A Coarse Grained Sentiment Analysis Method Based on Hierarchical Bert Neural Network

Abstract

The present invention relates to a coarse-grained emotion analysis method based on a hierarchical BERT neural network, which belongs to the technical field of Web mining and intelligent information processing; it includes the following steps: acquisition of corpus: obtaining corpus for coarse-grained emotion analysis; corpus preprocessing: including character cleaning, Clause segmentation and clause vector construction; construct sentence vector: use bidirectional long-short-term memory network, multi-layer perceptron and attention mechanism to calculate clause vector to generate sentence vector; gradient coordination mechanism optimization: introduce gradient coordination mechanism to solve coarse-grained sentiment analysis The problem of unbalanced data categories in medium; using hierarchical BERT neural network for coarse-grained sentiment analysis. Compared with the existing technology, the present invention constructs sentence vectors containing deep semantic information for review texts through a hierarchical BERT neural network, which improves the accuracy of coarse-grained sentiment analysis tasks and has broad application prospects in information recommendation, public opinion monitoring and other fields .

Description

technical field [0001] The invention belongs to the technical field of Web mining and intelligent information processing, and relates to a coarse-grained sentiment analysis method based on a hierarchical BERT (Bidirectional Encoder Representation from Transformers, full name Bidirectional Encoder Representation from Transformers) neural network. Background technique [0002] Coarse-grained text sentiment mining, referred to as coarse-grained sentiment analysis. Coarse-grained sentiment analysis aims to classify the sentiment polarity of review texts. Sentiment polarity includes three categories: commendatory, derogatory and neutral. Commentary text is divided into chapter-level commentary text and sentence-level commentary text. [0003] At present, coarse-grained sentiment analysis methods mainly include methods based on rule and statistical fusion, methods based on machine learning, and methods based on deep learning. [0004] The coarse-grained sentiment analysis metho...

AI Technical Summary

Problems solved by technology

Traditional language models are mostly shallow one-way neural networks, which are difficult to deeply describe or reflect the semantic information of the text

Second, the existing coarse-grained sentiment analysis methods lack deep mining of text semantic information when generating text vector representations, and lack of optimal processing for the imbalanced problem of classified data categories

Images