Twitter Sentiment Analysis

Sentiment analysis on multi-lingual tweets with accuracy 0.801

Task

This is an entity-level sentiment analysis dataset of twitter, which could be found by this link:
https://www.kaggle.com/jp797498e/twitter-entity-sentiment-analysis

Given a message and an entity. The task is to judge the sentiment of the message about the entity.
There are three classes in this dataset: Positive, Negative and Neutral. Messages that are not relevant regarded as Neutral.

Notebooks

• tf-idf-twi-sentiment-analysis.ipynb - Notebook with the best accuracy score, reached with LinearSVC (accuracy = 0.801)
• tf-idf-twi-sentiment-analysis_2.ipynb- Notebook with feature engineering and preprocessing steps and LinearSVC (accuracy = 0.774)
• LSTM.ipynb Notebook with LSTM Network (accuracy = 0.54)
• sentiment-analysis-using-transformers.ipynb - Notebook with Transformers (accuracy = 0.503)
• binary-sentiment-analysis-using-transformers.ipynb - Notebook with binary classification and Transformers - 2 classes: Positive, Negative (accuracy = 0.815)
• binary-pos-neutr-sentiment-analysis-transformers.ipynb - Notebook with binary classification and Transformers - 2 classes: Positive, Neutral (accuracy = 0.779)