Create README.md

README.md

@@ -0,0 +1,31 @@
+# Dataset Summary
+The Friends dataset consists of speech-based dialogue from the Friends TV sitcom. It is extracted from the [SocialNLP EmotionX 2019 challenge](https://sites.google.com/view/emotionx2019/datasets).
+
+# Supported Tasks and Leaderboards
+text-classification, sentiment-classification: The dataset is mainly used to predict a sentiment label given text input.
+
+# Languages
+The utterances are in English. 
+
+# Dataset Structure
+
+A data point containing text and the corresponding label.
+
+An example from the friends_dataset looks like:
+
+{
+    'text': 'Well! Well! Well! Joey Tribbiani! So you came back huh?',
+    'label': 'surprise'
+}
+
+# Dataset Creation
+
+The dataset contains 1000 English-language dialogues originally in JSON files. The JSON file contains an array of dialogue objects. Each dialogue object is an array of line objects, and each line object contains speaker, utterance, emotion, and annotation strings.
+        {
+            "speaker": "Chandler",
+            "utterance": "My duties?  All right.",
+            "emotion": "surprise",
+            "annotation": "2000030"
+        }
+        
+Utterance and emotion were extracted from the original files into a CSV file. The dataset was cleaned to remove non-neutral labels.