# BB_twtr at SemEval-2017 Task 4: Twitter Sentiment Analysis with CNNs and LSTMs

**Authors:** Mathieu Cliche  
**arXiv:** [1704.06125](https://arxiv.org/abs/1704.06125)  
**Date:** 2017-04-20  
**Summary page:** https://www.summarizepaper.com/en/paper/1704.06125v1/

## Key points

- Mathieu Cliche created a state-of-the-art Twitter sentiment classifier using CNNs and LSTMs networks.
- The system leverages a large amount of unlabeled data to pre-train word embeddings, which are then fine-tuned using distant supervision on a subset of the unlabeled data.
- The final CNNs and LSTMs are trained on the SemEval-2017 Twitter dataset, where the embeddings are fine-tuned again to further improve performance.
- Several CNNs and LSTMs are ensembled together to boost accuracy.
- This approach achieved first rank on all five English subtasks among 40 teams in SemEval-2017.
- Combining CNNs and LSTMs is an effective approach for sentiment analysis on social media platforms like Twitter.

## Summary

In the paper "BB_twtr at SemEval-2017 Task 4: Twitter Sentiment Analysis with CNNs and LSTMs," Mathieu Cliche describes their successful attempt at creating a state-of-the-art Twitter sentiment classifier using Convolutional Neural Networks (CNNs) and Long Short Term Memory (LSTMs) networks. The system leverages a large amount of unlabeled data to pre-train word embeddings, which are then fine-tuned using distant supervision on a subset of the unlabeled data. The final CNNs and LSTMs are trained on the SemEval-2017 Twitter dataset, where the embeddings are fine-tuned again to further improve performance. Several CNNs and LSTMs are ensembled together to boost accuracy. This approach achieved first rank on all five English subtasks among 40 teams in SemEval-2017. The paper provides detailed information about the methodology used for pre-training word embeddings, fine-tuning them using distant supervision, training CNNs and LSTMs on labeled data, and ensembling multiple models to enhance performance. This study demonstrates that combining CNNs and LSTMs is an effective approach for sentiment analysis on social media platforms like Twitter.