Evaluation triple extraction method based on multi-scale feature fusion

Abstract

The invention discloses an evaluation triple extraction method based on multi-scale feature fusion. The method comprises the following steps: 1) data preprocessing: performing operations such as word segmentation and stop word removal on sentences in a corpus; (2) word embedding is constructed, wherein words in sentences are converted into word embedding through pre-training vectors; 3) feature enrichment: extracting multi-scale local features by using CNNs with different convolution kernel sizes, and further capturing context information in sentences by using BiLSTM so as to obtain word representation with rich features; 4) constructing phrase-level word representation: designing a simple phrase perception representation selection mechanism, and selecting proper phrase-level word representation by judging the length of the phrase to which the current word belongs; and 5) evaluation triple extraction: constructing a corresponding grid representation, and further performing prediction and decoding to obtain a final evaluation triple. According to the method, complex statements in comments can be better understood, and the extraction accuracy of the evaluation triples is improved.

Description

technical field [0001] The invention relates to the field of emotion computing, in particular to an evaluation triplet extraction method, which has important significance in the fields of judging user's expressed emotion, assisting user's decision-making analysis and the like. Background technique [0002] In recent years, deep learning has achieved major breakthroughs and applications in academia and industry, and has attracted widespread attention from all walks of life. Deep learning has the advantages of good training effect and no need for complex feature extraction. It is widely used in the field of sentiment analysis. A series of neural network structures and their improvement methods have been proposed (such as: Deep Neural Networks, DNN), Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs, etc.) are used to solve the subtask of attribute-level sentiment analysis with promising results. [0003] With the rapid development of the era of big data, m...

AI Technical Summary

Problems solved by technology

Recent studies make full use of the interaction between tasks, and model the task end-to-end for the interaction between words, but the sparse features contained in sentences (such as: local features, phrase-level features) are often not obtained. full research

Images