Sentiment Analysis for Issues Monitoring Using Linguistic Resources

TALN 2005, Dourdan, 6-10 juin 2005 Sentiment Analysis for Issues Monitoring Using Linguistic Resources Ecaterina Rascu (1), Kai Schirmer (2) and Johann Haller (1) (1) (1) Institut für Angewandte Informationsforschung Martin-Luther-Str. 14, D-66111 Saarbrücken {kati;hans}@iai.uni-sb.de (2) Schirmer Media Research Lietzenburger Str. 90, D-10719 Berlin kaischirmer@web.de Mots-clés : Etude d’opinion, outil de veille économique, classification des opinions Keywords: Sentiment analysis, issues monitoring system, fine-grained sentiment classification Résumé L'identification et l'évaluation des avis, opinions ou jugements exprimés sur un sujet, une entreprise, ou un produit sont des tâches essentielles dans le domaine de l'analyse des médias. L'étude d'opinion est employée pour repérer de nouvelles tendances, mesurer le degré de satisfaction des clients ou pour alerter quand des tendances négatives risquent d'être défavorable à l'image de marque de l'entreprise. Dans cet article nous présentons un outil de veille économique qui permet de classifier très finement des documents publiés en ligne ainsi que d'identifier et d'évaluer les opinions exprimées dans des articles en ligne et des forums de discussions. Après la présentation des diverses composantes du système et des ressources linguistiques utilisées, nous décrivons en détail SentA, la composante d'étude d'opinions, et évaluons sa performance. Abstract Sentiment analysis dealing with the identification and evaluation of opinions towards a topic, a company, or a product is an essential task within media analysis. It is used to study trends, determine the level of customer satisfaction, or warn immediately when unfavourable trends risk damaging the image of a company. In this paper we present an issues monitoring system which, besides text categorization, also performs an extensive sentiment analysis of online news and newsgroup postings. Input texts undergo a morpho-syntactic Ecaterina Rascu, Kai Schirmer and Johann Haller analysis, are indexed using a thesaurus and are categorized into user-specific classes. During sentiment analysis, sentiment expressions are identified and subsequently associated with the established topics. After presenting the various components of the system and the linguistic resources used, we describe in detail SentA1, its sentiment analysis component, and evaluate its performance. 1 Introduction In recent years the tendency to exploit the huge amount of information available on the internet, especially in form of news articles or newsgroup postings for marketing and corporate communication purposes, has increased considerably. Text mining techniques have been developed to find relevant texts, classify them into meaningful clusters and also to extract specific information from them. One of the more sophisticated information extraction tasks, which has proved to be extremely valuable for companies, is the identification of sentiments towards a topic, a company, or a product in online news and newsgroup postings. This is essential for evaluating trends in public opinion, determining the level of customer satisfaction and taking preventive measures in case the level of dissatisfaction risks damaging the image of the company. Various approaches to automatic sentiment evaluation have been proposed in order to efficiently deal with the large amount of available data as well as to reduce the high costs associated with the manual evaluation of such information. On the one hand, statistical …

Content truncated to 500 words