With the gradual development of machine learning and deep learning techniques, scholars have made significant progress in classifying texts using well-performing neural networks. When conducting more in-depth research on short texts, finegrained sentiment analysis is the key to solving the current problem. As a result, multi-categorization text analysis at the comment text level has shifted to aspect-level judgments on statements in comments. In this project, aspect-level sentiment analysis is improved for the short text of Weibo comments based on the BERT model combined with an attention mechanism. The model proposed in this paper uses the attention mechanism to analyze the weights of the short text and precisely weight the summation and finally the category output of the short text can be obtained. For the characteristics of Weibo comments, such as few words and short sentences. This paper achieves good results by improving the lexical model. The algorithm in this paper conducts comparative experiments on the Chinese data set of Weibo users' comments. The experimental results show that the algorithm in this paper is better at aspect-level-achieve higher classification accuracy and recall rate than other algorithms on aspect-level sentiment classification tasks.