In point cloud analysis, point-based methods have rapidly developed in recentyears. These methods have recently focused on concise MLP structures, such asPointNeXt, which have demonstrated competitiveness with Convolutional andTransformer structures. However, standard MLPs are limited in their ability toextract local features effectively. To address this limitation, we propose aVector-oriented Point Set Abstraction that can aggregate neighboring featuresthrough higher-dimensional vectors. To facilitate network optimization, weconstruct a transformation from scalar to vector using independent angles basedon 3D vector rotations. Finally, we develop a PointVector model that followsthe structure of PointNeXt. Our experimental results demonstrate thatPointVector achieves state-of-the-art performance $\textbf{72.3\% mIOU}$72.3% mIOU on theS3DIS Area 5 and $\textbf{78.4\% mIOU}$78.4% mIOU on the S3DIS (6-fold cross-validation)with only $\textbf{58\%}$58% model parameters of PointNeXt. We hope our work willhelp the exploration of concise and effective feature representations. The codewill be released soon.