35 results about "Word group" patented technology

The application relates to the computer technical field, and provides a keyword matching method, an electronic device cluster and a program product, and further provides a computer readable storage medium. The keyword matching method provided by the application is applied to an electronic device, and the method comprises the following steps: extracting all continuous words satisfying the length of index word groups from a to-be-matched text, and generating one or more to-be-retrieved word groups; performing secret processing on the one or more to-be-retrieved word groups, and generating one or more ciphertext retrieval word groups; matching the one or more ciphertext retrieval word groups with ciphertext index word groups; when there is a ciphertext retrieval word group matching a ciphertext index word group, obtaining one or more keyword lengths; extracting continuous words satisfying the one or more keyword lengths from the to-be-matched text, and generating one or more to-be-matched word groups; performing secret processing on the one or more to-be-matched word groups, and generating one or more ciphertext matching word groups; and matching the one or more ciphertext matching word groups with ciphertext keywords, and obtaining a matching result. According to the method of the first aspect, the data processing amount occupied by the matching operation can be effectively reduced, and the execution efficiency of the matching operation is improved.

The present disclosure provides an updating method and device of an intent template library, an electronic device and a storage medium, relating to the technical field of computers, in particular to the technical field of artificial intelligence such as deep learning, natural language understanding, voice interaction. The specific implementation scheme is: obtaining a target sentence with an empty intent recall result from a historical dialogue library; determining the word slot label corresponding to each word group in the target sentence; based on a preset word slot arrangement rule, sorting the word slot labels corresponding to the target sentence to obtain a semantic template corresponding to the target sentence; performing intent recognition on the semantic template or the target sentence to determine a target intent corresponding to the semantic template; and storing the semantic template and the target intent in the intent template library. Thus, the automatic updating of the intent template library is realized, which not only expands the size of the intent template library and reduces the cost of updating the intent template library, but also improves the accuracy and reliability of the intent template library and optimizes the performance of the machine in the human-computer interaction process.

The present application relates to the technical field of information security, in particular to a Chinese character encryption and decryption method and device based on nine-square grid pinyin; the present application uses nine-square grid pinyin input method to correspond any encryption symbol with nine-square grid; when plaintext is sent by spelling pinyin, the output is the position and order of each time pressing the corresponding nine-square grid; the encryption symbol on the corresponding nine-square grid is matched according to the position and order of the corresponding nine-square grid; the pinyin of the plaintext is separated according to single Chinese character or word by combining the separation bit, and the ciphertext is formed; when the private key is known by both the sending and receiving parties, the corresponding pinyin and sentence content are determined according to the encryption symbol matching the letter on the corresponding character bit, combining the separation bit, and combining the meaning of the word group or the meaning of the sentence, and the decryption is realized. The whole method is based on nine-square grid pinyin, which reduces the learning cost and difficulty, and also reduces the decryption difficulty; meanwhile, the position is used for positioning, which increases the encryption type and possibility, and also increases the decryption difficulty.

A memory device having a plurality of groups of words is provided. Each group of the plurality of groups of words are stored in a contiguous section of a memory and each group of the plurality of groups of words includes a plurality of words. The plurality of words each have a plurality of elements with a same set of index values for a first dimension of a tensor and a same set of index vales for a second dimension of a tensor. The plurality of words have a different index value for a third dimension of the tensor.

The present application belongs to the technical field of data processing, and relates to a mental health detection method and device based on large model fine-grained enhancement and a medium. A large language model is used to convert a mental health scale into an extended questionnaire and generate a discrimination standard based on mental health diagnosis rules. For each group of situation, clue, thinking cognition and behavior data in the mental knowledge data set, the situation and clue are combined and mapped into a key vector, and the thinking cognition and behavior are combined and mapped into a value vector. An individualized question set is obtained based on the interview records of the extended questionnaire, the question and answer results of the individualized question set are combined with preset prompt words and input into the large language model to extract mental health representation evidence. The mental health representation evidence is mapped into a feature vector and the similarity with each key vector is calculated. The value vector corresponding to the K key vectors with the highest similarity is combined with the mental health representation evidence, and then the combined result is input into the large language model together with the discrimination standard to output the mental health score of the user to be evaluated and the determination reason.

The embodiment of the application provides a text error correction method and device, computer equipment and a storage medium, and belongs to the technical field of natural language processing. The method comprises the following steps: obtaining an original word group; performing spelling evaluation on the original word group to obtain spelling evaluation information; wherein the spelling evaluation information represents that the original word group is spelled correctly or incorrectly; if the spelling evaluation information represents that the original word group is spelled incorrectly, extracting a selected word group pair matched with the original word group from a preset word group library according to the original word group; wherein the selected word group pair comprises at least one selected homophone error word and at least one selected reference word group of the selected homophone error word; and correcting the original word group according to the selected reference word group. The embodiment of the application can improve the accuracy and efficiency of text error correction.

The application relates to a retrieval enhanced generation technology, aiming at the problem that current mainstream RAG methods are difficult to efficiently realize keyword mining and inter-word relationship in the financial field to improve the performance of the RAG method, and a financial field RAG keyword retrieval method based on inter-word relationship mining is provided, which comprises the following steps: collecting and counting a large number of vertical field documents and common questions, calculating the term (A+B) point mutual information of two adjacent words termA and termB, and storing all possible term combinations by using a prefix tree dictionary; simultaneously calling two fine-tuning BERT models, a word importance model and an abstract model, to respectively execute the functions of keyword discovery, i.e. word importance, and inter-word relationship mining, i.e. word closeness, construct a word segmenter enhanced for financial vocabulary, and obtain candidate combined keywords; calculating the comprehensive score of the multi-word combination, verifying the feasibility of the triple combination and the binary combination, and integrating the keywords with high importance and the mined binary or triple word groups or short sentences. The application significantly enhances the new information recognition capability in the financial field.

The application discloses a mixed expert model word grouping scheduling method and electronic equipment, and relates to the technical field of artificial intelligence, which comprises the following steps: in response to the arrival of word units in the form of a stream to a preset buffer, obtaining the embedding vector features, routing history and arrival time corresponding to each word unit to splice into the target feature vector corresponding to each word unit, and the preset buffer is configured with a preset waiting time threshold and a capacity upper limit value; clustering and grouping each word unit according to the target feature vector to generate at least one task group; determining the priority of each task group based on the average waiting time of each word unit in each task group and the historical hot index; under the condition of meeting the limitation of the preset buffer, scheduling each task group according to the priority to enable the mixed expert model to perform operation on each word unit. The method of the application improves the utilization rate of experts and the system throughput while ensuring the upper bound of delay through short window buffering and dynamic clustering.

The present application relates to a kind of fusion context-aware multi-bit watermark embedding and extraction method.Watermark embedding method includes: candidate word set generation and embedding timing determination based on Top-p strategy;Model confidence filtering based on Top-1 Gap;Pseudo-random word group based on watermark message and seed;Group refinement;Semantic consistency filtering and determinate sampleable set;Probability renormalization and sampling output.Watermark extraction method includes: input preprocessing and sequence initialization;Word element conditional judgment and bit decoding;Periodic sampling based on fixed interval.The embodiment systematically solves the triangular dilemma of "capacity, quality, robustness" difficult to consider in the prior art, realizes embedding traceable multi-bit information under the premise of almost not reducing text generation quality, and can effectively resist semantic maintenance type modification attack.

This invention discloses a data governance platform for automatic identification of sensitive data, relating to the field of sensitive data identification and governance technology. It solves the problem that due to differences in organizational structure, business systems, and data platforms, many data organizations operate independently, resulting in data non-sharing, duplicate data, disconnected information, and uneven data distribution, making sensitive data management inconvenient. The data governance platform for automatic identification of sensitive data includes a monitoring and display module, a transmission gateway, a sensitive data storage unit, a data receiving unit, a sensitive data comparison unit, and an operation unit. The monitoring and display module includes a display, a signal converter, and a signal receiver. This invention uses pre-defined sensitive word groups for preliminary analysis, identification, and classification of data. After identifying the sensitive word groups, the data is further segmented and compared, reducing the computational workload of data identification and improving the efficiency of sensitive data identification and governance.

The application relates to the technical field of natural language processing, in particular to a text training method of a forward prediction algorithm, which comprises the following steps: obtaining an original text data set, performing serialization annotation on each text sequence, and generating a standard prediction label; traversing each text sequence, extracting character-level features, word group-level features and semantic-level features at each sliding window position, and forming three-layer gradient feature vectors; inputting the three-layer gradient feature vectors into a forward prediction network to generate a prediction value of the center position of the sliding window; based on the prediction value and the standard prediction label, a multi-dimensional loss function containing a character-level loss term, a word group-level loss term and a semantic-level loss term is constructed, the three loss terms are weighted and fused to obtain a total loss value; and network parameters are updated according to the total loss value, and a forward prediction model is output. Through hierarchical feature extraction and multi-dimensional loss function weighted fusion, efficient training of a text prediction model with character-level precision and semantic understanding is realized.

The present disclosure provides a data processing method, which can be applied to the technical field of big data and the technical field of finance. The method comprises: in response to a user query request, obtaining a reference field by analyzing the user query request, the reference field comprising at least a target query period field and a target institution field to which the user belongs; based on the target query period field and the target institution field, reading target historical query log data from a database, the target historical query log data being generated before the target query period and being associated with the target institution; associating the target historical query log data with a plurality of predefined query semantic word groups to obtain a target query semantic word group; and associating the target query semantic word group with data labels of a plurality of labeled data to determine a plurality of target data from the plurality of labeled data. The present disclosure also provides a data processing device, equipment, storage medium and program product.

The application discloses an AI information pushing method and system based on personalized home-based aged care services, and relates to the technical field of personalized pushing. The method comprises the following steps: acquiring behavior data of historical published text information; determining positive sample text information from the historical published text information based on the behavior data; determining a preference value of the user for various text information based on the behavior data of the positive sample text information, determining target text information of interest based on the preference value, and selecting text information to be pushed accordingly; extracting high-frequency word groups of the positive sample text information and the pushed text information and performing similarity calculation to obtain a text similarity matrix; determining the weight of each similarity value in the text similarity matrix according to the word frequency, and determining the text similarity based on the weight; calculating the recommendation value of the text information to be pushed according to the text similarity, and selecting the text information to be pushed to push to the user. The application can accurately push text information of interest to the user.

The application provides a multi-modal large model task-driven target detection method and system based on meta learning, relates to the technical field of target detection, and comprises the following steps: obtaining multi-modal input data and performing preprocessing to obtain unified semantic representation; a general candidate generation network and a large language model are used to generate a candidate region set and a visual attribute word group set; region feature extraction is performed on each candidate target frame in the candidate region set to obtain a candidate visual embedding sequence; the visual attribute word group set is subjected to vectorization processing to obtain an attribute text embedding sequence; the candidate visual embedding sequence and the attribute text embedding sequence are jointly calibrated to obtain a calibrated candidate visual embedding sequence and a calibrated attribute text embedding sequence; a task-related score is calculated through a trainable scoring function, and all candidate frames in the candidate region set are screened to obtain a target detection result. The application realizes target detection guided by task description and has the ability of cross-task rapid adaptation.

Embodiments of the present specification provide a sample processing method and device, wherein the sample processing method comprises: obtaining a to-be-processed sentence; constructing a word group corresponding to the to-be-processed sentence, and determining a feature label corresponding to each word unit in the word group; constructing at least one negative sample and a target positive sample according to the feature label and the word group; calculating a sample similarity between the target positive sample and each negative sample, and determining a target negative sample in the at least one negative sample according to the sample similarity.

This invention discloses an environmental perception intelligent assistance method, device, and medium based on speech recognition, relating to the field of intelligent speech interaction technology. The method includes: acquiring continuous speech signals and obtaining speech segments through framing, spectral denoising, and silence detection; dividing syllable boundaries; extracting syllable sequences based on the syllable boundaries and converting them into pinyin; generating a candidate path set by combining a vocabulary dictionary; reading the candidate path set; matching the word texts in the candidate paths with the syllable sequences to obtain candidate word segments; grouping and filtering the candidate word segments to generate stable segments; sorting the stable segments in chronological order to obtain a stable segment set; determining a state word group splicing threshold based on the stable segment set; splicing and summarizing the stable segments to obtain a state word group set; determining a state sequence based on the continuity relationship between state word groups in the current speech segment and the previous speech segment; and sorting the state sequence in chronological order to obtain a state sequence set. This invention improves the accuracy of semantic information extraction.

Embodiments of the present application disclose a survey questionnaire pushing method and device, electronic equipment and storage medium. The method comprises: obtaining data information of a to-be-pushed survey questionnaire; preprocessing the data information to obtain at least one spliced word group; the preprocessing comprises synonym replacement and splicing processing; inputting the at least one spliced word group into a target survey questionnaire pushing model to obtain a target object corresponding to the to-be-pushed survey questionnaire, and pushing the to-be-pushed survey questionnaire to the target object. The technical scheme of the embodiments of the present application carries out word segmentation on a text question and an option corresponding to the text question, extracts key information of the survey questionnaire and discards word groups without actual meaning, splices at least one spliced word group, greatly simplifies the content of the survey questionnaire, obtains a target object corresponding to the to-be-pushed survey questionnaire according to the spliced word group, and improves the efficiency of survey questionnaire analysis and the accuracy of survey questionnaire pushing.

The application provides a visual question and answer model training method and a visual question and answer method. The visual question and answer method comprises the following steps: obtaining an image to be recognized and a corresponding question to be answered, performing text recognition on the image to be recognized, adding a plurality of word groups recognized to a general answer set of a visual question and answer model, obtaining a target answer set, respectively performing feature extraction on the image to be recognized and the question to be answered to obtain image features and first text features, inputting the image features and the first text features into the visual question and answer model to obtain a first answer. By adding the plurality of word groups in the image to be recognized to the general answer set, when the visual question and answer model is used to obtain an answer corresponding to the question to be answered, not only the answers in the general answer set are considered, but also the influence of the word groups in the image to be recognized on the question to be answered is considered, so that the accuracy of the output answer is effectively improved.

The application relates to a data encryption method and device, computer equipment and a storage medium. The method comprises the following steps: padding the to-be-compressed data until the data length of the padded to-be-compressed data reaches a preset length to obtain first data; iteratively expanding the first data, and selecting a first number of word groups from the first data after each round of iteration is completed; after a target word group is selected from the word groups of the first data and compression is performed on the target word group, the following steps are repeatedly performed until the first number of word groups are all compressed: obtaining a compression result, selecting a new target word group from the word groups of the first data, and compressing the compression result and the new target word group; and when it is judged that the total number of the word groups selected after the iteration reaches a second number, the iteration of the first data is stopped. The method can solve the problem of a long data encryption period.

This invention discloses a method and apparatus for evaluating the dissociation of reasoning diversity and optimizing self-consistent voting in AI products. Belonging to the fields of artificial intelligence and natural language processing, it generates model reasoning trajectories through temperature sampling. It calculates surface diversity indicators (including lexical diversity and triplet overlap) and answer diversity indicators (including the number of unique answers and answer information entropy) from two dimensions: vocabulary and answers. The surface diversity indicators are then corrected for length bias. Based on the correlation coefficient between the corrected surface diversity and answer diversity sequences, the dissociation relationship between the two types of diversity is quantified. This invention verifies that long reasoning trajectories in knowledge distillation models only improve text surface diversity and cannot enhance real-world reasoning and exploration capabilities. It innovatively uses answer information entropy instead of traditional lexical diversity as a self-consistent reasoning quality evaluation indicator, completing model capability evaluation and adaptive optimization of decoding strategies. This provides a scientific and precise technical basis for reasoning model selection and self-consistent strategy optimization.

Embodiments of the present disclosure provide a data cleaning method and device, electronic equipment and storage medium, and relate to the technical field of computer. The method comprises: obtaining a plurality of first published contents published in a first historical period and a keyword library to be cleaned; for each keyword group, if at least one of the plurality of first published contents includes the keyword group, the keyword group is taken as a candidate keyword group, and at least one first published content including the candidate keyword group is taken as target published content; according to the content type of each target published content corresponding to each candidate keyword group, a suspected non-malicious keyword group is determined from each candidate keyword group; and at least part of the keyword groups in each determined non-malicious keyword group are deleted from the keyword library. The embodiments of the present disclosure improve the cleaning effect of the keyword library and improve the overall accuracy of the cleaned keyword library.

The application discloses a medical data retrieval method, system and storage medium, and relates to the technical field of data processing. The method comprises the following steps: acquiring an intention word group and an entity word group based on a user retrieval text instruction; acquiring an expansion word group from a term library based on the intention word group; acquiring a plurality of first medical data from a medical database based on the expansion word group; acquiring at least one second medical data from each first medical data based on the entity word group; and outputting each second medical data. The application avoids dilution of the retrieval intention of the core disease, guarantees the recall rate of the core medical data, reduces the data processing range through two-stage retrieval of 'rough recall first and precise screening later', effectively solves the problems of result redundancy and large proportion of irrelevant data caused by traditional full-quantity segmentation and full-database matching, improves the retrieval accuracy and execution efficiency, and makes the retrieval result highly consistent with the real needs of the user.

The application discloses a data intelligent retrieval and access method and system in the field of space science, and the method comprises the following steps: according to the text input by a user, adopting a word segmentation model and a word embedding model respectively, mapping the text to corresponding key words and embedding vectors, and forming double-path recall retrieval results of a data set based on key word matching and embedding vector similarity matching; based on a rearrangement model, reordering the two kinds of retrieval results of the data set to form a query result of the data set granularity; according to the text input by the user, a data sample and a selected data set, adopting a pre-constructed machine feature extraction model, extracting a vector representation of user retrieval information, and based on vector similarity matching, retrieving data files in the data set to form a query result of the data file granularity; based on a pre-established optimal affine transformation function and a metric space, quantitatively calculating the data similarity with respect to a retrieval sample in terms of distance to obtain a retrieval result.

The application relates to the technical field of legal consultation services, in particular to an intelligent question and answer matching method based on machine learning, which obtains a legal consultation text, identifies a subject behavior object and an article, analyzes and combs segmented sentences, generates a semantic layered table, extracts a keyword group, integrates a semantic unit, constructs a semantic network, generates a mapping table, compares question and answer semantic logic, generates a matching corresponding set, examines and revises a text to generate a revised set, reviews consistency, and generates an intelligent matching result set. The application optimizes question and answer matching precision by processing the logical relationship of the legal text through semantic layered analysis, enhances semantic consistency identification and matching flexibility by combining keyword extraction and semantic network construction, corrects semantic deviation and logical inconsistency, ensures result connection and coherence, reduces mismatching compared with the prior art, improves question and answer accuracy, makes legal consultation more intelligent and accurate, and provides reliable matching results.

The application discloses a speech synthesis method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of speech synthesis. The original text needing speech synthesis is determined, and a current multi-sound character in the original text is recognized. The target pronunciation of the current multi-sound character in a word group is queried from a pre-generated multi-sound character dictionary according to the word group to which the current multi-sound character belongs in the original text. A speech synthesis marked text corresponding to the target pronunciation is generated. The target text used for replacing the original text to perform speech synthesis is generated according to the speech synthesis marked text and the original text. The speech synthesis quality of the text containing the multi-sound character can be improved.

The application provides a Chinese learning content adaptive recommendation method and system, which comprises the following steps: subject body and secondary theme structured labeling are carried out on public Chinese corpus, and the corpus is divided into three language units of single character, word group and syntax structure; frequency distribution features and context co-occurrence semantic vectors are extracted, continuous difficulty values are calculated, and corresponding HSK levels are mapped; a multi-granularity corpus vector library is constructed; recommendation scores are calculated based on learner mastery and target difficulty, and adaptive learning content is output. The application improves the corpus difficulty grading accuracy and content recommendation matching degree, and is suitable for multiple Chinese learning scenarios.

The application provides a software architecture recovery method and device. By introducing a phrase-level architecture topic semantic, the significance of semantic distinction between files is improved. By constructing a semantic prototype and reasoning a cluster-level semantic representation, hierarchical improvement from a file-level semantic to a cluster-level semantic is realized. By identifying cross-language interaction relationships and introducing a cross-language topology completion mechanism, the structural vulnerability of a cross-language boundary in a graph model is compensated. Finally, structural dependence, semantic association and cross-language topology information are fused to improve the accuracy and structural consistency of cross-language software system architecture recovery.

The application discloses a scoring prediction method based on end-to-end text filtering, comprising: obtaining a to-be-predicted review text, determining a user and a commodity corresponding to the review text; obtaining at least one historical review text of the user and at least one historical review text of the commodity; obtaining a word group level convolution feature vector corresponding to the review text and a word group level convolution feature vector corresponding to the historical review text; obtaining a feature vector corresponding to the review text and a feature vector corresponding to the historical review text; determining the similarity between the review text and the historical review text according to the feature vectors; filtering the historical review text based on the similarity, and calculating a similarity text feature vector according to the filtered historical review text; fusing the feature vector of the review text and the similarity text feature vector to obtain a global text feature vector; and determining the predicted score of the review text according to the global text feature vector. The application can improve the scoring prediction performance and precision of the review text.

The application relates to a kind of method and system for controlling measuring instrument, it is related to artificial intelligence and instrument automation control technical field.The system includes agent module and reasoning module, agent module includes word segmentation module, mapping module and at least one function agent module;Word segmentation module obtains the control information input by user, extracts the feature word group of control information;Word segmentation module inputs each feature word to the preset feature-semantic model, determines the category and semantics of control information according to the preset feature-semantic model;Mapping module determines the function agent module corresponding to control information according to the category and semantics of control information;Function agent module obtains control information, carries out format conversion based on the category and semantics of control information, to generate first instruction code;Reasoning module is used to obtain first instruction code, carries out code conversion to first instruction code, to generate second instruction code, and second instruction code is sent to agent module, to execute control information, realizes the control to measuring instrument.