Dialogue Relation Extraction (DRE) is a task that aims to identify all entity pair relations within a dialogue. However, it is difficult to establish direct associations between inter-sentence entity pairs and the lack of path information makes identifying inter-sentence entity pair relations challenging. To address this issue, we proposes an effective inference model that constructs an entity co-occurrence graph of dialogue documents to model inter-sentence entity pair associations, incorporates multiple entity pair information to enrich path semantics, and employs attention mechanism to capture the semantics associated with each relation in the path information. These enhancements lead to improved inter-sentence inference and increase the effectiveness of dialogue relation extraction.