Analisis Perbandingan Algoritma TF-IDF dan Word2Vec dalam Rekomendasi Film Berbasis Konten Letterboxd

Abstrak

The exponential growth of film streaming platforms has created significant challenges for users in discovering content that aligns with their preferences. Recommendation systems have proven to be a strategic solution. This study aims to compare the performance of Term Frequency-Inverse Document Frequency (TF-IDF) and Word2Vec algorithms in the context of content-based filtering recommendation systems, utilizing data from the Letterboxd platform. The research was conducted through a Systematic Literature Review (SLR) of 15 prior studies, with a selection process comprising the definition of inclusion and exclusion criteria, literature searches across academic databases, and comparative synthesis of findings. The review findings indicate that TF-IDF excels in weighting explicit terms within structured metadata such as genre, director, and cast, and can be implemented without a complex training process. Word2Vec, on the other hand, demonstrates superiority in capturing semantic relationships between words through dense vector representations, yet requires a large corpus and careful hyperparameter tuning. Based on the literature synthesis, TF-IDF combined with Cosine Similarity is deemed more optimal for a content-based film recommendation system using Letterboxd metadata, given the structured and explicit nature of the data. Recall@5 of 73% and Recall@10 of 80% reported in prior studies further support this conclusion

Identifier

• OpenAlex: https://openalex.org/W7172289997
• DOI: https://doi.org/10.55606/jutiti.v6i2.7244