SamWalker++: recommendation with informative sampling strategy

Recommendation from implicit feedback is a highly challenging task due to the lack of reliable negative feedback data. Existing methods address this challenge by treating all un-observed data as (dislike) but downweight confidence these However, treatment causes two problems: (1) Confidence weights unobserved are usually assigned manually, which flexibility and may create empirical bias on evaluating user's preference. (2) To handle massive volume data, most existing rely stochastic inference sampling strategies. since user only aware very small fraction items in large dataset, it difficult for samplers select xmlns:xlink="http://www.w3.org/1999/xlink">informative training instances really dislikes item rather than does not know it. above problems, we propose novel recommendation SamWalker SamWalker++ that support both adaptive assignment efficient model learning. models with social network-aware function, can adaptively specify different according users’ xmlns:xlink="http://www.w3.org/1999/xlink">social contexts . network information be available many recommender systems, hinders application SamWalker. Thus, further SamWalker++, require any side constructed pseudo-social network. In network, similar users connected specific nodes or community nodes. This way, one's benefit knowledge other users. We also develop fast random-walk-based strategies our draw informative instances, speed up gradient estimation reduce variance. Extensive experiments five real-world datasets demonstrate superiority proposed SamWalker++.