Spotting Trends: The Wisdom of the Few
0 likes1,670 views
This document discusses spotting trends on social media. It proposes that a small group of "trend spotters" and "trend makers" are responsible for identifying and spreading trends, not the most popular users. The authors analyze a dataset of 9,316 users, 6,395 items, and 21,252 votes to identify these influential users and introduce a trend-aware recommendation algorithm that leverages these trend setters to provide more relevant recommendations to other users. They evaluate their approach based on recall and precision metrics and conclude that distinguishing between popular and trending items improves recommendations over traditional popularity-based algorithms.
Spotting Trends: The Wisdom of the Few
- 1. Spotting Trends The Wisdom of the Few Xiaolan Sha Daniele Quercia Pietro Michiardi Matteo DellAmico
- 2. Our Work What are the trends one might be interested in?
- 3. What Is a Trend Popular Trending
- 4. Who Create Trends 鐔鐔 In鍖uentials Trend Makers ED (upload) BIN OM ESS A C OC PR Accidental Trend Spotters (vote)
- 5. Our Context SP Te E CI c De hnol og AL sig n- y Co -Sav ns v cio y us
- 6. Our Context A core group of users are main content contributors. 9,316 users, 6,395 items and 21,252 votes
- 7. Traditional item i p(u,i) = 1 if u likes i, otherwise 0 user u Items ALGORITHM Recommend Users Preference Matrix
- 8. Our Approach Trend Spotters Recommend Identify Trends Trends Trend Makers User Space Item Space Trend-aware Recommendation
- 9. A Special Few ? Features 鐔 Activity Trend Spotters Trend Makers Content Social Network SVM User Space Geography
- 10. A Special Few ? 鐔 if uploaders are trend makers of any quality Trends Logistic Regression Item Space if voters are trend spotters - only of high quality
- 11. Trend-aware trend i p(u,i) = 1 if u likes trend i user u SVD Recommend Trends Trends Trend Space Trend-aware Preference Matrix
- 12. Our Approach Trend Spotters Recommend Identify Trends Trends Trend Makers User Space Item Space Trend-aware Recommendation
- 13. Baseline item i p(u,i) = 1 if u likes i and item i is a trend, otherwise 0 user u Items ALGORITHM Recommend Users Preference Matrix
- 14. Evaluation Recall(N) = #hits / |T| Precision(N) = #hits / (N * |T|)
- 15. Summary Popular Trends Trend Spotters & Makers Exist ! Identify them improve the recommendation.
- 16. Spotting Trends: The Wisdom of the Few Xiaolan Sha, Daniele Quercia, Pietro Michiardi, Matteo DellAmico Thank You!