42 results about "Lexicon" patented technology

This application provides a text-guided method and system for detecting face forgery. The method includes obtaining a face image to be detected (whether it is genuine or fake), inputting the face image into a face detection model to obtain the face authenticity detection result. The face detection model includes: constructing a text prompt lexicon covering multiple granularities and generating multi-dimensional text prototypes; extracting visual features, optimizing the feature distribution of visual features to obtain global visual features; performing feature separation and enhancement on the global visual features; mapping the global visual features after feature separation and enhancement to predicted text features; applying similarity constraints to obtain cross-modal prototype matching results; applying discriminative constraints on different-dimensional text prototypes to obtain feature measurement learning results; and outputting the detection result based on the cross-modal prototype matching results and feature measurement learning results. This method solves the problem of poor generalization in face forgery detection and improves detection accuracy and generalization ability.

The present application relates to the technical field of intelligent logistics, and particularly relates to a parcel pile address full word association method and device, equipment and a storage medium, the method first carries out data preprocessing to the original address data set obtained, obtains a plurality of address full word entries, respectively allocates a unique address full word identifier to each address full word entry, and is one-to-one associated and bound with the delivery code region number set obtained, obtains a standardized address full word library, determines a parcel pile in response to the operation instruction of the courier, obtains at least one address full word entry and the unique address full word identifier corresponding to the address full word entry from the standardized address full word library, and constructs an association relationship database based on the corresponding unique address full word identifier and the number of the parcel pile, matches the obtained address full word to be matched based on the standardized address full word library, if the matching is successful, determines the parcel pile information corresponding to the address full word to be matched based on the association relationship database, and aims to improve the distribution efficiency.

The present application relates to the technical field of electronic digital data processing, and discloses a four-level index pinyin bucket word library generation method, acquires a Chinese word library text, extracts the first Chinese character and the first pinyin of each line of words as a classification key, and generates a triple; the triple is classified into a corresponding pinyin bucket according to the pinyin, the words in the bucket are sorted in descending order of word frequency, an independent Chinese character block is generated for each unique Chinese character, and the words are solidified according to the word length by using a first character multiplexing mechanism; a four-level static index of an initial letter statistical area and a pinyin index area and a Chinese character word index area and a word storage area is sequentially constructed, each index area uses fixed-length entries and absolute offset addressing; a metadata area containing a file length double-semantic field is constructed, a high-frequency-first deterministic truncation is performed on the word library according to a preset threshold, each data area is spliced and a data integrity check value is appended, and an embedded binary word library is generated. The problems of high storage redundancy and uncontrollable memory are solved, and the purposes of deterministic analysis, resource adaptation and high security are achieved.

1. Name of the product in this design: Lyrics translation module for word translation display and interactive graphical user interface on player page of electronic device. 2. Purpose of this design: An electronic device. 3. The key design features of this product are: the parts in the graphical user interface, and the parts not marked with dotted lines are the parts for which protection is required. 4. The picture or photo that best illustrates the key design points: Design 1 front view. 5. Design 1 is designated as the basic design. 6. Purpose of the graphical user interface: The overall appearance design of the interface is used to display the song playback page, display lyrics, control song playback, and display word translations based on the dictionary (words marked in Chinese lyrics will be switched to English words with Chinese translations, while words marked in foreign language lyrics will directly display the Chinese translations); the partial appearance design of the interface is used to display lyrics, control song playback, and display word translations based on the dictionary (words marked in Chinese lyrics will be switched to English words with Chinese translations, while words marked in foreign language lyrics will directly display the Chinese translations). 7. Human-computer interaction method of graphical user interface: In the main view of Design 1, the foreign language lyrics are displayed in the middle of the interface. The corresponding Chinese translation is displayed below the words marked by the dictionary in the lyrics. Users can click on the lyrics area to bring up the word list module. Design 2 uses the same human-computer interaction method as Design 1. In the main view of Design 3, the Chinese lyrics are displayed in the center of the interface. The words marked in the lyrics are switched to English words with Chinese translations. Users can click on the lyrics area to bring up the word list module. Design 4's main view is a landscape playback page, with foreign language lyrics displayed in the center of the interface. The corresponding Chinese translations are displayed below the words marked in the lyrics according to the dictionary. Users can click on the lyrics area to bring up the word list module. In the main view of Design 5, when the user clicks on the blank area in the middle of the interface, the lyrics are displayed in an immersive state, showing the interface changes from the main view of Design 5 to the interface change diagram of Design 5.

The application belongs to the technical field of data retrieval, and particularly relates to a traditional Chinese medicine auxiliary syndrome differentiation system based on semantic analysis. The system comprises a traditional Chinese medicine symptom preprocessing module, a dynamic inverted index construction module and a semantic retrieval matching module. The preprocessing module extracts symptom entities and degree modifiers of the inquiry text through a traditional Chinese medicine field dependency syntax analysis tree, maps them to a standard symptom word library and converts them into weight coefficients. The dynamic inverted index construction module establishes an inverted list with the standard symptom words as index keys, and appends the bias values of the basic syndrome vectors and the document syndrome vectors after the document identification. The semantic retrieval matching module generates a query vector with weight coefficients according to a symptom query request, calculates the inner product of the query vector and the bias values in the inverted list, eliminates the document identification below the preset threshold, and outputs the syndrome conclusion corresponding to the remaining documents. The application solves the semantic gap and the synonym omission problem caused by the non-standardized expression of traditional Chinese medicine.

The application relates to the field of medical software and discloses a risk modeling method based on deep uncertainty quantification and neural bias correction. In view of the cognitive uncertainty caused by the writing habits or fuzzy description of doctors in clinical medical records and the problem that the risk is underestimated due to keyword matching in the prior art, deep semantic features are extracted by a large language model, semantic cognitive uncertainty is quantified by means of Monte Carlo sampling, and potential negative description bias is automatically identified. Then, a neural style embedding vector is introduced, a residual network is combined, nonlinear risk compensation is performed on the uncertain features, and calibrated risk factors are output. The method also uses a variational autoencoder to realize privacy desensitization and feature dimension reduction, thereby realizing automatic positive compensation of risk while getting rid of the restriction of an artificial vocabulary, and significantly improving the insurance pricing accuracy and fairness under unstructured medical data.

The application discloses a kind of based on the automatic construction method and system of business ontology of information foraging theory, it is related to knowledge engineering and artificial intelligence field.The method includes: determining the target vector of business ontology and constructing the bidirectional smell word bank containing positive and negative smell words;Multi-source heterogeneous documents are cut into information patches, and the multidimensional smell intensity containing theme correlation component and smell word hit component is calculated;According to smell intensity, a foraging priority queue is constructed from high to low, and patches below the global stop threshold are skipped;The patches extracted are extracted ontology elements in turns, and whether to terminate and migrate is determined according to the comparison of instantaneous yield and average yield of environment;Ontology elements are fused into business ontology according to confidence, and the smell word bank is updated according to output, so that the word bank, smell intensity and ontology extraction form a foraging closed loop.The application concentrates limited computing power on high-value patches, reduces computing power consumption and processing delay under the premise of ensuring accuracy and recall rate.

The present application relates to the technical field of data processing, and relates to a sensitive word extraction method and device for an Internet community and a storage medium. The method comprises the following steps: obtaining a total sensitive word library and historical post and comment data; training using the total sensitive word library, the historical post and comment data and a first preset BERT model to obtain a sensitive word coarse extraction model; extracting sensitive words from post and comment data that has been manually audited using the sensitive word coarse extraction model to extract first target sensitive words; when it is determined through manual auditing that the first target sensitive words meet sensitive word extraction rules, storing the first target sensitive words and corresponding first target post and comment data in a sensitive word source post and comment library; training using data in the sensitive word source post and comment library and a second preset BERT model to obtain a sensitive word fine extraction model; and extracting sensitive words from online full-amount post and comment data using the sensitive word fine extraction model to extract second target sensitive words.

This invention discloses an intensity-aware emotion classification method based on a hierarchical knowledge-enhanced graph neural network. Its key features include: preprocessing the Chinese text to be classified by word segmentation, syntactic parsing, and dependency tree construction; extracting word-level contextual semantic vectors and global sentence representations using a pre-trained language model; identifying sentiment words and degree words in the text, calculating the dynamic sentiment intensity of lemmas, and adaptively expanding the sentiment lexicon; extracting a sentiment knowledge subgraph and aggregating knowledge embeddings, followed by seed-driven pruning to obtain a sparse dependency subgraph; fusing semantic and knowledge features to generate initial node representations, and learning these representations using a hierarchical intensity-aware graph neural network; finally, generating graph-level features through dual-path pooling, fusing global semantics, and outputting the emotion classification result through a multi-task framework. This invention solves the problem of ambiguous emotion classification caused by insufficient modeling of the combined intensity of "degree words + sentiment words" in existing technologies.

The application discloses a brand prompt library generation method, which comprises the following steps: collecting user real questions related to a brand from a public information platform to form a user question set; combining a term dictionary of an industry corresponding to the brand to extract the first N high-frequency segmented words corresponding to the user question set as a candidate question association word set; combining brand characteristics to analyze the user question set by using an intention recognition model to form a structured question association word classification, and screening and classifying the candidate question association word set according to the question association word classification to generate a structured question association word system; and generating a brand prompt library according to the question association word system and combining a preset brand knowledge graph. The application also discloses an electronic device and a computer program product for executing the above method.

This invention discloses a domain-adaptive training method and apparatus for a large-scale aerospace software model, belonging to the field of aerospace model training. The method includes: processing an aerospace-specific corpus that has undergone word segmentation and stop word filtering based on word frequency and inverse document frequency to obtain a core lexicon for the aerospace software domain; performing hybrid weighted mask sampling on the input text based on the core lexicon and the domain weight of each term in the lexicon, and training a basic model based on the sampling results to obtain a pre-trained model; performing low-rank adaptive fine-tuning updates on the pre-trained model's parameters based on structured instruction sets and enhanced thinking chain data to obtain an instruction fine-tuning model with engineering reasoning logic; and performing direct preference optimization on the instruction fine-tuning model based on high-quality preference pairs cleaned through multi-judge model collaborative voting to obtain a reliable large-scale software model that meets engineering requirements. This invention can effectively improve the reliability and security of the model output results.

This invention relates to the field of network and information security technology, specifically to a method for adversarial prompt purification and semantic perturbation defense against embodied intelligence. It involves constructing a jailbreak token lexicon and an embodied intelligence synonym lexicon to define attack characteristics. Then, it receives user commands, purifies them by removing obvious malicious elements, and standardizes the commands. Furthermore, it applies semantic perturbation defense to the purified commands by replacing words with synonyms. This disrupts the adversarial token sequence while preserving the original semantic intent, thereby reducing the likelihood of jailbreak attacks. This invention proactively weakens the effectiveness of jailbreak attacks at the input source, transforming potentially malicious prompts into neutral ones to eliminate hidden or confusing commands while retaining the meaning of the dialogue. Unlike direct rejection, this invention cleans up commands before the embodied intelligence model uses them.

This application discloses a game operation method, apparatus, device, and storage medium, relating to the field of game technology. The method includes: responding to a game launch operation, selecting target words from a game vocabulary database based on the player's profile as the target answer for the current round of the game; generating a first prompt word based on the basic attribute words of the target answer and displaying it on the game interface, the game interface having a question window and an answer window; acquiring the prompt question received from the question window, generating a corresponding prompt answer based on the characteristic attribute words of the prompt question and the target answer, and displaying it on the game interface; acquiring the game answer received from the answer window, and determining whether to continue the current round of the game based on the game answer and the target answer. Through the above technical means, the game can adaptively adjust the prompt words and vocabulary to cater to the player's flexible word association and logical reasoning thinking, making the game's prompting mechanism more flexible and varied, thus improving the game's flexibility and playability.

The application discloses a method for constructing a constraint lexicon and generating text based on atomized poetic primitives. The method constructs a poetic primitive library composed of multiple indivisible words, organizes it into five semantic categories: life, emotion, nature, time and space, and humanity, and binds an eight-dimensional emotion attribute vector to each primitive. According to the user's creative intention, a target poetic primitive set is selected and input into a large language model as a constraint condition to guide the generation of text that meets the target emotional attributes. The application converts the image combination logic of poetry creation into a computable primitive constraint, achieving atomic-level emotional control over AI-generated text.

This invention relates to the field of intelligent information processing technology, and more particularly to a self-evolving process entry routing closed-loop method and system based on user feedback. The method constructs a two-dimensional dynamic lexicon containing process keywords and fuzzy synonyms; it segments the user's natural language to extract core keywords; it calculates the first and second hit rates of the core keywords and the two-dimensional lexicon respectively, and then weights and fuses them based on the two-dimensional weight coefficients to obtain a comprehensive matching score; it returns recommendation results to the user based on the score and ranking threshold and obtains interactive feedback data; finally, it dynamically updates the underlying lexicon based on the feedback data, and adaptively adjusts the weight coefficients and ranking thresholds in conjunction with business operation requirements data. This invention breaks through the barriers of traditional single exact matching, and achieves a self-evolving routing mechanism without manual intervention through data closed-loop, significantly improving the routing accuracy in complex contexts and dynamic business conditions.

This disclosure relates to a speech recognition method, apparatus, device, vehicle, and medium. The method includes: performing speech recognition on voice navigation information to obtain at least one candidate recognition result; determining an acoustic score for each candidate recognition result and a language score for each candidate recognition result; identifying target hot words contained in each candidate recognition result according to a preset hot word library and obtaining a target hot word score for the predetermined target hot words; determining a reference score for each candidate recognition result based on the acoustic score, language score, and target hot word score, and determining the target recognition result based on the reference score. In the embodiments of this disclosure, based on the speech recognition results combining language and acoustics, client-side hot words are introduced to determine the final recognition result. Since client-side hot words reflect personalized search habits, the influence of similar sounds can be removed, further improving the accuracy of the recognition results.

The application provides a text quality inspection method and device, medium and product, and relates to the application of a large model in the field of financial technology. After obtaining the text to be inspected, the method performs quality inspection processing on the text to be inspected based on a text quality inspection model trained based on a training data set of a plurality of semantically associated text data (text data corresponding to synonyms, text data corresponding to homophonic words, and / or text data corresponding to abnormal sentence patterns), to obtain all abnormal text data in the text to be inspected. The method does not need to continuously update the rule library manually, generates a word library automatically through semantic association, and the text quality inspection model can identify implicit abnormal scenarios (such as homophonic words and implicit abnormal sentence patterns) that cannot be covered by a traditional rule library, thereby reducing manual intervention and significantly improving the comprehensiveness of quality inspection.

The present application relates to the technical field of natural language processing, and more particularly to a script visual asset structured generation method based on a large model, which comprises the following steps: obtaining a script original text and dividing it into scene text units; extracting deep semantic vectors by using a large-scale pre-training language model; mapping literary descriptions to a primary visual feature set based on a visual semantic feature library; encapsulating each component by using a structured mapping engine, and constructing a structured visual asset description model including global scene parameters, entity object parameters and dynamic interaction parameters; and detecting and outputting consistent structured data through a cross-scene logic checking mechanism. The present application realizes the automatic conversion of script text into parameterized data, and improves the semantic analysis depth and visual matching accuracy.

The application discloses a speech recognition method and related equipment, and relates to the technical field of data processing. The method comprises the following steps: in response to a speech recognition instruction, obtaining to-be-recognized speech and a preset hot word library; and using a preset ASR model to generate a speech recognition result based on the to-be-recognized speech and the preset hot word library. Before speech recognition, the application determines a topic distribution, a word distribution of a topic and a distinguishing degree corresponding to a word based on a preset LDA model and a preset IDF algorithm, and then generates high-quality hot words and the corresponding weight of each hot word. In the process of speech recognition, a more accurate speech recognition result is obtained based on the high-quality hot words.

This invention proposes an intelligent enterprise accounting voucher generation and compliance review system, comprising: a text processing module, including a target lexicon generation unit, which can extract several negative words, several modifier texts paired with each negative word, and several correction words from preprocessed enterprise accounting text data and generate a target lexicon; and a current text recognition module, which is used to calculate the part-of-speech relevance C of the current text based on the target lexicon, and to identify the current text as a negative word or a corrected word based on the part-of-speech relevance C and the context of the current text, so as to perform a correction operation based on the recognition result, thereby solving the problem that the existing system has insufficient ability to analyze the scope of negative words and the consistency of context before and after correction, which leads to the existing system easily misclassifying negative statements and ignoring the associated impact of correction operations, thus causing voucher generation errors or compliance review loopholes.

This invention discloses a method for risk identification of sensitive content on web pages, belonging to the technical field of network content security. This method establishes a massive sensitive word library to accurately identify sensitive words and privacy information on extracted pages, significantly improving performance to meet the needs of batch monitoring of web pages. By adjusting the scores of the sensitive word library, adding or deleting sensitive words, or fine-tuning the algorithm, the overall monitoring accuracy can be controlled. The method includes the following steps: establishing a sensitive word library; loading the sensitive word library and constructing the identification system context environment; reading the content of each valid page and formatting it before outputting the formatted page content; identifying the formatted page content against the sensitive word library to extract sensitive content metadata containing all data information containing sensitive words; and performing semantic analysis on the sensitive content metadata through unsupervised classification to obtain sensitive content results.

A method for censoring digital media can include obtaining a user input via a user interface in regard to a digital video, obtaining an audio file for the digital video, and sending the audio file and a word back to an artificial intelligence (AI) engine. In some embodiments, the method can further include converting the audio file to text and identifying occurrences of words from the word bank and timestamps corresponding to the occurrences via the AI engine. In at least one embodiment, the timestamps received from the AI engine can correspond to moments in the digital video where the words in the word bank are used. Additionally, the method can include manipulating a mute function of a player of the digital video based on the timestamps received from the AI engine

The application discloses a kind of field-oriented scientific and technological project duplication checking method and system, the method includes: respectively constructing project duplication checking contrast library, field dictionary and field stop word library;At least according to field dictionary and field stop word library, the inverse document frequency of each word in corpus is calculated;According to the inverse document frequency of each word in corpus, the keywords of materials in project duplication checking contrast library and declaration library are extracted;The repetition rate of the project to be checked and the project in duplication checking contrast library is calculated based on the extracted keywords, and the project with high similarity is determined according to the repetition rate.The field-oriented scientific and technological project duplication checking method and system of the application can help enterprise technology management personnel to compare the file to be checked with a large number of historical project information, find out the project with highly similar research content, can avoid the waste of enterprise technology resources caused by repeated project establishment, and ensure the fairness of project establishment.

The application discloses a personalized comment recommendation method based on implicit dimension mining, relates to the technical field of natural language processing and artificial intelligence, and comprises the following steps: screening network commodity comments and fine-tuning an M3E-base model to construct an M3E-base-TextDimension model; recognizing entities in the comments by using a large language model, and rewriting the comments to extract comment dimensions; combining user demands and the comment dimensions, screening a key dimension set by using the large language model to confirm demand dimensions; inputting the comment dimensions and the demand dimensions into the M3E-base-TextDimension model to generate comment Embedding and demand Embedding; and performing Top-N comment recommendation by calculating the cosine similarity of the comment Embedding and the demand Embedding. By constructing a self-defined stop word table and using multiple open source stop word libraries, the application realizes effective filtering of text data and improves the quality of subsequent text processing.

The application relates to a data processing system for updating a word library, which comprises a preset word library, a processor and a memory storing a computer program, and when the computer program is executed by the processor, the following steps are realized: a candidate text set is acquired, an intermediate text set is acquired, a specific word list corresponding to the candidate text set is acquired according to the candidate text set and the intermediate text set, a target text set corresponding to the specific word is acquired, a target label list is acquired, a candidate label set is acquired according to the target text set and the target label list, a key label set corresponding to the specific word list is acquired, each key label list in the key label set is subjected to a deduplication treatment, and a label corresponding to each specific word in the specific word list is acquired to update the preset word library. The application is not limited to a same text database in terms of a text source, can be compared with other text databases, the accuracy of the acquired specific word is improved, and the accuracy of the acquired updated word library is high.

The application discloses a domain text proofreading method and device, equipment and a storage medium, and relates to the technical field of text processing. The method comprises the following steps: in the case that a domain text to be proofread in a target domain is received, a domain dictionary corresponding to the target domain is called; the domain dictionary comprises a first dictionary in which exclusive expressions that need to be forcibly exempted in the target domain are defined, and a second dictionary in which disabled expressions that need to be forcibly corrected in the target domain are defined; the domain dictionary is implanted into a prompt word, and a large model is called to perform proofreading processing on the domain text, thereby generating a primary proofreading result; based on the domain text and the primary proofreading result, the large model is called to optimize the primary proofreading result, thereby obtaining a final proofreading result. Through model fine-tuning by using the domain dictionary, domain adaptation is realized in a low-cost manner. In combination with a proofreading mode in which primary proofreading and secondary verification optimization are combined, false detection and missed detection caused by adaptation deviation of the domain dictionary are corrected, and the proofreading performance on the domain text is comprehensively improved.

This application discloses a method and system for mnemonic word recognition and extraction based on a neural network model. It rapidly filters potential mnemonic word sequences through multi-stage text preprocessing, then performs initial screening based on length and dictionary completeness requirements to obtain a first set of filtered sequences. A dual-branch neural network, combined with an adversarial memory training system, simultaneously outputs standard types and true probabilities, achieving intelligent filtering. Furthermore, compared to traditional techniques that directly perform checksum algorithms and virtual wallet verification, this application avoids performing hash operations and blockchain network requests on a large number of invalid sequences, significantly improving processing efficiency. After intelligent filtering, a second filtering is performed using a corresponding checksum algorithm, and finally, availability is confirmed through virtual wallet address validity verification, ensuring both recall and accuracy.

This invention relates to the field of learning methods, specifically to an intelligent English word replacement method for mixed Chinese-English reading. The method includes the following steps: selecting imported articles, categorizing them according to different types, and choosing the desired article; selecting replacement options according to system prompts, choosing different word libraries, and then selecting the replacement ratio and order; replacing the article from pure Chinese to a mixed Chinese-English article, selecting comparative replacement translations; for the same word appearing multiple times in the same article, partially displaying the original Chinese text and partially replacing it with English. Through this mixed Chinese-English reading learning method, students learn English while reading Chinese, passively memorizing words, and can autonomously and selectively translate and replace languages ​​according to their preferences. Automatic replacement is also based on their language level and learning progress, avoiding comprehension difficulties while reading. This progressive learning approach helps increase learning interest and improve memory.

This application provides a method, system, apparatus, device, and medium for sensitive word detection. The method includes: acquiring a file to be detected, wherein the file includes multiple characters; sequentially inputting each character in the file into a sensitive word detection model; matching each character using the sensitive word detection model to obtain a sensitive word detection result for the file; wherein the sensitive word detection model includes at least one set of matching forms constructed from a sensitive word database. Some embodiments of this application can improve the speed and accuracy of sensitive word detection.