因果关系是事物间最重要的关联关系，是自动推理的基础。传统因果关系的获取基于自然语言处理、模式识别、统计和机器学习方法，虽然取得了一定进展，但在处理隐式因果关系等方面遇到瓶颈，成为自然语言处理领域的难题。深度学习强大的特征抽取能力为因果关系学习带来了新的机遇和挑战，但是传统的深度学习方法无法进行因果推理。针对这些问题，本项目首先基于深度学习和强化学习对候选因果实体进行抽取。接下来构建图网络推断候选实体间因果关系，尤其对于跨句和跨段的隐式因果关系能够进行有效的推断。其次，定义带有极性和强度的因果关系一般形式，将因果关系测度和因果关系推断过程相结合。最后，结合深度学习和符号表征形式化因果关系推断过程，从而弥补深度学习方法在可解释性上的局限。本项目对于推动人工智能新技术在因果关系学习领域的研究具有重要理论意义和应用价值。

Causality is the most important relationship in the universe, which constitutes the base of automatic inference. Traditional causal-effects relationship extractions are based on natural language processing, pattern recognition, statistics and machine learning. Although some progress has been made, a bottleneck has been reached in tackling implicit and ambiguous causality, which becomes a difficult problem in natural language processing. The powerful feature extraction ability of deep learning brings new opportunities and challenges to causality learning, but traditional deep learning methods cannot make causal inference. To address these problems, this research program first extracts the candidate causality entities based on deep learning and reinforcement learning methods. Next, we build GNN to infer causality relationships especially for the causality from implicit cross-sentence and cross-segment. Then, the research program defines the most universal format of causality relationship with polarity and strength and integrates it into causality inference process. Finally, this research program combines deep learning and symbolic representation to formalize causal relationship inference, thus making up for the limitations of deep learning methods in interpretability. This research program has important theoretical significance and application value for promoting the research of new artificial intelligence technique in the field of causality learning.