Data Science Recommender Systems Accuracy and Beyond

The project at hand is to build a recommender system that can provide users with personalized joke recommendations. In addition to optimizing for recommendations that are accurate, the recommender also focuses on maximizing user experience by adding some level of serendipity to the recommendations - meaning users will receive unexpected items with the potential to be enjoyable within their most highly recommended items. Initially, we investigate the performance of five recommendation algorithms: User-Based Collaborative Filtering (UBCF) using cosine and Pearson similarity Item-Based Collaborative Filtering (IBCF) using cosine and Pearson similarity A Random recommender algorithm (a baseline) For each method, we conduct a 3-fold cross-validation to evaluate precision and performance. After we establish the best performing algorithm, we promote novel items in the recommendation list by replacing a selection of the most recommended items with more novel items. In this approach we try to balance accurate recommendations with user engagement and discovery. If you find this work useful, please cite it using the DOI: 10.5281/zenodo.16777408