Aspect-based sentiment analysis employing linguistics content over social media for Web of Things

To implicitly reflect the relationship between traits and emotional expressions in a context, prevalent ways to uncover emotional configurations mostly rely on attention processes. Aspect-level sentiment analysis (ABSA) seeks to identify the emotional patterns of certain qualities in phrases while omitting the grammatical information of the Web of Things. The model includes a graph convolutional neural network (GCN) based on a dependency syntax tree and exploits syntactic structural data to directly correlate traits with their related emotional expressions, hence reducing classification interference brought to by duplicated inputs. The pretrained model BERT also picks up more grammatical knowledge by being guided by intermediary layer representations of different types of grammatical information. Each GCN layer's input is merged with the preceding layer's output and the guiding information from the BERT intermediate layer. Finally, the feature for sentiment classification is the representation of characteristics in the final layer of GCN. The experimental results on the SemEval 2014, Task4, Restaurant, Laptop, and Twitter datasets show that the proposed model provides better performance of classification accuracy than a substantial number of benchmark models.