User-Item Feature Graph for Content Based Recommendations of Japanese Anime Shows

Graph induced from 10000 Japanese animation shows, where nodes are shows and edges are a) user written show reviews with ratings and b) user-written show-to-show recommendations. Each show has textual and non-textual features. The data has been split for train/val/test on both warm start, where recommended items have existing interactions, and cold start, where they don't, scenarios.