Twitter Sentiment Analysis

Requirements

1. nltk
2. colab

Dataset

The original dataset, train.csv has 99899 tweets.

Pre-processing

Pre processing usually depends on the type of data under analysis. For twitter sentiment analysis, preprocessed steps followed are as follows:

1. Removing words containing a particular pattern eg: tweets contained user names like @user1

2. Removing punctuations, numbers and apostrophes

3. Tokenization

4. Fixing the word length and one could also perform spell correction eg: converting juuuusssttttt to just

5. Removing stop words and words with length less than 2

6. Lemmatization

7. Removal of rare and most frequently occuring words

8. Some manual corrections

Models Implemented

1. Naive Bayes (57.23 % acc for testing dataset)
2. Logistic Regression (57.24 % acc for testing dataset)
3. LSTM