Automatic Construction of a Knowledge System Using Text Data on the Internet

In order to be truly useful to humans, robots should be able to communicate smoothly with them. To gain this capability, robots must possess cognitive and speech functions similar to the consciousness function of humans. It is important for robots to be capable of a flow of consciousness and emotions, and for this purpose robots must first of all possess a so-called memory of knowledge. This paper defines a network of association and kansei (a Japanese term relating to emotions and feelings) values that model the memory of knowledge. It also describes the construction of an association-kansei database comprising the association-kansei networks. This database searches through text data on the Internet, statistically processes the data, automatically calculates the strength of associations between collected words, extracts emotional elements incidental to respective words, and stores the results in the database. This paper further introduces techniques used to handle idioms, separate words of different meanings and classify concepts using superordinate-subordinate relationships. Lastly, the paper describes evaluation experiments that confirm the usefulness of the constructed database.