In the natural language processing area, Aspect-Based Sentiment Analysis is utilized to analyse and determine sentiment polarity for each aspect within a document or sentence. A significant portion of research is the investigation of the correlation between sentiment aspects' polarity and their local context. However, only a limited work considers this point of analysing deep correlations among the global context as well as the sentiment aspect polarity. This research proposes a multilingual sentiment analysis model with an aspect-specific sentiment representation with GCN. Accordingly, considering languages like English, Chinese, Arabic, French and German. Initially, the input text undergoes preprocessing through tokenization, stop word removal and the stemming process. Subsequently, extracting features like embedding, proposed aspect-specific sentiment representation with GCN (ASSR-GCN), Bag-of-Words (BoW) and word length-based features. In ASSR-GCN feature, capturing a wide range of aspects and improving the extraction of significant terms. Then, the feature representations are fed into hybrid (ISN-LN) model that combines the strengths of Improved SqueezeNet and LeNet for effective sentiment analysis. According to the ISN classification model, it incorporates proposed feature splicing (PF) and improved multiscale Depthwise separable convolution module (IMDSC) to prevent the unnecessary parameter generation. To optimize the performance, an improved score-level fusion (ISFL) method is employed to combine the outputs of the hybrid model, which ensures high accuracy in classifying three sentiment categories, namely, Neutral, Positive, and Negative. The ISN-LN model achieved the highest accuracy of 0.956, precision of 0.935 and F-measure of 0.935.
Tummala et al. (Mon,) studied this question.