Aspect-Based Sentiment Analysis of Financial Headlines and Microblogs Using Semantic Similarity and Bidirectional Long Short-Term Memory

Financial headlines and microblogs usually have sentiment of finance and research further about aspect of sentiment analysis still needed. From dataset FiQA 2018 challenges 4 aspect represent about financial, which is corporate, economy, stock and market. This research proposes method to determine financial headlines and microblogs to financial aspect. A sentence form dataset pre-processed into keyword. Aspect categorization using semantic similarity calculate similarity measurement on word vector of sentence with term list of each aspect, the highest similarity value determines the aspect. Neural network is used for sentiment classification of headlines and microblogs, the method used word embedding and bidirectional long short-term memory (BLSTM). The sentiment classification results are combined with the aspect categorization results to determine which aspects have the highest positive sentiment. The highest aspect categorization performance is obtained combined semantic similarity and bidirectional long short-term memory which reach 88.1% and semantic classification accuracy reach 77.0%. The stock aspect has received more positive sentiment compared to the sentiment of other aspect, this can be used for started investment in stock.