In person webinar: join colleagues in the Library DisCo at 11:30am on 3/14.
This webinar will open at 11:30am on Monday, March 14th. To access the webinar at this time, click on the link; a download of “Zoom_launcher.exe” should start automatically. On a Windows computer, click on the downloaded file in your browser, then click “Run.” On a Mac, you will be prompted to download a .zip file of the Zoom Launcher. Once that is saved, double-click on the .zip file and then double-click “ZoomusLauncher” to install. The webinar room will be live at 11:30am on Monday.
We are in the process of implementing a new, patents pending inference engine for Stanford in collaboration with a Silicon Valley start-up that is a completely new way to extract meaning and knowledge across multiple sources. In effect this discovery environment produces inferential semantic relationships among millions of digital documents (articles, books, websites, — any digital text really) in a wide and expanding range of subject and genres with clear links to the documents if licensed and to information about the documents enabling purchase on demand, potentially a pay-per-view option.
The method, Hyper-Association of Related Inferences (HARI has been developed by an Italian mathematician named Ruggero Grammatica, a friend. The technology combines machine learning, natural language processing, and a succession of algorithms. HARI ingests enormous amounts of ie-texts efficiently, then uses algorithms to determine relationships among concepts, providing a tool for browsing or discovery. (This is at least a thousand times more difficult to put into words than to grasp when you see a demonstration.) There is an instantiation that has ingested and analyzed 22+ million Medline entries, though more interesting and extensive results arise when full texts are analyzed.
One way to think of it:
Google provides specific answers to specific queries. It’s like a library, such as the library of Congress, in which, instead of gaining access to the stacks, you write down the name of a book, give it to a librarian, and are then given the book for which you asked and only that book.
HARI, by contrast, is like browsing in library stacks. One will find specific information that was sought, but other connections among related concepts will be exposed, and some, perhaps many, of those will be unexpected.
The HARI engine is based on NLP, AI, and custom algorithms that allow the processing of huge amounts of text based information and the extraction of concepts (not keywords). The concepts are then correlated and identify relationships and inferences on the chosen topic, ultimately revealing the full texts containing the concepts. It is a semantically based process that can be “tuned” by the end user both in the front end of a search and once results are obtained.
This complements our existing list and link type of search engine tools that academic users, both students and faculty, typically use today. HARI returns a more in-depth, visual result from each search. At any moment the user can explore any of the concepts displayed and new streams of correlations are presented suggesting new paths of investigation. And through use of the links, a researcher might be directed to licensed source documents or to a pay per view interface. The benefits to publishers would be substantial; we wish to engage with you and others to elaborate and extend HARI dramatically. The benefits to students and professors as well as many other professionals who make use of information resources hiding in thickets of content are also dramatic.