The rapid expansion of the video game industry makes it difficult for users to find games that match their preferences. Previous studies have developed recommender systems to assist this process. However, many still face personalization issues. A key limitation is the underutilizing of user reviews, which often contain valuable emotional insights. As a result, recommendations may lack relevance to individual user needs. To address this, we propose a hybrid recommender system that integrates content-based filtering (CB), item-based collaborative filtering (CF), and sentiment analysis using BERT. This study introduces a novel integration of BERT-based sentiment analysis into a hybrid recommender system combining CB and item-based CF, aiming to enhance both personalization and emotional relevance. Sentiment analysis captures users’ emotional opinions, which are combined with CB and item-based CF similarity scores using grid search-optimized weighting. The system is developed using publicly available data from the Steam platform. Evaluation results show strong performance, achieving an NDCG@10 score of 0.9283 reflects a high overall ranking quality, an MRR@10 score of 0.9327 highlights that the first relevant item appears very early in the recommendation list, and a HitRate@10 score of 0.9980 indicates high effectiveness in identifying at least one relevant item. These results demonstrate that integrating sentiment analysis not only improves accuracy but also the personal alignment of the recommendations. This leads to a more engaging and satisfying recommendation experience.