Stocks Sentiment Analysis NLP LSTM
-Humungous data on social media about a stock which is impossible to go through and make decisions.
-Leverage technology to gain interesting insights and make better investment decisions.
-Based on the signals of the model investors can make confident and well informed decisions.
-Time saved by 95% by leveraging technology.
-Avoid decision fatigue and analysis paralysis.
-Collection of data from twitter through twitter API.
-Structuring and making sense of unstructured and noisy text data.
-Data cleaning and preprocessing.
-For positive class the words : long, bullish, buy call, high, hold are very common.
-For negative class short, bearish, buy put, low very common.
-Both the classes have some common words like volume which can imply both sentiments and add noise in the data.
-Pandas, Numpy, Matplotlib.py.
-Seaborn and Plotly Express.
-Wordcloud.
-NLTK.
-IEEE dataport + twitter.
-Data cleaning.
-Removing Punctuations and stopwords.
-Stemming and Tokenization.
-Split the data into 90% train and 10% test.
-Performed Stemming, tokenisation , encoding and embedding to reduce feature dimensions
-Built a simple custom LSTM network for Sentiment analysis with the following configuration
-Optimizer used was ADAM
-Loss taken was categorical cross entropy and metrics taken as accuracy.