Abstract
Letterboxd’s movie ratings influence millions, yet its scoring algorithm is opaque. We investigate the discrepancy between the platform’s displayed score and the true user rating average, which we define as Δ. Analyzing a corpus of 1,737 Brazilian films and over 1.3 million ratings, we uncover the factors driving this distortion. Our analysis reveals a systematic algorithmic compression that pulls extreme scores toward the mean, with a strong negative correlation (−0.903) between a film’s true rating and its Δ. Using K-Means, we identify four distinct rating distribution profiles (e.g., Polarized, Highly-Rated) and demonstrate that these profiles, along with genre, are significant predictors of the score adjustment. Niche genres like documentaries and musicals, which often exhibit polarized or extremely high ratings, are penalized most heavily. Furthermore, we find that popularity acts as a stabilizer; as a film’s rating count increases, the magnitude of Δ decreases. Taken together, these results indicate that Letterboxd employs a normalization mechanism that mitigates the influence of outlier patterns, potentially fostering more representative aggregate scores and enhancing comparability across films. This study proposes greater transparency in these algorithms that shape cultural consumption.

This publication has 10 references indexed in Scilit: