MELD stands for Multimodal EmotionLines Dataset, which was created by enhancing and extending the EmotionLines dataset.

The MELD dataset contains the same conversation instances in EmotionLines, but it also contains data from other modalities (audio, video, and text). The MELD dataset contains more than 1,400 conversations and 13,000 sentences from the TV series Friends. These conversations involve multiple speakers.

Each sentence in the conversation is labeled with one of seven emotions: anger, disgust, sadness, happiness, neutral, surprise, and fear. The dataset also provides sentiment annotations (positive, negative, and neutral) for each sentence.