Tagsplanations: explaining recommendations using tags
Comments:
Cindy Skach
Michael Atkinson
Reference:
Vig, J., Sen, S., and Riedl, J. Tagsplanations: explaining recommendations using tags. IUI '09. http://delivery.acm.org/10.1145/1510000/1502661/p47-vig.pdf?key1=1502661&key2=6271472031&coll=DL&dl=ACM&ip=165.91.4.169&CFID=16209705&CFTOKEN=47524431
Summary:
Tagsplanations is a design implemented on a movie recommender website that not only made recommendations to users but explained why the recommendation was made. This is the novel point about this design. It is split into both relevance of an item and preference of an item. For instance, if an item is similar in feature, the relevance would be high, but if the user preferred a certain type of item or brand, the preference would be high.An experiment was conducted in order to see how users responded to the different types of recommendations. The conclusions were that tag preference was more important than tag relevance. Concerning effectiveness, the two were about the same. For mood compatibility, relevance was rated higher.
Discussion:The thought of telling people why an item was put on the "items you may enjoy" list in my opinion would be a very beneficial idea. For books on Amazon I can usually understand why they place the items there because most of the books are about computer science. However when it comes to other items sometimes I am oblivious to the reason they appear on my recommendations. Thus, this design I see as benefiting users. The paper was well organized and it was easy to track what they did and how they did it due to the nice section headings.
Future work could include testing this design in more than one domain. Since they only tested it on a movie recommendation website, they could try a site that sells accessories.
I agree that the paper was straightforward and easy to understand. I also really liked how they displayed information in their design. They were able to condense it down into a way that was both easy to understand and informative for the user.
ReplyDeleteThis is a really interesting idea. Like you all have noted, some suggestions are not always well founded, however if we could see the reasoning behind it we could understand the connection. I would really like to see this out there soon...
ReplyDeleteI really like this idea. I agree that the choices recommending services make are often mysterious and I would like to know why they chose what they chose.
ReplyDeletePandora actually does a really good job of doing this. It fully explains all of the reasons why it recommends each song.
Community tagging probably prevents tag-system abuse. I wish they had discussed this in the paper.
ReplyDeleteI would personally like knowing why something is suggested to me. I usually ignore those sections because I don't think they actually get a good idea of what I like.
ReplyDelete