Twitter sentiment analysis of the dataset taken from Kaggle competition (https://www.kaggle.com/c/tweet-sentiment-extraction).
In the social media, Twitter has been the most engaging platform for a long time and many companies, political personalities, celebrities have their presence in twitter which makes it a great place for having discussions by millions of users about various topics ranging from laws passed by the parliament to the new movie releases. This makes it interesting to analyse the sentiments of the tweets to see if a tweet by a certain user has positive or negative impact on the community. Going a step further, we can also find which words of the tweet contribute to the sentiment. The analysis of public reaction can be easily done using the sentiment analysis and the keyword extraction of the tweets. In this project, the sentiment analysis of tweets using various deep learning algorithms is tested and their performance using different metrics was calculated. Also a method to perform keyword extraction of tweets was explored (https://www.kaggle.com/yutanakamura/dear-pytorch-lovers-bert-transformers-lightning). The dataset that was considered encompasses a broad set of tweets which are classified into 3 different types of sentiments, namely, positive, negative and neutral.
The deep learning algorithms tested are listed as follows:
- Multilayer Perceptron (MLP)
- Convolutional neural network (CNN)
- Recurrent neural network (RNN)
- Long short-term memory (LSTM)
- Gated recurrent unit (GRU)
- Bi-directional LSTM (Bi-LSTM)
- RoBERTa (Transformer architecture)
- BERT
- XLNet
Note - The report of the experiments is included in this repo.