YNU_DYX at SemEval-2019 Task 9: A Stacked BiLSTM for Suggestion Mining Classification

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 1272–1276 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 1272 YNU DYX at SemEval-2019 Task 9: A Stacked BiLSTM Model for Suggestion Mining Classification Yunxia Ding, Xiaobing Zhou∗, Xuejie Zhang School of Information Science and Engineering Yunnan University, Yunnan, P.R. China ∗Corresponding author:zhouxb@ynu.edu.cn, ynudyx@gmail.com Abstract In this paper we describe a deep-learning sys- tem that competed as SemEval 2019 Task 9- SubTask A: Suggestion Mining from Online Reviews and Forums. We use Word2Vec to learn the distributed representations from sen- tences. This system is composed of a Stacked Bidirectional Long-Short Memory Network (SBiLSTM) for enriching word representa- tions before and after the sequence relation- ship with context. We perform an ensemble to improve the effectiveness of our model. Our official submission results achieve an F1-score 0.5659. 1 Introduction Suggestions in the Oxford Dictionary are defined as ideas or plans for consideration. Some of the listed synonyms of suggestions are propos- al, proposition, recommendation, advice, hint, tip, clue. In general, other types of text and sugges- tions are easily distinguished by the definition of the suggestion (Negi and Buitelaar, 2015). Suggestion mining can be defined as the extrac- tion of suggestions from unstructured text, where the term ‘suggestions’ refers to the expressions of tips, advice, recommendations etc. We often see comments on products in product forums which are recommended or not recommended, and some users will consider whether to purchase the prod- uct based on these comments. Suggestion mining is also defined as automatic extraction of recom- mendations from a given text. These texts that ex- press user suggestions can usually be found in so- cial media platforms, blogs, or product online fo- rums (Negi and Buitelaar, 2017; Negi et al., 2016). Suggestion mining remains a relatively young area compared to Sentiment Analysis (Negi and Buitelaar, 2017), due to the lack of a large num- ber of tagged datasets. SemEval 2019 Task 9- SubTask A is mainly a binary classification, iden- tifying sentences which express suggestions in a given text. And we need to classify each sentence of given text, the categories being suggestions or non suggestions. This is similar to the polarity analysis of emotions, as positive or negative in- stances, respectively. In the past, classification problems in natu- ral language processing were solved by tradition- al methods, such as sentiment analysis (Nielsen, 2011; Go et al., 2009; Bollen et al., 2011; Mo- hammad et al., 2013; Kiritchenko et al., 2014) which were handled by classifiers such as Naive Bayes (McCallum et al., 1998)and SVMs (Gun- n et al., 1998). However, deep neural networks achieve increasing performance compared to tra- ditional methods, due to their ability to learn more abstract features from large amounts of data, pro- ducing state-of-the-art results in sentiment analy- sis. The SubTask-A is part of SemEval 2019 Task9: Suggestion Mining from Online Reviews and Fo- rums, and is concerned with classifying sugges- tions forum for Windows platform developers— suggestions or non suggestions. …

Content truncated to 500 words