Movie Feelings is a dataset of movie emotion features designed to systematically characterize the fine-grained emotions evoked by movies. It breaks through the limitations of traditional datasets that only classify positive and negative emotions or basic emotions. It is widely used in the construction of movie recommendation systems, audience rating prediction, emotion modeling research, and comparative evaluation of NLP and large language models in emotion understanding tasks. It can also be used to analyze the correlation between different emotion patterns and movie reviews, providing important data support for computational social science and film and television data analysis. This dataset contains 1,500 representative and culturally influential films, spanning from 1920 to 2024, covering 50 emotional states including doubt, fear, calm, disgust, solidarity, and joy. The dataset includes basic film information, rating data, plot text, and three independent emotional feature systems (F1/F2/F3): F1 is a continuous feature system based on film reviews, while F2 and F3 are the top 3 discrete emotional labels generated based on model inference.

Emotional Characteristic System

• F1 (Review-Based NLP, Emotion Features): Based on 880,000 movie review texts, it generates 50-dimensional continuous emotion intensity features (0–1) by matching and statistically analyzing an emotion thesaurus.
• F2 (Transformer, emotional features for plot understanding): Uses BART for zero-shot emotion classification and combines it with RoBERTa for semantic reordering to model the understanding of movie plot summaries, selecting the top-3 emotions from 50 emotions as binary features.
• F3 (GPT-4o, Generative Emotion Features): Based on GPT-4o, it directly infers emotions from the movie plot, generates Top-3 emotion labels from a limited set of 50 emotions, and converts them into binary features.