Polarized Extractive Summarization of Online Product Reviews
We have created a new sentiment classified dataset because the datasets available online are on complete reviews, and a review on a whole can not be given a single polarity because the review may contain both positive and negative aspects of user experience. With the goal of facilitating a more precise polarised summaries, we arepreparing a dataset containing sentence classifications as positive, negative, and neutral.
This dataset was obtained by crawling the phone reviews of verified users onAmazon’s website and splitting them into 12,000 sentences. Six popular brandssuch as Samsung, Oppo, Vivo, Redmi, Oneplus, and Apple were considered. Ineach brand, we captured 5 different configurations and crawled 100 reviews. In or-der to create a coherent dataset, the reviews were separated by the ”full stop”(.)and manually annotated by the polarity of the sentences. Our dataset containsthe following distribution of polarized sentences: Positive sentences(5982), Nega-tive sentences(4138), and Neutral sentences(1880).