research - Quantitative Research
Does consumer sentiment change which kinds of movies succeed? Six ways of representing a plot, tested on 5,388 films.
A completed INST664 study asking whether consumer sentiment changes which kinds of movies succeed. I compared six ways of representing film plot summaries (genre labels, LDA topics, SBERT embeddings with PCA and K-means, lexicon affect, and theory-driven semantic axes) and modeled within-year box-office percentile against content-by-sentiment interactions on 5,388 films (1978–2012), with Benjamini-Hochberg FDR correction within each method. There was no aggregate sentiment effect, but there were real content-conditional shifts, written up as a full paper.
Compared six plot representations: genre labels, LDA topics, SBERT embeddings with PCA and K-means, lexicon affect, and theory-driven semantic axes.
Modeled box-office percentile against content-by-sentiment interactions on 5,388 films (1978–2012), with a TMDb-augmented 6,961-film sensitivity check.
No aggregate sentiment effect, but real content-conditional shifts: teen-setting stories gained relative advantage in higher-sentiment periods (q = 0.02).