高级人工智能-Prominent streak discovery.pptVIP

Advanced AI课程 Autumn 2016 罗平 PROMINENT STREAK DISCOVERY 2-4 * Prominent Streak Discovery in Sequence Data. Xiao Jiang, Chengkai Li, Ping Luo, Min Wang, Yong Yu. KDD 2011, pages 1280-1288. Discovering General Prominent Streaks in Sequence Data. Gensheng Zhang, Xiao Jiang, Ping Luo, Min Wang, Chengkai Li. ACM TKDD, 8(2):article 9, June 2014. Prominent Streaks Prominent streaks stated in news articles: “This month the Chinese capital has experienced 10 days with a maximum temperature in around 35 degrees Celsius – the most for the month of July in a decade.” “The Nikkei 225 closed below 10000 for the 12th consecutive week, the longest such streak since June 2009.” “He (LeBron James) scored 35 or more points in nine consecutive games and joined Michael Jordan and Kobe Bryant as the only players since 1970 to accomplish the feat.” Concepts Streak Input: a sequence of values Streak [l, r], v is a triple: left-end ( l ), right-end ( r ), minimum value in interval [l,r] 3 1 7 7 2 5 4 6 7 3 [6, 8], 4 Streak dominance relation s1=[l1, r1], v1 dominates s2=[l2, r2], v2 iff r1 - l1 r2 - l2, v1 = v2 or r1 - l1 = r2 - l2, v1 v2 Prominent streaks (PS) A streak is prominent if it is not dominated by any other streaks. Example 3 1 7 7 2 5 4 6 7 3 Prominent Streaks are Skyline Points in 2-d Space 3 1 7 7 2 5 4 6 7 3 Tasks Task 1: discovery Find all prominent streaks in a sequence Task 2: monitoring Always keep prominent streaks up-to-date, when sequence grows (real-world sequences often grow) Solution Framework Skyline Operation [B?rzs?nyi et al. 2001] (many algorithms) Data Value Sequence Candidate Streaks Prominent Streaks 3 1 7 7 2 5 4 6 7 3 Candidate Generation Algorithms (brute-force, NLPS, LLPS) Candidate Generation: Number Of Candidates Brute-force Quadratic NLPS Superlinear LLPS Linear Local Prominent Streak Local dominance relat