Recently the database recommendation feature was switched on in MultiSearch. There has been a lot of dissatisfaction in the discussion list about its effectiveness. For example, Dutch Disease which is an economics term recommends MedLine and SocIndex. Privatization and prisons recommends SOCIndex and Environmental Sciences and Pollution Management.
I wondered what experiences others have had with this feature and whether we wish to continue with it.
Janette
I’m glad to see this issue raised as I have also been finding it produces some very odd suggestions. Like directing me to Sociology resources when I typed in ‘grove dictionary music’. At best, it brings up such general resources that it contributes nothing; at worst it is blatantly and embarrassingly wrong! I favour turning it off until they can refine it.
I hadn’t seen this feature yet, but one of the 2 recommended databases following an initial search on "google book settlement" is SportDiscus.
Awesome!
Dave C.
It has improved considerably since it was first implemented and I am sure Serials Solutions is watching discussions on the list and will continue with refinements to this feature. It must be very difficult to come up with algorithms to suit every possibility correctly.
I think they should be commended as it gives more exposure to Library resources, particularly for those who stick with their "favourite" databases; they may never have known SportDiscus existed!
It may be more helpful to send them suggestions for keyword/phrase triggers for a specific database.
Deirdre
While I think the basic idea behind the recommendations is a good one, I’d like to turn them off. We are sending people to MultiSearch touting it as a one-stop-shop for their general information needs…and the first thing they are confronted with after doing their search is not the search results. It is a notice from the one-stop-shop more or less telling them to go have a look somewhere else more "specialized" with a good idea lightbulb glowing cheerily.I think that this is counter to the message we are trying to convey.
Kathryn
I agree with Kathryn, a search for self-compacting concrete brought a recommendation for Compendex -very relevant, and Environmental sciences and pollution management – perhaps relevant for some aspects. However placing of this advice at the top of the search results is counterproductive.
Eldnet – the Engineering Library discussion list in the US has been receiving lots of comments about inappropriate recommendations.
Christine
I don’t suppose we can mess with the CSS? Because if it was a column on the right (about as wide as it’s now high) it’d be out of the way of the search results but still available for information for those who think that MultiSearch is the only tool they’ll ever need.
I don’t think the occasional mistake is too terrible – each recommendation explains what it is so students can easily see whether or not it’s going to suit them.
Another thing though is the wording – "We found a specialised collection" is so vague, the whole box could just be seen as ads for things that the students might have to pay for. (After all its formatting isn’t so different from Google Ads.)
Deborah
A message from Serials Solutions:
Summon Clients,
Thank you for all of the great discussion about the new Database Recommender capabilities released on Tuesday! Results-set analysis clearly affords more interpretive recommendations, but several of the suggestions appearing after release were more like creative association than valid recommendations. These types of recommendations were not at all consistent with our expectations or our quality assurance testing of the feature.
An immediate investigation uncovered an issue resulting in many of the odd recommendations. While not to delve into the technical details, the root cause related to an interplay between the new version of the Database Recommender code and the new re-index that also went live this week. The relevancy tuning of the recommender was performed while the re-index was in progress. Hence, tuning and testing was done on the previous and somewhat smaller index. Your feedback which prompted this investigation also provided good counter examples for tuning.
I am happy to report that we have resolved the issue in a release a few hours ago. As result you will encounter many fewer odd recommendations. Also, you will see fewer recommendations in general. Note that we are not done. Just as with search relevancy, we will continue to evaluate and tune the algorithms powering the Database Recommender. Please keep your feedback coming!
Deirdre