Abstract
In the context of the booming digital economy, recommendation systems, as akey link connecting users and numerous services, face challenges in modelinguser behavior sequences on local-life service platforms, including the sparsityof long sequences and strong spatio-temporal dependence. Such challenges can beaddressed by drawing an analogy to the forgetting process in human memory. Thisis because users' responses to recommended content follow the recency effectand the cyclicality of memory. By exploring this, this paper introduces theforgetting curve and proposes Spatio-Temporal periodic Interest Modeling (STIM)with long sequences for local-life service recommendation. STIM integratesthree key components: a dynamic masking module based on the forgetting curve,which is used to extract both recent spatiotemporal features and periodicspatiotemporal features; a query-based mixture of experts (MoE) approach thatcan adaptively activate expert networks under different dynamic masks, enablingthe collaborative modeling of time, location, and items; and a hierarchicalmulti-interest network unit, which captures multi-interest representations bymodeling the hierarchical interactions between the shallow and deep semanticsof users' recent behaviors. By introducing the STIM method, we conducted onlineA/B tests and achieved a 1.54\% improvement in gross transaction volume (GTV).In addition, extended offline experiments also showed improvements. STIM hasbeen deployed in a large-scale local-life service recommendation system,serving hundreds of millions of daily active users in core applicationscenarios.