Multi-Entity Polarity Analysis and Detection of Subjectivity in Financial Documents
Keywords:Anaphora Resolution, Detection of Subjectivity, Machine Learning, Sentiment Analysis, Web Data Anno- tation
Polarity analysis aims at classifying an author’s opinion as positive, negative, or neutral. However, given the sheer volume of information available on the web, manually carrying out such task is unfeasible. In particular, this type of analysis is useful for companies when making decisions related to the financial market, which is particularly prone to changes according to shifting moods and opinions. Most studies in the literature deal with this problem by considering that whole documents have a single, global polarity. However it is not unusual that documents have opinions on several entities with possibly different polarities. This suggests that the polarity classification should be performed in an entity level. We also noted that many financial documents may not emit any opinion. Therefore, in this paper we propose a supervised polarity classification method based on multiple models and detection of subjectivity, in order to deal with financial documents that cite multiple entities. Our results showed that the hierarchical, multiple-models approach significantly outperformed the global-model baseline.