Literate Versus Spatial Search
One of the major problems facing my research in interactive search visualization is the difference in search strategies for list based layouts versus other approaches. List based approaches dominate current methods of searching the Web for information. Anyone who’s used Google to find something online is familiar with the standard “10 relevant results on a page, in a list, with title and ‘snippet’ information”.
However, there’s an interesting tension between strategies involved with presenting information to users in this fashion. There is the TF/IDF and “PageRank” strategies that Google uses to filter and order (resp.) the relevant pages for each query. These are essentially two different information “spaces” that are used to relate documents to a query. TF/IDF essentially helps to match a set of documents that are “semantically close” to a query, while PageRank helps to identify pages that are close to the “center” of a network of linked pages (Hypothetically, these “central” pages have higher quality information).
In a sense, PageRank is a sort of filter or lens that you can’t turn off. You can’t really tell how much PageRank is affecting the order of items presented, and you also can’t tell if Google has “enhanced” a query by tailoring its results to a more “useful” set of documents.
In fact, most people probably don’t even spend much time following the results exactly as they are arranged. Instead, they quickly scan the list of titles and summaries looking for more keywords that suggest the kind of content they are interested in. In this sense, Google doesn’t even need to provide a perfectly valid semantic or document-link space as long as the user gets something useful in the top 10 results. This method of search is really more about reading, and is therefore a literate form of search.
However, this isn’t typically the way that we find things in the real world. Generally, we do so by using our senses to size up our environment. Is there a good smell in the air? Is there a group of people moving a certain direction? Maybe there’s something interesting that direction? All of these strategies involve notions of proximity/approximation, social awareness, and direction (resp.).
The strategies involve not knowing explicitly what you want or where you want to go, but in manipulating your environment gradually until you identify a source or local optimization of the resource in the surroundings. In order to maximize your efficiency in this environment, you want to have control over how you can move through these spaces.
In the Google example, this may mean going “against” the PageRank position, or only operating in a certain “radius” from the center of the network of document links. Perhaps you don’t want to find something “too popular”, or are looking for something new. Maybe you want to find other complimentary topics or content for the document space you’re already familiar with?
Many of these strategies involve more control over the information spaces than the search portals provide. In addition, they also involve the use of human generated meta-data to a greater extent. Sometimes you want to follow the herd, but sometimes it makes sense not to.
I’ve been interested in how to handle this scenario with a recent paper on “Visualizing Social Links in Exploratory Search” (here’s my author’s copy.).
One of the problems I had was how to overcome the familiarity and reliance on literate search strategies in a spatial interface. In general, if I use the two dimensions of an applet to display document candidates, it often becomes impossible to layout the “snippets” of text necessary to describe the documents in greater detail. We handled this situation by essentially removing these snippets from both the simple “list” interface and the “map”-like interface. The interfaces using maps fared slightly better than the list in this study, but none of them were particularly well liked due to the absence of snippet text.
So… the question is, how do we overcome or accommodate literate search expectations in a spatial environment? Tag clouds? Force directed snippet placement? Abbreviations? It seems like there should be a way to do it.