Recommender similarity can be phantom correlation from user activity
Original title: Do Similarity Measures Measure Similarity? How User Activity Creates Phantom Correlation in Recommender Systems
Independent consumption with unequal volume still yields strong item associations under a Rasch model.
The paper argues that a substantial share of item-item similarity in recommender systems comes from differences in how much users consume, not from shared taste. Even if every user picks items independently and differs only in total volume, standard similarity measures still find strong associations. The authors call this phantom correlation and show that independent consumption under varying activity is a Rasch model, then derive how association depends on item popularity for each measure. Comparing that baseline with real benchmark datasets indicates how much popular similarity measures reflect the artifact in collaborative filtering.
- Phantom correlation is defined as association with no preference structure.
- The formal baseline is a Rasch model of independent consumption.
- Association strength depends on item popularity and the similarity measure used.
- The theoretical baseline is compared with real-world benchmark datasets.
From the original article
1 min read at arxiv.orgItem-item similarity is one of the oldest and most foundational signals in recommender systems: two items are assumed to be similar when the same users consume them. Despite the sparsity and scale of interaction data, the implicit assumption across the field is that item co-occurrence reflects genuine shared preference, capturing what users like together.
By Marco Favier · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.