Wishful Thinking - Finding suggestions and ’buy’ wishes from product reviews

Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text, pages 54–61, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Wishful Thinking Finding suggestions and ‘buy’ wishes from product reviews J. Ramanand Krishna Bhavsar Niranjan Pedanekar BFS Innovations BFS Innovations BFS Innovations Cognizant Technology Solutions Cognizant Technology Solutions Cognizant Technology Solutions Pune, India Pune, India Pune, India ramanand.janardhanan @cognizant.com krishna.bhavsar @cognizant.com niranjan.pedanekar @cognizant.com Abstract This paper describes methods aimed at solv- ing the novel problem of automatically dis- covering ‘wishes’ from (English) documents such as reviews or customer surveys. These wishes are sentences in which authors make suggestions (especially for improvements) about a product or service or show intentions to purchase a product or service. Such ‘wishes’ are of great use to product managers and sales personnel, and supplement the area of sentiment analysis by providing insights into the minds of consumers. We describe rules that can help detect these ‘wishes’ from text. We evaluate these methods on texts from the electronic and banking industries. 1 Introduction Various products and business services are used by millions of customers each day. For the makers of these products & services, studying these customer experiences is critical to understanding customer satisfaction and making decisions about possible improvements to the products. Thanks to the ad- vent of weblogs, online consumer forums, and product comparison sites, consumers are actively expressing their opinions online. Most of these reviews are now available on the web, usually at little or no cost. Moreover, these are available for a variety of domains, such as financial services, tele- com services, consumer goods etc. Automated analysis of opinions using such re- views could provide a cheaper and faster means of obtaining a sense of such customer opinions, thus supplementing more traditional survey methods. In addition, automated analysis can significantly shorten the time taken to find insights into the cus- tomer's mind and actions. Sentiment analysis of texts such as product re- views, call center notes, and customer surveys aims to automatically infer opinions expressed by people with regards to various topics of interest. A sentiment analysis exercise classifies the overall opinion of a review document into positive, neu- tral, or negative classes. It may also identify senti- ments at a finer granularity, i.e. recognizing the mix of opinions about the topic(s) expressed in the text. However, industry analysts (Strickland, 2009) report some common problems with the results of these exercises: 1. The results (usually numerical scores split across positive, negative, neutral classes) are hard to meaningfully interpret. 2. These results are more useful to certain roles and domains. Brand, reputation, and service managers in media and retail industries find senti- ment analysis more useful than product managers or sales teams in various industries. 3. The results do not ‘indicate user action’ i.e. opinions do not help identify a future action of the author based on the comments. An example of this is: does the consumer indicate that he intends to stop …

Content truncated to 500 words