Effective Combination of Bert Model and Cross-Sentence Contexts in Aspect Extraction

The Aspect Extraction (AE) field investigates in collecting words which are sentiment aspects in sentences and documents. Despite the pandemic, the number of products purchased online is still growing, which means that the number of product reviews and comments is also increasing rapidly, so the role of the task is gradually crucial. Extract aspects in the text is a difficult task, that requires algorithms capable of deep capturing the semantics of the text. In this work, we combine two models of the two research groups, with the first using the BERT algorithm with multiple concatenated layers and the second using the strategies to enrich the dataset by itself in the training or testing phase. The source code is available on github.com, researchers can run it through scripts, modify it for further research also. https://github.com/leanhkhoi/AE_BERT_CROSS_SENTENCES