Time based Sentiment Analysis of Financial Headlines using Recurrent Neural Network

The sentiment of financial news headlines plays an important role in understanding market trends and investor behavior. This study proposes a Recurrent Neural Network (RNN)-based model for accurately classifying the sentiment of financial headlines into positive, neutral, and negative categories. Keeping in mind the time trend based behavior of the financial world, and the impact of certain keywords relevant only in the financial context, the RNN architecture captures the contextual nuances often overlooked by traditional methods. To address the domain-specific challenges of financial language and the inherent trends based on the time series based data, the model aims to incorporate embeddings that are fine-tuned on financial text along with a capacity to capture time based context. Experiments conducted on a dataset of financial stocks for a period from 2003 to 2020 help demonstrate the effectiveness of the proposed RNN compared to other benchmark methods. It provides a result with 97% accuracy and accurately captures the context of verbal and time based sentiment context.