查看文章

Opinion Target Extraction (OTE) identifies opinionated aspects in review sentences with the goal of achieving Aspect Based Opinion Mining (ABOM). The available review corpuses and datasets for this task contain a large number of opinionated sentences from wider class of products from different e-Commerce applications. This raised the need for constructing the dataset from these corpora for OTE in a new way called distant supervision. This is by selecting the review sentences based on the learned opinionated relations among the terms in the sentence. The obtained review sentences are then labelled and thus dataset is constructed. The attention model based review sentence selection that uses distant supervision has ignored the review sentences that do not contain opinionated aspect terms and treated them as noise. This has reduced the precision in terms of number of extracted aspects. In order to improve the precision of extracted aspect terms, the attention based review sentence selection is replaced with the learned corpus linguistic rules based review sentence selection. This takes into account the non opinionated aspects available sentences and these aspects are treated as neutral opinion targets. Finally, the extracted aspect terms are analyzed for opinion orientations. It is shown that the approach of corpus linguistics rules based review sentence selection outperforms the attention model based review sentence selection with distant supervision.