Interpreting the causes of emotions in dialogues is vital for emotion models aiming to comprehend human sentiments. Current research often simplifies this task to causal emotion entailment (CEE), which focus only on identifying causal utterances without providing interpretable explanations. To address this limitation, we introduce a Multi-hop Emotion Cause Reasoning (MH-ECR) framework. Leveraging the chain-of-thought reasoning capability of large language models (LLMs), this framework incorporates context theory from linguistics to guide the extraction and reasoning about critical information. Using this framework, we construct a Conversation Emotion Cause Reasoning Dataset (CECR-Data) containing the reasoning chains generated by prompting the LLMs. Additionally, we have fine-tuned Llama2-7B on the CECR-Data to develop MH-ECReasoners that are tailored for interpreting emotion causes. Experimental results show that our model successfully handles the traditional CEE task and provides a comprehensive understanding of dialogue content while accurately inferring the specific causes of emotions.