An AI based cross‐language aspect‐level sentiment analysis model using English corpus

Abstract

Accurate cross‐language aspect‐level sentiment analysis methods can provide accurate decision support for social networks, e‐commerce platforms, and other platforms, thereby providing users with higher quality services. However, actual data is very complex and contains a large amount of redundant information. Existing methods face challenges in extracting semantic association information and deep emotional features when dealing with this complex data. To address these issues, an aspect‐level sentiment analysis model (called Multi‐XLNet‐RCNN) is proposed that integrates multi‐channel XLNet and RCNN. First, a multi‐channel XLNet (Multi XLNet) network model is used to perform autoregressive encoding operations on different languages, fully extracting contextual information from the text and better characterizing the ambiguity of the text. Then, in the RCNN module, the contextual features output by the BiGRU layer are concatenated with the pre trained input features to extract deeper emotional features. Finally, in response to the issue of inconsistent aspect‐level information in sentence features extracted from different language channels, a multi head attention mechanism based on aspect class interaction is utilized to obtain a text attention emotion representation for a given aspect, thereby improving the accuracy of aspect‐level emotion classification. The experiment uses the public English corpus provided by SemEval 2016 as the source language, and Chinese comment data on Dianping and JD E‐commerce platforms as the target language. The experimental results show that the proposed Multi XLNet‐RCNN sentiment analysis method can achieve accurate aspect‐level Sentiment analysis, and the accuracy rates on the two data sets of Dianping and Jingdong E‐commerce can be as high as 0.851 and 0.792, respectively, superior to other advanced comparison models. This model has good application value in cross‐language analysis of social networks and e‐commerce platforms.