Textual Demand Analysis: Detection of Users’ Wants and Needs from Opinions

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 409–416 Manchester, August 2008 Textual Demand Analysis: Detection of Users’ Wants and Needs from Opinions Hiroshi Kanayama Tetsuya Nasukawa Tokyo Research Laboratory, IBM Japan, Ltd. 1623-14 Shimotsuruma, Yamato-shi, Kanagawa-ken, 242-8502 Japan {hkana,nasukawa}@jp.ibm.com Abstract This paper tackles textual demand analy- sis, the task of capturing what people want or need, rather than identifying what they like or dislike, on which much conven- tional work has focused. It exploits syn- tactic patterns as clues to detect previously unknown demands, and requires domain- dependent knowledge to get high recall. To build such patterns we created an unsuper- vised pattern induction method relying on the hypothesis that there are commonly de- sired aspects throughout a domain corpus. Experimental results show that the pro- posed method detects twice to four times as many demand expressions in Japanese discussion forums compared to a baseline method. 1 Introduction Increasingly we can access many opinions towards products, services, or companies through elec- tronic documents including questionnaires, call logs, and other consumer-generated media (CGM) such as Internet discussion forums and blogs. It is very important for companies to get insights from their customers’ opinions by analyzing such docu- ments in large numbers. The most popular way to utilize such data has involved sentiment analysis (SA) (Nasukawa and Yi, 2003; Yi et al., 2003), which is the task of rec- ognizing the writers’ feelings as expressed in pos- itive or negative comments. Typically, a SA sys- tem focuses on expressions to identify the strong c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. or weak points of the subjects as in (1) or in the writers’ evaluations as in (2). (1) I think the pictures are beautiful. (2) I don’t like this camera very much. Here we call them polar expressions because they convey positive or negative polarities. By counting the polar expressions related to products or services, one can quantitatively compare the goodness of competing services, find the draw- backs of specific products, and so on. In addition to polar expressions, there are other types of expressions that provide valuable infor- mation, especially for the supplier side rather than the consumer side. Examples (3) and (4) express the demands of the writers. (3) I’d be happy if it is equipped with a crisp LCD. (4) I’m waiting for a single-lens reflex less than 30,000 yen to come on the market. We call such expressions “demand expres- sions”, and the underlined phrases “demand tar- gets.” While sentiment analysis reveals evaluations of existing products or services, the task proposed here, textual demand analysis1, gives more direct suggestions to companies: things they should do to attract customers. For example, by investigating demand targets, companies can add new functions to products on the market or plan new services to satisfy customers. These activities should lead to positive evaluations in the future. Interestingly, demand expressions may be noise in sentiment analysis, because the demand …

Content truncated to 500 words