A Comparative Study of Open Domain and Opinion Question Answering Systems for Factual and Opinionated Queries

International Conference RANLP 2009 - Borovets, Bulgaria, pages 18–22 A Comparative Study of Open Domain and Opinion Question Answering Systems for Factual and Opinionated Queries Alexandra Balahur, Ester Boldrini, Andrés Montoyo, Patricio Martínez-Barco Natural Language Processing and Information Systems Group, Department of Software and Computing Systems. Apartado de Correos 99, 03080, Alicante {abalahur, eboldrini, montoyo, patricio}@dlsi.ua.es Abstract The development of the Web 2.0 led to the birth of new textual genres such as blogs, reviews or forum entries. The increasing number of such texts and the highly diverse topics they discuss make blogs a rich source for analysis. This paper presents a comparative study on open domain and opinion QA systems. A collection of opinion and mixed fact-opinion questions in English is defined and two Question Answering systems are employed to retrieve the answers to these queries. The first one is generic, while the second is specific for emotions. We comparatively evaluate and analyze the systems’ results, concluding that opinion Question Answering requires the use of specific resources and methods. Keywords Question Answering, Multi-perspective Question Answering, Opinion Annotation, Opinion Mining, Non- Traditional Textual Genres. 1. Introduction Recent years’ statistics show that the number of blogs has been increasing at an exponential rate. A research of the Pew Institute [1] shows that 2-7% of Internet users created a blog and that 11% usually read them. Moreover, researches in different fields proved that this new textual genre is a valuable resource for large community behavior analysis, since blogs address a great variety of topics from a high diversity of social spheres. A common belief is that they are written in a colloquial style, but [2] shows that the language of these texts is not restricted to the more informal levels of expression and a large number of different genres are involved. As a consequence, free expressions, literary prose and newspaper writing coexist without a clear predominance. When using this textual genre, people tend to express themselves freely, using colloquial expressions employed only in day-by-day conversations. Moreover, they can introduce quotes from newspaper articles, news or other sources of information to support their arguments, make references to previous posts or the opinion expressed by others in the discussion thread. Users intervening in debates over one specific topic are from different geographical regions and belong to diverse cultures. All the abovementioned features make blogs a valuable source of information that can be exploited for different purposes. However, due to their language being heterogeneous, it is complex to understand and formalize in order to create effective Natural Language Processing (NLP) tools. At the same time, due to the high volume of data contained in blogs, automatic NLP systems are needed to manage the language understanding and generation. Analyzing emotions and/ or opinions expressed in blog posts could also be useful to predict people’s opinion or preferences about a product or an event. One of the other possible applications is an effective Question Answering (QA) system, able to recognize different queries and give the correct answer …

Content truncated to 500 words