Financial sentiment analysis of tweets based on deep learning approach

Abstract

The volume of unstructured texts has increased dramatically in recent years due to the internet and the digitization of information and literature. This onslaught of data will only grow, and it will come from new and unusual sources. Thus, it will be necessary to develop new and inventive approaches and tools to process and make sense of this data. Investors in the financial markets can now get information faster than ever before thanks to the expansion of communication channels, in addition to the online availability of news and reports in text format through providers like Reuters and Bloomberg. This contains a plethora of information that is often overlooked by financial market data. In order to measure the sentiment of a text, predictive and deductive methods are applied, these methods aim at extrapolating new feautures from big data. The main objective of this study is to create and test a new system capable of predicting finance and non-finance related tweets. The convolutional neural network (CNN) and latent dirichlet allocation (LDA) algorithms are used in the proposed approche. The suggested model's correctness is tested against a benchmark financial dataset, and the results demonstrate that with a database of 1,000,000 data points, our model is 99% accurate.