Research And Implementation Of Multidimensional Opinion Mining Method Based On Reviews

Reviews in e-commerce and online media reflect consumer satisfaction with goods or services,they have reference value for merchants and other consumers’ decisionmaking.The large number and rapid growth rate of Internet review texts make it difficult to process manually,and the multi-dimensional nature of comments makes it difficult for processing algorithms.This article studies the multi-dimensional problems in the review text,it constructs the opinion mining model using neural networks,it also designs and develops the opinion mining system.The main work of this article is as follows(1)A text representation combining word vector and char vector is proposed as model input,this article introduces the attention mechanism into the Bi LSTM-CRF model,it constructs an optimized opinion mining model Bi LSTM-attention-CRF.The text representation of word vectors and char vectors can obtain richer semantic information;the Bi LSTM-attention-CRF view mining model can focus on the content related to the perspective dimension and dig out the potential information of the comment text.In the design of the verification experiment on the catering Chinese comment dataset of the AI challenger algorithm competition,the experimental results show that the Macro-F1 value increases by 0.03 compared with the use of word vector and char vector for combined text representation,and the accuracy rate is improved by 2%.The Bi LSTM-attentionCRF model improved its macro-F1 value by nearly 0.02 compared to the baseline model,and the accuracy rate improved by 2.3%.(2)This article proposes a sentiment analysis model that combines long short-term memory network and convolutional gating network.After the opinion dimension is mined,the emotional tendency of the corresponding dimension in the comment is obtained using dimension-based sentiment analysis.Aiming at the problem of contextual feature extraction,the extraction scheme of selecting opinion dimension words and right context is proposed,and based on the sentence features under the scheme,the sentiment analysis model of LSTM-GCAE is constructed.Experimental results show that compared with the GCAE model,the Macro-F1 value of LSTM-GCAE is increased by nearly 0.015,and the accuracy rate is increased by nearly 2%.(3)This article designs and develops an opinion mining system for online reviews,it realizes the online extraction and analysis processing of Internet reviews,it uses the proposed algorithm model for dimensional mining and sentiment analysis,then visualizes the results.