Sarcasm is usually emotional and topical. Mining the characteristics of sarcasm semantics in different emotional tendencies and topic expressions helps gain insight into the sarcasm cause. Most of the existing work detect sarcasm or topic label based on a supervised learning framework, which requires heavy data annotation work. To overcome the above challenges, inspired by the multi-task learning framework, this paper proposes an unsupervised knowledge-enhanced prompt method. This method uses the similarity interaction mechanism to mine the hidden relationship between the sarcasm cause and topic, which integrates external knowledge, such as syntax and emotion, into the prompting and generation process. Additionally, it identifies the sarcasm cause and topic simultaneously. Experimental results on a real-world dataset verify the effectiveness of the proposed model.