Monday, April 16, 2007

Personal recommendations

Regarding recommender systems, the most familiar are known to be related to audio content, i. e. personal radio. For instance, there's Pandora, which works based on manually generated metadata, and it also explains why a title was recommended. On the other hand, there's Last.fm, subtitled the social music revolution, which apparently uses some kind of collaborative filtering (a similar approach is already familiar from Amazon). In both cases, the service "learns" from user ratings to better serve the end user. Other recommendation approaches rely on communities or content analysis.

Now there may be several criteria regarding popularity of recommendation-based systems, but those that are really based on such a feature (unlike Amazon, which uses that as an additional feature to better serve their customers) seem to be dependent on two core issues: the quality of their recommendations, as experienced by the end user, and the required effort to handle available content (e. g. metadata management).

If IP based entertainment services are to succeed - as compared to good old radio or television -, personalization seems to be a must. I am sure that broadcasting companies, many of which are working on providing their programs in digital archives already, will understand this as an added value and, possibly, an opportunity to generate additional business.

No comments: