This is an illustration of WebSIS (the framework of "Web
Search using Image Snippet"). WebSIS attempts
to change the current style of Web
search, by trying to exploit image information to improve Web search. Roughly, we try to extract meaningful image snippets from the Web pages, and return them to
the user along with the text snippets (a "text snippet" is something like a rough
summarization of the texts. if you do not know what are "text snippets", you can
simply search something by using current search engines such as Google or Yahoo.
The results they output are text snippets).
We think image snippets could be very helpful, since it is usually more faster for
a people to recognize that whether an image is relevant to his/her search query
or not. Thus, by browsing
through the image snippets, the user could identify some
(we do not mean "all") relevant web pages more quickly. On the other hand, usually
in order to get a good search result, the user has to provide some feedback (such
as adding a refined search query). While in WebSIS, the user can simply choose some
"relevant image snippets", then the search result will be refined.
This is just an illustration instead of a real prototype system, which allows you to feel the usefulness of the image
snippets. At present the illustration system only contains 2572 Web pages collected
by posing
29 topics
to Google. So,
please constraint your query
to these topics; Otherwise the system may return nothing. Moreover, in current implementation, after you selecting
the relevant image snippets and pressing "submit", the system does not really refine
the search results using image information. Instead, it simply looks for the shared
words in the descriptions of the image snippets (assuming you have chosen multiple
image snippets) and then uses these words to expand the original query. (In fact
it is possible to use the visual information of the image snippets to refine the
search results, which is an ongoing research of us, some preliminary results can
be found in our IJCAI'07 paper.)
After trying the illustration,
please do us a
favor to fill in the feedback form (simply choose the "Feedback
To Us" in the next page) to tell us your opinions on WebSIS. Your opinions
will be valuable for us to improve our research.
Details of WebSIS can be found at:
X.-B. Xue, Z.-H. Zhou,
and Z. Zhang.
Improve web search using image snippets. In: Proceedings of
the 21st National Conference on Artificial Intelligence (AAAI'06), Boston, MA, 2006, pp.1431-1436. (we have also a
patent
pending)
This illustration was developed by Mr. Xiao-Bing Xue (xuexb@lamda.nju.edu.cn). For any bugs please feel free
to report to Mr. Xue.
If you are interested in using the proposal of WebSIS (especially in commercial
system), please contact Prof. Zhi-Hua Zhou (zhouzh@nju.edu.cn).