Accuracy- and consistency-aware recommendation of configurations.

Constraint-based configurators support users in deciding which components and features should be included in a configuration. Due to the increasing size and complexity of configurable products and services, recommender systems are used to personalize the interaction with configurators. Since basic recommendation approaches such as collaborative filtering do not take into account constraints between variable values, recommendations can induce inconsistencies between user requirements and the underlying configuration knowledge base. In this paper, we introduce a constraint-based configuration approach that integrates the results of model-based collaborative filtering (e.g., implemented as feed forward neural network) into constraint solving in such a way that the solver (configurator) is able to determine consistency-preserving and user-relevant configurations.