Learning to Learn a Cold-start Sequential Recommender

The cold-start recommendation is an urgent problem in contemporary online applications. It aims to provide users whose behaviors are literally sparse with as accurate recommendations possible. Many data-driven algorithms, such the widely used matrix factorization, underperform because of data sparseness. This work adopts idea meta-learning solve user’s problem. We propose a meta-learning-based sequential framework called metaCSR, including three main components: Diffusion Representer for learning better user/item embedding through information diffusion on interaction graph; Sequential Recommender capturing temporal dependencies behavior sequences; and Meta Learner extracting propagating transferable knowledge prior good initialization new users. metaCSR holds ability learn common patterns from regular users’ optimize so that model can quickly adapt after one or few gradient updates achieve optimal performance. extensive quantitative experiments datasets show remarkable performance dealing user Meanwhile, series qualitative analysis demonstrates proposed has generalization.