119 results about "Comprehension approach" patented technology

The invention discloses an image understanding method based on a depth residual error network and an LSTM; the method comprises the following steps: firstly building a depth residual error network model so as to extract image abstract features, and storing the features as a feature matrix; using a dynamic attention mechanism in a LSTM model to dynamically form a proper feature vector according to the feature matrix; finally using the LSTM model to form a natural language (English) according to the feature vector. The method uses the advantages of the depth residual error network on image feature extraction and LSTM advantages on time sequence modeling; the depth residual error network and the LSTM model can form an encode-decode framework so as to convert the image content information into the natural language, thus extracting the deep information from the image.

The invention discloses an indoor scene understanding method based on a 2D-3D semantic data set. The indoor scene understanding method comprises the main content of collection, arrangement, training and test division of data and comprises the following steps: firstly capturing an image, outputting a scanning region, an original color depth image (RGB-D) and a 3D texture grid, sampling the grid to generate point clouds, carrying out semantic annotation on the data, projecting each point label to a 3D grid and an image field, and representing certain regions with concentrated data to parts with similar buildings in aspects of appearances and architectural features, so as to define standard training and test division. According to the semantic data set disclosed by the invention, cross-modal study models and non-supervision methods can be developed and united by virtue of rules existing in large-scale indoor spaces; powerful prompts are provided for the detection of semanteme, layout, shields, shapes and modes; and the indoor scene understanding method is not limited by scale, diversity and number.

The invention provides an interactive-question semantic understanding method in intelligent customer services. The method includes the following steps that the conversation content between current intelligent customer services and clients is subjected to co-text language environment expression, wherein co-text language environment expression comprises event expression and language environment expression; according to the co-text language environment expression, a conversation semantic event graph is constructed; according to multiple conversation corpora of the intelligent customer services and the clients, a business logic tree is constructed; according to a determined finite state automata, an order state machine is established; according to the semantic event graph, logic decision branches are selected from the business logic tree; according to the logic decision branches and the order state machine, a semantic processing template is returned to the intelligent customer services, and semantic expression generation is carried out. According to the interactive-question semantic understanding method in the intelligent customer services, interaction questions and answers based on aflow diagram are achieved, the accuracy of client interrogation understanding of the intelligent customer services is increased, the consistency between the intelligent customer services and client conversations is guaranteed, and the work efficiency of the intelligent customer services is improved.

The invention provides a multi-task chapter-level event extraction method based on a multi-headed self-attention mechanism. The method comprises the following steps: converting single sentence-level event extraction into chapter-level event extraction of a packaged sentence set; carrying out the word embedding representation by utilizing a pre-trained language model BERT; taking all word embedding and position embedding in a single sentence as input, employing a convolutional neural network model for coding, and capturing the most valuable features in the sentence in combination with a segmented maximum pool strategy; utilizing a multi-head self-attention model to obtain chapter representation and attention weight fused with full-text semantic information; utilizing a classifier to obtain a predicted event type; taking event types as prior information, linking the event types to an input sequence for event element extraction, and extracting all related elements in the sequence by using a pre-training model in combination with a machine reading understanding method. The method can be used for text-level event extraction tasks, and the breakthrough of converting a sequence labeling problem into a machine reading understanding problem is achieved.

The invention relates to the field of artificial intelligence, in particular to a man-machine conversation understanding method and system for a specific field and relevant equipment and aims to improve accuracy of conversation understanding. The conversation understanding method of the man-machine conversation system comprises the steps as follows: receiving a word currently input by a user and mapping the word to a vector space; expressing a historical word vector, semantic annotation information and intent category information as vectors by using a semantic representation layer; acquiring asemantic label of the current word by using the semantic labeling layer; acquiring the intent category of the current word by using an intent recognition layer. Extra part-of-speech information is introduced during model training, the part-of-speech prediction layer is used for predicting the part-of-speech of the next input word, semantic information shared by semantic annotations, intent recognition and part-of-speech prediction is fully used and promoted mutually by joint processing of the three tasks; the system and the method have clear logic, high efficiency and high accuracy, and the technical problem that the existing man-machine conversation system cannot effectively perform real-time conversation understanding is properly solved.

The invention provides a dialogue command understanding method based on a knowledge graph. The conversation command understanding method based on the knowledge graph includes: based on a preset domainknowledge set, correspondingly constructing a domain knowledge spectrum library; converting an input dialogue command into a candidate list with a plurality of candidate matching sequences; converting each candidate matching sequence in the candidate list into a knowledge graph vector through a preset coding mode; forming a dialogue command vector related to the input dialogue command according to the knowledge graph vector, training the deep learning neural network model based on the dialogue command vector, and executing dialogue command semantic understanding operation through the traineddeep learning neural network model.

The invention discloses a DLLE (Linear Local Embedding of Difference) model-based data dimension reduction and characteristic understanding method, and belongs to the field of computer vision. The method comprises the steps of firstly, obtaining an image sequence through a visual sensor, then analyzing an input motion image sequence, extracting a foreground human body contour region through a background subtraction method, performing binarization, researching a periodic characteristic of a motion, performing key frame extraction on each motion sequence, and extracting a complete motion periodic sequence; performing manifold dimension reduction through a DLLE algorithm to obtain a low-dimensional eigenvector, and storing the low-dimensional eigenvector in a motion database; and performing identification through a nearest neighbor classifier by comparing a mean Hausdorff distance between a test sequence and a motion sequence in a training sample library. According to the method, the application of a differential function and category information-based neighborhood preserving embedding algorithm to human body motion identification is proposed; a DLLE model can not only keep a manifold local geometric structure during dimension reduction but also fully utilize category information of original high-dimensional data; and the extension from unsupervised extension to supervised extension is realized.

The embodiment of the invention discloses a global semantic understanding method and device, a computer device and a storage medium. The method comprises the following steps of acquiring a plurality of target texts in a display interface of a user terminal; detecting keywords in the target text according to the preset keyword list; querying the semantic comprehension results corresponding to the key words from the preset network resources; generating a synopsis of the key word and a network link according to the semantic understanding result, wherein the network link is one or more network page links obtained after the network query is carried out according to the key words. By acquiring the target text in the display interface of the user terminal, and then filtering out the key words inthe target text, querying the semantic understanding result corresponding to the key word in the preset network resources, and generating the semantic understanding result into the key word brief introduction and the network link to display to the user, the user can understand the key word in time and accurately, the query operation steps of the user are reduced, and the use is convenient and reliable.