Poster slam (2 minutes maximum) about my Doctoral Symposium presented at ACM Recommender Systems 2011

1. Predicting Performance in
Recommender Systems
ACM Conference on Recommender Systems 2011 – Poster slam
October 24, Chicago, USA IRG
IR Group @ UAM

2. Is it possible to predict the accuracy
of a recommendation?
Is it useful?
ACM Conference on Recommender Systems 2011 – Poster slam
October 24, Chicago, USA IRG
IR Group @ UAM

3. Hypothesis
Data available to a Recommender System
contains signals to predict the performance
ACM Conference on Recommender Systems 2011 – Poster slam
October 24, Chicago, USA IRG
IR Group @ UAM

4. We have defined several performance predictors
based on such signals
ACM Conference on Recommender Systems 2011 – Poster slam
October 24, Chicago, USA IRG
IR Group @ UAM

5. Application to dynamic recommender strategies
based on the expected performance
• Dynamic hybrid recommendation
• Dynamic neighbor selection in kNN
ACM Conference on Recommender Systems 2011 – Poster slam
October 24, Chicago, USA IRG
IR Group @ UAM

6. Some results
0.98 Standard kNN b) Neighbourhood size: 500
0.96 Clarity-enhanced kNN
• Good predictive power
0.94
0.92
0.90
MAE
0.88
0.86
• Dynamic recommenders
0.84
0.82
outperform static ones
0.80
10 20 30 40 50 60 70 80 90 10 20 30 40 50 60 70 80 90
% of ratings for training
ACM Conference on Recommender Systems 2011 – Poster slam
October 24, Chicago, USA IRG
IR Group @ UAM

7. Predicting Performance in
Recommender Systems
Alejandro Bellogín
Supervised by Pablo Castells and Iván Cantador
Escuela Politécnica Superior
Universidad Autónoma de Madrid
ACM Conference on Recommender Systems 2011 – Poster slam
October 24, Chicago, USA IRG
IR Group @ UAM