combining collaborative filtering with carbon footprint
play

Combining Collaborative Filtering with Carbon Footprint Calculation - PowerPoint PPT Presentation

Combining Collaborative Filtering with Carbon Footprint Calculation Joel Ross, Nitin Shantharam, Bill Tomlinson Department of Informatics University of California, Irvine ISSST 2010 May 18, 2010 Tuesday, May 18, 2010 Carbon Footprint The


  1. Combining Collaborative Filtering with Carbon Footprint Calculation Joel Ross, Nitin Shantharam, Bill Tomlinson Department of Informatics University of California, Irvine ISSST 2010 May 18, 2010 Tuesday, May 18, 2010

  2. Carbon Footprint The total amount of carbon dioxide directly and indirectly caused by an activity or accumulated over the lifetime of a product. Tuesday, May 18, 2010

  3. Online Footprint Calculators Tuesday, May 18, 2010

  4. Limitations of Carbon Calculators Interaction requires user time, effort, and knowledge Leads to a restricted scope Focus on the individual rather than the community Encourages purchasing offsets, not collective action Tuesday, May 18, 2010

  5. Collaborative Filtering A Definition: Collaborative Filtering is the process of filtering information for (or making predictions about) an unknown user based on information about a known group of users. Primarily used in recommender systems (e.g., Netflix, Amazon) Tuesday, May 18, 2010

  6. How does Collaborative Filtering Work? ?? Movie A Tuesday, May 18, 2010

  7. How does Collaborative Filtering Work? ?? Movie A Movie B Movie C Tuesday, May 18, 2010

  8. How does Collaborative Filtering Work? ?? Movie A Movie B Movie C Tuesday, May 18, 2010

  9. How does Collaborative Filtering Work? ?? Movie A Movie B Movie C Tuesday, May 18, 2010

  10. How does Collaborative Filtering Work? Movie A Movie B Movie C Tuesday, May 18, 2010

  11. Tuesday, May 18, 2010

  12. Tuesday, May 18, 2010

  13. Tuesday, May 18, 2010

  14. Tuesday, May 18, 2010

  15. Tuesday, May 18, 2010

  16. Tuesday, May 18, 2010

  17. Tuesday, May 18, 2010

  18. Algorithm Details User similarity determined through cosine similarity Normalize each variable 0 <= x <= 1 Construct an ordered [miles_driven, num_flights, electricity_ used, food_costs, ...] vector Determine cosine of vector angle Estimate is a simple average weighted by similarity: Tuesday, May 18, 2010

  19. Calculator Accuracy Tested with 397 users recruited through Mechanical Turk. Mean Absolute Error (MAE): The average amount that estimates deviate from the true value. U.S. Average Population Average Cosine Similarity as estimate as estimate estimate Avg. MAE: 0.145 Avg. MAE: 0.083 Avg. MAE: 0.078 (14.5% error) (8.3% error) (7.8% error) Tuesday, May 18, 2010

  20. Evaluating the Interaction 100 users compared Better Carbon to other major calculators 75% : Better Carbon as quick or quicker 62% : Better Carbon as easy or easier 30% : Better Carbon perceived as accurate as others 56% : Better Carbon created a stronger link between users and their communities "I liked the statements from my locality at the end of the Better Carbon - that brought it home." Tuesday, May 18, 2010

  21. Limitations of Better Carbon Still provides an estimate based on self-reported data Reducing an estimate is not the same as reducing impact Faster and easier... a problem? There may be benefits to manually working through a calculator Tuesday, May 18, 2010

  22. Future Work + Tuesday, May 18, 2010

  23. Summary: Better Carbon Better Carbon ( bettercarbon.com ) Uses collaborative filtering for carbon footprint calculation Generates estimates with better-than-average accuracy Is extendable; includes more factors without more user effort Provides a stronger social basis for carbon footprints Tuesday, May 18, 2010

  24. Summary: Better Carbon Better Carbon ( bettercarbon.com ) Uses collaborative filtering for carbon footprint calculation Generates estimates with better-than-average accuracy Is extendable; includes more factors without more user effort Provides a stronger social basis for carbon footprints Contact Information Acknowledgments Joel Ross Thanks to the Social Code Group. Informatics, UC Irvine This material is based in part upon work supported by the National Science Foundation under Grant jwross@uci.edu No. 0644415, by the Alfred P . Sloan Foundation, ics.uci.edu/~jwross and by Amazon Web Services. Tuesday, May 18, 2010

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend