33results about How to "Rich semantic features" patented technology

The invention discloses a sentence-to-semantic matching method and device for intelligent interaction, belonging to the fields of artificial intelligence and natural language processing. The technical problem to be solved by the present invention is how to capture semantic features and sentence interaction information to realize intelligent semantic matching of human-computer interaction sentence pairs. The technical solution adopted is to construct and train a sentence-pair semantic matching model composed of a multi-granularity embedding module, a multi-level alignment matching network module, a feature enhancement module, and a label prediction module to realize the multi-level alignment and matching representation of sentence information. Dimensional maximum pooling and interactive generation of matching tensors of sentence pairs to determine the matching degree of sentence pairs to achieve the goal of intelligent matching of sentence pairs. The device includes a sentence-pair semantic matching knowledge base construction unit, a training data set generation unit, a sentence-pair semantic matching model construction unit and a sentence-pair semantic matching model training unit.

The present invention provides a semantic similarity matching method, device and storage medium. The method includes: importing a first sample to be analyzed and a second sample to be analyzed, respectively performing Word vectorization processing to obtain the first word vector matrix corresponding to the first sample to be analyzed and the second word vector matrix corresponding to the second sample to be analyzed; construct a training network, and use the training network to respectively The first word vector matrix and the second word vector matrix are subjected to feature fusion processing to obtain a first fusion vector matrix corresponding to the first word vector matrix and a second fusion vector corresponding to the second word vector matrix matrix. The present invention solves the problems of feature loss, insufficient interaction between sentences, and disappearance of network gradients, enriches the semantic features of sentences, makes the information interaction between sentences more accurate and rich, and can capture more semantic information of sentence pairs.

The invention relates to image processing technology, in particular to a low-resolution pedestrian detection method, system and storage medium. The method of the present invention includes a training process and a testing process. The training process first determines the training set and the parameters of the training process; then inputs pictures in sequence according to the batch size, extracts the multi-scale features of the training pictures, and reconstructs and enhances the shallow features. Form a new multi-scale detection framework; finally perform frame classification and position regression, calculate training loss and backpropagation, and update weight parameters. The test process first determines the test set, uses the model obtained during the training process as the test model of the algorithm, sequentially inputs test pictures in small batches, extracts multi-scale features, reconstructs and enhances shallow features, and then classifies and positions frames return. The present invention uses a deep learning network to reconstruct shallow features, and at the same time improves the effectiveness of shallow features, so as to enhance the ability to detect low-resolution pedestrians.

The invention discloses a text generation method and device for a construction industry information service question answering system, belonging to the technical fields of artificial intelligence and natural language processing. The technical problem to be solved in the present invention is how to prevent the lack of inter-layer features in the encoding process of the generative question answering system and the feature loss in the decoding process, so as to ensure the accuracy of text generation. The technical solution adopted is: by constructing and training the The text generation model composed of the embedded module, the encoder module, the hidden state mutual information module, the hidden variable mutual information module and the decoder module realizes the multi-layer encoding of the original text and obtains the text representation and hidden information of the original text; The hidden state of the hierarchical encoding, the hidden variables of the original text and their sampling information are used to calculate the mutual information and maximize their mutual information; the original text and the target text are encoded, and the correlation information between the two is obtained through the attention mechanism, and finally the achieve the purpose of text generation.