Combining Embedding-Based and Semantic-Based Models for Post-hoc Explanations in Recommender Systems

TL;DR

The paper tackles the interpretability gap in embedding-based recommender systems by proposing a hybrid framework that combines knowledge-graph embeddings with ontology-driven semantic reasoning for post-hoc explanations. The approach splits tasks into high-accuracy top-n recommendations using relation-type subgraphs and per-relation embeddings, and instance-level explanations via RDF subgraphs and SPARQL-based reasoning, with a similarity-based explanation metric $Sim(a_{s1},a_{s2})$ used to quantify user-item alignment. Key contributions include a two-track architecture, per-relation KG embedding via node2vec, a LambdaMART ranking step, and flexible explanation formats (radar, tables, natural language) demonstrated on vehicle-domain data. This work advances practical explainability in recommender systems by delivering interpretable rationale alongside strong recommendations, improving user trust and potential adoption in e-commerce settings, and is scalable to real-world datasets. $I$, $U$, $D$, and $G=(E,R)$ formalize the data and knowledge structures at the core of the method, with $r(u,i)$ and $Sim(a_{s1},a_{s2})$ underpinning ranking and explanations respectively.

Abstract

In today's data-rich environment, recommender systems play a crucial role in decision support systems. They provide to users personalized recommendations and explanations about these recommendations. Embedding-based models, despite their widespread use, often suffer from a lack of interpretability, which can undermine trust and user engagement. This paper presents an approach that combines embedding-based and semantic-based models to generate post-hoc explanations in recommender systems, leveraging ontology-based knowledge graphs to improve interpretability and explainability. By organizing data within a structured framework, ontologies enable the modeling of intricate relationships between entities, which is essential for generating explanations. By combining embedding-based and semantic based models for post-hoc explanations in recommender systems, the framework we defined aims at producing meaningful and easy-to-understand explanations, enhancing user trust and satisfaction, and potentially promoting the adoption of recommender systems across the e-commerce sector.

Figures (4)

• Figure 1: Schematic diagram of the approach for building a post-hoc explanation in a recommender system, employing embedding-based and semantic-based models.
• Figure 2: Snapshot of a section related to instances of users and items - vehicles within the ontology-based knowledge graph
• Figure 3: Feature importance derived from the training process
• Figure 4: Three different formats for explanations to end-users