Metonymy

Problem:

To me, Wikipedia is a modern world wonder. However, directed exploration within Wikipedia is problematic. From the user’s perspective, falling down a rabbit hole of links in a depth-first search is easy — surfacing and understanding the clusters of information that an article sits within is not.

Solution:

Metonymy is an information periscope for readers. Keeping track of the user’s trail through Wikipedia, Metonymy uses degree-of-interest calculation to suggest locally and globally relevant Wikipedia articles.

Making sense of Wikipedia:

Saving Pages

Before knowing where you want to go, it’s important to know where you’ve been. Metonymy tracks all pages visited within Wikipedia, and allows users to star pages for later. (Honestly, it blows my mind that this isn’t already part of Wikipedia, though it is available with other Chrome extensions.)

Cluster Based Recommendation

Metonymy generates recommendations by calculating the term frequency — inverse document frequency of articles within two steps from the in-group and out-group, respectively. The user is able to build an in-group by toggling on saved articles, and the remaining saved articles serve as the outgroup. This filters out the articles that are highly connected within all of Wikipedia, leaving only the articles that are highly-referenced near the in-group and specific to the selected articles.

The goal is to empower users to triangulate their way to new Wikipedia articles, allowing a sort of “information alchemy” based on what articles the user puts in the in-group.

Designing with Algorithms

Designing an extension for Wikipedia is entirely about information architecture. The approach I took was to use algorithms as a method of interaction design: prototyping new methods of information access and retrieval, and evaluating the output and usability of the recommendation algorithm.

Ultimately, the dream is to build a system for myself that takes my knowledge as input and gives me optimal information as output. This leaves me with two real questions:

  • Has Metonymy changed the way I use Wikipedia?
  • Do I trust the results Metonymy gives me?

To the first question, Metonymy has made structured browsing easier. I no longer browse my way into dead-ends, because each recommendation is algorithmically intersting, in relation to either my current article or the boundary of my knowledge. However, I still have control over the recommendations so I have a greater trust in their output.

Further Work

There is still a lot of work to be done on Metonymy. The algorithm is loosely based on the degree-of-interest function found in “Search, Show Context, Expand on Demand”: Supporting Large Graph Exploration with Degree-of-Interest. As I built it, however, the algorithm runs in O(n^2), and starts to slow down noticeably with more saved links.

I’d like to do more work on the UI, providing a graphical, node-based view similar to the one in Apolo.

Finally, building the graph out of explicit links between articles is just the tip of the connection iceberg. There are many other types of connections that could be examined, such as authorship and crowdsourced trails throughout articles.