Recommender systems are popular tools for assisting customers in navigating through huge archives. Based item/product catalogue, a set of historical user actions upon these items and - perhaps - some meta data concerning users and/or products recommendations are derived by finding similarity relations, behavioral patterns, etc.
In the context of Big Data this mining of useful patterns is getting more complex. Possible strategies to handle this challenge are (i) parallelizing computation and (ii) adapting models to handle large data (e.g. pruning the calculation base).
Solutions tackling the topics above should be developed and evaluated in the context of a real recommender system.
- Adapting recommender algorithms (to be defined in cooperation with supervisor and student) for parallel computation using Map/Reduce
- Development of collaborative filtering models dealing with huge data
This is an applied project together with Smart Engine.