KEMB-Rec: Knowledge-Enhanced Explainable Multibehavior Recommendation With Graph Contrastive Learning

In the era of Internet of Things (IoT), intelligent recommendation systems are crucial components for users to locate the items they require. Existing recommendation systems overlook the diversity of user behaviors and rely solely on utilizing a singular form of user-item interaction data. Multibehavior recommendation (MBR) works to solve this problem by utilizing multityped user behaviors to mine the heterogeneous relations between users and items to improve recommendation accuracy. Nevertheless, there are still challenges to be overcome, including capture of differences and commonalities between different types of behaviors, learning of users’ personalized behavioral patterns, consideration of semantic knowledge, and building of users’ trust in algorithms. In light of the aforementioned considerations, we propose a knowledge-enhanced explainable MBR model (KEMB-Rec) with graph contrastive learning, comprising two modules. The first is the user behavior-aware module, which mines user’s behavior pattern using the user behavior hyper meta-graphs and captures the differences and commonalities between different behaviors through graph contrastive learning. The second is the semantic knowledge-aware module, which is based on single behavior interaction graphs to mine the semantic relational knowledge, and makes full use of it to represent users and items. Then, we design contrastive learning task and recommendation task, and the two tasks are optimized jointly. At the same time, effective recommendation explanations are provided by mining paths and semantics between users and items as a way to enhance user trust and satisfaction. The proposed KEMB-Rec is evaluated in real-world data sets, with results indicating that KEMB-Rec outperforms various baselines.