Abstract: The research on humanoid problem solving and its understanding of the problem has attracted attention from scholars all over the world. Nowadays, it has gradually become a hot spot in artificial intelligence applications. However, although the theory of automatic problem solving has made progress, it is hard to improve further that is rooted in the accuracy of problem meaning understanding. This paper reviewed the current research progress in mathematical word problem analysis models and methods, meaning representation, and semantic understanding. Through analysis, this paper further pointed out that the lack of interpretability of deep learning, the lack of standardized datasets and large-scale commonsense knowledge bases, and the lack of visualization of the solution process were the challenges faced in current research.