Towards the Extraction of Customer-to-Customer Suggestions from Reviews

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2159–2167, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Towards the Extraction of Customer-to-Customer Suggestions from Reviews Sapna Negi Insight Centre for Data Analytics National University of Ireland Galway {firstname.lastname}@insight-centre.org Paul Buitelaar Insight Centre for Data Analytics National University of Ireland Galway Abstract State of the art in opinion mining mainly focuses on positive and negative senti- ment summarisation of online customer reviews. We observe that reviewers tend to provide advice, recommendations and tips to the fellow customers on a variety of points of interest. In this work, we target the automatic detection of sugges- tion expressing sentences in customer re- views. This is a novel problem, and there- fore to begin with, requires a well formed problem definition and benchmark dataset. This work provides a 3- fold contribu- tion, namely, problem definition, bench- mark dataset, and an approach for detec- tion of suggestions for the customers. The problem is framed as a sentence classifi- cation problem, and a set of linguistically motivated features are proposed. Analy- sis of the nature of suggestions, and clas- sification errors, highlight challenges and research opportunities associated with this problem. 1 Introduction Opinion mining mainly deals with the summari- sation of opinions on the basis of their sentiment polarities (Liu, 2012). However, on a closer obser- vation of such opinionated text, we can discover other facets of opinions. For example, a hotel re- view sentence like, Make sure you bring plenty of sun tan lotion-very expensive in the gift shop, would be labeled as a neutral sentiment in cur- rent opinion mining methodologies, since it would only be interested in collecting opinions about the hotel. In the case of aspect based sentiment anal- ysis, the sentence does not comprise of hotel re- lated aspects, and thus will again be labeled as neutral/objective. While such sentences are gen- erally ignored in sentiment based opinion sum- marisation, these can be very useful information to extract from the reviews. In hotel reviews, such suggestions range from tips and advice on the re- viewed entity, to suggestions and recommenda- tions about the neighbourhood, transportation, and things to do. Similarly, in product reviews sugges- tions can be about how to make a better use the product, accessories which go with them, or avail- ability of better deals. We refer to such sentences as customer-to-customer suggestions (CTC). Another type of suggestions, which can appear in the reviews, are the ones aiming at manufactur- ers or service providers, suggesting new features and improvements in products or services. For ex- ample, An electric kettle in the room would have been a useful addition. Recently, there have been some works on extracting the suggestions for im- provements from reviews (section 3), but they did not focus on CTC suggestions. Also, suggestions for improvement discuss only about the reviewed entity and its aspects, unlike suggestions to cus- tomers, which might also include other topics of interest. Suggestion mining and …

Content truncated to 500 words