Dataset Folder contains the twitter tweets.
Train file contain tweet_id, sentiment and tweet_text.
- tweet_id : unique for every tweet.
- sentiment : three types - negative, neutral and positive.
- tweet_text : tweets over which we have to analyse the sentiment.
test_sample data has two columns : tweet_id and tweet_text
- tweet_id: unique for every tweet
- tweet_text: sentiment over which we have to predict whether this text is negative, neutral or positive.
We were only able to predict the result and don't have the actual result of test sample as this analysis was performed over a Kaggle competition where sentiment result for tweet_text was not available.
Kaggle link : https://www.kaggle.com/c/sentiment-analysis-of-tweets/data
Three approaches were taken:
- Logistic Regression
- LSTM with Glove word embedding
- Bidirectional LSTM with Glove word embedding
Accuracy acheived on Kaggle test_sample:
- Logistic Regression - 65%
- LSTM with Glove word embedding - 67.5%
- Bidirectional LSTM with Glove word embedding - approx 67.