Difference between revisions of "MyMediaLite"
Zeno Gantner (talk | contribs) |
Zeno Gantner (talk | contribs) |
||
| Line 12: | Line 12: | ||
Recommenders may use the collaborative data, plus additionally user/item attributes and user/item relations. | Recommenders may use the collaborative data, plus additionally user/item attributes and user/item relations. | ||
| − | MyMediaLite includes evaluation routines for rating prediction and item prediction; it can measure [[MAE]], [[RMSE]], [[AUC]], [[precision at N]], [[MAP]], [[NDCG]], and [[MRR]]. | + | MyMediaLite includes evaluation routines for rating prediction and item prediction; it can measure [[MAE]], [[RMSE]], [[CBD]], [[AUC]], [[precision at N]], [[MAP]], [[NDCG]], and [[MRR]]. |
It also comes with command line tools that expose most of the library's functionality, accepting data in a simple text-based input format compatible to the one used by [[Mahout]]. | It also comes with command line tools that expose most of the library's functionality, accepting data in a simple text-based input format compatible to the one used by [[Mahout]]. | ||
Revision as of 05:12, 17 January 2012
MyMediaLite is a lightweight, multi-purpose library of recommender system algorithms, written in C#. Using Mono, it runs on all major computing platforms.
It addresses the two most common scenarios in collaborative filtering:
- rating prediction (e.g. on a scale of 1 to 5 stars), and
- item prediction from implicit feedback (e.g. from clicks or purchase actions).
MyMediaLite is free software (open source software), it can be used and distributed under the terms of the GNU General Public License (GPL).
All recommender implementations in MyMediaLite are based on the same efficient and scalable infrastructure; for each recommendation task, there is an API and a base class that make it easy to implement new recommenders. Recommenders may use the collaborative data, plus additionally user/item attributes and user/item relations.
MyMediaLite includes evaluation routines for rating prediction and item prediction; it can measure MAE, RMSE, CBD, AUC, precision at N, MAP, NDCG, and MRR. It also comes with command line tools that expose most of the library's functionality, accepting data in a simple text-based input format compatible to the one used by Mahout.
Additional features:
- saving and reloading of recommender models
- real-time online updates for many recommenders
- attribute-based diversification of recommendation lists
Literature
- Zeno Gantner, Steffen Rendle, Christoph Freudenthaler, Lars Schmidt-Thieme: MyMediaLite: A Free Recommender System Library, RecSys 2011
External links
- http://ismll.de/mymedialite
- MyMediaLite at github.com
- early-stage Java port of MyMediaLite
- MyMediaLite at mloss.org
- Twitter: @mymedialite
- presentation at FOSDEM 2011
- blog post about MyMediaLite at MSDN Channel 9's Coding4Fun blog
- interview about MyMediaLite at Danny Bickson's blog
