情報処理学会 第75回全国大会講演要旨
5D-7
Sentiment Classification by Capturing Users' Preferences across Different Targets
Sentiment classification is used to extract and classify opinion from text which is always expressed towards some target. However, users tend to have similar sentiment on one target while they have same opinion on other targets. For example, users who complimented "iPod" and abused "Samsung"s Nexus 7 tablet", may be more likely to buy "iPad min". These user preferences haven’t been considered and need additional data than only text. At the same time, annotated datasets frequently give additional information (e.g. user information, target information) which has still not been well utilized until now.
This paper proposed a way of using user and product information to help finding user’s preference between targets for helping sentiment classification. Compared with baseline, proposed results tested on both reviews and tweets datasets show obvious increase of precision for sentiment classification task.
This paper proposed a way of using user and product information to help finding user’s preference between targets for helping sentiment classification. Compared with baseline, proposed results tested on both reviews and tweets datasets show obvious increase of precision for sentiment classification task.
