1302-GC History-Aware Popularity Calibration in Recommender Systems

Location

https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php

Document Type

Event

Start Date

20-11-2026 2:00 PM

Description

Popularity bias makes recommenders favor a few popular items and neglect the long tail. Calibrated Popularity re-ranks lists to match each user’s own Head/Mid/Tail mix, but uses one strength for all users. We ask whether that strength should depend on rating history. Using MovieLens 1M (1,000,209 ratings, 6,040 users) and Spark ALS in Scala, we compare unadjusted, fixed-calibration, and history-aware recommendations. Phase 1 is complete: the ALS baseline is built and evaluated. Its lists are 64% Tail, versus 14% Tail in users’ histories, with overall UPD 0.327. Phase 2 will test whether history-aware calibration improves relevance at equal UPD, especially for short-history users.

This document is currently not available here.

Share

COinS
 
Nov 20th, 2:00 PM

1302-GC History-Aware Popularity Calibration in Recommender Systems

https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php

Popularity bias makes recommenders favor a few popular items and neglect the long tail. Calibrated Popularity re-ranks lists to match each user’s own Head/Mid/Tail mix, but uses one strength for all users. We ask whether that strength should depend on rating history. Using MovieLens 1M (1,000,209 ratings, 6,040 users) and Spark ALS in Scala, we compare unadjusted, fixed-calibration, and history-aware recommendations. Phase 1 is complete: the ALS baseline is built and evaluated. Its lists are 64% Tail, versus 14% Tail in users’ histories, with overall UPD 0.327. Phase 2 will test whether history-aware calibration improves relevance at equal UPD, especially for short-history users.