Search Results and Representations

Last time we spoke about letting users interact with search results, either by letting them go somewhere useful, or doing something useful.

But often search is part of a larger exploratory or investigative activity that is long-lived. In these situations, one consideration that helps me think about systems design is the relationship between REPRESENTATION and RESULTS.  Informally, we might say:

  • Representation: the structure of the set of information I am processing with the help of search
  • Results: the pieces of information that I dealing with, through my discovery process including search.

We can think about two types of relationship between these:

  • the representation is well-known, and results are sought that fit into it
  • the representation is not well-known, but evolves from working with results.

As an example of the first case, consider technology watchers researching a new technology and analysing its likely impact.  They probably have evolved a standard approach to this, and access the same resources and authorities over and over, before applying their particular brand of analysis.  Those of us who like to bake regularities into software love this kind of situation.  Regarding the task as a high-level pattern match, we might be able to partially automate this with saved searches, auto-categorize the results in a pre-structured repository, and allow the researchers to add annotations.  And if we find that the researchers present their output (Market Analyst Reports) in a different structure, we could consider viewing their annotations by their research framework or their output framework.

There are some oversimplifications here.  The technology watcher’s approach is actually less rigid than the example indicates; in particular, it will evolve over time as their domain changes.  I found this myself when I was trying to get up to speed on SharePoint 2013 Information Architecture, and deliberately looked for differences via What’s New, Deprecated Features, etc. So in addition to populating a framework with facts, our solutions should help us shape the framework itself.  This assessment may be informed by experience, change in the number of hits in a category, “buzz” about what’s hot and what’s not, or how effective our analysis is relative our competitors.

As an example of the second case, consider the detective work as portrayed in many TV shows.  The detective is trying to find out “whodunnit”, with a set of facts that grows from someone being murdered all the way to the crime being solved.  The detectives stand around a whiteboard with boxes and arrows showing parties involved, their relationships, and timelines. In addition to people and places, the detectives use investigative constructs that have evolved over the years, such as “alibi”, “means”, “motive”, “opportunity”, as well as trust attributes, based on whether  the information came from a DA or a “snitch”.  These don’t jump out when doing information modelling based on physical entities, but do jump out when doing user research. Hypotheses are constructed and deductive logic applied to sharpen focus or rule people out.  The detectives create a representation of the crime, supported by evidence.  For the next crime, they do it all over again.  The constructs will be the same, but their instances and combinations will be completely different.

The requirements for a system to support this type of activity seem to be those that make the whiteboard so effective:

  • the ability to create and modify entities of known types, for example the victim and other participants, using conventional iconography
  • the ability to create and modify relationships of known types between these entities; these could of various kinds, including space-time trajectories, business relationships, interpersonal relationships, beneficial relationships; these could also be real and hypothetical.

Search for the YouTube video “sensemakeing III” for a discussion.

What has this got to do with search? Well, watching the TV shows, we see both broad based exploratory searches (“interview everyone who was at the reception”), as well as very detailed information seeking searches (“George, find out if Mr. Smithers bought his shoes from Pringles”).  But to a large extent, we have crossed over into specialised sense-making applications and won’t discuss these further.

Stay tuned.