The task of inductive knowledge graph completion (KGC) aims to predict missing facts involving entities unseen during training. Since polysemous relations are prevalent in knowledge graphs, existing approaches often present relations with entirely different embeddings under different contexts. However, since the semantics of polysemous relations are multiple but related, neglecting their inherent correlations may degrade the performance of existing methods. To address this issue, this paper proposes a novel Path-based Polysemy Relation Message Propagation Network (PPRMP) for inductive KGC based on the paths between entities. Our method is built upon a Message Passing Neural Network (MPNN) framework, which can efficiently compute relation information along the paths connecting a source entity and target entities through iterative message propagation. To model the polysemy of relations, it learns a global embedding for each relation and computes context-dependent semantic variations conditioned on the query relation during multi-hop path reasoning. Then, it combines the global and context-dependent representation to yield contextualized embeddings that reflect multiple yet related semantics. Extensive experiments show that our PPRMP outperforms existing state-of-the-art methods in two benchmark datasets.