63 results about "Semantic clustering" patented technology

The invention discloses a semi-supervised automatic aspect extraction method based on domain information. The semi-supervised automatic aspect extraction method comprises the steps of network information crawling, information pre-processing, keyword extraction, comment document recombination and fine-grit mark LDA learning. The invention further discloses a semi-supervised automatic aspect extraction system based on the domain information. The semi-supervised automatic aspect extraction system based on the domain information comprises a network information crawling module, an information pre-processing module, a keyword extraction module, a comment document recombination module and a fine-grit mark LDA learning module. By the adoption of the semi-supervised automatic aspect extraction method and system based on the domain information, all extracted aspects of a commodity are more clear and more definite, and the differences between the aspects are more clear; a generated aspect structure (order and content) generated through the semi-supervised automatic aspect extraction method and system can be kept consistent with a commodity aspect structure which is predefined in a seed word set, so that the semi-supervised automatic aspect extraction method and system have the advantages that semantic clustering can be conducted on different expressions used by a consumer for description of the same commodity aspect, and human interference can be reduced in the process of opinion mining of the commodity.

The invention discloses a session data processing method and device, a knowledge base building method and device, a knowledge base optimizing method and device and an interaction method and device. The session data processing method includes the steps that semantic clustering processing is carried out on a first interrogative sentence of each piece of to-be-processed session data, and one or more first groups are obtained; semantic clustering processing is carried out on second interrogative sentences of all the to-be-processed session data belonging to the same first groups, and one or more second groups are obtained; the semantic clustering processing process is constantly repeated till one or more leaf groups meeting preset conditions are obtained, in other words, semantic clustering processing is carried out on (i+1)th interrogative sentences of all to-be-processed session data belonging to the same ith groups, and one or more (i+1)th groups are obtained; based on the semantic clustering processing result, a scene session sequence is established through time-sequence arraying according to all the leaf groups and all the corresponding higher-level groups. By means of the scheme, the situation that corresponding content can not be fed back to a user can be reduced, and the user experience can be improved.

The invention discloses an emergent event abstract extracting method based on a sparse learning theory. The method comprises the following steps: acquiring emergent event topics, and performing query extension on the emergent event topics to obtain an extension topic lexical item set of the event topics; retrieving events according to extension topic lexical items after the query extension of the events to obtain document sets relevant to the event topics; performing feature selection and semantic clustering on the document sets of the topics in sequence by a non-negative matrix decomposing method to obtain topic clustering results of the events; and extracting a representative sentence from the clustering results to serve as a finial abstract result according to a maximal marginal relevance (MMR) method. Through the technical scheme provided by the invention, when the emergent events burst, latest condition information of event development is provided for a user in the presence of massive redundant news report streams.

A taxpayer industry classification method based on noise label learning comprises the steps that firstly, text information to be mined in taxpayer industry information is extracted for text embedding, and feature processing is conducted on the embedded information; secondly, non-text information in the taxpayer industry information is extracted and coded; thirdly, a BERT-CNN deep network structure conforming to the taxpayer industry classification problem is constructed, and the number of layers of the network, the number of neurons of each layer and the input and output dimensions are determined according to the processed feature information and the target category number; then, the constructed network is pre-trained through comparative learning, nearest neighbor semantic clustering and self-label learning in sequence; finally, a noise modeling layer is added on the basis of the constructed deep network, modeling is carried out on noise distribution through network self-trust and noise label information, and model training is carried out based on noise label data; and finally, the deep network in front of the noise modeling layer is taken as a classification model, and taxpayer industry classification is performed based on the model.

Semantic clustering techniques are described. In various implementations, a conversational agent is configured to perform semantic clustering of a corpus of user utterances. Semantic clustering may be used to provide a variety of functionality, such as to group a corpus of utterances into semantic clusters in which each cluster pertains to a similar topic. These clusters may then be leveraged to identify topics and assess their relative importance, as for example to prioritize topics whose handling by the conversation agent should be improved. A variety of utterances may be processed using these techniques, such as spoken words, textual descriptions entered via live chat, instant messaging, a website interface, email, SMS, a social network, a blogging or micro-blogging interface, and so on.

The invention provides a labelling image scene clustering method based on vision and labelling character related information. The method comprises the following steps of: dividing a training image and a test image respectively by using a NCut (Normalized Cut) image dividing algorithm; constructing a vision nearest-neighbour graph G(C)(V, E) of all images {J1, ., Jl} PCtrain for learning, wherein in a training image set, each image has one group of initial normalized labelling character weight vectors; spreading the labelling character of each training image among the vision nearest neighbours, receiving the accepted images according to the degree of normalized EMD (Earth Mover's Distance) among the accepted images; for each training image, normalizing the accumulated labelling character weights; after the vision characteristics of the image are converted into a group of labelling characters with weights, carrying out the scene semantic clustering by using a PLSA (Probabilistic Latent Semantic Analysis) model; learning each scene semantic vision space by using a Gaussian mixture model; and carrying out the scene classification by using the vision characteristics. With the invention, the coupling precision between the vision characteristics of the image and the labelling character can be increased, and the method can be directly used for the automatic semantic labelling of the image.

A system and method for providing an easy-to-use interface for verifying semantic tags in a steering application in order to generate a natural language grammar. The method includes obtaining user responses to open-ended steering questions, automatically grouping the user responses into groups based on their semantic meaning, and automatically assigning preliminary semantic tags to each of the groups. The user interface enables the user to validate the content of the groups to ensure that all responses within a group have the same semantic meaning and to add or edit semantic tags associated with the groups. The system and method may be applied to interactive voice response (IVR) systems, as well as customer service systems that can communicate with a user via a text or written interface.

The invention discloses a method for detecting urban infrastructure emergencies based on clusters. The method includes the steps that firstly, calls of citizens for reflecting urban infrastructure problems are answered and recorded in a work order database; after the database is screened, semantic keywords are extracted for screened-out work orders through the Chinese word segmentation technology, the causal relationships between the semantic keywords are sorted, and then address keywords of the work orders are extracted; semantic clustering is conducted on the work orders; spatial clustering is conducted on each cluster on the semantic level; time clustering is conducted on each cluster on the spatial level, accordingly clusters on the time level are acquired and identified as the urban infrastructure emergencies, and distribution of the urban infrastructure emergencies is displayed on a map through a visual design scheme of root nodes of the urban infrastructure emergencies; thus, the urban infrastructure emergencies are detected, and users can more quickly and sensibly make decisions so as to solve the problems in urban infrastructure.

A semantic textual information recognition system is provided. The system includes a memory configured to store instructions and results. The system includes a processor operatively coupled with the memory configured to receive a plurality of text elements along with respective text element coordinates, wherein the processor includes a spatial reconstruction module configured to identify the plurality of text elements on an information axis based on the text element coordinates. The processor also includes a semantic clustering module configured to determine a plurality of semantic clusters of the plurality of text elements using the plurality of text elements on the same information axis and a semantic data model. The processor further includes a rank clustering module configured to generate a plurality of rank clusters by ranking the plurality of semantic clusters. The processor further includes a machine learning module configured to update the semantic data model based on the feature set.

A tracking relation recovery method based on syntax comprises the following steps of sentence segmenting, part-of-speech tagging, block analyzing, noise reducing, clustering and cluster mapping. An index term which can most likely describe the self characteristic is extracted from a product through a syntactic analysis technology, the tracking relation between different software products is recovered by semantic clustering, and thus the problems that a traditional requirement tracking maintenance method can be affected by noise in the product easily, and is high in cost, poor in expandability and the like are solved.

The invention discloses a cross-modal semantic clustering method based on a bidirectional CNN, and the method comprises the steps: firstly carrying out the preprocessing of data, and carrying out the pre-training of a text sample of a training set; then constructing a cross-modal retrieval network, training the cross-modal retrieval network through the training set, and calculating a loss function of the network; carrying out back propagation, and optimizing the connection weight through a selected optimizer and corresponding parameters; performing training for multiple rounds to obtain a final network model; and finally, testing the trained model through the test set, and calculating various evaluation indexes. According to the method disclosed by the invention, the accuracy and the efficiency of cross-modal retrieval are improved by utilizing clustering of semantic information. According to the method, the loss of the sample and the clustering center in the target space, the distribution difference loss of the categories in different modes and the discrimination loss are designed to help semantic clustering, so that the recognition capability among different categories is enhanced, and the correlation among different modes is enhanced.

The invention discloses an adversarial signal detection method based on multichannel feature reconstruction, and the method comprises the steps: firstly collecting a signal data set, inputting the signal data set into a feature extraction depth receiver, calculating an embedded feature map and a corresponding semantic cluster, and inputting the embedded feature map and the corresponding semantic cluster into a multichannel feature encoder; the multi-channel feature encoder comprises a private semantic encoder, a public semantic encoder and a noise encoder; training the encoder, and reconstructing the encoder to obtain reconstructed features; and inputting the reconstructed features into a meta-classifier, training the meta-classifier, and completing distinguishing of normal samples and adversarial samples. According to the method, the early shooting features are extracted through the noise feature encoder, multi-feature reconstruction is carried out on the signal features, and the difference between an adversarial sample and a normal sample is enhanced. According to the adversarial signal detection device based on multichannel feature reconstruction, noise features of adversarial samples are extracted, so that smaller adversarial disturbance can be detected more accurately.

The invention discloses a method for establishing word vector improved model based on semantic embedding. The method comprises the following steps: 1) a bidirectional long-short-term memory network training stage; 2) a context vector calculation stage: inputting a sentence and a target word t into the bidirectional long-short-term memory network trained in the step 1) to obtain a context vector; 3) a context vector semantic clustering stage: a, calculating the current context vector and the similarity of each semantic cluster center of the word t by using cosine similarity; b, calculating theprobability P of the class cluster to which the current context vector belongs by using a Bayesian non-parametric statistical model; c, maximizing a P value and solving a class cluster corresponding to the value; c, offsetting the class cluster center to which the current context vector belongs. According to the method, the neural network and the Bayesian non-parametric statistical method are utilized to solve the problem that a current word vector model cannot solve the problem that one word has multiple meanings.

The invention provides a rapid identification method and device for large-scale intention, and electronic equipment. The method comprises the steps of performing semantic vector conversion on intention category information input by a conversation of a historical user; performing semantic clustering on each intention after semantic vector conversion; establishing an index for a result after semantic clustering, wherein the index is used for searching an intention category corresponding to dialogue input in a preset intention database; obtaining dialogue input of a current user in real time, andperforming search matching by using the index; and inputting the semantic vector and the search matching result input by the dialogue into a sorting model, and sorting to determine an intention recognition result. According to the method, a clustering index-deep matching recall-sorting scheme is adopted, so that the manpower is greatly saved, the thousand-level maximum-scale intention recognitionis realized, the intention recognition efficiency is improved, and the user intention recognition accuracy is also improved.

The invention provides a visual relation detection method and device based on a high-order semantic structure of a scene graph, and the algorithm comprises the steps: predicting the types and positions of all objects in a picture, outputting a visual feature vector corresponding to each object, and carrying out the pairing operation of every two of all detected objects; based on a pairing result, extracting a combined visual feature vector, and encoding the position to obtain a position code; inputting the categories of all the objects into a hierarchical semantic clustering algorithm, and processing to obtain a high-level semantic feature vector corresponding to each object; performing semantic coding on the output of the hierarchical semantic clustering algorithm; generating a relation classifier weight; and combining the visual feature vector, the joint visual feature vector and the position code into a unified feature vector, carrying out point product operation on the unified feature vector by using the relation classifier weight, and finally obtaining a relation conditional probability between every two objects as a scene graph.

A method and apparatus are provided for automatically acquiring grammar fragments for recognizing and understanding fluently spoken language. Grammar fragments representing a set of syntactically and semantically similar phrases may be generated using three probability distributions: of succeeding words, of preceding words, and of associated call-types. The similarity between phrases may be measured by applying Kullback-Leibler distance to these three probability distributions. Phrases being close in all three distances may be clustered into a grammar fragment.

A system is provided for decomposing events from managed infrastructures. A first engine is configured to receive message data from a managed infrastructure that includes managed infrastructure physical hardware that supports the flow and processing of information, the at least one engine is configured to determine common characteristics of events and produce clusters of events relating to the failure of errors in the managed infrastructure. Membership in a cluster indicates a common factor of the events that is a failure or an actionable problem in a physical hardware of the managed infrastructure directed to supporting the flow and processing of information. The first engine is configured to create one or more situations that is a collection of one or more events or alerts representative of the actionable problem in the managed infrastructure. A second engine is configured to determine one or more common steps from events and produces clusters relating to events. The second engine determines one or more common characteristics of events and produces clusters of events relating to the failure or errors in the managed infrastructure. The system is configured to use data-driven fault localization, more particularly using semantic clustering.