|
Okay. So what I'm about to say will prove beyond nary a doubt that I'm a big dweeb. Yesterday, I saw one of the best demos I've ever seen--Flamenco, a search interface that exploits faceted metadata. In this instance it's an image browser (that link goes to the actual working prototype) for architects and city planners. They typically use such images for inspiration, collage-making, pinning up on walls, that kind of thing. The interface has been tailored to meet their specific needs and assumes familiarity with the jargon of the profession. Now typically a 'great demo' means that the technology is all whiz-bangy... That wasn't the case here(for that I go to CHI and see what the Media Lab kids are doing). It was a great demo because:
a) the thing actually works - it looks good and operates well
b) it addresses a real need--better tools for navigating complex information spaces are definitely required, and Flamenco is a great option
c) it's predicated on extensive user research--the interface was developed and iterated through user testing, and is successful because of it... it's not a technology looking for a reason to be
d) it provides a great 'first step' that others will be able to solidly build upon This paper (.PDF) describes the design and evaluation of Flamenco. (I'm on the review committees for both the CHI-AIGA forum and DIS2002, and this paper is better than anything I read for either of those conferences.) Take the time to read it, and, while doing so, play with the interface. Marti was gracious enough to post her talk (PowerPoint). It summarizes the paper, and also throws in a few other ideas that occurred to her. Stuff that excites me:
- Epicurious users showed a preference for the "enhanced search" over the "simple search". And that the design of Flamenco bore this out for image searching... All this fairly rigorous researchsuggests that users aren't afraid of facets, and, in some cases, prefer it. Expose them! (It also suggests that this is true in fairly well-defined domains such as recipes and architectural image browsing... It seems that familiarity is necessary...)
- Wondering about how a person approaches a task is reflected in how they traverse the facets. Marti pointed out that they didn't do any real research on this, but it's a question she wants to pursue. She surmised that a task such as "I want to make a summer pasta" would be represented as an ingredient + occasion search. The idea being that facets can be very task-oriented, thanks to their flexibility.
- Validation for smart user-centered processes. Marti strongly believes that, in this instance, the "search problem" is an interface one, not an algorithm one. And her work is bearing that out.
- The exploratory nature of the interface pretty much eliminates errors. There is no "starting over", no dead ends. You just keep clicking and refining until you find what you want. Anyway, a big congrats and thumbs up to the folks at SIMS for their efforts here. I'm hoping we start seeing it influence others very soon.
7 comments so far. Add a comment.
Previous entry: "Something to get from the library."
Next entry: "Memories Can't Wait."
Comments:
COMMENT #1
I just looked at the Flamenco pages the other day and that thing wasn't up yet. It is impressive, and beats the hell out of other image-search interfaces I've seen (I've worked in a University slide library that had 6 million images in a database--searches literally could take hours to run). There are some problems: it's easy to get a set of results that's overwhelming both to the user and to the web browser; try clicking on "designers: people" in the first group of terms. But it's amazingly easy to refine your search and even change the direction of your search midstream. IOW, you can start out looking for one thing and change your mind several screens later without actually having to start over. It's clearly something for domain experts to use. The first screen is an overwhelming mass of detailed terms; that wide exposure of terms is part of the reason it works so well of course. I didn't think Epicurious' system was as successful, which makes me wonder if complex facet-based search systems are really most appropriate in situations where users are domain-knowledgeable. Wine.com's simpler set of facets seems to be easy to use for me as a wine non-expert.
Posted by Andrew @ 02/13/2002 01:22 AM PST [link to this comment]
COMMENT #2
i like the way Flamenco incorporates the referent structure into the interface itself -adding both a dynamic sorting/filtering interface and also a useful layer of context for viewing the images themselves. call it breadcrumbing squared. or cubed. or something. anyway, Yahoo, eat your heart out. i do agree with andrew that this particular execution seems more appropriate for an expert system; probably too many features, too densely packed in for a general audience. but as useful as the facet browsing may be, what i really want (i think) is a way of getting at the visual information encoded in the picture itself - stuff that may or may not end up in someone else's descriptive classification. what i would _really_ like to see: a browsing interface like this wrapped around an image analysis engine like QBIC which could analyze and describe the contents of the images programmatically - e.g., show me things with blue in them, or pictures with water, or even just "pictures like this one." if you could integrate the metadata browsing with an engine that analysed the contents of the images themselves ... now that would be a pretty nifty search product. still and all, kudos to the Berkeley kids - it's quite a demo.
Posted by alex @ 02/13/2002 10:04 AM PST [link to this comment]
COMMENT #3
This interface is powerful because it reflects the way visual researchers users research. The MO in a slide collection has always been to search by period or artist. Subject classificaton was only made possible when electronic indexes started to become available. This model explodes the index to show ALL the access points (descriptors). See for instance, this record for a Josef Albers work. Each indexing/access point "People, Period, Styles, Concepts, etc." is listed with hyperlinks to find records with similar indexing. This is the power of indexing. Why doesn't this happen more often on the web? Because if you don't capture/index on those facets, you can't use it for searching/browsing. Exciting stuff, and interesting that they chose to list all of the facets on the search page.
Posted by Michael Angeles @ 02/14/2002 07:44 AM PST [link to this comment]
COMMENT #4
I got a chance to play with it today. I really like it. It's definitely given me lots of ideas on how to implement a project I've been working on...but the one problem I think that many sites will have to face when starting similar projects is the initial time and effort to tag each of the objects in the collection with the attributes. Also what happens if after the collection has been in production new facets are discovered, what kind of effort/time will be needed to revisit all the objects in the collection? Tagging content will always be the biggest issue when you have a very large information collection. Developing the controlled vocabulary will be the first hurdle. Just imagine if they just applied this to books/journals in a library since they already have subject headings and attributes.
Posted by Madonnalisa @ 02/14/2002 05:54 PM PST [link to this comment]
COMMENT #5
I have been waiting to see a matrix implementation like this one for some time. It is heading in a very liberating but complex direction. As two of the comments above indicate however there remain challenges.
Issues related to the amount of metadata tagged to an information object are tied to the potential manipulation an object will under go in its'life cycle. It will require different metadata for resource discovery, life cycle (records)management and resource exchange purposes. There will even be systems metadata tied to it that no one may ever see. Metadata can be multipurposed to serve these various needs but I don't think we will ever see "a little" metadata associated information objects. In a real way the associated metadata increases the value of the object. In a metadata strategy for managing information holdings this can represent a significant investment component.
The second issue of domain familiarity when searching arises from the diversity of target audiences. The content of a site may require a highly technical thesaurus interface as to aid one of its' audiences and a very general generic vocabulary to aid another. This is of more interest in the context of resource discovery and exchange than other processes.
It will be interesting to see how these challenges are met when dealing with sematic shift in language (street jargon as an example) or on sites that need to work across multilingual boarders. Many nation's populations are multilingual and from a government service and program delivery point of view it takes on special significance
The last comment I have to make has to do with the responsibility the search client has in finding the information they want/need on the web. Much to do is made over facilitating resource discovery. For sales sites this is particularly important, but I think there may be a down side in making the searcher unreasonably lazy in their demands for information at their finger tips.
Posted by Cecil Somerton @ 12/20/2002 12:11 PM PST [link to this comment]
COMMENT #6
I have been waiting to see a matrix implementation like this one for some time. It is heading in a very liberating but complex direction. As two of the comments above indicate however there remain challenges.
Issues related to the amount of metadata tagged to an information object are tied to the potential manipulation an object will under go in its'life cycle. It will require different metadata for resource discovery, life cycle (records)management and resource exchange purposes. There will even be systems metadata tied to it that no one may ever see. Metadata can be multipurposed to serve these various needs but I don't think we will ever see "a little" metadata associated information objects. In a real way the associated metadata increases the value of the object. In a metadata strategy for managing information holdings this can represent a significant investment component.
The second issue of domain familiarity when searching arises from the diversity of target audiences. The content of a site may require a highly technical thesaurus interface as to aid one of its' audiences and a very general generic vocabulary to aid another. This is of more interest in the context of resource discovery and exchange than other processes.
It will be interesting to see how these challenges are met when dealing with sematic shift in language (street jargon as an example) or on sites that need to work across multilingual boarders. Many nation's populations are multilingual and from a government service and program delivery point of view it takes on special significance
The last comment I have to make has to do with the responsibility the search client has in finding the information they want/need on the web. Much to do is made over facilitating resource discovery. For sales sites this is particularly important, but I think there may be a down side in making the searcher unreasonably lazy in their demands for information at their finger tips.
Posted by Cecil Somerton @ 12/20/2002 12:13 PM PST [link to this comment]
COMMENT #7
I have been waiting to see a matrix implementation like this one for some time. It is heading in a very liberating but complex direction. As two of the comments above indicate however there remain challenges.
Issues related to the amount of metadata tagged to an information object are tied to the potential manipulation an object will under go in its'life cycle. It will require different metadata for resource discovery, life cycle (records)management and resource exchange purposes. There will even be systems metadata tied to it that no one may ever see. Metadata can be multipurposed to serve these various needs but I don't think we will ever see "a little" metadata associated information objects. In a real way the associated metadata increases the value of the object. In a metadata strategy for managing information holdings this can represent a significant investment component.
The second issue of domain familiarity when searching arises from the diversity of target audiences. The content of a site may require a highly technical thesaurus interface as to aid one of its' audiences and a very general generic vocabulary to aid another. This is of more interest in the context of resource discovery and exchange than other processes.
It will be interesting to see how these challenges are met when dealing with sematic shift in language (street jargon as an example) or on sites that need to work across multilingual boarders. Many nation's populations are multilingual and from a government service and program delivery point of view it takes on special significance
The last comment I have to make has to do with the responsibility the search client has in finding the information they want/need on the web. Much to do is made over facilitating resource discovery. For sales sites this is particularly important, but I think there may be a down side in making the searcher unreasonably lazy in their demands for information at their finger tips.
Posted by Cecil Somerton @ 12/20/2002 12:14 PM PST [link to this comment]
Add A New Comment:
|