MiCRO: Multi-interest Candidate Retrieval Online
Frank Portman, Stephen Ragain, Ahmed El-Kishky

TL;DR
MiCRO is a generative framework designed for personalized candidate retrieval in social media, effectively modeling multi-interest user preferences and temporal item relevance to handle ephemeral content and shifting trends.
Contribution
It introduces a novel generative statistical model that captures multi-interest user preferences and temporal dynamics, improving retrieval in rapidly changing social media environments.
Findings
Strong empirical performance on large-scale datasets
Effective modeling of ephemeral and trending content
Open-sourced datasets for future research
Abstract
Providing personalized recommendations in an environment where items exhibit ephemerality and temporal relevancy (e.g. in social media) presents a few unique challenges: (1) inductively understanding ephemeral appeal for items in a setting where new items are created frequently, (2) adapting to trends within engagement patterns where items may undergo temporal shifts in relevance, (3) accurately modeling user preferences over this item space where users may express multiple interests. In this work we introduce MiCRO, a generative statistical framework that models multi-interest user preferences and temporal multi-interest item representations. Our framework is specifically formulated to adapt to both new items and temporal patterns of engagement. MiCRO demonstrates strong empirical performance on candidate retrieval experiments performed on two large scale user-item datasets: (1) an…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsRecommender Systems and Techniques · Data Mining Algorithms and Applications · Information Retrieval and Search Behavior
