Abstract: Emotion cause extraction(ECE) is a new research direction of affective computing. As a kind of fine-grained sentiment analysis, its purpose is to find out the part of the given document text that triggers the emotion, which is also called tracing to the source of an emotion. ECE has a high academic research value and a wide range of application scenarios in reality, because of its involvement in the fields of linguistics, psychology and other related domains. Recognizing the emotion cause in a document is more useful than just only identifying the emotion. For example, it can help people understand the source of stress and better control their emotions. Although most of the research in affective computing focuses on emotion recognition, emotion prediction and emotional information extraction, many scholars have turned to analyze the causes behind emotions, and have produced rich results in recent years. We make a comprehensive review and analysis of automatic ECE from texts in multiple perspectives, starting from the problem definition and the classification of ECE. Then we review the main methods for ECE especially based on deep learning. After that, the benchmark datasets and the evaluation metrics for ECE task are detailed summarized. Finally, we discuss the deficiency of the existing works on ECE and forecast the future challenges.