Abstract: Decrease in the cost of acquiring 3D point cloud data coupled with the rapid advancements in GPU computing power have resulted in an increased demand for 3D point cloud semantic segmentation in numerous 3D visual applications, including but not limited to autonomous driving, industrial control, and MR/XR, which further advances the development of deep learning methods in 3D point cloud semantic segmentation. Recently, many novel deep learning network architectures, such as RandLA-Net and Point Transformer, have been proposed and have achieved notable improvements in semantic segmentation accuracy while decreasing the computational load. However, previous research on 3D point cloud semantic segmentation methods has focused primarily on relatively early works, whose approaches have been gradually abandoned over the years and cannot accurately reflect the current research status. Moreover, the existing methods have been categorized based on their input data types, making it difficult to compare the segmentation performance of different techniques and not providing a comprehensive view of the relationship between methods using different network architectures. Therefore, this paper reviews the mainstream 3D semantic segmentation methods developed in the last three years using different deep learning network architectures and is organized into three levels. First, the two principal 3D point cloud data acquisition methods, including their customary datasets and metrics to evaluate model performance, are introduced. Second, a systematic review of 3D semantic segmentation methods based on different network architectures is organized, followed by a statistical analysis of the evaluation of performance between different models on two 3D segmentation datasets—S3DIS and ScanNet. The analysis of model performance on these two commonly used datasets includes model structure relevance, strengths, and limitations. Finally, an insightful discussion of the remaining methodological and application challenges and potential research directions is provided. This paper offers an extensive overview of the recent three-year research progress in 3D point cloud semantic segmentation and summarizes various network architecture pipelines, elucidates their fundamental operations, compares the model performance across multiple architectures, discusses their notable strengths and limitations, most importantly, concludes the current challenges and promising research directions for future investigations. Furthermore, this paper enables researchers to effortlessly identify the relevant research and research hotspots among different 3D point cloud semantic segmentation methods based on the analyses presented and aims to update the reviews on 3D point cloud semantic segmentation methods with a better viewpoint and highlight key properties and contributions of proposed methods, providing promising research directions for the main challenges.