Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Recommender Systems


Published on

Based on:
Recommender Systems by Prem Melville & Vikas Sindhwani

Presented by:
Vijayindu Gamage
Udith Gunaratna
Pubudu Gunatilaka

Published in: Technology, Education
  • Be the first to comment

Recommender Systems

  1. 1. Based on: Recommender Systems by Prem Melville & Vikas Sindhwani Presented by: Vijayindu Gamage Udith Gunaratna Pubudu Gunatilaka
  2. 2. LOGORecommender Systems
  3. 3. LOGORecommender Systems Structure of a Recommender System
  4. 4. LOGORecommender Systems Classification  Collaborative Filtering  Neighborhood-based Collaborative Filtering  Item-based Collaborative Filtering  Model-based Collaborative Filtering  Content Based Recommending  Hybrid Approaches
  5. 5. LOGORecommender Systems Neighborhood-based Collaborative Filtering Basic Steps  Assign a weight to all users with respect to similarity with the active user.  Select k users that have the highest similarity with the active user – (neighborhood)
  6. 6. LOGORecommender Systems  Compute a prediction from a weighted combination of the selected neighbors’ ratings.
  7. 7. LOGORecommender Systems Neighborhood-based CF - Problem LESS users … neighbors are EASY to find !
  8. 8. LOGORecommender Systems Neighborhood-based CF - Problem MANY users … neighbors are HARD to find !
  9. 9. LOGORecommender Systems Item-based Collaborative Filtering  Proposed in 2003  DOES NOT match similar users  DOES match similar items  Leads to faster online systems  Results in improved recommendations
  10. 10. LOGORecommender Systems Item-based Collaborative Filtering  Pearson correlation is used  Rating for item i for user a is predicted
  11. 11. LOGORecommender Systems More Extensions Highly correlated neighbors based on very few co-rated items Significance Weighting  multiply the similarity weight by a significance weighting factor  Default Voting  assume a default value for the rating for items that have not been explicitly rated  Inverse User Frequency  Universally loved/hated items are bad
  12. 12. LOGORecommender Systems Model-based Collaborative Filtering  Uses statistical models for predictions  Based on data mining and machine learning algorithms  Latent factor and Matrix factorization models have emerged as a state-of-the-art methodology  Netflix Prize competition
  13. 13. LOGORecommender Systems Content-based Recommending Pure collaborative filtering recommenders treat all users and items as atomic units Can make a better personalized recommendation by knowing more about a user or an item  Demographic information  Movie genres  Literary genres
  14. 14. LOGORecommender Systems Content-based Recommending User liked & Movie Genre Recommendation
  15. 15. LOGORecommender Systems Content-based Recommending Focused on recommending items with associated textual content 2 approaches  Treat as an Information Retrieval (IR) Task  Treat as a Classification Task
  16. 16. LOGORecommender Systems Hybrid Approaches Used to leverage the strengths of content-based and collaborative recommenders. Merging the list results to produce a final list. Content-boosted collaborative filtering
  17. 17. LOGORecommender Systems Evaluation Metrics Evaluation matrix is used to measure the quality of a recommender system. These systems are typical measured using predictive accuracy metrics 1. Mean Absolute Error (MAE) 2. Root Mean Squared Error (RMSE)
  18. 18. LOGORecommender Systems Mean Absolute Error 
  19. 19. LOGORecommender Systems Root Mean Squared Error (RMSE) 
  20. 20. LOGORecommender Systems Challenges and Limitations Sparsity Cold-Start Problem Fraud  push attacks  nuke attacks
  21. 21. LOGORecommender Systems Sparsity User ratings matrix is typically very sparse Effects collaborative filtering systems The problem system has a very high item- to user ratio. The system is in the initial stages of use. Solution - making assumptions about the data generation process
  22. 22. LOGORecommender Systems Cold-Start Problem New items and new users pose a significant challenge to recommender systems. New item problem – content-based approach to produce recommendations for all items, New user problem selecting items to be rated by a user so as to rapidly improve recommendation performance with the least user feedback
  23. 23. LOGORecommender Systems Fraud Push attacks  Increase the rating of their own products Nuke attacks  Lower the ratings of their competitors Item-based collaborative filtering is more robust to these attacks Content based methods are unaffected by profile injection attacks.
  24. 24. LOGORecommender Systems Content based or Collaborative filtering Advantages of CF over CB CF can perform in domains where there is not much content associated with items CF can also preform when content is difficult for a computer to analyze. CF system has the ability to provide serendipitous recommendations.