Correlator, a whole new way of searching for information, is up and running for you to try. Currently it displays results only from the English Wikipedia, but its technology can be applied to any collection of English documents (such as news, blogs, etc).
In short, given a query Correlator will return to you a page automatically assembled from snippets of the collection. Then, you can dig deeper into related people, organizations, locations, events, or general semantic concepts such as nationalities, materials, food, works of art, etc. Correlator is specially suited for searching concepts that are distributed across several entries. For instance try: Dinosaurs in Argentina, transistor or Picasso and Peace.
Besides looking at the different entries that discuss these concepts, you can check the most prominent names (people, organisations, etc.) for this topic (e.g. Pablo Picasso), generate interesting timelines on the fly (e.g. black holes) as well as interesting maps (e.g. emergency landing). Notice that in all those cases Correlator collects information from many entries of Wikipedia, not just one. For example, when looking at the black holes timeline we find information from the entry Timeline of blackhole physics as well as the entry Blackholes in fiction.
All of the entities we show (companies, cities, foods...) were detected automatically using state of the art natural language processing. (A more detailed explanation is available in our previous post). Because automatic semantic tagging is difficult, the computer will tag many words incorrectly. We use novel entity ranking algorithms to filter all the information and bring up the most important (and hopefully most correct). Furthermore, in the Names pane you can turn on the "Correction Mode" and eliminate incorrect entities. This information will be used to further improve our tagging and ranking algorithms.
We hope you enjoy Correlator. We would like to thank here all the people that were involved in making Correlator possible:
- Concept: Hugo Zaragoza, Carlos (chato) Castillo, Jordi Atserias
- Research and initial prototype:
- Yahoo! Research Barcelona: Jordi Atserias, Massimiliano Ciaramita, Peter Mika, Borkur Sigurbjornsson, Carlos (chato) Castillo and Hugo Zaragoza
- External: Giuseppe Attardi (U. Pisa, in sabbatical with us), Henning Rode (U. Twente, intern with us), Daniel Julia (pimpampum, contractor).
- Public demo development: Jordi Atserias, Cesar de Pablo Sánchez (U. Carlos III, intern with us), Janet George, Jacob Leatherman, George Levchenko, Tejaswi Kasturi and Pras Sarkar
- Special thanks to Domenico Dato (Tiscali Research) and Sebastiano Vigna (U. Milan)
Hi all, ah first comment! I like Correlator! Would like to see results in a grid that expand on hover. A little like the tag cloud you do for Names. Keep up the good work ... and do it fast.
Comment by anon1 (June 25, 2009, 6:31 pm)Basically the results look like an on-the-fly thesaurus - you can see related words, but unfortunately you have no idea *how* they are related, or how important the links between the words are. A good thesaurus will at least give you a sense of the hierarchy of the terms - whether they are more specific or less specific for instance. I got no conception from the spatial layout what it meant - and this is in my specialist subject where it should have been obvious. Sometimes it is not at all clear what the words themselves mean. Clicking on an unfamiliar word just gets more words. So rather than giving more info, you get a lot of raw data, and is not this app designed to help with information overload? I only saw links to Wikipedia - did you mean to design a front end for that website, or were you thinking of offering a guide to authoritative sources? Possibly an interesting concept but I saw nothing to excite me or even entice me back - I can already search Wikipedia directly and indirectly. Information has to be organised as well as presented nicely, and it has to be organised in a way that is blindingly obvious for it to be useful. The interface looks good, but it doesn't deliver beyond that.
Comment by Jayarava (June 27, 2009, 3:08 am)Tell us what you think, leave a comment: