Commonsense Reasoning Using Theorem Proving and Machine Learning
Commonsense Reasoning Using Theorem Proving and Machine Learning
复制标题
DOI：
10.1007/978-3-030-29726-8_25
复制
发表时间：
2019-08
期刊：
影响因子：
--
通讯作者：
Sophie Siebert;C. Schon;Frieder Stolzenburg
中科院分区：
文献类型：
--
作者：
Sophie Siebert;C. Schon;Frieder Stolzenburg
常识推理对计算机来说是一项困难的任务。目前的算法在基准测试中得分约为80%。然而，这些方法通常使用缺乏可解释性的机器学习。因此，我们在这里提出了一种与自动定理证明相结合的方法。自动化定理证明允许我们以可解释的方式获得新知识，但不可避免地受到现有背景知识的不完整性的影响。我们通过使用机器学习来缓解这个问题。在本文中，我们提出了我们的方法，它使用一个自动定理证明，现有的大本体与背景知识，和机器学习。我们提出了第一个实验结果，并确定了训练数据量不足和缺乏背景知识的原因，我们的系统没有从基线中脱颖而出。
Commonsense reasoning is a difficult task for a computer to handle. Current algorithms score around 80% on benchmarks. Usually these approaches use machine learning which lacks explainability, however. Therefore, we propose a combination with automated theorem proving here. Automated theorem proving allows us to derive new knowledge in an explainable way, but suffers from the inevitable incompleteness of existing background knowledge. We alleviate this problem by using machine learning. In this paper, we present our approach which uses an automatic theorem prover, large existing ontologies with background knowledge, and machine learning. We present first experimental results and identify an insufficient amount of training data and lack of background knowledge as causes for our system not to stand out much from the baseline.