<p>Geological data exhibit significant multimodal characteristics, and the accurate association between spatial entities and descriptive texts is crucial for geological knowledge discovery and Geographic Information System (GIS) data integration. This study proposes a deep Entity‑Text Matching Model designed to simultaneously address two critical tasks: binary classification matching and one‑to‑many ranking retrieval. First, the attribute textualization is applied to achieve semantic representation of geological spatial entities, coupled with Geohash spatial encoding. Subsequently, the pre-trained language model is employed to extract deep semantic features. Finally, the Enhanced Sequential Inference Model (ESIM) is introduced for fine-grained semantic interactions inference between descriptive texts and spatial entities. Experiments were conducted on a real-world geological dataset show that the model achieves an F1 score of 0.962 in binary classification and an NDCG@10 of 0.927 in ranking retrieval, significantly outperforming existing baselines. Ablation studies further confirm the necessity of each module. This work provides a unified and effective solution for cross-modal retrieval in geology.</p>