Opinion Mining in Newspaper Articles by Entropy-Based Word Connections

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1828–1839, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Opinion Mining in Newspaper Articles by Entropy-based Word Connections Thomas Scholz and Stefan Conrad Heinrich-Heine-University Institute of Computer Science D-40225 D¨usseldorf, Germany {scholz,conrad}@cs.uni-duesseldorf.de Abstract A very valuable piece of information in news- paper articles is the tonality of extracted state- ments. For the analysis of tonality of newspa- per articles either a big human effort is needed, when it is carried out by media analysts, or an automated approach which has to be as accu- rate as possible for a Media Response Anal- ysis (MRA). To this end, we will compare several state-of-the-art approaches for Opin- ion Mining in newspaper articles in this pa- per. Furthermore, we will introduce a new technique to extract entropy-based word con- nections which identifies the word combina- tions which create a tonality. In the evalua- tion, we use two different corpora consisting of news articles, by which we show that the new approach achieves better results than the four state-of-the-art methods. 1 Introduction The Web keeps many potentially valuable opinions in news articles which are partly new online articles or uploaded print media articles. Many companies or organisations such as political parties or even dis- tinguished public figures perform a Media Response Analysis (MRA) (Watson and Noble, 2007) in order to analyse the output of their effort in public rela- tions. So, an opinion-oriented analysis of news arti- cles is important, because the tonality (Watson and Noble, 2007; Scholz et al., 2012a) is the key indi- cator of a MRA. A purely manual solution implies a big human effort for so-called media analysts, be- cause they have to read and rate approx. 200 to 800 news articles each week. As a consequence, an automated Opinion Mining solution is very attractive. At the same time, Opin- ion Mining in newspaper articles appears to be dif- ficult, because not all parts of news articles are as subjective (Balahur et al., 2010) as reviews, for ex- ample. Also, different parts of one article can con- tain different opinions (Watson and Noble, 2007). Therefore, we work with extracted statements of news articles, in which a sequence of consecutive sentences has the same tonality value. At the same time, some approaches focus more on differentiating only between positive and negative news and leave out neutral examples (Taboada et al., 2011; Scholz et al., 2012b). Conversely, we have noticed that even if the used words in the news domain are quite similar, the tonality which the words express can be differ- ent, especially if neutral examples are involved (cf. section 3.1). We propose this task formulation: Problem definition: Let s ⊆ d be a statement and document d represents a newspaper article. The task is to determine the tonality y for a given statement s, consisting of k words: t : s = (w1, w2, ..., wk) �→ y ∈ {positive,neutral,negative} (1) Normally, a statement …

Content truncated to 500 words