53 results about "Computational semantics" patented technology

The invention discloses a semantic query expansion method based on domain knowledge, which comprises the following steps: taking concept expression and a knowledge tree system as the basis to construct the domain knowledge; performing primary semantic analysis on query phases input by users to form a semantic item list; utilizing results of the primary semantic analysis and taking the domain knowledge as the basis to construct a semantic map with expansion types and expansion weights; respectively computing semantic distances between each vertex and an initial vertex in the semantic map; determining an expandable item of each item in the semantic item list according to the semantic distances; and finally, combining all expandable items according to AND / OR logic relations to obtain a semantic item set representing the query intension of the users, and submitting the semantic item set to a searching system for searching. In the semantic query expansion method based on the domain knowledge, the computing time is short, the domain knowledge is fully utilized, and newly-added expanded semantic items and the original query phases have definite semantic relations, and the recall ratio and the precision ratio of the searching system can be improved effectively.

The invention discloses a method for detecting code similarity based on semantic analysis of a program source code, which relates to computer program analyzing technology and a method for detecting complex codes of computer software. The method solves the prior problems of low similarity detection accuracy and high computing complexity on the codes of different syntactic representations and similar semantemes, and also solves the problem of incapability of realizing large-scale program code similarity detection. The method comprises the following steps: resolving two segments of source codes to be detected into two control dependence trees of a system dependence graph respectively and executing basic code standardization respectively; utilizing a measure method to extract candidate similar code control dependence trees of the control dependence trees which are subjected to the basic code standardization; executing an advanced code standardization operation on extracted candidate similar codes; and computing semantic similarity to obtain a similarity result so as to finish the code similarity detection. The method is applied to source code piracy detection, software component library query, software defect detection, program comprehension and the like.

Determining semantically equivalent text or questions using hybrid representations based on neural network learning. Weighted bag-of-words and convolutional neural networks (CNN) based distributed vector representations of questions or text may be generated to compute the semantic similarity between questions or text. Weighted bag-of-words and CNN based distributed vector representations may be jointly used to compute the semantic similarity. A pair-wise ranking loss function trains neural network. In one embodiment, the parameters of the system are trained by minimizing a pair-wise ranking loss function over a training set using stochastic gradient descent (SGD).

The invention relates to the technical field of movement behavioral analysis and prediction in a sense network environment, and specifically to a movement space-time trajectory analysis method in the sense network environment. The movement space-time trajectory analysis method in the sense network environment comprises data reception of receiving trajectory movement position data generated by a positioning device and resolving the data format into a data format applicable to data treatment; semantic treatment of performing clustering operation on the semantic trajectory data; space-time correlation of performing characteristic analysis and statistics on clustered semantic trajectory data in a time domain and a space domain, and performing time-space correlation analysis in combination with the time domain and the space domain; correlation similarity analysis of calculating space-time correlation similarity of the semantic trajectory and performing analysis and calculation on the correlation among different space domains and different movement objects; outputting a result. The movement space-time trajectory analysis method in the sense network environment solves the problem of continuous treatment and mutual correlation of time and space dimensions in a traditional transactional database, and meets the need from a sense network application service to real-time analysis of trajectory movement data.

The present invention discloses a statistics-based machine translation method and apparatus and an electronic device, a semantic similarity-degree calculation method and apparatus and an electronic device, and a word quantization method and apparatus and an electronic device. The statistics-based machine translation method comprises: according to a feature that affects a translation probability and that is of each candidate translation and a pre-generated translation probability prediction model generating a translation probability of a sentence to be translated into each candidate translation, wherein the feature that affects the translation probability at least comprises a semantic similarity-degree between the sentence to be translated and the candidate translation; and selecting a preset number of candidate translations whose translation probabilities rank top as a translation of the sentence to be translated. By adoption of the statistics-based machine translation method provided by the present application, the semantic level of the natural language can be reached deeply when the machine translation model is constructed, and the deviation of semantics between the translation and the source text is avoided, so as to achieve the effect of improving translation quality.

The invention relates to a quick multi-keyword semantic sorting search method for protecting data privacy in cloud computing. A domain weighted scoring concept is introduced in document scoring, and keywords in different domains such as a title, an abstract and the like are endowed with different weights to be distinguished; a retrieval keyword is subjected to semantic expansion, semantic similarity is calculated, a three-factor sorting method is designed by combining the semantic similarity, the domain weighted scoring and a correlation score, and a cloud server can perform accurate sorting on search results and return a sorting result to a search user; and for the defect of low query efficiency of a searchable encryption scheme, a vector block segmentation marking matching algorithm is designed, and a document vector created by a vector space model is subjected to block segmentation to generate a marking vector with a relatively small dimension number. According to the method, the query efficiency can be improved, the index creation time can be shortened, and semantic ciphertext keyword search is realized.

The invention discloses a generation method of single document summaries. The method includes the steps of clustering paragraphs of a document to be summarized, and defining each class as a semantic block; calculating similarity of each sentence pair in the semantic blocks to score one sentence with the other sentence, and defining the sentence with highest score as a sentence expressing partial core content in each semantic block; and connecting the sentences to generate a summary according to emergency sequence of core sentences. Word similarity and named entity recognition are introduced to single document summaries, so that extracting precision of summaries is higher. Clustering speed is increased by means of single pass. Single document summaries can be extracted accurately. The generation method is high in accuracy of extracting news and announcement documents.

A zero-sample classifying method based on class transfer comprises the steps of acquiring a vision characteristic of C kinds of training samples, a class semantic characteristic of the training sampleand a true label matrix; calculating a semantic similarity matrix by means of cosine similarity or Gaussian similarity through the class semantic characteristic; calculating a diagonal matrix of a class semantic similarity matrix; calling a Sylvester equation in an MATLAB toolset for obtaining a mapping matrix; inputting the vision characteristic of the training sample, the corresponding class semantic characteristic and the true label matrix into a target function, continuously adjusting the value of a model regularization parameter, calculating the least value of the target function, and finishing model training; and in a testing period, inputting the vision characteristic of the testing sample and the corresponding semantic characteristic, calculating scores of the classes, and determining the class with highest score as the predicated class of the testing sample. The zero-sample classifying method based on class transfer has advantages of sufficiently digging the semantic relationbetween different classes, realizing knowledge transfer between a known class classifier and an unknown class classifier, and realizing high convenience in application in image classification.

The invention provides a news sentence clustering method based on semantic similarity. The method includes the following steps: preprocessing news sentences of a corpus, and extracting available words; utilizing the available words to train a continuous bag-of-words model to obtain an initial word vector of each available word; utilizing an initial sentence vector of each news sentence and the initial word vectors of the left and right adjoining available words of a certain available word in the news sentence to train the continuous bag-of-words model in an iterative manner to obtain a currentword vector of each available word in the news sentence and a final sentence vector of the news sentence; merging an average value of the word vectors of all the available words, one-hot vectors of high-frequency words and the final sentence vector of each news sentence to obtain a semantic vector of the news sentence; and calculating distances between the semantic vectors to obtain the semanticsimilarity between the different news sentences, and clustering the news sentences of the corpus in accordance therewith. The invention also provides an electronic device and a computer-readable storage medium.

The invention discloses a relational query method implemented on a large-scale data set and belongs to the field of semantic webs. The method comprises the following steps: 1) calculating connective subgraphs (only containing a same type of tags) in a semantic data directed graph G; 2) merging the connective subgraphs, and dividing the directed graph G into a plurality of subgraphs; 3) calculating a strongest connective subgraph C in each subgraph subjected to merging, and calculating a bipartite graph of the strongest connective subgraph C; 4) storing the shortest path of all the subgraphs Cinto a path set RS; 5) recording the tags (containing two points of a non-redundant tag path) in each divided subgraph so as to obtain a tag set of each subgraph; and 6) judging whether paths conforming to the query conditions exist in the directed graph G by using the tag set, if so, returning path query results, otherwise, carrying out traversal among the subgraphs, determining subgraphs which can reach a target node according to the path set RS, and then, returning the path query results by using the tag set of the subgraph. The method disclosed by the invention is used for supporting the relational query of mass data, and is strong in expandability.

The invention relates to a semantic segmentation method and system based on edge dense reconstruction for streetscape understanding, and the method comprises the steps: carrying out the preprocessingof an input image of a training set, enabling the image to be standardized, and obtaining preprocessed images with the same size; extracting general features by using a convolutional network, then obtaining three-level context space pyramid fusion features, and extracting coding features by using the two parts of cascade connection as a coding network; acquiring semi-input size encoding features by using the encoding features, acquiring edge features based on a convolutional network, and reconstructing image resolution by taking a dense network fused with the edge features as a decoding network in combination with the semi-input size encoding features, and acquiring decoding features; calculating semantic segmentation loss and auxiliary supervision edge loss, and training the deep neural network by taking minimization of weighted sum loss of the semantic segmentation loss and the auxiliary supervision edge loss as a target; and performing semantic segmentation on the to-be-segmented image by using the deep neural network model, and outputting a segmentation result. The method and the system are beneficial to improving the accuracy and robustness of image semantic segmentation.

The invention discloses a text similarity measuring method based on semantic analysis and a semantic relation network. First, two texts are input and are preprocessed, and a preprocessed result is a collection of vocabularies; second, the lexical semantic similarity of the preprocessed result of the two texts is calculated, and the semantic relation network in constructed according to a calculation result; third, flow medium numerical values of nodes in the semantic relation network are respectively calculated, and characteristic collection of the two texts is obtained; fourth, a bipartite graph is constructed according to the characteristic collection of the two texts, and a path weight number between two parts of the bipartite graph is set; fifth, the similarity between the two texts is calculated by utilizing a bipartite graph optimal matching method. The text similarity measuring method based on the semantic analysis and the semantic relation network can be used for the technical field of data mining and information retrieval technology such as text clustering and information retrieval. Compared with other existing text similarity calculating methods, the accuracy degree of text similarity calculation is greatly improved.

A word sense disambiguation method based on a hidden Markov model includes step 1) of training corpus, using a SemEval 2007#task5 test corpus set to parse a to-be-disambiguated sentence; performing word segmentation on the sentence; step 2) of finding ambiguous words in the sentence after the word segmentation, extracting a target ambiguous word and segmented words on the left and right of the target ambiguous word; training corpora, and calculating semantic class vocabulary transition probability and semantic class transition probability; step 3) of extracting the number of sentences containing the ambiguous word from manually annotated corpora, calculating the observation probability, and calculating the observation probability of words on the left and right of the ambiguity word; step 4) of using the value trained by the previous corpus to calculate the state transition probability, the extracted initial state probability, the observation probability and the state transition probability as parameters of the hidden Markov model, and using the well constructed disambiguation model to disambiguate the sentence in the test corpus; and step 5) of using a similarity calculation methodto verify the accuracy of disambiguation results.

The invention discloses a search method for relevance feedback images based on an ant colony algorithm and a probability hypergraph, which comprises a training stage and a search stage. The training stage comprises the following steps of extracting low-level features of images, studying a dictionary and performing the high level representation of the images in a library image. The search stage comprises the following steps of extracting low-level features of sample images; performing the high level representation of the sample images; constructing an affinity matrix; initializing or updating a pheromone matrix; labeling positive correlation images and negative collation images for search results of all sample images in the image library; calculating a semantic pheromone matrix; calculating an affinity enhancing probability; enhancing the affinity matrix by using the ant colony algorithm; constructing a hypergraph; and returning the search result and finishing search, or updating the pheromone matrix for the next search. According to the method, an efficient and accurate image search technology is provided for image search and a higher use value is realized.

The invention relates to a text sequence iterative method for semantic understanding. The text sequence iterative method for semantic understanding aims at solving the problems that in the prior art, a method for crawling searching of the engine retrieval number is adopted on the aspect of semantics similarity computation, the efficiency is low, recurrence cost is oversize, depending on an ontology base and a semantic resource, manual checking or purely manual construction is needed, the expandability aspect needs improving, and the practical application is limited to some extent. According to the technical scheme, the method comprises the steps that 1, an original text triad and a background knowledge base triad are extracted; 2, a real number vector of the original text triad and a real number vector of the background knowledge base triad are calculated; 3, the semantic similarity is calculated; 4, the weighted value of the real number vector of the background knowledge base triad is calculated; 5, a sequence iterative model is constructed; 6, text classification is conducted on a text through a support vector machine, and estimation is conducted on classification performance. The text sequence iterative method for semantic understanding is applied to computer fields.

A method and system for conducting semantically weighted searches in a governed corpus of terms is provided. A search expression having a plurality of terms for performing a search in the governed corpus of terms is received. The governed corpus of terms comprises a plurality of corpus expressions each comprising a plurality of terms, each term within the governed corpus of terms being associated precisely with a single concept within a lexical database. At least one concept of the lexical database is assigned to each term in the search expression based on a syntactical analysis. A semantic similarity is calculated between pairs of concepts of the search expression and one of the corpus expressions. A total semantic similarity is calculated between the search expression and the one of the corpus expressions by aggregating the semantic similarities of the pairs of concepts based on an order of significance of the terms.

The invention discloses a Spark-based multi-feature combined efficient Chinese text clustering method. the method comprises the following steps of: uploading mass data sets into an HDFS file system byutilizing high fault tolerance and high data access throughput of the HDFS file system, carrying out data preprocessing and submitting the data sets to a Spark cluster; and after text set preprocessing is completed, respectively calculating a semantic similarity and a word frequency statistics-based cosine similarity of a dimensionality-reduced text, combining the two similarities to obtain a final text similarity, and carrying out text clustering by utilizing the obtained text similarity and combining a maximum distance method. According to the method, semantic information and word frequencystatistics information are combined to ensure that the text similarity calculation is more correct and the number of iterations is greatly decreased at the same time.

The invention discloses a pedestrian detection method. The method comprises the following steps: extracting a shot pedestrian image to obtain a visible part boundary block diagram; inputting the pedestrian image into a detector, inputting the output of the Nth detection layer in the detector and the visible part boundary block diagram into a semantic segmentation model, and outputting a semantic attention diagram and semantic segmentation loss by the semantic segmentation model; inputting the semantic attention map into the (N + 1) th detection layer of the detector, and outputting a prediction boundary frame diagram and detection loss by the detector; Calculating the sum of semantic segmentation loss and detection loss to obtain total loss; if the total loss is greater than a preset lossthreshold, adjusting internal adjustable parameters of the detector and the semantic segmentation model according to a preset rule, and repeating the detection operation; otherwise, outputting and displaying the prediction boundary block diagram. According to the method, missing detection or false detection caused by inter-class shielding is reduced through the semantic segmentation model, the pedestrian detection accuracy is ensured, the number of training objects is small, and the cost is low. The invention further discloses a device and equipment based on the method and a computer readablestorage medium.

The invention relates to an unstructured data management method for the field of film and television materials. The method is used for carrying out distributed storage and query and visual managementon unstructured data in the field of film and television materials. The distributed storage specifically comprises the steps of: establishing knowledge ontology of the field of film and television materials, wherein the knowledge ontology comprises a domain knowledge base KS and a tag classification library LS; and determining a storage path of the unstructured data, storing the data in an HDFS (Hadoop Distributed File System) according to the storage path and storing description information of the unstructured data of the film and television materials in a database. The query and visual management specifically comprises the steps of: expanding the search condition of a user by calculating the semantic similarity and correlation degree; searching according to the expanded search conditionby utilizing an ElasticSearch search engine in combination with a Jena inference algorithm, obtaining a search result and feeding back to the user; and displaying the searched data by utilizing a visual tool. Compared with the prior art, the method disclosed by the invention has the advantages that the method is suitable for the unstructured data management; the search condition can be expanded; the query is flexible; the visual displaying is realized and the like.

The invention relates to the field of text modeling, and provides a patent text modeling method based on word2vec and semantic similarity. The method includes: crawling the patent text set and performing preprocessing; calculating the TF-IDF value of each word in the patent text set sorting and selecting to obtain feature word set; The text set is imported into word2vec model and the word vector is obtained by training. Cosine similarity is calculated to get the similar word set wordC_1; Word2vec similarity was calculated to get the similar word set textC_1. The similarity of word set textC_1and textC_1 was calculated to get the similar word set textC_1. The text set is imported into the text processing system for training, and the semantic similarity is obtained. The similar word set wordC_2 is selected. The semantic similarity is calculated to get the similar word set textC_2. The semantic similarity is calculated to get the similar word set textC_2. The extended word set textC_f isobtained by calculating the mixed similarity. Weights are calculated to form new text identifiers to complete the modeling. The invention adds part of the information between words to the traditionalvector space model from the statistical angle of the word2vec and the semantic angle of the semantic similarity, reduces the sparsity of the text matrix to a certain extent, and the clustering effectis more remarkable and stable, and has stronger text identification ability.

In order to realize the automatic acquisition of domain concepts, the invention provides a domain ontology concept automatic-acquisition method based on a Bootstrapping technology. The method comprises the following steps of: first, in order to solve the problem that ontology concept in a composite form cannot be extracted, extracting compound words based on mutual information and left and right information entropies; then extracting candidate domain concepts based on the decision condition of the candidate concepts with co-occurring sentence frequency; evaluating the candidate concepts by applying a method for combining M evaluation with T evaluation, extracting the domain concept conforming to evaluation criterions, and choosing the domain concept with a higher evaluating value as a key concept for a new round of learning process; in order to avoid missing the domain concepts with low frequency of occurrence and similar meanings, introducing a semantic factor, and extracting the domain concepts with similar meanings by calculating semantic similarity; and finally, providing a detailed algorithm realization flow.

The invention discloses an image-text conversion method and device, an intelligent interaction method, device and system, a client, a server, a machine and a medium. The image-text conversion method comprises the following steps: constructing a sample data set; training a deep learning model and a language characterization model by using the sample data set; when the user input is an image, extracting the visual features of the input image through the deep learning model, calculating the similarity of the visual features, and selecting the text corresponding to the image with the highest similarity as the corresponding output of the input image; and when the user input is the text, extracting semantic features through a language representation model, calculating semantic feature similarity, and selecting an image corresponding to the text with the highest similarity as corresponding output of the input text. By constructing the special data set and adopting various conversion models, richer and more efficient interaction functions are provided for users, and real deep interaction is realized.

The invention provides a knowledge graph-based semantic similarity calculation method for financial industry software testing. The method comprises the following steps: S1, carrying out word segmentation operation on a financial text; s2, selecting a word segmentation combination most related to the text subject; and S3, calculating the semantic similarity of the segmented word combination by using the knowledge graph and using the concept IC weighted minimum path length. The natural semantic detection algorithm based on the knowledge graph comprises the following steps: firstly, carrying outword segmentation on a financial text by utilizing multiple word segmentation algorithms to obtain a word segmentation combination; calculating the concept distance between the word and the text keyword to measure the similarity between the word segmentation combination and the text topic; and finally, selecting the word segmentation combination with the minimum concept distance sum to carry out semantic similarity detection. The information IC of the concept is used in the knowledge graph to weigh the shortest path length between the concepts, and better performance is shown in accuracy compared with other methods.

The invention relates to a sentence similarity assessment method based on a deep semantic model and semantic role labeling. The method comprises the steps that a text character string is mapped to a feature vector in the low semantic space, and the cosine similarity is used for measuring the similarity between two sentences; an existing semantic role is reserved, and other semantic roles are processed in a unified mode; according to the similarity between predicates, predicate pairing is performed on sentence pairs to obtain predicate matching pairs, and a similar calculated value between semantic roles is further obtained; the multiple semantic roles of each of the multiple predicates of each sentence are subjected to semantic collocation, the semantic role similarity is calculated, the similarity calculated through the deep semantic model and the similarity calculated on the basis of the semantic roles are linearly combined to be adopted as the final sentence similarity. On the basisof the semantic roles, the Pearson's correlation coefficient is increased by 2.226%, and is 0.226% higher than that of the top one result on the SemEval2017 evaluation official website.

The invention discloses an SRC (Semantic Relational Closeness) calculation method, which comprises the following steps of: 1, carrying out word segmentation processing on a file, filtering word sets obtained after the processing, and filtering away non-nouns and common use nouns without actual meanings; 2, carrying out weighted processing on the filtered word sets according to positioning formation in a text, and calculating co-occurrence frequency weight values of the filtered word sets; and 3, calculating the SRC according to the average weight value between two words. The SRC calculation method has the advantages that the distance between the two words and the context relationship between the two words are sufficiently considered, and the word relevancy judging accuracy is improved.

The invention discloses an improved computing method for resolving ontology concept semantic similarity based on semantic distance. The method comprises the following steps: computing the semantic distance d1 between the ontology concepts (g1, g2) through an initialized ontology concept module and computing a distance factor d2 based on the semantic distance so as to construct the semantic similarity sim1(g1, g2); constructing the semantic similarity sim2(g1, g2) according to the corresponding depth and density of two ontology concept nodes; and finally obtaining the concept semantic similarity sim2(g1, g2). Compared with the traditional method for resolving the semantic similarity based on the information theory method and the semantic distance method, the accuracy of the semantic similarity resolved by use of the method disclosed by the invention is higher; and the semantic distance computation has a certain theory preciseness; the obtained result is more accurate by integrating the information theory and the semantic distance method; the multi-inheritance problem of the node in the ontology tree is solved; this method for computing the semantic similarity is more close to the experience point of an expert on the quantization concept; the ontology inference effect is better improved, and more extensive application research value is realized.