Preference reasoning with soft constraints in constraint-based recommender systems

Abstract

A recommender system (RS) supports online users in e-commerce by proposing products that are assumed to be both useful and interesting. Knowledgebased recommendation systems form one branch of these online sales support systems that is particularly relevant for high-involvement product domains like consumer electronics, financial services or tourism. A constraint-based RS is a specific variant of a knowledge-based RS that builds on a CSP formalism for problem representation and solving. This article formalizes the different variants of a constraint-based recommendation problem based on consistency and the empirical part compares the performance of different constraint-based recommendation mechanisms in offline experiments on historical data.

Multi-criteria Ratings for Recommender Systems: An Empirical Analysis in the Tourism Domain

“…The original publication is available at www.springerlink.com! In the RS literature different interpretations of knowledge are common [1], for instance, explicitly encoded knowledge bases that contain sets of logical sentences [2] and case-based recommendation approaches that require domain knowledge for defining similarity functions [3]. The first two families of algorithms tackle recommendation as a Machine Learning task, where from known training data models are learned and subsequently employed to predict unseen or withheld data in a second step.…”

“…The first two families of algorithms tackle recommendation as a Machine Learning task, where from known training data models are learned and subsequently employed to predict unseen or withheld data in a second step. In contrast knowledge-based recommendation in the sense of [2] does not follow an inductive learning approach but requires explicitly coded expertise, for instance, in the form of business rules and constraints that should constitute plausible heuristics and procedures in the eyes of experienced sales personnel. For more details on recommender systems and the genesis of the research field see [1,4].…”

PB-ADVISOR: A private banking multi-investment portfolio advisor

“…In particular, when compared with [30], the increase in F1 by 14% highlights the benefits of the proposed system. This increase is more than patent when compared with [79] (PB-ADVISOR improves by 160% F1 figures) and [80], (174%). However, PB-ADVISOR, due to the intrinsic complexity of the functional scenario is below the achievements of [71] (62%) and [14] (20%), where recommendations are more accurate.…”

“…In this research, the best results obtained within the experiments performed are 9.21% for precision, 27.58% for recall and 13.81% for F1. In this context of e-commerce, Zanker et al [80], formalize the different variants of a constraint-based recommendation problem based on consistency and compare the performance of different constraint-based recommendation mechanisms in offline experiments on historical data using Precision and Recall metrics. In this case, the Precision, Recall and F1 values obtained by the authors are 23.80%, 9.12% and 13.09%, respectively.…”

“…Both Experts (Table 5) 51.70 0.258 PB-ADVISOR vs. Correct cell (Table 6) 14.70 0.233 PB-ADVISOR vs. Any expert cell (Table 7) 22.90 0.359 Portfolio [30] 32.00 0.320 Conversion rates [79] 27.58 0.138 E-Commerce [80] 9.12 0.131 Consumer Behaviour Prediction [71] 91.00 0.968 Movies [14] 29.00 0.450 provide different recommendations for the same client in the same context. The precision of PB-ADVISOR with the investment categories is near 83%.…”