So, here they are. They all refer to situations where there is a set of resources which can be filtered according to different criteria.

Zoom–in

The most common pattern: Given a set of resources, select a subset by  adding an additional (AND-connected) filter criterion.

All photos -> zoom in (Europe) -> photos taken in Europe
Photos taken in Europe -> zoom-in (Cities) -> photos taken in Europe AND taken in cities

Zoom–out

The complementary pattern: Given a filtered view, select a superset by either

• Removing a (AND-connected) filter Photos taken in Europe AND taken in cities -> zoom out (cities) -> photos taken in Europe
• or: Adding a OR-connected filter Photos taken in Europe -> zoom–out-OR (Africa) -> photos taken in Europe OR Africa
• or: Replacing a filter with a more general version Photos taken in Germany -> zoom–out-UP (Germany) -> photos taken in Europe

In either way, you end up with a more general query, that yields at least all of the original results, and usually more.

Shift

Zoom-in and zoom-out can be combined into a shift navigation mode. Here, one part of the filters is replaced by a concept that is neither more general nor more specific than the original one.

Photos taken in Germany -> shift (France) -> photos taken in France

Pivot

Another combination of zoom-in and zoom-out is the pivot navigation. It is very common in web applications like e.g. delicious.com. Given a filter setting and its results, you can jump to a fresh query consisting only of one of the occurring metadata terms.

Photos taken in Germany -> pivot (Moritz) -> photos taken by Moritz

This is often accomplished by adding clickable links to the results’ metadata items.

Querying by examples

This corresponds to a pivot on a number of resources and metadata fields at the same time. The most specific concepts that apply to a whole item selection are collected and used in a new query.

Clinton, Bush, Obama -> query by examples -> American presidents

Obviously, the generalization capabilities here depend a lot on the metadata structure. In some database, the example above might generalize to “persons” or “males”, in others to “Male american presidents after 1980″.

Related set

Demonstrated in parallax and humboldt, this navigation mode follows the same metadata link on the whole result set to construct a new one.

Photos taken in Europe -> related set (photographer) -> Photographers of photos taken in Europe

I realize this is quite dry material and could use some illustrations, examples, references. Nevertheless, I hope it spawns some new thoughts in those of you thinking about search and browsing in web applications!

Today was a good day, so I thought I would share its results immediately, instead of fine-tuning forever — who knows when I find the time anyways!

I built a little facet browser for the New York Times Article Search API - an impressively fast faceted search engine covering over two million articles. So, give it a spin!

Some caveats:

• Don’t look for the page navigation — there is none. Pure laziness, will update it soon.
• The initial counts are based on a search for “the” (which I figured would appear in all articles). Unfortunately, only the top 15 or so values per facet are returned, so you cannot click, e.g. the year 2008 in the beginning. Will fix.
• The API has a request limit of 5000 queries per day. So if your requests don’t work — come back tomorrow morning :)
• Unfortunately, the API seems to support only one value per facet. So, all facets are single-select.(fixed, see comments).

The code is based on my totally revamped elastic lists prototype. I used this project as a little sandbox experiment of how easy customization is possible, and especially how to make a switch from a fully client-based to a server–based filtering model.

FIND ’08: 2nd international workshop on Dynamic Taxonomies and Faceted Search

Torino, Italy, September 1–5, 2008 (in conjunction with DEXA 2008)

Important Dates:

Submission of papers: 15 March 2008 Notification of acceptance: 20 April 2008 Camera-ready copies: 15 May 2008

http://www.sig-find.org/find08/index.html

Workshop description

Current search paradigms for the web, direct access via search engines and navigational access via static taxonomies, have recently been strongly criticized. A third approach, dynamic taxonomies or facetedsearch, focuses on user-centered conceptual exploration and is finally gaining acceptance to the extent that it is now the de facto standard in product selection for e-commerce.

Dynamic taxonomies work on multidimensional taxonomies (usually organized by facets) and provide a single, coherent visual framework in which users can focus on one or more concepts in the taxonomy, and immediately see a conceptual summary of their focus, in the form of a reduced taxonomy derived from the original one by pruning unrelated concepts. Concepts in the reduced taxonomy can be used to set additional, dependent foci and users iterate in a guided yet unconstrained way until they reach a result set sufficiently small for manual inspection.

The access paradigm supported is aconceptual exploration, far more frequent in “search” tasks than the retrieval by exact specification supported by search engines and database queries. The underlying model is simple and easily understood by users, offers substantial benefits over traditional approaches and has an extremely wide application range and a potential for important extensions. This 1-day workshop will discuss all aspects of dynamic taxonomies and faceted search, from underlying technology to user interfaces and commercial use, including demo systems and applications.

Topics of Interest:

Topics of interest include, but are not limited to the following areas: • Multidimensional/faceted taxonomy design • Automatic or semiautomatic multidimensional/faceted taxonomy creation from meta-data, from structured and unstructured documents or by social tagging • Extensions of dynamic taxonomies • Efficient and scalable indexing and retrieval • User interface issues • Emerging applications • Evaluation approaches and metrics

Hand in your papers!

A real wow-project has gone into version 2: Exhibit. It is part of SIMILE, focussing on “Semantic Interoperability of Metadata and Information in unLike Environments”, which provides a whole toolbox of pragmatic semantic web applications.

Exhibit itself presents a “a three-tier web application framework written in Javascript, which you can include like you would include Google Maps.” An exhibit application typically consists of a content presentation area and several widgets for filtering, sorting and grouping the content presentation, following the faceted browsing paradigm also used in my elastic lists. Currently, the available widgets cover plain checkbox lists, maps, and timelines, and a live search.

From my first experiments with the tool, I can say it is really ridiculously easy to create custom views on existing data sets: Data can be imported “from a Google Spreadsheet, Excel spreadsheet, EditGrid spreadsheet, BibTex files or any JSONP data source” without much effort.

Applications are built by loading the exhibit script and then assigning “Exhibit roles” to layers in your java-script: For instance,

<div ex:role="facet" ex:expression=".discipline"
ex:facetLabel="Discipline" />

will create a filtering widget looking for values of the “discipline” property in your data set and display them as a list to be used for filtering. It couldn’t possibly be easier to build client-side facet browsing applications.

Other nifty are the integrated browser history for filtering steps and the ability to export filtered views as e.g. HTML, but also tab-separated text files, RDF, etc.

Check out the Getting started page to get an impression.

The down side, of course is, that all data has to be loaded on the client before the application can start. So we are rather speaking about hundreds than thousands of data items. Also, I would of course be interested in building custom widgets, however, this looks a wee bit more complicated from first looks into the code. And of course UI-wise, I would have some suggestions as well, but anyways, I find the project quite impressive already as it is!

Kudos to David Huynh and the rest of the team at MIT. More info can also be found in David Huynh’s PhD thesis.

It is a demonstration of the “elastic list” principle for browsing multi-facetted data structures. Click any number of list entries to query the database for a combination of the selected attributes. If you create an “impossible” configuration, your selection will be reduced until a match is possible.

The example data is based on the Noble prize winners dataset used in the Flamenco facet browser.

Elastic lists enhance traditional facet browsing approaches by • visualizing relative proportions (weights) of metadata values by size • visualizing unusualness of a metadata weight by brightness • and animated filtering transitions.

In unfiltered view, the brightness shows a trend measure, indicating a rising number of prices of the last years.

In filtered views, a brighter background indicates a higher weight of the metadata value compared to the overall distribution.

If, for instance, you click “peace” as in the example above, you will see that “female” and “Switzerland” are much brighter, indicating that the proportion of women and Swiss is much higher in this context than compared to the whole data set. That’s interesting information and could also be used to characterize the result set of a keyword query or any other collection in terms of its “characteristic” metadata values. Besides that, it fosters understanding of how metadata values are correlated with each other, which is often interesting information itself.

You can also switch on little sparklines to see the temporal distribution of each metadata value: