A study on machine creativity via concept combination and education tools for creativeness

有人说，创造力源于人类。本研究的目的是在机器上实现这种创造力，在本研究中，我们假设创造力是一个概念，从已知的知识组合。换句话说，随机性是机器创造力的关键思想。我们把这个想法应用到我们的故事制作系统中。我们有两种类型的故事制作系统。一个人向用户询问问题，然后从人类的回答中创建一个故事。另一个是使用词库，在没有人工帮助的情况下输出故事情节。通过这些系统，我们展示了用户帮助的意义，以及它对实现机器创造力的强大作用。对于相关研究，我们研究了日语中的“Shiritori”问题。Shiritori是一种由两个以上玩家玩的文字游戏。然后，我们提出了两种类型的算法Shiritori ;此外，本文还提出了一种基于词与随机性关系的概念网络构建算法、一种基于词连接的文本切分方法、一种基于多Agent环境下的评价函数学习算法机器的创造力可以通过自动对话系统来实现。我们还提出了一个回复选择系统，词选择方法，通过决策树学习和会话分析系统。

It is said that creativity is originated in human beings. This study aims to realize such creativity on machines.We assume, in this study, that the creativity is a concept occurred from combinations of well-known knowledge. In other words, randomness is a key idea for machine creativity. We apply this idea to our story making system.We have had two types of story making systems. One asks queries to users, and a story is created from human replies. Another uses thesaurus and outputs a plot of a story without human help.With these systems, we have shown a meaning of user's help and how it is powerful for the implementation of machine creativity.For a related study, we have studied a "Shiritori" problem in Japanese. Shiritori is a word game played by more than two players. Then, we have proposed two types of algorithms for Shiritori ; one outputs the longest word chain that has the maximum words, and another outputs the longest word chain that has the maximum characters.In addition, we have proposed an algorithm for making a concept network from relations of words and randomness, a text segmentation method by words connection, a learning algorithm of an evaluate function on a multi-agent environment, and so on.Machine creativity can be realized by an automatic conversation system. We have also proposed a reply selection system, word selection method via decision tree learning and a conversation analyzing system.