Sentiment Analysis of Financial News Using Unsupervised and Supervised Approach

Sentiment Analysis aims to extract sentiments from a piece of text. In addition to numeric data, sentiments are being increasingly favored as inputs to decision making process. However extracting meaning automatically from unstructured textual inputs involves a lot of complexities. These often depend on the domain from which the text was taken. Our work focuses particularly on extracting sentiments from financial news. We have proposed and implemented a framework using unsupervised and supervised techniques. We have proposed a hybrid approach of using seed sets for calculating the semantic orientation of news articles in a semi-automatic way. This approach produces better results than the standard techniques used in unsupervised sentiment analysis. Then we performed the experiment using the supervised approach with machine learning classifier Support Vector Machine (SVM). We compared the results with those produced from the standard unigram and unigram + bigram approaches and found that the proposed approach produces better precision.