This proposal focuses on developing methods to automate the time-consuming process of selecting good overview and review articles from on-line new sources. The goal is to select articles understandable by persons outside the field and use them on the AITopics information portal maintained by the AAAI. The AITopics website is a widely used authoritative source about Artificial Intelligence that is used by a wide variety of people seeking information about AI: novices, students, the press, etc. AITopics is intended to provide explanatory, accessible material about multiple subtopics that make up the field of AI. This project will develop new methods for this finding/filtering such articles and build on methods already used by AITopics. The resulting software will be open-source and is intended in the long run to be applicable to other scientific discipline beyond AI, particularly by their professional societies.
The project will pursue two thrusts to generalize the current NewsFinder program, which was developed as a part of AITopics and currently finds many types of news stories about AI. The first thrust of this project is to enable it to find overview and review articles automatically. The second thrust is to enable the program to learn from data collected from readers and administrators.
Federal Award ID: 1239062 Report Submission Period: 05/01/2012 to 04/30/2013 Principal Investigator: Bruce G. Buchanan, Ph.D. The information portal AITopics (on the web at www.aitopics.org) is a compendium of introductory articles about artificial intelligence that have been previously published in reliable sources. It has been maintained manually by volunteers at the Association for the Advancement of Artificial Intelligence (AAAI), and our NSF-funded effort has been directed at reducing the time required of volunteers without reducing the quality of information provided. We have developed a program, called NewsFinder, that finds interesting news stories about AI in current online periodicals to post on the website daily. We have also built a virtual archive of classic publications within AITopics that lists books and articles of historical significance to the field, with links to online text when they are available. We continued working toward our long-term goal of building a content management framework that is useful in many sciences. Several major improvements in the AITopics site were implemented this year. Starting with a complete reimplementation of the portal we added the following features: • Each of the 20 primary topics (e.g., machine learning, robots) is subdivided into several second-level topics, creating an index of about 180 terms • Each of about 7000 articles in the site may be found by searching for key words in the title or summary, as well as by looking through the hierarchical index • About 2000 titles of classic books and articles have been indexed and added to the site, with links to online full text for well over half of them • Several out-of-print titles have been digitized and added to the site • Additional editing software was created to simplify the task of managing the site, all of which are applicable to managing information about different subject matter domains. With NSF funding, we hired a student intern to work with us full time over the summer to design and implement the new features. He co-authored a paper for the Innovative Applications of AI session of the AAAI national conference, and participated in the conference exhibit about the site.
