Most existing temporal knowledge graph reasoning (TKGR) methods primarily model structural dependencies and temporal dynamics from historical facts, while fact-aligned event semantics, such as hierarchical event types and background contexts, are often underexploited. This may result in incomplete semantic characterization of relations and entities, making it difficult to distinguish semantically similar event relations and contextually plausible entity candidates. To investigate this issue, this paper proposes ES-TKGR, an event-semantics-enhanced model for TKGR, which constructs and integrates event semantics aligned with TKG facts to model implicit dependencies among facts. Specifically, ES-TKGR consists of three key components. First, the Event Type Hierarchical Constraint Encoding (ETHCE) module incorporates hierarchical event type semantics to strengthen relation modeling by introducing a hierarchical distance-aware prototype contrastive objective over relation representations, while promoting semantic interaction among facts through explicit type-aware connections. Second, the Static Knowledge Graph Encoding (SKGE) module incorporates event background semantics by constructing a global static graph from historical facts and aligning it with background knowledge, thereby capturing time-independent global interaction patterns. Third, an adaptive fusion module integrates dynamic features from ETHCE and static features from SKGE to obtain unified representations. To evaluate future entity and relation prediction under fact-aligned event semantics, we construct four ICEWS-based TKG datasets aligned with event semantic information. Experimental evaluation demonstrates that ES-TKGR outperforms or remains competitive with recent baselines, underscoring the effectiveness of fact-aligned event semantics for enriching entity and relation representations and modeling implicit fact dependencies in TKGR.
