Enhancement of collaborative filtering algorithm with the use of energy-based approach in dating applications

ABSTRACT: The online dating industry is fast evolving, and providing accurate and mutually satisfying matchmarking recommendations has created some challenges. This paper would present an enhanced Energy-based Collaborative Filtering (EBCF) algorithm specific for dating apps, tackling three types of problems predicting mutual interest, understanding dynamic user preferences over time, and the problem of over-recommendations for hyperactive users. Traditional EBCF algorithms are generally biased to a perspective from a single user and thus forget about the reciprocal nature of dating matches. One proposes to combat it with a Bidirectional Energy Scoring mechanism when deciding if an interest is present and by whom. To acknowledge the temporal dynamics in user preferences and give more importance to recent interactions. The researchers introduce a penalty that will reduce the effect of overfitting the model on users who interact the most while at the same time enhancing diversity and inclusively in recommendations. Evaluations of the EBCF algorithm were done on the OkCupid dataset with Mean Squared Error (MSE), Precision, and Recall as metrics. The results evidently show that proposed improvements to matchmaking have resulted in a large increase in accuracy, and Precision improved from 62.29% to 69.59%, with Recall climbing from 67.71% to 94.41% over the baseline approach. Prediction errors are lower and recommendation distribution is balanced showing further improvement in recommendation quality in terms of relevancy and contribution to mutual satisfaction. All of these advances demonstrate that enhanced EBCF has the potential to shift the dating app paradigm through a marked improvement in accuracy and inclusion.

Filipiniana