This document summarizes two recent papers on sentiment analysis and emotion detection using emojis: 1. The first paper presented a method using millions of emoji occurrences on Twitter to learn representations that can detect sentiment, emotion, and sarcasm without requiring labeled training data. It achieved state-of-the-art results on several benchmarks. 2. The second paper introduced an interpretable emoji prediction model using label-wise attention LSTMs. This approach is less biased towards frequent emojis and provides visualizations of the attention weights to analyze model predictions. It was shown to better predict infrequent emojis compared to standard attention models.