A Tiny Tweak Could Make Your Streaming Suggestions Smarter

Researchers asked whether sequential recommenders really need to learn positional embeddings to encode the order of user interactions, or whether that ordinal signal can be replaced by a structural one derived from the item space. Their proposal: use Laplacian positional embeddings in SASRec. They build an item co-occurrence graph from training interactions, compute eigenvectors of its symmetric normalized Laplacian, and use them as frozen graph-derived positional embeddings. The backbone architecture and training objective stay unchanged. Tested on four public sequential-recommendation benchmarks, this simple replacement improved SASRec on most ranking metrics and remained competitive with strong positional and temporal encoding baselines. The takeaway, according to the article: item-item graph structure can be an effective substitute for standard ordinal positional embeddings in sequential recommendation.