Generating Popularity-Aware Reciprocal Recommendations Using Siamese Bi-Directional Gated Recurrent Units Network

Abstract

Reciprocal Recommender Systems (RRS) use bilateral preferences to satisfy the needs of both the parties involved. Popularity bias is a critical problem in RRS, which arises when the RRS tend to prefer few popular users over others and thus exhibit biased behavior towards popular users. It can have adverse effect on both the parties involved. Popular users may become swamped by the requests received from a large number of users and cease to acknowledge, which can make it difficult for other users to make contact. To address this challenge, we propose Popularity-aware Siamese Bi-directional Gated Recurrent Units (PSBiGRU) with the proposed popularity-aware reciprocal score (ParS)-based re-ranking that uses semantic similarity between explicit user profiles. The proposed model is evaluated on two reciprocal environments, namely, online recruitment and online dating. Experimental findings demonstrate that PSBiGRU surpasses the compared state-of-the-art methodologies and illustrate its viability.

Keywords

• Reciprocal recommender system
• Siamese neural network
• online recruitment
• online dating
• popularity bias