42 results about "Context window" patented technology

A computer-implemented system and method is disclosed for retrieving documents using context-dependant probabilistic modeling of words and documents. The present invention uses multiple overlapping vectors to represent each document. Each vector is centered on each of the words in the document, and consists of the local environment, i.e., the words that occur close to this word. The vectors are used to build probability models that are used for predictions. In one aspect of the invention a method of context-dependant probabilistic modeling of documents is provided wherein the text of one or more documents are input into the system, each document including human readable words. Context windows are then created around each word in each document. A statistical evaluation of the characteristics of each window is then generated, where the results of the statistical evaluation are not a function of the order of the appearance of words within each window. The statistical evaluation includes the counting of the occurrences of particular words and particular documents and the tabulation of the totals of the counts. The results of the statistical evaluation for each window are then combined. These results are then used for retrieving a document, for extracting features from a document, or for finding a word within a document based on its resulting statistics.

The invention is suitable for the technical field of computers, and provides a word vector training method and system. The method includes: performing word vector training on each training target word in a training sample document, and meanwhile acquiring a window word of each training target word in a context window in the training sample document; predicting the occurrence probability of each window word by using a Skip-gram model; updating a word vector corresponding to the each window word in a word vector library and an intermediate vector corresponding to each non-leaf node in a code path corresponding to each training target word in a Huffman tree; updating whole document vectors of the training sample document through a preset formula; calculating increasing local input vectors of a CBOW model, and then calculating mixed stitching vectors of the CBOW model; setting the mixed stitching vectors as input of a projection layer of the CBOW model; predicting the occurrence probability of each training target word; and finally updating the word vector of each training target word and the intermediate vector corresponding to each non-leaf node in the Huffman tree. The method and the system improve the accuracy of the word vector of each training target word.

The invention provides a method for identifying an airport target in a remote sensing image by fusing scene information and depth features. The method comprises the steps of generating target candidate boxes of the airport on the image in a sliding window manner according to a plurality of preset sizes; constructing a deep convolutional neural network feature extractor, adding corresponding internal window and context window for each target candidate box for realizing learning and extraction of features, internal features and context features of regional images of the candidate boxes, and performing combination to obtain a fused description feature; performing type determination of the target candidate boxes based on a support vector machine (SVM) to obtain type attributes of the target candidate boxes and probabilities of belonging to corresponding types; and performing locating precision processing of the target candidate boxes to obtain an identification result of the airport target in the remote sensing image. By applying the method, the position and size of the airport can be quickly and accurately identified in the high-resolution remote sensing image; and the method is suitable for research on identification of the airport in the remote sensing image in various illumination conditions and various complex backgrounds.

Provided are an apparatus and method for recognizing speech using an attention-based content-dependent (CD) acoustic model. The apparatus includes a predictive deep neural network (DNN) configured to receive input data from an input layer and output predictive values to a buffer of a first output layer, and a context DNN configured to receive a context window from the first output layer and output a final result value.

A computer-implemented system and method for extracting Human Generated Lists from an electronic database is described. The system searches for objects of the same class within a context window to identify Human Generated Lists and stores them to an archive. The archive may be used to generate a relationship network. The system generates variable length data vectors to represent the relationships between the objects within each Human Generated List. This relationship network can then be queried to discover relationships between the objects in the Human Generated Lists and to provide related objects as recommendations.

The invention discloses a graph node multi-tag classification method based on depth learning. The method comprises the steps that a graph data module is loaded, and graph data are analyzed and stored in the form of a dictionary; a walk path module is generated to complete random walk in the graph data, and a generated walk path is returned; a node eigenvector module is generated, and the walk path returned by the previous step, the specified vector representing dimension and the context window size are used as input to call a word2vec algorithm to calculate the eigenvector representation of each graph node; a training data module is generated, and nodes of a certain percentage are randomly selected from all graph nodes as training node data; for each node, the eigenvector and the tag sequence corresponding to the node are taken to form a two-tuple set as a training sample; and finally, a depth confidence network model is built. The graph node multi-tag classification algorithm proposed by the invention has higher correct rate than a traditional multi-tag classification algorithm.

The invention discloses a word multi-prototype vector representation and word sense disambiguation method based on CRP clustering, which comprises the following steps: the text in the massive text corpus is purified and pretreated to obtain plain text, CRP algorithm is used to cluster the context window representation of target polysemous word in the text corpus set. The target polysemous words inthe text corpus set are marked according to the clustering classification, and the polysemous words are trained on the marked text corpus set to obtain the multi-prototype vector representation of the polysemous words; 2, the target short text is preprocessed to obtain a short text word sequence, a target polysemous word in a word sequence is identifued, the contextual window of the target polysemous words is used to represent the similarity between the centroids of clusters corresponding to the words in the text corpus, and the word vector corresponding to the maximum similarity clusters isused as the word vector representation of the specific meaning of the polysemous words in the context to disambiguate the meanings of the polysemous words. The invention solves the problem of polysemyexpression in word expression and the problem of ambiguity identification in word meaning expression.

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.

The grouping of windows based on user-defined contexts is disclosed. A computer-implemented method of grouping windows based on user-defined contexts in accordance with an embodiment includes: assigning a user-defined context to a plurality of elements in an environment; grouping the plurality of elements into at least one group based on the user-defined context assigned to each of the plurality of elements; and displaying a context window tab corresponding to each group of elements.

The invention relates to the technical field of speech processing and specifically provides a method and a device for estimating a noise power spectral density of a speech signal. The method comprisesthe following steps: extracting a time context window feature from a noise speech signal, inputting the time context window feature to a pre-trained speech existence probability estimator which outputs an estimated speech existence probability corresponding to a current time frame; correcting the estimated speech existence probability according to a Bayes rule to determine a speech existence probability; determining the noise power spectral density corresponding to the corresponding time frame based on the speech existence probability according to a recursive smoothing formula. Through the technical scheme of the invention, the estimation accuracy of the noise power spectral density is improved in the case of small computing resources, thereby being beneficial to effectively eliminating noise signals, minimizing distortion in the speech processing process and improving speech enhancement performance.

Selections from a long list of elements on a Graphical User Interface (GUI) are presented. When a context window is presented for the selections, different cues are provided in the context window to provide feedback regarding the selections that are not in the current view, and to also facilitate rapid navigation to those selections on the GUI. Interaction with the GUI widgets within the context window, and interaction with the context window itself, facilitates the navigation toward the other selections that are not in the current view. Interaction with the GUI widgets also facilitates the preview of the selections that are not in the current view, thus making it easier to determine what has been selected without having to navigate to the view that contains those selections.

The invention provides a mobile browsing method. The method comprises the following steps: clicking the 'link key' in the 'link window' of a mobile phone browser to perform 'down key' operation; clicking 'content key' to execute the functional operation corresponding to the default link in 'link window' and switch to the 'content window'; clicking 'link key' in the 'content window' of the mobile phone browser to switch to the 'link window' and automatically perform once 'down key' operation; and clicking 'content key' to perform the operation of scrolling down the screen or execute 'PageDown' link operation.

The invention provides an emotion tendency analysis method based on environmental element embedding and deep learning, which comprises the following steps: S1, collecting text data for training, and obtaining a word segmentation text; S2, training word vectors of the word segmentation text through Word2vec and Gloss, and then obtaining environment element embedding to serve as word vector representation of text semantics in a mode of expanding word vector characteristics of the word segmentation text; S3, automatically learning contexts to extract emotion comment objects by utilizing a neuralnetwork with BLSTM and a dynamic context window fused; S4, based on a local attention mechanism, training word vectors of the text semantics through BLSTM to obtain sentence-level feature vectors; S5, training sentence-level feature vectors through the convolutional neural network to obtain global text-level feature vectors; and S6, classifying the global text-level feature vector by using the multi-classification function Softmax to obtain the emotional tendency of the text data. According to the method, the text data sentiment tendency judgment accuracy is improved.

A computer-implemented system and method for extracting Human Generated Lists from an electronic database is described. The system searches for objects of the same class within a context window to identify Human Generated Lists and stores them to an archive. The archive may be used to generate a relationship network. The system generates variable length data vectors to represent the relationships between the objects within each Human Generated List. This relationship network can then be queried to discover relationships between the objects in the Human Generated Lists and to provide related objects as recommendations.

The invention discloses a biomedical named entity recognition method and system based on general language characteristics. The method comprises the following steps of performing the named entity labeling on a biomedical text; extracting general language characteristics of the biomedical text; selecting the size of a context window and constructing a feature template by using general language features; inputting the labeled corpus and the formatting feature template into a conditional random field for model training to obtain a conditional random field model for biomedical field named entity recognition, performing biomedical named entity recognition on the biomedical text to be recognized by using the model, and finally outputting a recognition result; and an online biomedical named entityrecognition system is built by using a B / S architecture. According to the biomedical named entity recognition method and system based on the general language characteristics, the named entity recognition effect in the biomedical field is improved to a certain extent, and the universality and the use convenience of the system are improved.

The invention relates to a method and a device for determining the weights of key words of multimedia resources. The method comprises the following steps: determining a multimedia resource list corresponding to a target user according to historical data of multimedia resources watched by the target user; for each multimedia resource in the multimedia resource list, determining the weights of all the key words according to the occurring times of all the key words of the multimedia resources in context windows corresponding to the multimedia resources, wherein the context windows corresponding to the multimedia resources comprise a plurality of multimedia resources adjacent to the multimedia resources in the multimedia resource list. The method and the device have the advantages that the accuracy and the efficiency for determining the weights of the key words of the multimedia resources can be improved, and the consumption for computing resources can be reduced, so that the intelligent recommendation and searching effect of the multimedia resources can be improved and the user experience can be improved.

The invention relates to a microblog hot topic discovery algorithm based on BTM and GloVe similarity linear fusion, which is characterized by comprising three stages of data acquisition and preprocessing, modeling and clustering, and comprises the following steps of: performing data acquisition and preprocessing, modeling obtained data, and clustering the modeled data; the invention provides a microblog hot topic discovery algorithm based on BTM and GloVe similarity linear fusion in order to solve the problem that a distance function of a K-means algorithm can influence a microblog hot topic clustering result. The GloVe model only trains non-zero elements in the word and word co-occurrence matrix rather than the whole sparse matrix to utilize statistical information, and the sparsity problem faced by the TF-IDF algorithm in the document-word vector matrix construction process is effectively relieved. The GloVe model is combined with a global matrix decomposition method and a local context window method at the same time, the trained word vector can carry more semantic information, and the problem that one word has multiple meanings and cannot be well solved by a BTM topic model canbe relieved to a certain extent.

A word semantic similarity solution method based on a context window comprises the steps of inputting words to be compared in a statistical method module; determining a context range of the words to be compared; finding out two sentences with maximum weight in the range; calculating similarity between the two sentences; and finally, solving the similarity between the words to be compared according to the similarity of the sentences. By the word semantic similarity solution method, very valuable quantitative description is provided for determination of an effective range of the context, and the defect of previous subjective description is overcome; the position of description capability of the context to the key word is gradually reduced from the near to the distant, and the word semantic similarity solution method conforms to ordinary knowledge of people; the linearity and the signal-to-noise ratio of a weight contribution value are better, and simple subsequent calculation is facilitated; the normalization curve accuracy of the weight contribution value is higher; the influence of a sentence constituent relation in a left window and a right window of a key word on defining of an effective window in the context is considered; and the solution of word semantic similarity by applying a context window technology is achieved, and calculation precision and accuracy are higher.

The invention provides a word vector generation method and device, a computer storage medium and electronic equipment. The method comprises the steps of: obtaining an associated text composed of associated text units of a context window of a target text unit in each statement, mining a text set composed of all the associated texts by utilizing a document topic generation model to obtain topic distribution vectors corresponding to the target text units of all the statements; and configuring the same semantic tags for the target text units with relatively high topic distribution vector similarity, configuring different semantic tags for the target text units with relatively low topic distribution vector similarity, and training a word vector model by using the training corpus set added withthe semantic tags to obtain a word vector of each semantic tag, wherein the topic distribution vector corresponding to the target text unit can reflect the context of the target text unit, the semantics of the target text unit in different statements can be distinguished according to the similarity of the topic distribution vector, and a word vector model is trained based on the distinguishing result to obtain word vectors of different semantics.

The invention discloses a condition random field based telecom field named entity recognition method. The condition random field based telecom field named entity recognition method includes the steps:converting corpus into an input format of a condition random field CRF model, and using a word based marking model to mark the corpus; selecting the size of a context window and selecting features from a candidate feature set to construct a feature template; defining the feature template of the condition random field CRF model, inputting the acquired corpus and the acquired feature model into thecondition random field CRF model, and acquiring a telecom field named entity recognition CRF model, using the telecom field named entity recognition CRF model to perform telecom field named entity recognition on a telecom text to be recognized, and acquiring an output shaft; and restoring the recognized telecom field named entity from the acquired output result. The method can extract the telecomfield named entity through an automatic method, can improve the efficiency of telecom field named entity recognition to a certain extent, and can ensure the good accuracy and the good recall rate ofthe telecom field named entity recognition result.

An information system using active context technology allows a user to view and interact with a large amount of data disposed within multiple objects with relative ease, minimal time, and minimal interaction. An active context object allows an active context window to display labels, identifiers, and links for objects that are actively (e.g. presently) contextually associated with each other. Contextual association of objects can be determined by a user, inferred by the nature of each object, or automatically determined by an information system. In another embodiment, an active context window is persistently displayed to a user to allow the user to easily obtain information about contextually associated objects and easily allow the user to navigate among contextually associated objects. In one embodiment, the active context object can be transmitted to other users to allow each user to access contextually related objects and, thus, contextually related information.

The invention discloses a spatial semantic similarity calculation method based on sliding window sampling. The method comprises the following steps: preprocessing corpus data containing spatial information, performing projection processing on coordinates in the preprocessed corpus data by adopting a preset equal-area projection method to obtain an actual space range, determining a context window,performing sliding sampling, and finally performing similarity calculation on every two words in a word set of the whole corpus. According to the method, a model capable of measuring the spatial semantic similarity of the words is constructed by mining the spatial semantic similarity relationship between the words. The method is superior to a traditional text similarity model and a geographic space similarity model in the aspect of comprehensively considering spatial correlation and text correlation. As a new angle of understanding human natural language by integrating human spatial thinking and spatial perception, a traditional natural semantic similarity model is effectively supplemented, and the accuracy of an intelligent geographic information retrieval and recommendation system is effectively improved.

The invention provides a text recognition method and device based on a Chinese word vector model and a storage medium. The method comprises the following steps: constructing a first Chinese word vector model and a second Chinese word vector model based on a target word, a context window word and a target word tone n-gram; training the first Chinese word vector model and the second Chinese word vector model based on a corpus to obtain the trained first Chinese word vector model and second Chinese word vector model; and obtaining a semantic vector of an input text by using the trained first Chinese word vector model and outputting a result. According to the invention, the trained second Chinese word vector model is used to verify the identification result; the recognition accuracy is improved, the tone element is used when the word vector is constructed, the information contained in the Chinese word vector is enriched, the influence of feature noise on the word vector is avoided, the semantic and intonation features of the target word are reserved to the maximum extent, a negative sample is introduced, and a loss function is constructed.

Selections from a long list of elements on a Graphical User Interface (GUI) are presented. When a context window is presented for the selections, different cues are provided in the context window to provide feedback regarding the selections that are not in the current view, and to also facilitate rapid navigation to those selections on the GUI. Interaction with the GUI widgets within the context window, and interaction with the context window itself, facilitates the navigation toward the other selections that are not in the current view. Interaction with the GUI widgets also facilitates the preview of the selections that are not in the current view, thus making it easier to determine what has been selected without having to navigate to the view that contains those selections.

The invention provides a word vector generation method and device, a computer storage medium and electronic equipment. The method comprises the steps of: obtaining a target sequence pattern which contains a target text unit in the training corpus set through sequence pattern mining, wherein the support degree of the target sequence pattern is not less than a support degree threshold value; for each statement, if the context window of the target text unit in the statement contains the target sequence mode meeting the length condition, configuring a semantic tag for the target text unit to obtain target text units corresponding to the same target sequence mode, wherein the target text units corresponding to the same target sequence mode carry the same semantic label, and the target text units corresponding to different target sequence modes carry different semantic labels; and obtaining a word vector of each semantic label by training a word vector model. According to the scheme, the target sequence mode capable of reflecting different contexts of the target text unit is mined, so that different semantic tags are configured for the target text unit with different contexts, and a plurality of word vectors representing different semantics of the target text unit are obtained.

A solution method for determining a context range of a key word comprises the steps of initializing a statistical method module; performing normalization curve processing by solving a weight contribution value of each position in a context adjacent to an input key work C; and finally, calculating n values of a left window and a right window of a context window in a key word C context. By the solution method, very valuable quantitative description is provided for determination of an effective range of the context, and the defect of previous subjective description is overcome; the position of description capability of the context to the key word is gradually reduced from the near to the distant, and the solution method conforms to ordinary knowledge of people; and an extremely valuable theoretical basis is provided for solving word and sentence similarity calculation better.

A system and a computer program product for displaying contextual semantic data in a graphical overlay includes a processor activating a first lens corresponding to a first domain, activating at least one first domain context within the first lens, and responsive to activating the at least one first domain context within the first lens, displaying the first domain context within the first lens. At least one second domain context is activated and displayed throughout the entire graphical overlay while simultaneously displaying the at least one first domain context within the first lens. Alternatively, the at least one second domain context is displayed within an activated second lens, which corresponds to a second domain, while simultaneously displaying the at least one first domain context within the first lens. By simultaneously displaying the first and second domain contexts, a cross-domain interaction between the first and second domains is identified.

The invention is suitable for the technical field of computers, and provides a word vector training method and system. The method includes: performing word vector training on each training target word in a training sample document, and meanwhile acquiring a window word of each training target word in a context window in the training sample document; predicting the occurrence probability of each window word by using a Skip-gram model; updating a word vector corresponding to the each window word in a word vector library and an intermediate vector corresponding to each non-leaf node in a code path corresponding to each training target word in a Huffman tree; updating whole document vectors of the training sample document through a preset formula; calculating increasing local input vectors of a CBOW model, and then calculating mixed stitching vectors of the CBOW model; setting the mixed stitching vectors as input of a projection layer of the CBOW model; predicting the occurrence probability of each training target word; and finally updating the word vector of each training target word and the intermediate vector corresponding to each non-leaf node in the Huffman tree. The method and the system improve the accuracy of the word vector of each training target word.