A Light-Weight Approach to Recipient Determination When Recommending New Items

Recommending new items can be a challenging problem in practical applications. Since no past user interactions (e.g., ratings) for such items are available, collaborative filtering techniques cannot be directly applied. Even if item feature information and a content-based approach are used as a fallback, the new items might not show up in the recommendation lists of many users. A possible approach in such situations is to determine a limited set of users to whom the new items are recommended even if there are other, longer existing items that might have a higher assumed relevance. In this paper we present a light-weight approach to this recipient determination problem in the context of the recommendation of job offers on a business social network. At its core, the method uses a nearest-neighbor technique and a set of additional domain-specific heuristics to assess the relevance of new job offers for a user. The evaluation in the context of the ACM RecSys 2017 challenge showed that the method, despite its simplicity, led to good results both in the offline and the online setting.