The "Affective Text" task, part of SemEval-2007, focuses on classifying emotions and valence (positive/negative polarity) in news headlines. The task aims to explore the connection between emotions and lexical semantics. The dataset consists of news headlines from major sources, with six predefined emotion labels (Anger, Disgust, Fear, Joy, Sadness, Surprise) and valence annotations. The task was conducted in an unsupervised setting, with teams allowed to use any resources, including a set of words from WordNet Affect. The evaluation included fine-grained and coarse-grained metrics, with inter-annotator agreement studies showing moderate correlation. Five teams participated, each with unique approaches, such as rule-based systems, word-space models, and machine learning techniques. The results indicate that emotion annotation is challenging, suggesting room for future improvements.The "Affective Text" task, part of SemEval-2007, focuses on classifying emotions and valence (positive/negative polarity) in news headlines. The task aims to explore the connection between emotions and lexical semantics. The dataset consists of news headlines from major sources, with six predefined emotion labels (Anger, Disgust, Fear, Joy, Sadness, Surprise) and valence annotations. The task was conducted in an unsupervised setting, with teams allowed to use any resources, including a set of words from WordNet Affect. The evaluation included fine-grained and coarse-grained metrics, with inter-annotator agreement studies showing moderate correlation. Five teams participated, each with unique approaches, such as rule-based systems, word-space models, and machine learning techniques. The results indicate that emotion annotation is challenging, suggesting room for future improvements.