- So this leads to a few questions:
- Has this 'trust effect' in recommenders been studied at all? I'd be very interested to se some good pysch research on the topic.
- Assuming it really does exist, how do we, as developers of automated tools integrate such a thing into recommendation engines? Is it really as simple as making the 'best' recommendations possible? (I think probably no...)
- Putting together (1) and (2) what effect does 'developing a relationship over time' have on the viability of a recommendation engine?
Anyway, I believe this is one of the critical issues to the success of music recommenders in improving beyond 'sounds like' and 'others like you enjoy' type of things to a more dynamic and comprehensive recommendation service.
2 comments:
Nice post - I haven't really thought much about trust in music rec systems, but it's an important point. Seems there has been some work on trust aware rec systems and CF. I'd be keen to learn more...
that paper looks interesting, I'll have to give it a read. Yeah, when I wrote this, I was thinking a great deal on the building of a relationship over time and how this might affect the acceptability of a recommender's misses and this sort of thing. I think a really good DJ (radio or club or whatever) or music editor can, over the long term actually change the taste of some segment of the audience if that audience sufficiently trusts the recommenders taste (as an extension perhaps of the listeners own). Anyway, this may dove tail into your thoughts on expert versus mob decisions as well. To rephrase, how do you steer the mob? I think I need to expand this into a whole post, and sort out how to do some experiments on the topic.
Post a Comment