Albeit, the implicit feedback based recommendation problem - when only theuser history is available but there are no ratings - is the most typicalsetting in real-world applications, it is much less researched than theexplicit feedback case. State-of-the-art algorithms that are efficient on theexplicit case cannot be straightforwardly transformed to the implicit case ifscalability should be maintained. There are few if any implicit feedbackbenchmark datasets, therefore new ideas are usually experimented on explicitbenchmarks. In this paper, we propose a generic context-aware implicit feedbackrecommender algorithm, coined iTALS. iTALS apply a fast, ALS-based tensorfactorization learning method that scales linearly with the number of non-zeroelements in the tensor. The method also allows us to incorporate diversecontext information into the model while maintaining its computationalefficiency. In particular, we present two such context-aware implementationvariants of iTALS. The first incorporates seasonality and enables todistinguish user behavior in different time intervals. The other views the userhistory as sequential information and has the ability to recognize usagepattern typical to certain group of items, e.g. to automatically tell apartproduct types or categories that are typically purchased repetitively(collectibles, grocery goods) or once (household appliances). Experimentsperformed on three implicit datasets (two proprietary ones and an implicitvariant of the Netflix dataset) show that by integrating context-awareinformation with our factorization framework into the state-of-the-art implicitrecommender algorithm the recommendation quality improves significantly.
展开▼
