What is it about?
Recommender systems have helped companies like Amazon to give consumers access to a much greater variety of products. Products that are rare and less known are also referred to as the `long tail'. Amazon's success is due to making this long tail accessible. Nonetheless, recommenders are trained based on people's preferences, and people generally like popular products more. Therefore recommenders steer people in first instance to things that are popular. This problem is known in jargon as `congestion', and this is especially problematic in settings where items with limited availability are recommended, such as jobs, for which typically only 1 person is hired. In this paper, we propose a new approach called ReCon for reducing congestion in job recommendation systems. ReCon uses the Optimal Transport theory to ensure a more equal spread of job recommendations over job seekers. ReCon is a multi-objective method that reduces congestion while training the job recommendation model.
Featured Image
Photo by JESHOOTS.COM on Unsplash
Why is it important?
The method introduced in this paper reduces congestion in job recommendation, while maintaining an acceptable level of relevance of the recommendations. ReCon is an in-processing approach that optimizes an Optimal Transport loss together with a recommendation model loss to reduce congestion. ReCon offers advantages over post-processing approaches in real-life scenarios that require incremental updates of the model.
Perspectives
Read the Original
This page is a summary of: ReCon: Reducing Congestion in Job Recommendation using Optimal Transport, September 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3604915.3608817.
You can read the full text:
Contributors
The following have contributed to this page