A day in the life of BBCi Search

I am speaking next week at an internal BBC Search & Findability conference, where I will be doing an as-yet-unfinished presentation on "What search tells you about your users" - which I plan to make available on currybetdotnet in due course.

A lot of it will be based on some research I did over Christmas into the behaviour of BBCi search users on one particular day, which I have been presenting as 'A Day in the Life of BBCi Search'.

When dealing with the large quantity of data generated by the BBCi Search service the easiest way to analyse it has been to aggregate it by search keyword, and extrapolate user behaviour from these searches. However I realised that the keywords that are 'popular' barely scratch the surface of the number of unique searches made every day. I therefore wanted to find a way that I could extract patterns from the data by other means. The conclusion I came to was that I needed to limit the dataset to a sample of usage from a finite period of time, that was both small enough to hand classify, but large enough to be representative.

The study concentrated on looking at separate samples of searches from different times of a single day. This allowed for the analysis of changing patterns of use over time. I hand-classified around 20,000 individual searches to determine their regional bias and their relevancy to the BBC. I also looked at whether they included spelling mistakes, natural language queries or URLs, and how the service reacted to that. And I looked at the number of words users tended to input.

The PowerPoint I normally give isn't much use out of context - I need to be there waving my arms and pointing at the various charts and graphs - so I have put together an article - ' A Day in the Life of BBCi Search' - which hopefully does the job of the pointing and the waving.