Bundle recommendation: State-of-art
Generating and Personalizing Bundle Recommendations on Steam? This paper helps understand semantics of what constitutes a ‘good’ bundle, in order to recommend item sets to users on the basis of constituent products. A new dataset from a video game distribution platform provides both ‘traditional’ recommendation data (rating and purchase histories between users and items), as well as bundle purchase information. Features such as bundle size and item compatibility are shown to be more effective when combined with traditional matrix factorization techniques,?
Product bundle identification using semi-supervised learning A multi-modal classifier leverages listing's title, image, and attributes as features for bundle identification. Challenges include? a small set of manually-labeled clean examples and a larger set of noisy-labeled examples, in conjunction with class imbalance due to the relative scarcity of bundles. Experiments with manually-labeled and/or the noisy-labeled data for training demonstrate only moderate performance. A self-training ensemble-based, semi-supervised algorithm with a greedy model selection instead performs better over two different meta-categories.
There is however no guarantee that products in a bundle may actually be related to each other. Traditional approaches that use mostly transactional data are extended herein by incorporating both domain knowledge from product suppliers in the form of hierarchies by learning a deep similarity model that leverages textual attributes.?
A high-quality bundle generalizes frequent items of interest, and diversity across bundles boosts the user-experience and eventually increases transaction volume. This problem is formalized as structured prediction decomposed by determinantal point processes (DPPs). Two datasets used are generated from co-purchase records and the other two extracted from real-world online bundle services.
Related posts