Film recommendation systems have become an integral part of the modern entertainment industry, helping users find content that suits their preferences amid advances in digital technology. The recommendation system generally uses several methods such as collaborative filtering (CF), content-based filtering (CBF), and hybrid filtering. One of the main problems in the development of recommendation systems is the cold start problem, which occurs when the system must recommend items to new users or when user preference information is very limited. This issue can reduce the accuracy of recommendations and provide a less than satisfactory user experience. This research will be carried out by comparing several word embeddings such as Word2Vec, Glove, BERT, XLNet, RoBERTa, TF-IDF and GPT-2 which will be applied in hybrid filtering. The use of word embedding techniques such as Word2Vec, Glove, BERT, XLNet, RoBERTa, TF-IDF, and GPT-2 is becoming increasingly important in improving the performance of recommendation systems by understanding the context and semantics of textual data. Previous research has shown that the use of word embedding can improve the accuracy of recommendations. For example, research using RoBERTa produced an MAE of 0.0012 and an RSME of 0.001. This study aims to determine the effect of word embedding on the film recommendation system using hybrid filtering. By using a combination of word embedding techniques such as Word2Vec, Glove, BERT, XLNet, RoBERTa, TF-IDF, and GPT-2, it is expected to improve the accuracy of recommendations and provide insight into the effectiveness of each approach. This research also aims to provide guidance for the development of a better recommendation system in other areas in the future.