IoT security issues are a huge burden for users who lack relevant knowledge or skills when they are faced with processing large amounts of information in a short period of time. In order to allow the public to effectively understand IoT security, this study collected 2018–2023 IoT-related literature from the Web of Science as a research dataset to extract key information and conduct subsequent analysis for the public to use easily. Feature extraction is one of the important steps in this study. This study proposed a MPNet_CTM model to build topic model, which applied the masked and permuted net (MPNet) to process IoT security articles and used correlated topic model (CTM) to process topic model. In terms of topic numbers, this study applied correlated topic model and the consistency score to build the best topic numbers. In verifying effectiveness of topic model, there are five claasifiers used for classification and evaluation, including decision tree, random forest, XGBoost, Adaboost, and CNN. After BERT and MPNet natural language processing, this study compares BERT + LDA, BERT + CTM, MPNet + LDA, and MPNet + CTM. The results show that the MPNet + CTM based on XGBoost classifier is the best performance. Finally, this study uses VOSviewer to visually present the connection strength of the extracted words, and to demonstrate the keyword strength and connection of IoT information security. The results can help IoT users better understand the trends and challenges of IoT information security.