From theaters to streaming: The influence of algorithms on film culture and consumption

Abstract

This study explores the socio-cultural evolution of cinema in the digital age, emphasizing the shift from collective theatrical viewing to personalized, algorithm-driven consumption on streaming platforms and social media. Between January and December 2024, we collected and pre-processed 331,354 English-language user reviews from Letterboxd via a Python-based web-scraping pipeline. Reviews were cleaned (stop-word removal, lemmatization, tokenization), truncated to 512 tokens, and then subjected to sentiment classification using the CardiffNLP “Twitter-XLM-RoBERTa-base-sentiment” model. Subsequently, we developed a prototype recommendation system employing collaborative filtering, deep-learning embeddings, and cosine-similarity metrics to simulate real-world content suggestions. Our sentiment analysis reveals a near balance of positive (108,280), neutral (116,011), and negative (107,063) reactions to films depicting artificial intelligence, underscoring both audience fascination and ethical ambivalence. The recommendation model demonstrates how personalized suggestions can accelerate content discovery while fostering filter bubbles that reinforce existing preferences and limit serendipitous encounters. Despite these algorithmic constraints, online film communities (e.g., Letterboxd) sustain cinema’s traditional social function through collective discussion and user-driven curation. These findings contribute to debates on media democratization, algorithmic governance, and cultural stratification, and they offer practical insights for designing more transparent, user-centric recommendation engines that balance personalization with diversity.