Hybrid Filtering for Apartment Recommendations with Sentiment and Similarity-Based Rating Augmentation

Urbanization in Jakarta has increased the demand for affordable and well-located housing, driving more residents toward apartment living. However, the rising volume of listings makes it difficult for tenants to find options that suit their specific needs. This study proposes a hybrid recommendation system for the apartment rental market, combining content-based and collaborative filtering to support users with varying levels of interaction history. Preferences such as location, budget, facilities, and lifestyle are collected through a structured form and matched using TF-IDF and cosine similarity. To improve collaborative filtering, user ratings are refined through rule-based sentiment correction and enriched using similarity-based augmentation. The system applies Singular Value Decomposition (SVD) to model user-apartment relationships. Evaluation results show that CF performs well with sufficient history, while the hybrid model provides balanced recommendations for users with limited or changing needs. Compared to prior studies in other domains that rely on complex deep learning methods, this approach achieves competitive accuracy, reaching an F1@10 score of 0.7500, using interpretable techniques and explicit inputs. User feedback confirms its effectiveness in aligning results with budget, location, and facility preferences. Future work includes dynamic hybrid weighting, real-time behavioral feedback, and extend the model with deeper architectures as data grows for real-world deployment.