Listen to this post
Well the latest news about Google is that it’s trying to predict what you are searching for instantly. The results come up as you are still typing what you are looking for. The early verdict is “clever, not psychic.” I wonder if it will turn into a kind of game–what is Google thinking I’m searching for?–very like when people allow their phones to complete the words they are typing, even (and especially) when it’s the wrong words.
The impulse to know more than users do about what users are searching for can get us into trouble. I’m not saying that users of library resources always know what is possible, when conducting a search. But it is possible for information specialists to go too far in assuming they think they know what people are looking for–or how they should be looking for it. Information scientists are trained to organize information, and to retrieve it from information systems, but are not always working with systems that are intuitive to users. Thinking that information science provides the best way to organize information can get in the way of library users finding the information they need–or even, perceiving the library as a good place they can go for resources. If the organizing structures of the information resources are too challenging, people can give up.
I wonder about Google Instant. Will people respond well to Google presuming that it can know so much from so little initial data? Will Google build in the right kind of flexibility. so that people can find what they are actually looking for, not just what Google thinks they are looking for?
Like humans, Google is the sum of it’s past experiences. Their machine learns with every search, result, and click. We are teaching it. Not bad for a 12 year old.
I think that an important distinction is being missed here: that between the value of information scientific research in developing useful paradigms and algorithms to better mine data – particularly in MARC-based systems where hooks (words, subject headings, etc.) are sparse – on the one hand, and how and to what tasks such research is directed.
I agree that trying to predict what a user has in mind is a dicey proposition. I know that word processors, application development tools, and other applications that automatically finish words for me are a pain in the neck. It takes me more time to undo their “helpfulness” than to just type in the needed terms.
A better approach is to take what the user has put in him/herself and then use information technology tools – such as probabilistic retrieval, utilization of multiple thesauri (including call numbers, tables of contents, etc.) and expand the searched elements to come up with appropriate searches. Then, if nothing is found (or if the desired items are not found), what the user has put in can be analyzed by the system – using appropriate tools to suggest alternative searches (even down to suggesting alternate spellings).
As for machine learning, I’m afraid that in multi- and cross-domain searching, it will be hard to construct heuristic rules that can be useful in improving search result quality (at least not without some additional serious work on the back end to cluster terms in an appropriate manner, hardly a trivial task).