Using Contextual Information for Service Recommendation

Recommender systems have successfully supported the effective and efficient selection of one product out of the many which meet consumer's needs. Our work extends this work to the area of service recommendation, where we demonstrate the need for using multiple criteria regarding service qualities, and the need to consider multiple contextual dimensions regarding the expected use of that service. This motivates our proposed approach, which uses collaborative filtering and considers both multiple ranking criteria and a number of different context dimensions. The expected sparsity of ranking when handling the contextual information is dealt with by introducing the metric of concept similarity for different context types, and showing how this metric can help reuse data between contexts. At the end of the paper, two rounds of experiments are described. The first shows that considering context produces better predictions, and the second round is used to test our approach for handling sparsity by reusing data between contexts using the similarity metric.