Friday, June 11, 2004

Tweaking search

Since there was a "spanner" in our intranet search, we have finally got around to tweaking some of the configuration options, improving the UI and are now looking to start gaming the results based on what we can garner of the algorithms that are used. Luckily, while running through the config, I discovered that we could reveal the actual document scores returned by the engine as opposed to simple matches. In an effort to "improve" our results, we have set up an intricate series of "Lorem Ipsum" documents (so as not to skew search terms for any other intranet searches) with the word "Conundrum" (which returns 0 results at present) secreted in various combinations of location within the document (title, meta etc...). Fingers crossed, when we compare our current results with any tweaks that we make, we should be able to instantly see the effect on our "Lorem Ipsum" search documents. Fun, fun, fun! (hmmm)

Update 14/06/2004

The plan was highly successful, and exposed some of the workings of the weighting process of which we were not 100% aware - for the record, it looks like we are over-skewing results with keyword occurrences in the title (among other things). We have adjusted the configuration files, and will see what results we get back with the same search term tomorrow. At the very least the two sets of results will be cross-comparable.

No comments: