The Zero-Shot Object Navigation (ZSON) task requires the agent to find the target object in unfamiliar environments without prior explicit training. Existing ZSON methods prompt LLM with the semantic of each frontier, which lacks sufficient spatial and semantic information during the exploration process for reasoning. They also utilize the semantics between the observed environment and the target as the sole source for exploration decisions. Inspired by the phenomenon that humans driven by curiosity, in addition to the target, during exploring the environment, this paper proposes USCE, a novel ZSON framework that unifies semantic and curiosity exploration. USCE represents the observed scene by both RRT-Pruning topology map constructed by the shortest path from agent position to each frontier waypoint and semantic map to preserve the refined information of the environment. Benefiting from the topological and continuous semantic representation of the scene, USCE fully exploit the common-sense reasoning ability of LLM. Extensive evaluation on MP3D and HM3D validates that USCE surpasses existing benchmarks in both Success Rate and exploration efficiency (absolute improvement: + 3.2% SR and + 0.6% SPL on MP3D, + 1.8% SR and + 1.4% SPL on HM3D).