A Study Of Aspect-Based Sentiment Analysis Methods Incorporating Part-of-Speech

Aspect-based sentiment analysis methods are often classified by researchers as sentiment lexicon based,traditional machine learning based and deep learning based,depending on their underlying principles.In recent years,deep learning has become the dominant research approach for aspect-based sentiment analysis tasks as it outperforms other methods.However,existing deep learning models rarely exploit the Part-of-Speech information sufficiently.In order to improve the performance of aspect-based sentiment classification by making full use of the Part-of-Speech information of text,this paper proposes two aspect-based sentiment analysis models incorporating Part-of-Speech information based on deep learning methods.A corresponding Web system is also designed to improve the applicability of the models.The specific work is as follows.(1)This paper proposes an Aspect-based sentiment analysis model combining Part-of-Speech and LSTM.In order to make use of the grammatical information carried by the lexical sequences,the model first annotates the Part-of-Speech of the text to be processed,trains the Part-of-Speech vector using the annotated lexical sequences,and then concatenates the word vector with the Part-of-Speech vector as the input vector of the model.In this way,the Part-of-Speech information is effectively incorporated into the deep learning model,and the model combines LSTM and attention mechanisms to extract important information that facilitates model prediction.Experiments conducted on the publicly available Restaurant14 and Laptop14 datasets show that the performance of this model is improved compared to the comparison model.In addition,the validity of the Part-of-Speech information was verified by ablation experiments.(2)This paper proposes an Aspect-based sentiment analysis model that combines Part-of-Speech and GCN.In order to more fully exploit part-of-speech information,this paper incorporates part-of-speech information into a deep learning model based on graph convolutional networks,which can use syntactic dependencies between words to learn rich textual information including part-of-speech information.In addition,a data enhancement strategy for aspect-based sentiment analysis tasks is proposed in order to overcome the problems of small training data size and uneven data distribution.Experiments on publicly available datasets validate that the model has good performance for sentiment classification.(3)Based on the model proposed in this paper,an Aspect-based sentiment analysis system is designed,which enables users to handle single-text and multi-text sentiment classification tasks more easily and visualise the processed results.