Article Preview
Top1. Introduction
In the last years, recommender systems have proved their effectiveness to tackle information overload phenomenon (Ricci et al., 2015) which continues to increase with the huge amount of information daily delivered on the Internet. Generally, a recommender or recommendation system (RS) begins by identifying the preferences of the user through her interactions with the RS, and finishes by suggesting recommendations according to these preferences.
There are several approaches to develop the recommender systems, mainly, Collaborative Filtering (CF), Content-Based Filtering (CBF), Knowledge-Based Filtering (KBF) and Hybrid approaches between them (Bobadilla et al., 2013; Ricci et al., 2015; Sohail et al., 2017). In turn, KBF encompasses other development approaches (Aggarwal, 2016a; Sohail et al., 2017), mostly, Case-based (Smyth, 2007) and Constraint-based (Felfernig & Burke, 2008; Felfernig et al., 2015). For both, knowledge-based recommendation (KBR) exploits the explicit user requirements and the deep knowledge about items domain to compute recommendations (Erdenizel al., 2019).
While, the literature abounds with research works that deal with the classical approaches (i.e., CF and CBF), there is little emphasis on knowledge-based recommendation especially for its kind of constraint-based recommender system (CBRS, in short), and only few works exist in this field, which has let CBRSs underutilised in large scale comparatively to collaborative and content-based recommender systems. The recommendation task in classical approaches is formulated around matrix-completion problem, while this is abstracted to satisfy various kinds of constraints in CBRSs. The CBRS approach can overcome some weaknesses of CF and CBF recommender systems such as cold-start and real-time user preferences.
The rarity in research about CBRSs is mainly due to difficulties that meet developers to design the knowledge base about a specific application domain, and which requires absolutely the intervention of domain experts.
In KBR scope, the so-called knowledge acquisition bottleneck (Felfernig et al., 2015) has remained a big challenge, and, for many years, has braked the development of CBRSs to be widespread applications. It means that the knowledge engineers must work to clearly encode the knowledge of domain experts into a formal and executable representation; briefly, they have to design knowledge base with domain experts. The knowledge base about users and items is considered as the backbone in any software architecture proposal for CBRSs, which can be represented by diverse formalisms such as, databases, cases, vectors, ontologies (Bobadilla et al., 2013; Bouraga et al., 2014; Sohail et al., 2017).
Very recently a real-world source of knowledge (Felfernig & Burke, 2008) about Tourism domain has been established by different experts of this domain. It is named “DATAtourisme” (http://www.datatourisme.fr), and based on ontological models benefiting from the advantages of knowledge representation by ontologies (extensibility, expressiveness with semantic richness, sharing consensual conceptualization, reasoning capabilities …) (Boudaa et al., 2017).
The present work gets into the Knowledge-Based Recommendation area and aims to explicit how to build CBRSs. This trend is motivated firstly by the scarcity of researches in this field. Secondly, by the availability of the well-formed knowledge source of DATAtourisme that will facilitate the task of constructing the knowledge base for our case. As far as the authors are aware, no research exists using ontologies in the CBRSs field.
This work details a generic software architecture for the CBRSs development which is validated by implementing a prototype mobile application called DATAtourist (similar to DATAtourisme appellation). The latter is built upon a knowledge base over tourism by using DATAtourisme ontology, and is evaluated under varied usage scenarios.
The remainder of this paper is organized as follows. Theoretical background on CBRSs and DATAtourisme ontology are given in Section 2, while the related work are presented in Section 3. Section 4 introduces the design of our CBRS (DATAtourist recommender system) by detailing the proposed generic software architecture. This architecture is validated by implementing a mobile application and testing it according to three different execution scenarios in Section 5. In section 6, the evaluation of DATAtourist has been discussed, whereas, lessons learnt through this work are given with some research issues in Section 7. Finally, Section 8 concludes this paper by indicating our future research.