With the wide application of IoT devices, the proliferation of vulnerability information poses a serious challenge to cyber security. Vulnerability classification, as an important application of natural language processing technology in cybersecurity, aims to extract key features from vulnerability descriptions and classify them to help security managers quickly identify and fix vulnerabilities. Traditional dimensionality reduction methods such as TF-IDF, despite reducing algorithmic complexity, have limited understanding of semantics and context and are prone to ignore some important information. In recent years, statistical methods such as information gain and chi-square test have been introduced to improve feature extraction, but there are still problems such as high-dimensional sparse data dependency. The rise of deep learning techniques, especially the application of word embedding and pre-trained language models, has dramatically improved the accuracy and robustness of vulnerability classification and alleviated the limitations of traditional methods. Combined with artificial intelligence techniques, vulnerability classification is developing in a more efficient and intelligent direction, providing strong support for IoT cybersecurity.