With the rapid development of information technology, Internet has become the main carrier of public opinion spreading. All kinds of public opinion events come out one after the other, which can be quickly spread on different media in a short time and receive extensive attention and participation from large-scale Internet users. It may also trigger a strong reaction in the Internet space and the real society, and even induce large-scale mass incidents. Therefore, quick monitoring and effective early warning of online public opinion events become more and more important. In response to the growing scale of Internet information and the expanding scope of its dissemination, how to discover online public opinion events and push the precise personalized monitoring information has become the focus of current public opinion applications. Aiming at the problem that the interest is hard to capture in case of the sparsity of user interaction, a topic recommendation model based on HKN(hierarchical knowledge network) is proposed. The model expands the semantics and increases potential information association between topics by using hierarchical knowledge. It models the hierarchical knowledge, topics, and users to obtain corresponding embeddings. With those embeddings, we use a multi-layer perceptron matching model to predict the CTR(click through rate). Experimental results show that the HKN model outperforms multiple baseline algorithms by 6.7% and 4.9% on the average of F1(the balanced F score) and AUC(the area under curve) metrics CTR value respectively.