238 results about "Word order" patented technology

The invention discloses a short text clustering method based on deep semantic feature learning. The method includes the steps that dimensionality reduction representation is performed on original features under the restraint of local information preservation through traditional feature dimensionality reduction, binarization is performed on an obtained low-dimension actual value vector, and error back propagation is performed with the binarized vector being supervisory information of a convolutional neural network structure to train a model; non-supervision training is performed on a term vector through an outer large-scale corpus, vectorization representation is performed on all words in text according to the word order, and the vectorized words serve as implicit semantic features of initial input feature learning text of the convolutional neural network structure; after deep semantic feature representation is obtained, a traditional K-means algorithm is adopted for performing clustering on the text. By means of the method, extra natural language processing and other specialized knowledge are not needed, design is easy, deep semantic features can be learnt, besides, the learnt semantic features have unbiasedness, and good clustering performance can be achieved more effectively.

A method for searching patent documents by applying degree of similarity and a system thereof are disclosed. The method for searching patent documents by applying degree of similarity comprises receiving at least one search keyword from a user of the service; searching a document previously stored in a database, by the search keyword; and evaluating a degree of similarity to the search keyword on the document that is searched by the search keyword, wherein the degree of similarity is evaluated by measuring at least one degree among a degree of appearance frequency of the search keyword in the document, a degree of proximity between the search keywords, and a degree of word order between the search keywords. Therefore, patent documents may be searched by arranging the documents according to a degree of similarity to a search keyword, and not according to whether the keyword is included in the patent documents.

The invention relates to the technical field of log audit and safety management, and aims at providing a word segmentation algorithm-based log parsing method and a word segmentation algorithm-based log parsing system. The word segmentation algorithm-based log parsing method comprises the following steps: performing segmentation on a log, performing word sense analysis on segmentation results, performing word sense filtration on obtained segmentation results with word sense tagging, performing feature extraction on the obtained filtered segmentation results with the word sense tagging, performing feature matching on obtained word sense order feature codes, and performing semantic parsing on obtained semantic parsing rules; the word segmentation algorithm-based log parsing system comprises a segmentation module, a word sense analysis module, a word sense filtration module, a word order feature extraction module, a feature matching module and a semantic parsing module. According to the word segmentation algorithm-based log parsing method and the word segmentation algorithm-based log parsing system disclosed by the invention, the difficulty and complexity of log parsing are greatly reduced, and therefore the efficiency of performing parsing rule development on the log is increased; the word segmentation algorithm-based log parsing method and the word segmentation algorithm-based log parsing system can be better adapted to certain changes of a log format.

A system and method is provided for generating a translingual parsing model and for using the model to machine translate a source language to a destination language. The system and method includes receiving a sentence in the source language, and with the translingual parsing model, searching among and statistically ranking candidate parses each having elements labeled with destination language words, syntactic labels, and role labels indicating relationships between the elements. A statistically high ranked parse is selected and rearranged using the syntactic and role labels, in accordance with word order conventions of the destination language to generate a translingual parse of the source language sentence.

The invention discloses a real-time feedback method and a system of long voice continuous recognition and a recognition result. The method comprises the following steps: extension possibility from a sentence end point to an identification start point in an identification network is increased; in the process of decoding, when an identification path extends, the identification path can not only generate a single sentence, but also generate a plurality of continuous sentences, and large space video signals are separated through overall optimization of acoustics and linguistic probability; common parts of optimal history path of all active nodes are detected at fixed period, and identification word order which is fixed at the present moment is obtained; updated local identification result is fed back to a user in real time; and decoding space which is corresponding to identification parts and ensured is recycled. When the real-time feedback method and the system of the long voice continuous recognition and the recognition result are used, endpoint detection algorithm is not needed, continuous and sustainable identification of long voice signals are achieved, and human-computer interaction experience is obtained for users.

The invention relates to a deep learning multi-classification emotion analysis model combined with an attention mechanism, belongs to the technical field of natural language processing. The invention analyzes weakness of an existing CNN network and an existing LSTM network in the aspect of text emotion analysis, and provides the deep learning multi-classification emotion analysis model combined with the attention mechanism. According to the model, local features extracted by a CNN network and word order features extracted by an LSTM model are fused by using the attention mechanism, and emotion features extracted by the CNN network and the LSTM network are spliced on a classification layer by using an idea of an integrated model to serve as emotion features finally extracted by the model. Through a contrast experiment, the accuracy of the model is obviously improved.

The invention discloses a construction method of hybrid neural network model for dialogue generation. The construction method of hybrid neural network model for dialogue generation includes the steps: acquiring a data set in a mode of dialogue statement pairs, and constructing a glossary; generating a word embedded table; initializing the convolution neural network with special structure, generating a vocabulary recommending table corresponding to the input statement, determining whether real output is provided, and if so, training the parameters of the convolution neural network in the step; initializing the recurrent neural network with special structure, using the last step to output, generating a vocabulary identity list with word order, determining whether real output is provided, and if so, training the parameters of the recurrent neural network in the step; after the training result satisfies the set index, saving the glossary and the word embedded table, and saving the parameters of the convolution neural network and the recurrent neural network, thus completing construction of the whole model. The construction method of hybrid neural network model for dialogue generation solves the problems that a current neural network dialogue model is slow in the training speed, low in the accuracy and general in statement generation because the glossary is too long.

The invention discloses a word2vec improvement method of related factor training combining parts of speech and word orders. The method provides a model of Structured word2vec on POS. The model includes a CWindow-POS (CWP) model and a Structured Skip gram-POS (SSGP) model. According to each of the two models, part-of-speech label information and the word orders are used as influencing factors for combined optimization, and part-of-speech correlation information is utilized for modeling inherent syntactic relationships between words in context windows; and weighted calculation is carried out on context word sequences through part-of-speech correlation weights, then vector inner-product calculation is carried out according to the word position orders, and a stochastic gradient descent (SGD) algorithm is used for learning the related weights and word embedding in a combined manner. The method directionally embeds words according to the position orders thereof, realizes combined optimization for word vectors and weighted matrices related to the parts of speech, and has high efficiency in all of word analogy tasks, word similarity tasks and qualitative analysis.

An integrated retrieval method for multi-language information retrieval relates to multi-language information retrieval method, solving the problems of source language information loss caused by the multi-language information retrieval of the existing separation mode, a lot of noise and low accuracy of retrieval result, specifically comprising the following steps: step one, translating the source language inquiring key word input by the user into the key word of target language; step two, dividing the key word of target language into three relation modes according to the word order of each word, the decoration and collocation relation of each word, and word distance of each word that are precision matching mode, common display mode and independent mode; step three, obtaining the condition probability of precision matching mode, condition probability of common display mode and condition probability of independent mode in the inquiring file D; step four, calculating the file generating inquiring probability in the inquiring file D; step five, calculating the similarity of source language inquiring key word and inquiring file character vector; step six, calculating the condition probability of multi-language information retrieval; step seven, returning the retrieval result. The method is suitable for cross language information retrieval.

The invention provides a fault-tolerant text query method and equipment. The method comprises the following steps: calculating the text similarity between a first index word submitted by a user and each entry in a lexicon; detecting a result entry of which the text similarity is more than or equal to a first threshold value; and according to the magnitude of the text similarity, arranging the result entry; and providing the arranged result entry to the user for selection. The invention has the advantages that when a user carries out text query and the index word input by the user has wrongly written characters, or redundant words or lacks of one or more words, or the word order in the index word is reverse or above comprehensive faults exist, the invention can furthest find the target entry of the user and has high fault tolerance.

The invention discloses a spoken language evaluation method based on grammatical analysis and a spoken language evaluation system. The spoken language evaluation method is characterized in that syntax content of a voice segment can be extracted, and a syntactic tree can be established according to the syntax, and then the node similarity between the syntactic tree and the preset syntactic tree can be calculated, and the word order evaluation result of the voice segment can be acquired; the predicate verb of the voice segment can be acquired, and the predicate verb in the preset predicate verb corpus can be identified to acquire the tense evaluation result of the voice segment; the grammatical evaluation result of the voice segment can be acquired according to the word order evaluation result of the voice segment and the tense evaluation result of the voice segment. According to the invention, the spoken language evaluation can be carried out from three aspects, namely fluency, accuracy, and grammatical evaluation, and then by combining the spoken language voice quality evaluation index and the spoken language content evaluation index, the evaluation model is more objective, comprehensive, and accurate.

The invention relates to a drug entity relationship extraction method and system based on an attention mechanism neural network. The method comprises the following steps: (1) analyzing the text content of a pharmaceutical document, using sentences as basic units for sentence segmentation, and performing vectorization representation on each word in the sentences; (2) inputting a vectorized representation result into a recurrent neural network, extracting association characteristics of words in the sentences according to a front-back bidirectional word sequence through the recurrent neural network, and identifying all medicine entities; (3) obtaining inter-word importance weights in the sentences through an attention mechanism neural network, and combining the inter-word importance weights with the output in the step (2); And (4) inputting a result obtained in the step (3) into a convolutional neural network, and predicting a category relation between every two medicated entity words through the convolutional neural network. According to the classification method for increasing the attention mechanism concerned entity class information weight, the influence caused by wrong dependencyanalysis results in long sentences can be reduced, and the accuracy of extracting the pharmacochemical entity relationship is improved.

The invention discloses a text document representation method and a device based on depth learning topic information enhancement. The method comprises the following steps: S1, data preprocessing operation is carried out on the corpus document in the form of text. S2, a text sequence layer is designed, and the context information of each word in the word order is embedded into the representation vector of each word in the document. S3, the sequence elements are transitioned to higher-level topic information through the attention layer. S4, in the topic layer, a representation of the current document D in all topic directions is generated. S5, the similarity between all the topic information is limited. S6, the topic representation vector is fused into the semantic representation vector Repof the document D at the presentation layer. 7, that parameters of the Rep are updated by a classify and an objective function, the method can efficiently embed the context semantic information and the potential topic information of a text sequence into a document representation vector, and the presentation vectors enhanced by the topic information can significantly improve the performance of a text mining model use the Rep.

The invention provides a method for calculating short text semantic similarity. The calculating method comprises the steps of segmenting to-be-calculated short text; extending the segmented words through a continuous bag of words; eliminating ambiguities of the extended words through machine translation; calculating importance of the words having undergone ambiguity elimination, and weighting a word order of the words; calculating a semantic distance of the words having undergone word order weighting, and calculating text similarity according to the semantic distance. By implementation of the method, the similarity of the short text can be calculated quickly and accurately.

The invention discloses an intelligent question answering method and system based on a pet knowledge graph, and the method comprises the steps: constructing a named entity dictionary, abstracting questions, and facilitating the classification of the questions. A method of combining word2vec with Levenshtein Distance is provided to realize entity linking, and experiments show that the method is effective. Texts are trained by constructing a text classifier based on Naive Bayes, and the improved TF-IDF naive Bayes classification algorithm is provided, the distribution situation of feature wordsin a text set and the category distribution situation are considered, and the improved TF-IDF effectively improves the text classification effect. Through the result of the text classifier, the intention of the natural language question is determined, and the natural language question is matched with the corresponding word sequence graph. The word order graph is converted into a similar SQL querystatement of the OrientDB, and querying is performed in a graph database storing the knowledge graph. Finally, the constructed intelligent question and answer system based on the knowledge graph is displayed in an example, and experiments show that the system has a relatively high application value in question and answer application in the field of pets.

The invention discloses an electric power customer service work order sentiment quantitative analysis method based on Word2Vec, and relates to an electric power customer service work order analysis method. A traditional sentiment analysis method cannot effectively discriminate the sentiment intensity. The method of the invention comprises the steps of combining the power customer service work order text features; classifying and sorting the historical electric power customer service work orders and the unsatisfied work orders, cleaning data, combing based on the Baidu word bank to form an initialized multivariate emotion word bank; carrying out the work order text word segmentation by adopting a reverse maximum matching algorithm; based on the Word2Vec neural network, constructing the positive words, negative words, degree adverbs and a word vector of a word order fused with customer appeal semantics; performing the machine learning training through the historical customer service workorder to generate a learning model fusing appeal emotion, expanding a part-of-speech corpus based on the part-of-speech affinity-consanguinity relationship in the model, performing emotion quantization calculation by adopting a similarity word sequence matrix quantization algorithm, and completing customer service work order emotion quantization analysis, thereby effectively distinguishing emotion intensity differences, and determining an emergency degree.

The embodiment of the invention provides a short text template mining method and device, electronic equipment and a readable storage medium. Through the method, problems concerning synonyms and a wordorder in a template can be effectively solved, and an accurate short text template easy to use can be generated. The method comprises the steps that keywords are extracted from a problem text to forma segmented word sequence; the keywords are clustered according to meanings to obtain keyword clusters; the keywords in the segmented word sequence are replaced with the keyword clusters containing the keywords to obtain a word cluster sequence; and the word cluster sequence in an optimal arrangement mode is selected to serve as the short text template.

The invention discloses a text classification method for customer service online quality inspection. The text classification method comprises the following steps: step 1, converting voice into text; step 2, annotating data step 3, preprocessing data; step 4, selecting a feature coding mode (including one-hot (word2vec) + position coding, n-gram); and step 5, selecting a classification model (startspace, CNN). The invention aims to solve the problem of how to consider an entity sequence relationship when online customer service quality inspection is carried out by adopting a classification method at present, and hope to fully consider the sequence relationship under the condition of limited data volume, so as to improve the effect. The method has the advantages that the word order characteristics are introduced, the data volume requirement is not increased, and the method is simple and easy to implement.

The invention discloses a machine translation optimization method capable of exploring more reference translation version information. The machine translation optimization method mainly includes the steps of acquiring word alignment information from a source end to a target end by GIZA++, dividing reference translation versions into phrase blocks according to the word alignment information, constructing sub-graphs for every translation version according to a source-end word order, merging multiple sub-graphs, representing multiple juxtaposed reference translation versions into a reference translation version graph, linking the different reference translation versions to acquire more information, linking a reference translation version to be evaluated with the reference translation version graph through source languages, and selecting a proximate route of the translation version to be evaluated to finish final translation version quality evaluation. The machine translation optimization method evaluates the machine translation versions more sufficiently by extending the reference translation version information through graphs, and can better aid the system in parameter learning when participating in training.

The embodiments of the invention provide an address resolution method and device, and relates to the technical field of information processing. The method comprises the steps of segmenting an address to be matched to generate a first segmentation result, and segmenting each standard library address to generate a second segmentation result; calculating the semantic score of third segmentation in the first segmentation result to generate a first semantic vector, and calculating the semantic score of the third segmentation in the second segmentation result to generate a second semantic vector; generating a first word order vector according to the third segmentation and the first segmentation result, generating a second word order vector according to the third segmentation and the second segmentation result, calculating a semantic similarity according to the first semantic vector and the second semantic vector, and calculating a word order similarity according to the first word order vector and the second word order vector; and selecting a standard library address matched with the address to be matched from the plurality of standard library addresses according to the semantic similarity and the word order similarity. The address resolution method and device are simple, fast and high in working efficiency.