Comment #1
Comment #2
Reference:
Title: Creating Collections with Automatic Suggestions and Example-Based Refinement
Author: Adrian Secord, Holger Winnem¨oller, Wilmot Li, Mira Dontcheva
Venue: UIST 2010; New York University & Adobe Systems Inc.
Summary:
This article discusses two systems that create collections from media devices. The musical program was SongSelect, and the photo program was PhotoSelect.
SongSelect has a query based interface. A user can type in "a lot of rock, no U2," etc. The program recognizes key words such as "a lot, some, no," etc, and the program will create a playlist from that. The user can also specify the number of songs or overall length of the playlist. From there, the interface has a suggestion widget that does exactly as it describes. The user can drag and drop the suggestions onto the playlist. Also, the interface has two tabs for history and the user's library. A user can drag and drop from the library, or use previous searches/queries from these tabs.
PhotoSelect is very similar to SongSelect. It still uses the query based system to generate collections, and has two panes with the library on one side and the collection being generated on the other. The keywords are mostly the same, and the suggestion widget is still there. However, they changed a bit about the suggestion widget for PhotoSelect. PhotoSelect uses a grid structure to display the photos, so no metadata is visible.
They implemented these programs with a client-server model. The server was implemented in C++ as a web server and uses the Geocode constraint-solving toolkit.
Discussion:
Discussion:
I actually enjoyed this read because it actually discusses an interface and how it works. However, this article is incredibly long. While I appreciate the detail, the length was pretty brutal. Nonetheless, the two programs described sounded pretty cool to me. I really liked how they used plain words for their query language. This was probably my favorite article to read, in spite of the length, because it actually described an interface. On top of that, the interfaces sound cool. I would really like to try out these two programs. I think they would be very easy to use for even the most basic of user. The queries are easy to figure out, and the entire library is always present as well. All in all, these are definitely worth keeping an eye on in the future.
A lot of power, no peculiar syntax. Take note, Google; I know you are reading this.
ReplyDeleteWell I'm glad you managed to summarize the article succinctly, because I'm afraid to read it myself now that you mention the length.
ReplyDeleteAnyway, both applications sound pretty cool, and I'd like to try them out for myself, too.
I still have to wonder, what's up with the client-server model...sound like evil empire stuff to me. BTW...over 70,000 songs...wtf.
ReplyDeleteI would assume that the client-server model is necessary because of the processing power needs of the selection algorithm. That probably means that if this system was used by a real company that they would look in at what you were listening to.
ReplyDelete70,000 songs sounds like a pretty huge number, but it's probably quite reasonable for a music hobbyist. I'm actually quite curious what the upper bounds are on track numbers for some people.