Cold-start Bundle Recommendation via Popularity-based Coalescence and Curriculum Heating

TL;DR

The proposed CoHeat (Popularity-based Coalescence and Curriculum Heating) addresses the highly skewed distribution of bundle interactions through a popularity-based coalescence approach, which incorporates historical and affiliation information based on the bundle's popularity.

Abstract

How can we recommend cold-start bundles to users? The cold-start problem in bundle recommendation is crucial because new bundles are continuously created on the Web for various marketing purposes. Despite its importance, existing methods for cold-start item recommendation are not readily applicable to bundles. They depend overly on historical information, even for less popular bundles, failing to address the primary challenge of the highly skewed distribution of bundle interactions. In this work, we propose CoHeat (Popularity-based Coalescence and Curriculum Heating), an accurate approach for cold-start bundle recommendation. CoHeat first represents users and bundles through graph-based views, capturing collaborative information effectively. To estimate the user-bundle relationship more accurately, CoHeat addresses the highly skewed distribution of bundle interactions through a popularity-based coalescence approach, which incorporates historical and affiliation information based on the bundle's popularity. Furthermore, it effectively learns latent representations by exploiting curriculum learning and contrastive learning. CoHeat demonstrates superior performance in cold-start bundle recommendation, achieving up to 193% higher nDCG@20 compared to the best competitor.

Figures (6)

• Figure 1: (a) Extremely skewed distribution of bundle interactions in real-world datasets (data statistics are summarized in Table \ref{['table:datasets']}). (b-c) For an unpopular bundle, user-bundle view provides insufficient information while user-item view provides sufficient information.
• Figure 2: Performance comparison of CoHeat with competitors on Youshu, NetEase, and iFashion datasets, evaluated by Recall$@20$. We mark cold-start methods as orange, and warm-start methods as red. The cold-start methods typically sacrifice warm setting performance to excel in cold settings. The warm-start methods show poor performance in cold settings. CoHeat demonstrates superior performance over existing methods in both cold and warm settings.
• Figure 3: Overview of CoHeat (see Section \ref{['sec:proposed']} for details).
• Figure 4: Learning mechanism of CoHeat (see Section \ref{['subsec:discussion']} for details).
• Figure 5: Performance comparison by cold bundle ratio.

Theorems & Definitions (4)

• lemma 1
• lemma 2
• lemma 3
• lemma 4