What is it about?
The problem of maximum inner product search (MIPS) is essential in recommender systems (RS). However, it does not consider diversity, although this is also an important factor for user satisfaction in RS. This work proposed a reasonable way of incorporating diversity into the MIPS problem. Moreover, we propose an efficient algorithm that can scale a large number of items, so that users can obtain a recommendation list in an interactive manner.
Featured Image
Photo by Alexander Grey on Unsplash
Why is it important?
Diversity is still an important factor, because it prevents users from leaving services. How to effectively incorporate diversity is not trivial. We address this issue. In addition, as the number of items in recommendation applications is growing, dealing with such bigdata is an essential requirement in modern systems. Our algorithm overcomes this challenge while satisfying the diversity requirement.
Perspectives
This is the first work that shows how to "simply" incorporate diversity into the MIPS problem. One merit of our problem formulation is that how to generate vectors representing user-item and item-item is black-box. That is, applications can employ existing embedding techniques based on their requirements. Therefore, we hope that this work helps applications that require diversity without changing embedding models.
Daichi Amagata
Read the Original
This page is a summary of: Solving Diversity-Aware Maximum Inner Product Search Efficiently and Effectively, September 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3523227.3546779.
You can read the full text:
Resources
Contributors
The following have contributed to this page
