Hybrid RNN at SemEval-2019 Task 9: Blending Information Sources for Domain-Independent Suggestion Mining

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 1199–1203 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 1199 Hybrid RNN at SemEval-2019 Task 9: Blending Information Sources for Domain-Independent Suggestion Mining Aysu Ezen-Can SAS Inst. aysu.e.can@gmail.com Ethem F. Can SAS Inst. ethfcan@gmail.com Abstract Social media has an increasing amount of in- formation that both customers and companies can benefit from. These social media posts can include Tweets or be in the form of vocaliza- tion of complements and complaints (e.g., re- views) of a product or service. Researchers have been actively mining this invaluable in- formation source to automatically generate in- sights. Mining sentiments of customer reviews is an example that has gained momentum due to its potential to gather information that cus- tomers are not happy about. Instead of read- ing millions of reviews, companies prefer sen- timent analysis to obtain feedback and to im- prove their products or services. In this work, we aim to identify informa- tion that companies can act on, or other cus- tomers can utilize for making their own expe- rience better. This is different from identify- ing if reviews of a product or service is neg- ative, positive, or neutral. To that end, we classify sentences of a given review as sug- gestion or not suggestion so that readers of the reviews do not have to go through thou- sands of reviews but instead can focus on ac- tionable items and applicable suggestions. To identify suggestions within reviews, we em- ploy a hybrid approach that utilizes a recurrent neural network (RNN) along with rule-based features to build a domain-independent sug- gestion mining model. In this way, a model trained on electronics reviews is used to ex- tract suggestions from hotel reviews. 1 Introduction With the growth of social media usage, the inter- est in text mining approaches has increased. One task that has gained momentum recently is senti- ment analysis where the goal is to determine opin- ions/emotions from a text input, generally a prod- uct or service review. Different approaches have been proposed for sentiment analysis task such as multilingual models to be used with limited data (Can et al., 2018) and sentiment lexicons (Banea et al., 2008). Twitter posts also has been one source of reviews to be mined in terms of sen- timent (Pak and Paroubek, 2010; Ezen-Can and Can, 2018; Tellez et al., 2017). The task of suggestion mining is similar to sen- timent analysis in that the input is the same (e.g., customer reviews). However, the output of a sen- timent analysis model and a suggestion mining model is different. While sentiment classifiers fo- cus on grouping reviews as positive or negative, suggestion mining models identify the reviews that have suggestions/actionable items/advice to other people/service providers. In this paper, we present a suggestion mining model that takes product/service reviews and clas- sifies each sentence in a given review as sugges- tion or not suggestion. To that end, we employ a hybrid LSTM …

Content truncated to 500 words