55 results about "Semantic code" patented technology

A feed-forward bit-rate control system that selects and formats DQUANT values in real time for optimal semantically-encoded image transmission by video-compression codecs complying with the ITU-T standards H.263 or H.263E. The system uses a prequantization process to reduce the computational load imposed by the quantizer optimization calculations required for feed-forward semantic coding bit-rate control. The H.263 CBPY and MCBPC coding tables are reordered so as to be indexed using an orthogonal flag-based index values and the RLA coding table is block-indexed using clipped amplitude values, for faster access into these tables. The system also uses a rate-assist process to assure that DQUANT values are provided in each edge macroblock that follows a transition to and from each semantically-defined region when the edge macroblocks have four motion vectors by reducing the number of motion vectors in the macroblock to one.

The invention discloses a multi-label text classification method based on statistics and a pre-training language model. The multi-label text classification method comprises the following steps: S1, preprocessing training corpora needing to be classified; S2, establishing a label acquisition model based on a statistical method and a language model; S3, processing the obtained label data; S4, establishing a multi-label classification model based on a pre-training language model, and performing model training by utilizing the obtained label data; S5, performing multi-label classification on the text data to be classified by using the trained multi-label text classification model. According to the method, a statistical method and a pre-trained language model label obtaining method are combined, the ALBERT language model is used for obtaining the semantic coding information of the text, a data set does not need to be manually labeled, and the label obtaining accuracy can be improved.

The invention belongs to the field of computer natural language processing, and relates to a method for carrying out high-quality syntactic analysis on human natural language sentences, in particular to a syntactic analysis method based on sliding semantic string matching. The method is characterized in that when a rule base is built, hierarchic flattening transformation is firstly carried out on a usual phrase syntax tree, then semantic code labeling is carried out on the chunking information of each layer, and therefore the chunking rules of context-sensitive grammar of N elements are extracted; in the syntactic analysis process, the optimal chunking rules are matched to carry out stacking chunking type analysis through a sliding semantic string matching model; errors in the lower layer are found and corrected in the higher layer through an error correction template, and heuristic backtracking in the stacking chunking type syntactic analysis is achieved; template information is directly added to a semantic template index, and a machine can learn new syntactic rules immediately. The method solves the problems that a PCFG type syntactic analysis level can hardly be further improved and the correct chucking rules are hard to choose in the stacking chunking type syntactic analysis, and improves the existing syntactic analysis level.

The invention discloses an answer acquisition method based on machine reading understanding for open domain questions and answers. A BERT-based semantic coding module and an information interaction attention network are adopted to deeply capture potential semantic representations of problems and documents, effectively extract and fuse information between the problems and the documents, and captureglobal features of the problems and the documents; and an answer acquisition module based on Pointer Networks is adopted, and the attention weight is used as a pointer, so that the start and stop positions of the predicted answer can be positioned more accurately. The invention provides a reading comprehension-based answer acquisition method for open domain questions and answers. Empirical evaluation is performed on a CMRC 2018 data set. Experimental results show that the method can reach the standard level of open domain questioning and answering tasks, and excellent performance is obtained.

The invention provides a multi-model malicious code detection system based on reliability probability interval. Each machine learning detection model corresponds to a distribution of the underlying data, and various threshold-based detection models can be integrated into the statistical platform, so that the distribution of the semantic code data is detected from the multi-angle view, and the model degradation problem caused by the concept drift is relieved. The detection system changes the prediction mode of 0 or 1 of the existing machine learning detection model, calculates the score based on the existing detection model, carries out statistical analysis, and establishes a isotonic regression function for the score distribution of the sample and the label of the sample. For an unknown sample, according to the score given by the existing detection model, the calculated isotonic regression function is input, the reliability probability interval of a certain label can be given, and theprobability interval can relieve the problem of over-fitting of the fixed threshold to the training data set, the adaptive ability of the detection model to the current dynamic data is improved, and the concept drift phenomenon is found in advance.

The present invention discloses a website performance optimization method and apparatus. The method comprises: obtaining an optimization request; extracting a dom structure of a page of a to-be-optimized website according to the optimization request; encapsulating the dom structure to obtain the dom structure component; and carrying out redundant code removal, upward compatible or backward compatible semantics code generation, and intra-line style removal operations on the dom structure in the dom structure component, and generating a processed dom script. According to the method and apparatus disclosed by the present invention, the technical problems of huge consumption of the time, the manpower and financial resources due to secondary development or re-escalation of some old projects of the website in the prior art are solved.

The invention discloses a high spectral image classification method based on morphological characteristics and dictionary learning. The method comprises the following steps of extracting morphological characteristics from high spectral images, a process of dictionary learning, coding the characteristics, and classifying the images. The method is applied to the field of high spectral image classification, the structural relation of space information in the high spectral images is taken into full consideration, high-level semantic mapping is constructed on the basis of the space relation information, high-level semantic codes which can maintain the structure information of the characteristic space effectively are obtained and used for a high spectral image classification task, and the problem that semantic gaps exist between the high level meanings and bottom characteristics of the high spectral images is overcome. The method has substantial effect in high spectral image classification, and has higher application values.

The invention provides a construction method of a medical intelligent question-answering system based on an attention mechanism. The method comprises the following steps: firstly, collecting medical question and answer data, carrying out vectorization representation by using a Word2vec technology, and for each question, selecting a sentence with the maximum semantic similarity from answer data asa standard answer to construct a medical knowledge question and answer training library; constructing a deep learning model of the medical intelligent question-answering system based on the attentionmechanism, wherein the deep learning model comprises semantic coding and splicing of questions and answer data by using bidirectional LSTM, and adding the attention mechanism; and inputting the spliced semantic code into a bidirectional LSTM for feature fusion, and finally outputting a standard answer by a softmax function.

The invention discloses an unreal information detection method based on a BERT model and an enhanced hybrid neural network. The method comprises the steps of preprocessing a to-be-detected text; performing convolution and pooling operation on the input matrix by using a CNN network, and splicing the input matrix into a feature sequence; taking the feature sequence as the input of a BiLSTM network,and comprehensively capturing the deep semantic features of the text from the front direction and the rear direction by using a forward LSTM unit and a backward LSTM unit respectively; generating a semantic code containing attention distribution by utilizing an attention layer, and optimizing a feature vector; and finishing classification detection of the feature vectors by utilizing a classifierof the output layer, and judging whether the feature vectors are the non-real information. According to the method, the CNN, the BiLSTM and the attention mechanism are combined, the detection precision of the unreal information is high, local phrase features and global context features of the text of the unreal information can be extracted, text keywords can also be extracted, and the unreasonable influence of irrelevant information on a detection result is reduced.

The embodiment of the invention provides an interface element processing method and device. The method comprises the steps of obtaining interface description data; wherein the interface description data comprises interface elements and element feature data corresponding to the interface elements; arranging the interface elements according to the element feature data, and obtaining structured semantic codes describing the logical relationship between the interface elements; obtaining a recommended code variable name of the interface element according to the element characteristic data; and naming interface elements in the structured semantic code by adopting the recommended code variable name. The structured semantic code obtained by the embodiment of the invention has a brother relationship, a parent-child relationship and other logical relationships, and interface elastic layout is realized. In addition, the recommended code variable names of the interface elements are obtained according to the element feature data and used for naming the interface elements of the structured semantic codes, and the layout habit of a programmer can be better met.

The embodiment of the invention provides a text semantic coding method. The method comprises the steps: carrying out semantic encoding on a text through a character encoder and a word encoder, splicing results of character encoding of the text, splicing results of word encoding of the text, and obtaining character semantic encoding and word semantic encoding containing contexts; importing the character semantic codes and the word semantic codes obtained after splicing into an interactive matching layer based on a multi-head attention mechanism; and determining the fused word meaning code obtained by the interaction matching layer as the word meaning code of the text. The embodiment of the invention further provides a text semantic coding system. The embodiment of the invention provides anencoder based on multilayer word fusion. According to the encoder, after semantic encoding is carried out on characters and words, interaction is carried out on the obtained character and word semantic codes, then the interacted character and word semantic codes are fused through a gating unit trained in a self-adaptive mode, and deeper character and word semantic information is contained and serves as final text semantic representation.

The invention discloses a word conversion method, which is used for improving the accuracy of automatic translation of natural language. The method comprises the following steps: a semantic item corresponding to a work is obtained according to a word which is expressed by a first language characters; a semantic code is obtained according to the semantic item corresponding to the word or a semantic item which is expressed by a second language characters is obtained; a word which is expressed by the second language characters is obtained according to the obtained semantic code or the semantic item expressed by the second language characters; and the words expressed by the second language characters are organized in a sentence according to a syntax rule of the second language. The invention also discloses a device and a system for realizing the above method.

The invention discloses a Chinese named entity extraction method based on a multi-annotation framework and fusion features. The Chinese named entity extraction method comprises the steps that: firstly, Chinese characters are coded based on a pre-training language model; then, word information and word segmentation mark information are introduced for each Chinese character through dictionary matching, and dictionary features are constructed; on the basis, according to the meaning of the Chinese character in the matched word, phonetic notation is carried out on the Chinese character by using Chinese pinyin software, and pinyin features are constructed; thirdly, the dictionary features and the pinyin features are fused into Chinese character codes on the basis of a click-ride attention mechanism, Chinese character semantic codes combining the dictionary features and the pinyin features are obtained, and the recognition capacity for the Chinese named entity boundary is improved; and finally, the advantages of sequence labeling and index labeling are combined, two labeling tasks are jointly learned through a multi-task learning model, and the accuracy of Chinese named entity extraction is improved.

A machine-translation apparatus using multi-level verbal-phrase patterns includes: a simple sentence generation unit for generating an input simple sentence; a basic verbal-phrase pattern-matching unit for trying a match of a semantic code of each case component of the input simple sentence with basic verbal-phrase patterns; a default verbal-phrase pattern matching unit for trying a match of a size and case prepositions of the input simple sentence with default verbal-phrase patterns having a verb identical to that of the input simple sentence; a default word-order matching unit for trying a match of a word-order of the input simple sentence with default word-order verbal-phrase patterns having a case component structure identical to that of the input simple sentence; and a default preposition matching unit for generating a target sentence of an input sentence with default preposition patterns having a context identical to that of the input simple sentence.

The invention discloses a method for recommending relevant search data based on a user operation behavior. The method includes: acquiring operation data corresponding to a user operation behavior according to search data; generating an association probability graph between the search data and the operation data according to the search data and the operation data; generating semantic codes of the search data and the operation data according to the association probability graph; and recommending search data associated with the search data according to the association probability graph and the semantic codes. The search data and the operation data have high semantic, and then the associated search data content close to the search data can be searched. The invention also provides a relevant search device based on the user operation behavior.

The invention provides a facial reconstruction method based on a non-supervision automatic encoder. The main content of the facial reconstruction method based on a non-supervision automatic encoder includes a semantic code vector, a decoder based on a parameter model and a loss layer. The process of the facial reconstruction method based on a non-supervision automatic encoder includes the steps: scene description is given in a semantic code vector mode; the parameter decoder generates a composite image corresponding the face, and a reverse image is formed through standard reverse propagation, and then end-to-end training without supervision is realized, and the parameter decoder includes an image forming model, a lighting model, image formation and reverse propagation; and a loss function is defined by three items, and the loss layer includes dense luminosity calibration, sparse landmark alignment, statistical regularization and reverse propagation. The facial reconstruction method based on a non-supervision automatic encoder can encode the details of the face, such as posture, shape, expression, skin color and scene lighting, is more exquisite, does not need supervision and allows end-to-end learning. Compared with a network of synthesizing face data training, the network can be preferably popularized to real data.

The invention relates to 'one code' Chinese and English stenographing technology which is approved and supported by technological innovation fund of Ministry of Science and Technology in 2013. A normal keyboard and a stenograph are combined, clicking and double-clicking are combined, Chinese stenographing and English stenographing are combined, shifting is avoided, pinyin with word blocks (phonetic configurational codes), pinyin with semantic codes (phonetic semantic codes), word blocks with strokes (cconfigurational codes) and five-key single-stoke (stoke codes) can be adopted for typing while double-clicking phonetic configurational stenographing and phonetic semantic stenographing and double-clicking English can be adopted for typing, and no-repeated-code typing of 6763 words in GB2312 and 7000 Chinese characters in the 'Contemporary Chinese Language Universal Character List' can be realized. The method is simple to learn and even pupils can learn soon, operations are convenient, eight keys at most are doubly clicked with both hands, and typing speed is extremely high. The keyboard and software can be used together or independently and are generally applicable to devices such as PCs (personal computers), tablet computers, mobile phones and the like and universally suitable for everyone.

The invention provides a text rewriting method and device and electronic equipment, and relates to the technical field of computers. According to the text rewriting method, a bidirectional recurrent neural network is adopted to process a text, semantic recognition can be carried out on the text to be rewritten in combination with context content, and the accuracy of semantic understanding is improved; meanwhile, the semantic codes recognized by the bidirectional recurrent neural network are processed by adopting a gated recurrent unit network, so that the training efficiency can be improved while the rewriting performance of the bidirectional recurrent neural network is ensured; besides, according to the text rewriting method provided by the embodiment of the invention, the plurality of rewriting texts obtained after network processing of the gating loop unit are scored, and then the rewriting text meeting the requirements is selected from the plurality of rewriting texts as the targettext according to the scores, so that the diversity of text rewriting results can be improved.

The invention relates to a Chinese-Vietnamese bilingual multi-document news viewpoint sentence recognition method based on a sentence association graph, and belongs to the technical field of natural languages. Aiming at a Chinese-Vietnamese bilingual multi-document news viewpoint sentence recognition task, the invention provides a viewpoint sentence recognition model combining sentence associationfeatures and semantic features. The method comprises the following steps: constructing a Chinese-Vietnamese bilingual multi-document association undirected graph fusing event elements and emotion elements; obtaining sentence association features of the Chinese-Vietnamese bilingual; obtaining semantic code representation of the sentence; carrying out dimensionality reduction on the obtained semantic codes to obtain sentence semantic features of the Chinese-Vietnamese bilingual; and performing joint calculation by utilizing the sentence association features and the sentence semantic features toobtain viewpoint sentence recognition features, classifying the viewpoint sentence recognition features by adopting a classifier, optimizing the classifier by adopting a binary classification cross entropy loss function, and realizing viewpoint sentence recognition by adopting the optimized classifier. The Chinese-Vietnamese bilingual multi-document news viewpoint sentence recognition method caneffectively improve the accuracy of Chinese-Vietnamese bilingual multi-document news viewpoint sentence recognition.

The invention provides a character recognition method and device, and an electronic device. The method comprises the steps of extracting a feature map from a target character picture; performing sliding window processing on the feature map to obtain a spatial-temporal feature sequence; and inputting the spatial-temporal feature sequence into a pre-trained spatial-temporal attention model, so thatthe spatial-temporal attention model performs encoding processing on the spatial-temporal feature sequence to obtain a spatial-temporal semantic code, and decodes the spatial-temporal semantic code tooutput recognized text information in the target text picture. According to the technical scheme, the character information contained in the target character picture can be recognized based on the information of the two dimensions of the target character picture, and the recognition rate is effectively increased.

The embodiment of the invention discloses a music genre classification method and device and a storage medium. The scheme comprises: collecting positive and negative samples of preset genre music; extracting first lyric text data of music in the positive and negative samples; preprocessing the first lyric text data; extracting audio data of music in the positive and negative samples; preprocessingthe audio data; compressing audio data, and extracting a first audio feature in the compressed audio data; and extracting semantic codes of the first lyric text data according to the trained preset language model, combining the semantic codes with the first audio features, inputting the combined data into a music classification model generated based on the language model for training, and determining the music genre of the to-be-classified music according to the trained music classification model. According to the scheme provided by the embodiment of the invention, the genre of the to-be-classified music is determined through the trained classification model, and the accuracy of music genre classification is improved.

The invention relates to a Chinese character input method and a terminal. Firstly Chinese character codes corresponding to Chinese characters are pre-stored, wherein the Chinese character codes consist of Chinese pinyin and auxiliary semantic codes of the Chinese characters, and the auxiliary semantic codes are partial letters of foreign language words corresponding to the Chinese characters. Themethod comprises the steps of S1: receiving an input signal and inputting the Chinese pinyin; S2: displaying the Chinese character codes corresponding to the Chinese pinyin in a to-be-selected text region and the Chinese characters corresponding to the Chinese character codes; and S3: receiving a selection signal and inputting the selected Chinese characters to a text input region. By implementingthe method, the Chinese pinyin and the corresponding foreign language words are organically combined, so that the corresponding Chinese character quantity in a pinyin input process is greatly reduced; and the foreign language and the Chinese language are combined, so that foreigners can better learn the Chinese language.

The invention discloses a data privacy protection method based on dynamic context, which is characterized in that the method comprises the following steps: (10) semantic coding: storing and coding thesource data object by adopting an XML document semantic coding scheme; the user's personal bipartite privacy graph and dynamic knowledge context are provided according to the encoding scheme. (20) user login: obtaining the visual range of the user and obtaining the dynamic knowledge context; 30) access operation, wherein that query is taken as a basic operation, the path of the visible range is locate, and the access operation result is determined accord to the matching judgment of the bipartite graph of privacy; (40) dynamic knowledge context updating: updating the dynamic knowledge contextaccording to the superposition of user query operations and the accumulation of acquired prior knowledge. The data privacy protection method of the invention has the advantages of high information security, high data utilization efficiency and good real-time resistance to inference attack.

The invention relates to a dependency syntax-based similarity calculation model and system and a system building method, and the method comprises the steps of coding question semantics which comprisesshortest dependency path syntax coding, syntax tree-based expression and pre-trained word vector semantic coding; performing semantic coding on candidate query graphs corresponding to the questions so that answer query graphs are coded by pre-training word vectors to obtain query graph semantic codes; splicing pre-trained word vector semantic codes, shortest dependency path syntax codes and expression based on a syntax tree to obtain a question dependency syntax; and performing a mutual attention mechanism on the query graph semantic code and the question dependency syntax, performing information interaction, and performing similarity calculation to obtain semantic similarity. The performance of the system can be improved.

The invention provides a machine translation method, a training method, a corresponding device and electronic equipment. The machine translation method comprises the steps of obtaining a semantic codecorresponding to a source language input text; and based on the one or more target languages, respectively decoding the semantic codes to obtain output texts of the one or more target languages. According to the technical scheme provided by the invention, the semantic code corresponding to the source language input text only needs to be obtained once; therefore, the semantic codes can be repeatedly utilized, the output texts of different target languages are obtained through decoding based on the corresponding target languages, multiple times of calculation in the coding process are avoided,the overall calculation amount can be greatly reduced, and the processing efficiency of multi-language translation is effectively improved.

The embodiment of the invention provides an information transmission method in a semantic communication system. The method comprises the following steps that: a sending end firstly performs semantic information extraction to generate a semantic representation sequence, adds information of a semantic knowledge base into the generated semantic representation sequence, and then performs semantic coding to generate a semantic coding sequence; channel coding is performed on the semantic coding sequence to generate a wireless signal, and the wireless signal is transmitted to a receiving end through a wireless channel; the receiving end carries out channel decoding and semantic decoding on the received wireless signal; the receiving end feeds back to the sending end according to whether the channel decoding and the semantic decoding are correct or not; and the sending end decides whether retransmission is needed or not and the retransmission mode according to the feedback result until the receiving end obtains a correct semantic decoding result. According to the technical scheme of the embodiment of the invention, the efficiency of data transmission and feedback retransmission can be improved, and the waste of computing resources and transmission resources is avoided.

The invention provides a deep active learning text classification method based on a pre-training model. The method comprises the steps of: combining the pre-training model trained on a large number ofgeneral texts and utilizing the pre-training model to obtain semantic codes of the texts as input features of a classifier; then constructing a classifier to start training on an initial training set, based on the initial model and the to-be-marked sample selection strategy and the data supplement strategy provided by the invention, carrying out iteration continuously under participation of manual marking until the maximum number of iterations is reached or the marking budget is exhausted, and subsequently embedding the obtained model into a specific product to obtain a final product; and performing inference procedures. By means of the mode, high recognition accuracy can be obtained with low data labeling cost under the condition that diversity of training samples is guaranteed.

The invention discloses a paragraph semantics-based middle and primary school test question segmentation and extraction method and system. The method comprises the following steps: reading charactersof each paragraph in a test paper document, and analyzing structural semantics of each paragraph by adopting a structural semantic regular expression matching rule; recording the paragraph structure of the whole test paper by adopting paragraph structure semantic coding specifications to form a paragraph structure semantic string; performing structure correction and structure division on the paragraph structure semantic string; and according to the record of the structure division, extracting a fixed format from the test paper document to form a formatted document of a single test question. According to the invention, a large number of test paper documents commonly seen in middle and primary schools are classified and concluded according to structural modes; according to the test paper document structure analysis method, a plurality of most common structure models are abstracted, and the document structure analysis method with high adaptability is designed, so that structured splittingand test question information segmentation and extraction of the test paper document are realized, and the test paper document structure analysis method has high expandability and wide test paper model applicability.

A machine-translation apparatus using multi-level verbal-phrase patterns includes: a simple sentence generation unit for generating an input simple sentence; a basic verbal-phrase pattern-matching unit for trying a match of a semantic code of each case component of the input simple sentence with basic verbal-phrase patterns; a default verbal-phrase pattern matching unit for trying a match of a size and case prepositions of the input simple sentence with default verbal-phrase patterns having a verb identical to that of the input simple sentence; a default word-order matching unit for trying a match of a word-order of the input simple sentence with default word-order verbal-phrase patterns having a case component structure identical to that of the input simple sentence; and a default preposition matching unit for generating a target sentence of an input sentence with default preposition patterns having a context identical to that of the input simple sentence.