A Contextual Classification Strategy for Polarity Analysis of Direct Quotations from Financial News

Proceedings of Recent Advances in Natural Language Processing, pages 434–440, Hissar, Bulgaria, 12-14 September 2011. A Contextual Classification Strategy for Polarity Analysis of Direct Quotations from Financial News Brett Drury LIAAD-INESC Portugal brett.drury@gmail.com Ga¨el Dias HULTIG, Portugal DLU/GREYC, France ddg@di.ubi.pt Lu´ıs Torgo Fac. Sciences LIAAD-INESC, Portugal ltorgo@inescporto.pt Abstract Quotations from financial leaders can have significant influence upon the immediate prospects of economic actors. Indiscreet or candid comments from senior business leaders have had detrimental effects upon their organizations. Established polarity classification techniques perform poorly when classifying quotations because they display a number of complex linguistic features and lack of training data. The proposed strategy segments the quotations by inferred “opinion maker” role and then applies individual polarity classification strategies to each group of the segmented quotations. This strategy demonstrates a clear advantage over applying classical classification techniques to the whole cor- pus of quotations. While modelling con- textual information with Random Forests based on a vector of unigrams plus the “opinion maker role” reaches a maximum F-measure of 52.85%, understanding the “bias” of the quotation maker previously based on its lexical usage allows 86.23% F-measure for “unbiased” quotations and 71.10% F-measure for “biased” quotations with the Naive Bayes classifier. 1 Introduction Quotations from business leaders or government ministers can have profound effects upon the im- mediate and future prospects of economic actors. This phenomenon was demonstrated in a 1991 speech by Gerald Ratner at the Institute of Di- rectors. He described his company’s products as “crap” (Ratner, 2007) and that a pair of ear- rings sold by his company were “cheaper than a prawn sandwich but probably wouldn’t last as long” (Ratner, 2007). His company (Ratners) lost 500 million pounds in value and had to change its name to Signet to distance itself from his speech. There are other, less colourful, examples of quota- tions impacting the financial prospects of an eco- nomic actor. Mervyn King, the governor of the Bank of England, declared in 2008 that “now seems likely that Britain is entering a recession”1. The day after, the British Pound promptly lost value on the foreign exchange markets. Quotations are arguably an important source of information for researchers trying to determine the financial prospects of an economic actor. How- ever, analysing quotations in terms of conveyed opinion is a non-trivial task because (1) quotations may contain metaphors, euphemisms, slang, ob- scenities, invented words or negations and (2) their polarity mainly depends on the context of the quo- tation and in particular, the opinion maker. Most of the strategies proposed so far for po- larity classification or opinion mining have been focusing on positively or negatively labelling word/phrases (Hatzivassiloglou and McKeown, 1997; Strapparava and Mihalcea, 2008), sentences (Hatzivassiloglou and Wiebe, 2000; Turney, 2002) or texts (Pang et al., 2002; Chesley et al., 2006) in- dependent of their context. Recently, the context of opinion has been ad- dressed and research literature has revealed two different approaches. The first approach was pro- posed by (Al Masum Shaikh et al., 2007). The …

Content truncated to 500 words