Text Sentiment Analysis Combining Part-of-speech Skipping And Multi-attention Interactive Network

With the continuous expansion and rapid popularization of the Internet,the way,time and place people go online are no longer restricted by a single mode.Internet users can express their opinions on the products they have purchased,services they have enjoyed,and social hot spots on the Internet anytime and anywhere.Therefore,it will generate a large number of emotional network comments.How to efficiently and accurately mine useful emotional information from a large number of network comments is worthy of people's deep thinking.This thesis conducts in-depth study and further research on the research status of text sentiment analysis field and deep learning theory,and improves on the basis of existing analysis models to improve the speed and accuracy of text sentiment analysis.The main research is as follows:1.In order to solve the problem of long and short-term memory network text sentiment analysis model on the input text sentence feature analysis training time,low speed,and insufficient text feature extraction,which limits the accuracy of text sentiment analysis,a Skip-LSTM-CNN fusion part of speech is proposed.Sentiment analysis model.First,before the hidden layer of the long and short-term memory network updates the current features,the feature encoder is used to extract feature information from the text of the current feature,forward sequence and backward sequence,and combined with part of speech features to determine whether to skip the current feature;then analyze the skip After reading the semantic feature sequence,the convolutional neural network is used to further extract local features of the text for sentiment analysis,and finally the sentiment result is output with positive or negative labels through the classification layer.Experiments on Rotten Tomatoes and IMDb datasets show that compared with traditional sentiment analysis models and existing skipping models,the text sentiment analysis model proposed in this thesis has achieved good results in both performance and speed.2.Aiming at the problems of short comment text length,sparse content,insufficient text feature extraction,and ignoring the influence of target words on context modeling in target sentiment analysis tasks,a target sentiment analysis model based on a dual-channel interactive attention network is proposed.First,fusion of part-of-speech information is based on the input of a single text word vector,and an internal attention is used to preprocess the text target word;then,the text target feature and its context are respectively input to the Bi-directional long and short-term memory network and combined with the location information to obtain The deep semantic expression of the text;finally,the target and context are interactively learned through dual-channel interactive attention,so as to focus on the most important emotional features in the text.Experiments are carried out on the Restaurant and Laptop datasets of Sem Eval2014 and Twitter datasets.The experimental results can effectively prove the effectiveness of this model on target-based sentiment analysis tasks.