Abstract:For many recommender systems, the primary data source is a historical record of user clicks. The associated click matrix is often very sparse, as the number of users x products can be far larger than the number of clicks. Such sparsity is accentuated in cold-start settings, which makes the efficient use of metadata information of paramount importance. In this work, we propose a simple approach to address cold-start recommendations by leveraging content metadata, Metadata Alignment for cold-start Recommendation. We show that this approach can readily augment existing matrix factorization and autoencoder approaches, enabling a smooth transition to top performing algorithms in warmer set-ups. Our experimental results indicate three separate contributions: first, we show that our proposed framework largely beats SOTA results on 4 cold-start datasets with different sparsity and scale characteristics, with gains ranging from +8.4% to +53.8% on reported ranking metrics; second, we provide an ablation study on the utility of semantic features, and proves the additional gain obtained by leveraging such features ranges between +46.8% and +105.5%; and third, our approach is by construction highly competitive in warm set-ups, and we propose a closed-form solution outperformed by SOTA results by only 0.8% on average.

What problem does this paper attempt to address?

### Problems Addressed by the Paper The paper aims to address the cold-start problem in recommendation systems. Specifically, the cold-start problem refers to how to provide relevant recommendations when new users or new items first appear on the platform. Traditional collaborative filtering techniques rely on past user interaction data, which is unavailable for new users or new items, thus existing optimized systems perform poorly in cold-start scenarios. ### Main Contributions 1. **Proposed a new algorithm (MARec)**: - MARec combines embeddings learned from item and customer metadata with the user-item click matrix and provides closed-form update equations. - The algorithm achieves state-of-the-art performance on multiple benchmark datasets, with training times several orders of magnitude faster than the best baselines. 2. **Outperformed existing methods on four cold-start benchmark datasets with different sparsity and scale characteristics**: - Performance improvements range from +8.4% to +53.8%. 3. **Investigated the practicality of large language model (LLM) embeddings**: - By leveraging semantic features, additional performance improvements range from +46.8% to +105.5%. 4. **Approached state-of-the-art performance in warm-start settings**: - Proposed a closed-form solution, with average performance on warm-start datasets only 0.8% lower than the state-of-the-art results. ### Method Overview MARec consists of three main components: 1. **Backbone Model**: - Learns low-dimensional representations of user-item interaction data and their reconstruction. - Can choose well-performing collaborative filtering algorithms from the matrix factorization and autoencoder families. 2. **Embedding Model**: - Encodes dense representations of item metadata, bringing items with similar interaction patterns closer in the embedding space. 3. **Alignment Model**: - Aligns interaction data with item-item similarity estimated through the embedding model, thus fusing metadata representations with click history. ### Experimental Results - **Cold-Start Datasets**: - Experiments were conducted on Amazon Video Games, Netflix, MovieLens10M, and MovieLens Hetrec datasets, which have different sparsity and scale characteristics. - MARec significantly outperformed existing methods on all datasets. - **Warm-Start Datasets**: - Experiments were conducted on Pinterest and MovieLens1M datasets, validating the effectiveness of MARec in warm-start scenarios. ### Conclusion MARec effectively addresses the cold-start problem in recommendation systems by incorporating metadata information and also performs well in warm-start scenarios. The method not only achieves significant performance improvements on multiple benchmark datasets but also has high training efficiency, making it practically valuable.

MARec: Metadata Alignment for cold-start Recommendation

CoMeta: Enhancing Meta Embeddings with Collaborative Information in Cold-start Problem of Recommendation.

Learning to Learn a Cold-start Sequential Recommender

Improving the Performance of Cold-Start Recommendation by Fusion of Attention Network and Meta-Learning

Content-Aware Few-Shot Meta-Learning for Cold-Start Recommendation on Portable Sensing Devices

Task Aligned Meta-learning based Augmented Graph for Cold-Start Recommendation

RGMeta: Enhancing Cold-Start Recommendations with a Residual Graph Meta-Embedding Model

Cold-Start Sequential Recommendation with Collaborative Boosted Meta Transitional Learning

Cold-start Sequential Recommendation via Meta Learner

Solving the Cold-Start Problem in Recommender Systems with Social Tags

AdaMO: Adaptive Meta-Optimization for cold-start recommendation

CMML: Contextual Modulation Meta Learning for Cold-Start Recommendation

Metric Learning for Tag Recommendation: Tackling Data Sparsity and Cold Start Issues

AdaML: an Adaptive Meta-Learning Model Based on User Relevance for User Cold-Start Recommendation.

An autoencoder-based recommendation framework toward cold start problem

Cross-Domain Meta-Learner for Cold-Start Recommendation

Learning to Warm Up Cold Item Embeddings for Cold-start Recommendation with Meta Scaling and Shifting Networks

Online Item Cold-Start Recommendation with Popularity-Aware Meta-Learning

Attention-Based Adaptive Model To Unify Warm And Cold Starts Recommendation

An improved sequential recommendation model based on spatial self-attention mechanism and meta learning

Task-Difficulty-Aware Meta-Learning with Adaptive Update Strategies for User Cold-Start Recommendation