Automated Recommender System Integration in Model-Based Ecosystems

Recommender systems (RSs) are a special class of information systems devoted to assisting end-users by providing valuable items according to the context of the application. Nevertheless, implementing those systems is time-consuming as developers must elicit the proper technology that may vary according to the context. Even though existing frameworks have been proposed to facilitate the deployment of those systems, automated integration with existing model-based systems is still an open challenge. In this paper, we propose a model-driven engineering approach to automatically configure filtering-based RSs by relying on automated transformations and weaving modelling. In particular, we rely on two different metamodels to represent the application domain (variable) and the generic components of filtering-based RSs. Then, we define a weaving model to map each domain entity to a specific RS element e.g., algorithm or data encoding, thus realizing a conceptual link between the context and the system to be deployed. The proposed framework is then able to generate domain-specific model-based recommendations, that can be integrated in model-based systems. We used the proposed framework to implement two different types of filtering-based RSs evaluated using two different scenarios, i.e., movies and tourism recommendations. Our findings suggest that the proposed approach can be adopted to simplify the data ingestion of RS-based filtering and automate the generation of model-based recommendations.