Dialogue-Oriented Review Summary Generation for Spoken Dialogue Recommendation Systems

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 64–72, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Dialogue-Oriented Review Summary Generation for Spoken Dialogue Recommendation Systems Jingjing Liu, Stephanie Seneff, Victor Zue MIT Computer Science & Artificial Intelligence Laboratory 32 Vassar Street, Cambridge, MA 02139 {jingl, seneff, zue}@csail.mit.edu Abstract In this paper we present an opinion summari- zation technique in spoken dialogue systems. Opinion mining has been well studied for years, but very few have considered its appli- cation in spoken dialogue systems. Review summarization, when applied to real dialogue systems, is much more complicated than pure text-based summarization. We conduct a sys- tematic study on dialogue-system-oriented review analysis and propose a three-level framework for a recommendation dialogue system. In previous work we have explored a linguistic parsing approach to phrase extrac- tion from reviews. In this paper we will de- scribe an approach using statistical models such as decision trees and SVMs to select the most representative phrases from the ex- tracted phrase set. We will also explain how to generate informative yet concise review summaries for dialogue purposes. Experimen- tal results in the restaurant domain show that the proposed approach using decision tree al- gorithms achieves an outperformance of 13% compared to SVM models and an improve- ment of 36% over a heuristic rule baseline. Experiments also show that the decision-tree- based phrase selection model can achieve ra- ther reliable predictions on the phrase label, comparable to human judgment. The pro- posed statistical approach is based on do- main-independent learning features and can be extended to other domains effectively. 1 Introduction Spoken dialogue systems are presently available for many purposes, such as weather inquiry (Zue et al., 2000), bus schedules and route guidance (Raux et al., 2003), customer service (Gorin et al., 1997), and train timetable inquiry (Eckert et al., 1993). These systems have been well developed for laboratory research, and some have become commercially viable. The next generation of intelligent dialogue sys- tems is expected to go beyond factoid question answering and straightforward task fulfillment, by providing active assistance and subjective recom- mendations, thus behaving more like human agents. For example, an intelligent dialogue sys- tem may suggest which airline is a better choice, considering cost, flight duration, take-off time, available seats, etc.; or suggest which digital cam- era is the most popular among teenagers or highest rated by professional photographers; or which res- taurant is a perfect spot for a semi-formal business meeting or a romantic date. Luckily, there are enormous amounts of reviews published by general users on the web every day. These are perfect resources for providing subjec- tive recommendations and collective opinions. If there exists a systematic framework that harvests these reviews from general users, extracts the es- sence from the reviews and presents it appropriate- ly in human-computer conversations, then we can enable dialogue systems to behave like a human shopping assistant, a travel agent, or a local friend who tells you …

Content truncated to 500 words