The deeper penetration of business-to-consumer e-commerce requires that customer decision support systems (CDSS) serve a wider range of users. However, a significant weakness of existing e-shopping assistance programs is their inability to aid non-professional consumers (non-prosumers) in buying highly differentiated products. This paper proposes a novel framework that infers product recommendations with minimal information input. At the heart of the proposed framework is the feature-usage map (FUM), a Bayesian network-based model that encodes the correlations among a product’s technical specifications and its suitability in terms of its using scenario (usage). It also incorporates a query-based lazy learning mechanism that elicits a product’s rating score from product reviews and constructs its corresponding FUM in an on-demand manner. This mechanism allows the knowledge base to be enriched incrementally, with no need for an exhaustive repository of FUMs pertaining to all possible usage queries a user may invoke. The effectiveness of the proposed framework is evaluated through an empirical user study. The results show that the framework is able to effectively derive product ratings based on specified usage. Moreover, this rating information can also be incorporated into a conventional buying guide system to deliver purchase decision support for non-prosumers