Personalization and recency in retrospective music recommendation

Abstract A personalized music ranking need not contain a strong personal signal. We examined the contributions of account-specific factors and recency to retrospective recommendation of catalogue items absent from earlier matched histories. The training data comprised 35,072 account–item pairs from 729 accounts and 11,729 items. Two model families, three weighting schemes and five seeds produced 30 fits, compared with popularity under chronological validation and final evaluation involving 27 and 19 accounts, respectively. Rankings covered the full eligible catalogue; targets outside the training catalogue remained in metric denominators. Popularity was selected in validation. The personalized challenger achieved final NDCG@10 of 0.0271 versus 0.0153 for popularity, a paired difference of 0.0118 with a descriptive 95% account-bootstrap interval of [-0.0313, 0.0666]. One account benefited, one lost and 17 tied; factor and bias-only recommendation sets were closely aligned. Recency contrasts did not establish a factor-specific benefit. A later exploratory validation analysis found substantially higher NDCG@10 for ItemKNN (0.2621) than popularity (0.0433), while longer training did not consistently improve factor models. These results do not establish a reliable personalization advantage in this small final cohort. The study contributes an auditable comparison that distinguishes account-specific effects, temporal emphasis and catalogue constraints, and identifies item neighbourhoods as a priority for independent chronological evaluation.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-17
DOI
https://doi.org/10.21203/rs.3.rs-11743861/v1
Primary Topic
Recommender Systems and Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Personalization and recency in retrospective music recommendation

Samer Abaddi
Research Square
Recommender Systems and Techniques
preprint

Personalization and recency in retrospective music recommendation

Samer Abaddi
preprint en

Abstract

Abstract A personalized music ranking need not contain a strong personal signal. We examined the contributions of account-specific factors and recency to retrospective recommendation of catalogue items absent from earlier matched histories. The training data comprised 35,072 account–item pairs from 729 accounts and 11,729 items. Two model families, three weighting schemes and five seeds produced 30 fits, compared with popularity under chronological validation and final evaluation involving 27 and 19 accounts, respectively. Rankings covered the full eligible catalogue; targets outside the training catalogue remained in metric denominators. Popularity was selected in validation. The personalized challenger achieved final NDCG@10 of 0.0271 versus 0.0153 for popularity, a paired difference of 0.0118 with a descriptive 95% account-bootstrap interval of [-0.0313, 0.0666]. One account benefited, one lost and 17 tied; factor and bias-only recommendation sets were closely aligned. Recency contrasts did not establish a factor-specific benefit. A later exploratory validation analysis found substantially higher NDCG@10 for ItemKNN (0.2621) than popularity (0.0433), while longer training did not consistently improve factor models. These results do not establish a reliable personalization advantage in this small final cohort. The study contributes an auditable comparison that distinguishes account-specific effects, temporal emphasis and catalogue constraints, and identifies item neighbourhoods as a priority for independent chronological evaluation.

Research Square
Identity (US)
Recommender Systems and Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon