Jazz Odyssey

An article in today’s New York Times discusses the different types of recommendation systems. Recommendation systems are programs which attempt to recommend products a consumer may be interested in using combinations of explicit data (such as user solicited ratings) and implicit data (such as a record of the consumers previous purchases). The program then compares the collected data to similar data collected from others and creates a group of purchase suggestions for the consumer.

These systems have been around since the early days of ecommerce, with Amazon.com’s system being the one that most consumers are familiar with. The article takes a look at the current state of these systems, showing their potential as well as some of the negative issues, such as random or inappropriate/offensive recommendations. In general, the size of the database will determine the relative efficiency of a recommendation system (i.e., with Amazon.com’s millions of customers and transactions, their enormous database and years of purchasing history has enough critical mass to achieve a reasonable level of accuracy….smaller sites with fewer customers have less data to rely on and therefore are more prone to make recommendations that are less accurate). The article also talks about some next generation systems such as Liveplasma.com, which combines existing recommendation system technologies with more compelling graphical presentations and Pandora, which also adds the z-axis of a “panel of experts” as an additional data point to help make recommendations with even greater accuracy.

The next logical step in this technology is the integration of social networking…one such project already underway is Last.fm a clever combination of music profiling, iTunes playlist sharing and community that is growing quickly, and very much worth checking out.