Knowledge-aware Recommendations: Exploring the Interplay between Utility, Explanation Quality, and Fairness in Path Reasoning Methods

Adopting Knowledge Graphs (KGs) in recommender systems has engendered the emergence of sophisticated techniques, such as path reasoning, designed to navigate KGs and model complex relationships. KGs enable the representation of intricate connections, while path reasoning approaches adeptly learn to traverse these graphs, constructing detailed user-product relationships by discerning reasoning paths linking recommended products with those previously experienced by users. These identified paths are subsequently converted into well-articulated textual explanations, facilitating a deeper and more comprehensive understanding for the users. Despite its potential, the field is hindered by disparate and insufficient evaluation protocols, complicating efforts to assess the impact of existing methodologies. In this paper, we summarize our previous work on replicating and evaluating three state-of-the-art path reasoning recommendation approaches, originally presented at prestigious conferences, using a standardized protocol based on two publicly available datasets and benchmarking them against other knowledge-aware techniques. Our analysis encompasses recommendation utility, explanation quality, and fairness considerations for both consumers and providers. This investigation offers a comprehensive overview of the progress in the field, emphasizing key challenges and potential avenues for future exploration. Source code is available at https://github.com/giacoballoccu/rep-path-reasoning-recsys.