An agent that learns by reading the web

To create an AI agent that can learn facts from web text Description: Much information on the web exists as unstructured natural language text. For example: "Stanley Kubrick, director of 2001, also worked on the film AI" From this, we can infer various things: 'Kubrick directed 2001'; 'Kubrick is a director'; and possibly: 'Kubrick is a Person' The latter inference requires an ontology within which we know that directors are people, and which could be extended by learning. An example of this type of work is the TwitterBot being developed in the NELL project at CMU, see [login to view URL] An interesting extension of such work would be to enable the system to learn via feedback from users. A system like NELL could tweet a for example "I think that Kubrick is a Robot" , and then followers could correct the system in response. The aim of this project is to investigate the current work in this area, and to implement and evaluate a system that can learn facts by reading text on the web. URL: [login to view URL]