Abstract: Self-organizing feature map algorithm is one of the most important types of unsupervised learning method based on neural network technology but it tends to get stuck in local minima and yield the topological defect problem which limits its practical usage. Based on K-means algorithm, the paper presents a new self-organizing feature map algorithm to simplify its algorithm structure and speed up its calculation. First, a new simplified calculation structure is designed for the new algorithm based on analyzing working principle of self-organizing feature map algorithm; Second the presented algorithm combines self-organizing feature map algorithm and K-means algorithm and presents some improvements including redefining the information representation method and distance between two seeds and redesigning new calculation flow. Finally, the presented algorithm is realized and applied to evaluate university art course teaching evaluation and the realization results indicate that the presented algorithm has better performance in evaluation speed and accuracy and the algorithm can be used for system evaluation practically.