Research on Discovery and Creative Process Modelling in Logic

建立了一个新的框架，将高阶概念学习问题归结为新谓词引入问题，特别是通过给出高阶谓词对应的新谓词的定义作为PROGOL的背景知识，证明了高阶概念学习可以在一阶概念学习的框架下实现.本文提出了一个利用数据库作为PROGOL输入源的通用框架，通过半自动地生成正例、反例、背景知识、模式声明和类型信息。在此基础上，建立了一个实验系统DB-XML，实现了对数据库的数据挖掘，并在真实的业务环境中，利用DB-XML开发了一个电子邮件分类系统的知识库，实现了对设备故障咨询邮件的自动分类。我们以关键词词库作为背景知识DB-SQL，成功地自动提取了电子邮件分类规则，并对PROGOL进行了并行化以提高归纳逻辑程序设计的效率的尝试，在Prolog上建立了一个实验系统。特别地，我们证明了在算法中采用并行定理证明器MGTP（Model Generation Theorem Prover）可以使程序变得非常简单

We built a new framework for reducing higher order concept learning problem into the problem of a new predicate introduction.Particularly, we showed that we can achieve the higher order concept learning in the framework of first order concept learning by giving the definition of a new predicate corresponding to the higher order predicate as a background knowledge of PROGOL.Furthermore, we proposed a general framework of utilizing database as a source of input to PROGOL by semi-automatically generating positive examples, negative examples, background knowledge, mode declaration and type information. We built an experimental system called DB-Amp to achieve data mining in database based on the framework.We developed a knowledge base for an e-mail classification system which outomatically classifies e-mails concerning to equipment malfunction consultation in a real business environment by using DB-Amp. We gave keyword thesaurus as background knowledge DB-Amp and succeeded in sutomatically extracting e-mail classification rules.We also made an attempt to parallelize PROGOL to increase efficiency of inductive logic programming and built an experimental system in Prolog. Especially, we showed that we could make the program very simple by adopting a parallel theorem prover MGTP (Model Generation Theorem Prover) in the algorithm