses an RAG content generation method and system based on dynamic slicing and self-adaptive fusion, and relates to the field of natural language process.The method comprises the steps that in response to a user query instruction, semantic fusion processing is conducted on a plurality of related text blocks, and a preliminary text sequence is traversed; detecting a logic breakpoint between adjacent text blocks through a context association identifier, recalling a middle connection text block corresponding to the logic breakpoint, and inserting the middle connection text block into a sequence to generate a fusion text stream; performing adaptive information density adjustment on the fused text stream to obtain optimized context content; generating structured prompt information based on the optimized context content and the user query instruction; and calling the target large language model to generate a user query result based on the structured prompt information. The answer accuracy and logic continuity of the target large language model for user query are effectively improved, and the utilization efficiency of model computing resources is optimized.

The invention discloses a real-time video processing method and system based on artificial intelligence, and belongs to the technical field of computers. Through deep fusion of multi-source information and a regional adaptive coding mechanism, the problem of collaboration between content generation and network resource allocation in a high-interactivity live broadcast scene is solved. In the feature extraction link, video, audio and text streams are synchronously processed, so that three-dimensional perception of a live broadcast scene is realized; in the multi-mode driving instruction set generation process, audio event features and text interaction features are subjected to semantic association, so that content generation and real-time interaction are highly synchronized, and response delay caused by information splitting in a traditional scheme is avoided; according to the regional differentiation coding processing, resource allocation is dynamically adjusted according to the regional importance of visual semantic feature recognition and the network state, and the coding quality of a digital human region is preferentially guaranteed, so that the stability of core visual elements is maintained under the network fluctuation condition, and the intelligent level of real-time video processing and the user experience are improved.

The invention relates to the technical field of financial science and technology, and discloses a security transaction instruction intelligent analysis method and system based on a large model, and the method comprises the steps: carrying out the compliance screening of an instruction text flow and an instruction feature chart, and obtaining an initial text flow and an initial feature chart; analyzing a transaction feature factor corresponding to the initial feature chart, extracting a space-time structure feature vector from the initial feature chart, and analyzing a structure description text corresponding to the space-time structure feature vector to generate a chart analysis text corresponding to the initial feature chart; calculating segmentation entropy loss corresponding to the initial segmented text, and performing segmentation iterative optimization on the initial text stream to obtain a target segmented text; generating an instruction target text corresponding to the security transaction instruction, calculating a structured likelihood between instruction tags, and determining an association aggregation tag in the instruction tags; and analyzing the transaction intention corresponding to the security transaction instruction, and generating an analysis result corresponding to the security transaction instruction. According to the invention, the accuracy of security transaction instruction analysis can be improved.

The invention relates to a multi-language voice content recognition method and system, and belongs to the technical field of voice signal processing, and the recognition method comprises the steps: collecting original audio stream data, executing the noise reduction filtering processing, and segmenting the original audio stream data into a plurality of audio segments; extracting acoustic feature vectors of the audio clips; inputting the audio clip into the speech recognition model group, and obtaining a text clip and a confidence score; fusing all the text fragments to generate a preliminary recognition text, segmenting the preliminary recognition text into text blocks, and translating the text blocks into a target language text block by block through a streaming translation model in combination with a semantic feedback tag received in real time; executing bidirectional translation verification on the text blocks of which the confidence scores are lower than a set threshold value; inputting the translated text stream and the acoustic feature vector into a semantic analysis model, and fusing an output result to obtain a semantic feedback tag; and constructing a cross-modal association graph through the graph neural network, outputting a key information abstract and triggering an alarm signal. The speech recognition accuracy can be improved while the recognition efficiency is guaranteed, and risk response is carried out.

The invention relates to the technical field of retrieval enhancement, in particular to a big language model-based dynamic knowledge retrieval enhancement method, which comprises the following steps of: acquiring theoretical regulations and actual operation data of an unmanned aerial vehicle, and driving a model generation request; monitoring the reasoning text stream in real time, intercepting a logic triple to perform causal verification, and constructing negative feedback prompt blocking and resetting for unsupported logic; executing double-track system dynamic retrieval aiming at regulations and data, evaluating content conformity by utilizing reflection Token and executing conflict resolution correction; and generating a fault hypothesis by adopting tree decoding, constructing an anti-fact query to retrieve the evidence of the security, and pruning the path with the evidence of the security to determine a final path. According to the method, illusion is effectively blocked through real-time logic verification and a negative feedback mechanism, double-track retrieval and anti-fact pruning are combined, strict alignment between the reasoning process and physical facts and theoretical regulations is ensured, and the accuracy and credibility of fault diagnosis are remarkably improved.

The invention discloses a voice real-time question and answer processing method based on a large model and a domain knowledge base, which belongs to the technical field of computer data processing, and comprises the following steps: acquiring a voice input stream of a user, carrying out intelligent sound wave deconstruction on the voice input stream, generating a text stream, and carrying out dependency syntactic analysis and semantic role labeling; generating intention information and key information, dynamically accessing a domain knowledge base, executing predictive loading, generating a preloaded data subset, performing fusion processing by combining the intention information, the key information and the preloaded data subset, generating a fusion result, performing accuracy verification and correction, generating a natural language answer, and converting the natural language answer into voice output. The technical scheme of combining knowledge predictive loading driven by intention analysis, context fusion and answer traceability correction is adopted, and low-delay response, high-precision intention understanding and high-factuality answer of voice questions and answers can be achieved.

The embodiment of the invention discloses a teaching note generation method and device based on a large language model and a medium, belongs to the technical field of artificial intelligence, and solves the problems that classroom notes generated in the prior art are lack of content structuring and knowledge relevance, poor in readability and difficult to be used for effective review. The method comprises the following steps: acquiring a teacher voice signal in a classroom teaching environment in real time through pickup equipment, and converting the teacher voice signal into an initial text stream through a preset voice recognition model; taking the initial text flow as query input, and performing semantic similarity retrieval on the initial text flow and a preset course knowledge base to obtain a plurality of related knowledge fragments; generating an enhanced prompt based on the initial text stream and the plurality of related knowledge fragments, and generating a structured classroom note based on the enhanced prompt and a preset large language model; and outputting the structured classroom note to a user interface, and optimizing a classroom teaching note generation process based on feedback information and modification information of the user.

The invention relates to the field of computer vision and artificial intelligence, in particular to a multi-modal fusion document image classification method based on multi-branch deep convolution and hierarchical semantic modeling. According to the method, visual and text information of a document is co-processed through a double-flow architecture; a visual semantic feature of a document image is efficiently extracted from a visual flow by adopting a lightweight convolution operation and multi-scale feature fusion mechanism; an original text is obtained from a text stream through an optical character recognition technology, text error correction and word embedding extraction are carried out through a fine tuning language model, word-level and sentence-level context modeling is realized by inputting a hierarchical semantic coding network, and document-level features are generated. And high-precision document image classification is realized by fusing the double-flow feature vectors and combining a full-connection layer. Experimental results show that the method has an excellent multi-modal feature fusion effect, relatively high classification precision and strong robustness to complex document scenes in a document image classification task, and is suitable for application scenes of various document image classification.

The present application relates to the technical field of data compression, in particular to a data compression method and system, comprising: an LZ module, a text stream module, a sequence stream module, an interval entropy encoding framework, the LZ module is used for screening repeated data of original data, generating a text stream and a sequence stream; the LZ module screens repeated data through a hash verification method, specifically: calculating a hash value for the input data block, determining repeated data based on hash value comparison to avoid direct data content comparison; the text stream module is used for entropy encoding the text stream; the present application innovatively proposes an interval entropy encoding framework, constructs a more optimal probability model through artificial intelligence and mathematical methods, trains a more optimal finite state machine, and uses the optimal state machine in the compression and decompression process, efficiently solving the problem of slow decompression speed of neural network type methods.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for table detection using text streams. One of the methods includes detecting, in a text stream and using column identification data, text for one or more cells in a table; creating, using the text for at least some of the one or more cells in the table, a data structure for the cell a) that associates two or more values from the table and b) for use by a downstream system as part of a natural language analysis process of data from the text stream; and storing, in memory, the data structure.

The invention relates to a material processing method and system based on sensitive content detection and a medium, and belongs to the technical field of multimedia content security. The material processing method comprises the following steps: receiving an input original material, and analyzing the input original material into an image stream, a video stream and a text stream; executing HSV color space conversion, and extracting texture features; performing semantic analysis on the text stream to generate a word segmentation sequence and a named entity recognition result; random frequency noise is injected based on the saturation channel, and local binary pattern features are extracted from the saturation channel after noise injection; generating a semantic vector according to the word segmentation sequence and a corresponding named entity recognition result, and calculating a risk entropy value; and establishing a space-time mapping relationship between the image stream / video stream and the text stream, constructing a cross-modal incidence matrix, executing a grading decision according to an output result, executing a material processing operation, and feeding back a processing result to the cross-modal incidence matrix for weight updating. The sensitive content identification accuracy and timeliness can be improved.

The application discloses a new word and a first occurrence event recognition method and device, wherein the method comprises the following steps: obtaining hot-line work order text data according to a preset period, and preprocessing the hot-line work order text data to obtain a structured text stream; performing word segmentation and sliding window processing on the text stream to obtain an initial phrase candidate set; performing statistical access screening on the initial phrase candidate set, and performing alias merging on the candidate phrases passing the statistical access by using consistency rules of specified dimensions to obtain a word candidate set; updating a current space-time reference baseline based on the word candidate set, and recalculating historical full-amount data according to a specified fixed period to re-calibrate the space-time reference baseline; and determining whether there is a new word in the word candidate set and whether there is a first occurrence event based on the re-calibrated space-time reference baseline. Based on the unified space-time reference baseline, the new word and the first occurrence can be accurately and timely determined.

The model graph receives the data cue as input. The data cue is segmented into a plurality of segments. An instance of the model graph is generated for each segment of the data cue. Each instance of the model graph is also pruned according to policy information associated with a corresponding data cue segment of the model graph instance. Each instance of the model graph generates an intermediate output. A final output for the model graph of the entire data cues is generated based on the combination of the intermediate outputs.

This application relates to the field of computer technology, and in particular to a method, system, and storage medium for automatically generating stock market news. The method includes: acquiring transaction data, order data, and an external unstructured text stream of a target security within a continuous trading window; performing time-series slicing on the transaction data to generate price and volume sequences; identifying price abrupt changes and volume surges based on the price and volume sequences, and calculating the time offset between them; if the time offset falls within a preset causal relationship time window, extracting feature news segments from the external text stream; performing dependency parsing on the segments to extract subject entities and predicate actions to generate driving factors; and filling slots in the price sequence according to a preset news narrative graph to generate causal news. This application helps to automatically and accurately reveal the causal relationship between news events and stock market price and volume changes, generating high-quality stock market news.

The invention discloses an efficient edge reasoning system and method based on a lightweight multi-modal large model, and relates to the technical field of large model edge reasoning, and the method comprises the steps: S1, carrying out the time discretization and alignment segmentation of an original video frame, an audio signal and a text stream, and obtaining a lightweight feature vector, performing splicing and linear projection to obtain unified semantic feature vectors, and performing arrangement to obtain a unified semantic feature sequence; s2, performing memory reading based on the unified semantic feature vector and the memory state to obtain a memory reading feature, and performing calculation to obtain a surprise degree and a routing decision mark; s3, constructing a unified semantic feature sequence based on the routing decision mark and the short-term context window, and executing forward reasoning to obtain core semantic representation and a task output log value vector; and S4, constructing a write-in value vector based on the routing decision mark and the core semantic representation, performing surprise calculation on the write-in intensity, and obtaining the memory state of the next time step through low-rank outer product addition and two-norm truncation processing.

A system and method provide real-time voice-based order processing in an order terminal. A schema is generated that defines all possible orders, all possible options for each of the possible orders, and an output format for defining a user order. Audio signals are received representing an order spoken by a user. The received audio signals are transcribed via an artificial intelligence transcription model trained to recognize speech in real time to generate a transcribed text stream. The transcribed text stream and the generated schema are provided to a large language model (LLM) for processing the transcribed text stream to identify order information based on the generated schema. The LLM provides the identified order information in an output arranged according to the output format in the generated schema. The output from the LLM is received and a current transaction list is updated based on the identified order information in the output.

The invention provides a communication information content intelligent auditing method and system based on deep learning, and relates to the technical field of communication information auditing, and the method comprises the steps: firstly obtaining a to-be-audited session text flow in an enterprise-level instant messaging system, carrying out the message intention evolution trajectory modeling of each complete session thread, and obtaining a message intention evolution trajectory model; and extracting message intention labels and connecting in series to form an intention evolution state transition sequence. And carrying out path similarity comparison on the illegal business intention track and a preset illegal business intention state transition graph library, and identifying a suspected illegal intention track. And then constructing a session context evidence chain surrounding the core violation intention, inputting the session context evidence chain into a pre-trained message content deep analysis network to carry out multi-level violation feature extraction, generating a comprehensive violation confidence score, and generating a final content auditing interception instruction. According to the method, the accuracy and the intelligent level of content auditing of the enterprise-level instant messaging system can be effectively improved.

The invention provides a secret leakage early warning method and device in a writing process, and belongs to the technical field of data security. The method comprises the following steps: modeling sensitive information to be protected, generating encrypted sensitive features by using a homomorphic encryption technology, and constructing a sensitive feature fingerprint database; capturing a text stream in real time in a writing process of a user, and performing semantic analysis on the text stream by using a large language model to extract text features; in the encryption domain, the text features and encryption sensitive features in a sensitive feature fingerprint database are compared and calculated, so that the leakage risk level is evaluated; and when the risk level exceeds a preset threshold value, sending a real-time risk prompt. The invention also provides a device for realizing the method. According to the method, the risk analysis is carried out in the encryption domain, so that the risk of secondary leakage is fundamentally eliminated, and the data privacy security is enhanced; meanwhile, the text stream is analyzed in real time, so that active intervention can be performed at the moment when a secret divulging behavior occurs, and pre-defense is realized; and the semantic comprehension capability of the large language model is combined, so that the detection accuracy is improved.

The application provides a three-mode redundant circuit netlist compiling method and device based on state machine conversion, belongs to the technical field of anti-radiation integrated circuit netlist compiling processing, and comprises the following steps: generating a netlist ASCII text stream and a keyword-regular set by inputting core source data suitable for an anti-radiation scene, triggering a lexical analyzer, forming an ordered lexical unit sequence through lexical analysis, constructing a state transition graph and checking the state transition graph, integrating the graph and a fragment, extracting a syntax tree, converting the syntax tree into a flow graph data structure, finally generating an IR intermediate representation mapped with SoC hardware, and outputting the IR intermediate representation. The three-mode redundant circuit netlist compiling method and device based on state machine conversion are used, so that the problems of low recognition efficiency, large cost and easy deletion of redundant logic in the process of processing a three-mode redundant circuit netlist by a traditional hardware compiler are solved.

The invention discloses a multi-mode AI interaction method and device, a medium and a product, and the method comprises the steps: obtaining user portrait data, real-time scene data and user memory data; a fusion feature vector is generated; determining a target style, and combining the target style with the fusion feature vector to obtain a plurality of candidate multi-modal interaction contents; carrying out sorting processing on the plurality of candidate multi-modal interaction contents, and determining target multi-modal interaction contents; outputting the target multi-modal interaction content, and obtaining a user voice stream, a user image stream, a user video stream and a user text stream; performing timestamp alignment processing on the user voice stream, the user image stream, the user video stream and the user text stream to obtain multi-modal data; performing emotion recognition processing on the multi-modal data to determine a current emotion state; and matching and determining an interaction strategy by using the current emotional state, generating an adaptive interaction response and outputting the adaptive interaction response. By implementing the scheme, the overall intelligent level of man-machine interaction and the user experience are enhanced.

This invention discloses a method for recommending music in live streaming rooms, comprising: real-time collection of multimodal data from the live streaming room; the multimodal data including the streamer's video stream, audio stream, bullet screen text stream, and live streaming environment parameters; fusion analysis of the multimodal data to construct a real-time emotion-scene profile of the live streaming room; based on the emotion-scene profile, matching music from a pre-set music library or generating a matching background melody through a lightweight AI composition module; dynamically adjusting the music volume according to the streamer's voice energy to achieve adaptive mixing output; the invention further discloses a music recommendation system for live streaming rooms, a music recommendation device for live streaming rooms, and finally a computer-readable storage medium. This invention is the first to deeply integrate dynamic emotions and scene types during live streaming, breaking through the traditional static recommendation logic and enabling music to be adjusted in real time according to the live streaming atmosphere.

The invention discloses an outbound call real-time interruption processing method and device, computer equipment and a storage medium. The method comprises the following steps: when a preset verbal skill is played, monitoring and converting an audio stream into a text stream in real time through VAD and ASR services; inputting the processed data into a streaming interruption model for node configuration analysis; judging whether interruption is supported or not; if yes, when the confidence coefficient of the streaming interruption model exceeds a threshold value, stopping playing the preset verbal skill and switching to a waiting / recognition mode until the VAD confirms that the speaking of the user is finished; if not, monitoring the audio stream without interrupting the playing until the VAD confirms that the speaking of the user is ended; obtaining the whole dialogue content of the round; inputting the data into a whole-section hanging-up model and a high-risk model for posterior judgment; and performing corresponding processing, and temporarily storing the whole dialogue content of the round. By means of the method, high-precision intention recognition can be guaranteed, meanwhile, low-delay response speed can be provided, and unnecessary false triggering conditions are effectively reduced.

The invention discloses an interview evaluation method and system based on multi-modal information, and relates to the technical field of artificial intelligence, and the method comprises the following steps: starting an interview client, synchronously collecting a video stream and an audio stream of a candidate by the interview client, and generating a corresponding interactive text stream in real time; wherein the interactive text stream, the video stream and the audio stream corresponding to the same timestamp form multi-modal interview data; and locally caching and preprocessing an interactive text stream, a video stream and an audio stream in the multi-modal interview data to obtain multi-modal time sequence data. According to the method, by setting the adaptive buffer area and the priority, the contradiction between delay and packet loss can be intelligently balanced while the alignment and integrity of the multi-modal time sequence data can be ensured, high-quality and high-reliability input can be provided for the deep neural network model, and finally, accurate and steady remote automatic interview evaluation is realized.

The invention discloses an intelligent collection method and device based on audio analysis, and is applied to the technical field of data processing.The method comprises the steps that multi-mode audio analysis serves as the core, and basic information such as debtor credit data and legal document templates is collected and stored in a structured mode; call audio is transferred into a text stream with a timestamp in real time through an ASR technology, and audio framing, voice activity detection and other features are synchronously extracted. Through semantic, acoustic feature and dialogue rhythm three-dimensional analysis, a communication strategy and a switching threshold are determined by a multi-modal fusion and strategy decision engine, an adaptive verbal skill is generated, and a conversation link is established through a third-party platform. The debtor feedback is analyzed in real time, the verbal skill is dynamically adjusted until an effective result is obtained, finally key information such as committed repayment is extracted, a compliance legal document is automatically generated and sent out through one key after manual auditing, and whole-process intelligent collection is achieved.

The invention relates to a low latency decompressor. An example method of low latency decompression includes receiving a data read request to read data stored in a memory in a compressed storage format; and in response to receiving a data read request, accessing the compressed data sequence, dividing the compressed data sequence into three separate streams for parallel processing, the three separate streams including (i) a text stream, (ii) a historical cache stream, and (iii) a historical buffer stream, for each data sequence in the text stream, determining a text decompression block offset for the data sequence, for each data sequence in the historical buffer stream, a decompression block offset for each data sequence in the historical buffer stream is determined using one or more historical cache pointers associated with the data sequence, and for each data sequence in the historical buffer stream, a decompression block offset is determined via the historical buffer, and generate a data output in response to the data read request.