Trip@dvise: an Application Framework for Case-Based Personalized Travel Recommendations

This paper illustrates the motivations and presents the architecture of Trip@dvise, a software framework that supports the development of travel recommender systems. Trip@dvise has been designed to fully support the complex recommendation functions that emerged from the development of two travel advisory applications and to ease the integration of additional innovative web components, such as user log tracking and adaptive user interface. Recommender applications developed with this framework can access and select products and services stored in legacy data repositories, support the user in the query definition task, and recommend single products, or dynamically built coherent bundles of products. Recommendations, in the form of ranked lists of items (products or bundles), are computed exploiting a case-based model of the interaction session and a case base of travel plans developed by a community of travellers. Trip@dvise has been validated in the development of two recommender systems, NutKing and Dietorecs. It supports the user needs, eases the integration of heterogeneous data sources and additional third party components (e.g. user log tracking and adaptive user interface), and allows fast prototyping of innovative recommendation techniques based on machine learning and case-based reasoning