To whom it may concern,
Knowing that Safari has several books concerning Objective C, today I tried to find them using the search. After several tries, I gave up, used Google to find the books, and ended up typing in the author names.
This is obviously sub-optimal. Here are the three biggest problems with the search.
Problem the first: inappropriate stemming of search terms. If you're searching a corpus you don't know much about, then sure, some stemming is fine. If you're indexing technical books, then stemming 'Objective' to 'Object' is a poor choice. If you're going to insist on doing stemming, why not return results that contain the term unstemmed higher in the rankings?
There must be a huge archive of previous customer searches. Tokenize these in to words, and create a stop list against common ones to avoid stemming on them.
Problem the second: not specifying between different result concepts. Again, if you're Google, you don't have much choice but to return URLs (although even they make a fair stab at returning different objects: videos, images, products, etc). If you're a book publisher, you have much more scope for returning dramatically more useful results.
Specify your use-cases here. The search form actually allows people to search individual result objects, ish. There's a drop-down option for Authors, Titles, etc. However, it's an untitled form element, and the selected option there is "Entire Site". I'd suggest that the user expectation there is that clicking on that is going to offer sub-site search specializations, rather than meta-data specializations. But this is a bit of a tangent:
As a member of the search public, what am I looking for? One of: whole books based on title; whole books based on category; whole books based on author; individual sections from books on topic. Returning these mixed together makes the results confusing, and largely irrelevant. Split them out. Let's see: First three categories that match your search (more on this in a minute); First three authors that match your search; First three titles that match your search; First three chapters that match your search. This covers all your bases, whatever your user was searching for.
Problem the third: not indexing categories. There's a category called "Objective C". It didn't come up when searching for Objective C. It's also not obviously navigable to from the Safari homepage without performing a different search first, and then clicking through the returned nav on the left (where it isn't highlighted, but other categories are). If I "Browse the Safari Library", I can't drill down. What kind of browsing is that?
People may well be searching for a category. Why not index the category name itself, rather than just returning the categories the top books come from? With the stemming issues mentioned above, it means you may as well not even have the category classifications.
There are a relatively small number of books on Safari. Using a naive word-matching search when you know so much about your content already is far from ideal. You have obvious distinct objects that people are searching for - ignoring this and treating it like you're searching flat and homogeneous content is the reason the search is so totally broken.
-P
