Fine-grained, aspect-based semantic sentiment analysis within the economic and financial domains

The application of sentiment analysis in financial and economic applications has attracted great attention in recent years. News and social media represent a valuable source of information, that is timely available and potentially able to improve the forecast of economic and financial time series. Despite many successful applications of sentiment analysis in these domains, the range of natural language processing techniques employed is still very limited. In this work, we detail the technical presentation of a fine-grained aspect-based semantic sentiment analysis algorithm and check its performance with respect to a humanly annotated data set. The proposed approach is completely unsupervised and relies on a large custom-specific domain lexicon and on a thorough semantic polarity scheme, allowing a better interpretation and explanation of the analysis. Our method shows promising re-suits, with the proposed algorithm assigning a similar sentiment score as human annotators in the large majority of cases.