I’ve started preparing some work on search term frequency in our collection database.
The system is set up to track only successful searches – which we define as those that result in a user selecting an item from the search results. Taking figures generated last week, the database has served up over 1.87 million successful searches since launch (June 2006), whilst nearly 5 million objects have been viewed. Obviously users are getting to objects via direct links or using third party searches (Google etc, or our Opensearch feed) to get directly to records.
Of these 1.87 million searches there are only 19,352 unique terms.
Obviously there are few factors at play here. Firstly, there will always be clusters of popular terms – see Google Trends.
But what about the influence of interface?
Our current search and objects pages are set up with multiple (perhaps maximal) pivot points, or ways to get to other results and parts of the collection. The search/home page features a large randomised tag cloud which displays user-entered keywords. Clicking on one of these will result in a search result for that term.
The search result page now shows ‘related’ search terms as hyperlinks to searches for those terms.
The object record page shows (if they exist), user keywords with hyperlinks to a search for that word/phrase; the top three search terms related to that object (if the object has been viewed more than 30 times); as well as subject and object categories.
Each of these sets of hyperlinks are encouraging users to click them – probably before they manually type another search term in the large search box. Why type when the site you are using is making suggestions for you?
This requires further examination and a cross refencing of search terms against user keywords and also some heat tracking with a set of test users.
Here are the top 20 search terms as of November 2006 (excluding object numbers).