Optimized long short-term memory-based stock price prediction with sentiment score

Sentiment analysis examines the emotional content of a statement, such as views, assessments, feelings, or attitudes about a topic, human, or object. Emotions can be categorized as either unbiased, good, or bad. It determines how people feel about the company online through social media. Based on the sentiments, the problem of solving the stock price prediction model is advantageous as it involves the sentiment score evaluated from the text information. This work introduces a new stock price prediction considering sentiment scores from text info in this concern. For that, we have considered news data and stock data. Moreover, this work falls under bigdata perspective by increasing the data size. The proposed model includes two major steps: feature extraction and prediction. Feature extraction takes place under two scenarios: features from news data and features from stock data. Features like Bag of words, n-Gram, TFIDF, and Improved cosine similarity are extracted from the news data, and features like improved exponential moving average and other existing technical indicator-based features such as ATR, TR are extracted from stock data. Both the feature sets are fused to determine the final prediction results. Particularly, this final observation involves the sentiments from the given news data. For this, optimized LSTM model is used, where the optimal training process will be carried out by a new Harris Hawks Induced Sparrow Search Optimization via tuning the optimal weights. The proposed model is the combination of Harris Hawks Optimization Algorithm and Sparrow Search Algorithm, respectively. Finally, the performance of proposed work will be evaluated over the other conventional models with respect to different measures.