what s really new on the web
play

What's Really New on the Web? Identifying New Pages from a Series - PowerPoint PPT Presentation

Sep 20, 2022 •176 likes •473 views

What's Really New on the Web? Identifying New Pages from a Series of Unstable Web Snapshots Masashi Toyoda and Masaru Kitsuregawa IIS, University of Tokyo Web as a projection of the world Web is now reflecting various events in the real

  1. What's Really New on the Web? Identifying New Pages from a Series of Unstable Web Snapshots Masashi Toyoda and Masaru Kitsuregawa IIS, University of Tokyo

  2. Web as a projection of the world • Web is now reflecting various events in the real and virtual world • Evolution of past topics can be tracked by observing the Web • Identifying and tracking new information new information is important for observing new trends new trends – Sociology, marketing, and survey research War Online news Tsunami weblogs Sports BBS Computer virus

  3. Observing Trends on the Web (1/2) • Recall (Internet Archive) [Patterson 2003] – # pages including query keywords

  4. Observing Trends on the Web (2/2) • WebRelievo [Toyoda 2005] – Evolution of link structure

  5. Periodic Crawling for Observing Trends on the Web WWW WWW Crawler Crawler Comparison Comparison Archive Archive T 1 T 2 Time T N

  6. Difficulties in Periodic Crawling (1/2) • Stable crawls miss new information – Crawling a fixed set of pages [Fetterly et al 2003] ↑ Can identify changes in the pages ↓ Overlook new pages – Crawling all the pages in a fixed set of sites [Ntoulas et al 2004] ↑ Can identify new pages in these sites ↓ Overlook new sites ↓ Possible only on a small subset of sites • Massive crawls are necessary for discovering new pages and new sites

  7. Difficulties in Periodic Crawling (2/2) • Massive crawls make snapshots unstable unstable – Cannot crawl the whole of the Web • # of uncrawled pages overwhelms # of crawled pages even after crawling 1B pages [Eiron et al 2004] – Novelty of a page crawled for the first time – Novelty of a page crawled for the first time remains uncertain remains uncertain • The page might exist at the previous time • “Last-Modified” time guarantees only that the page is older than that time

  8. Our Contribution • Propose a novelty measure novelty measure for estimating the certainty that a newly crawled page is really new – New pages can be extracted from a series of unstable snapshots • Evaluate the precision, recall, and miss rate of the novelty measure • Apply the novelty measure to our Web archive search engine

  9. Basic Ideas • The novelty of a page p p is the certainty that p p appeared between t 1 and t t- -1 t – p p appears when it can first be crawled and indexed – – p p is new when it is pointed to only by new links – – If only new pages and links point to p p , p may also be novel p • The novelty measure can be defined recursively and can be calculated in a similar way to PageRank [Brin and Page 1998] • Reverse of the decay measure [Bar-Yossef et al 2004] – p p is decayed if p p points to dead or decayed pages –

  10. Novelty Measure • N(p ): The novelty of page p p (0 1) • N(p): – 1: The highest certainty that p p is novel – 0: The novelty of p p is totally unknown (not old) • Pages in a snapshot W(t ) are classified into W(t) old pages O(t ) and unknown pages U(t O(t) U(t) ) • Each page p in U(t ) is assigned N(p U(t) N(p) )

  11. Old and Unknown Pages Crawled pages: W(t W(t) ) U(t) ) U(t ? Crawled pages: W(t ? W(t- -1) 1) ? ? O(t) ) O(t t-1 t

  12. How to Define Novelty Measure If all in-links come from pages crawled last 2 times( L (t) ) L 2 2 (t) Crawled last 2 times L 2 (t) L 2 (t) New p ) 1 N(p) N(p t-1 t

  13. How to Define Novelty Measure If some in-links come from O(t) O(t)- -L L 2 (t) 2 (t) q ? New p ) 0.75 N(p) N(p t-1 t

  14. How to Define Novelty Measure If some in-links come from U(t ) ? U(t) q ? p ) ? N(p) N(p t-1 t

  15. How to Define Novelty Measure Determine the novelty measure recursively q ) 0.5 N(q) N(q 50% New p ) (3+0.5) / 4 N(p) N(p t-1 t

  16. Definition of Novelty Measure • : damping factor – probability that there were links to p p before t-1

  17. Experiments • Data set • Convergence of calculation • Distribution of the novelty measure • Precision and recall • Miss rate

  18. Data Set • A massively crawled Time Period Crawled pages Links Jul to Aug 17M 120M 1999 Japanese web Jun to Aug 17M 112M 2000 archive Oct 40M 331M 2001 – 2002: .jp only Feb 45M 375M 2002 Feb 66M 1058M 2003 – 2003 : Japanese Jul 97M 1589M 2003 pages in any domain Jan 81M 3452M 2004 May 96M 4505M 2004 Time Jul 2003 Jan 2004 May 2004 49M 61M 46M |L2(t)| 23M 14M 20M |O(t) - L2(t)| 25M 6M 30M |U(t)| 97M 81M 96M |W(t)|

  19. Convergence of Calculation • 10 iterations are sufficient for 0 < 3000000 Total difference from the previous iteration delta=0 delta=0.1 2500000 delta=0.2 2000000 1500000 1000000 500000 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Number of iterations

  20. Distributions of the Novelty Measure 20,000,000 2003-07 delta=0.2 18,000,000 2003-07 delta=0.1 • Have 2 peaks on 0 and MAX 16,000,000 2003-07 delta=0.0 14,000,000 – cf. Power-law of in-link distribution Number of pages 12,000,000 • Depend on the fraction of L 2 (t) 10,000,000 and U(t) 8,000,000 6,000,000 • Not change drastically by delta 4,000,000 except for the maximum value 2,000,000 0 =0 <=0.1 <=0.2 <=0.3 <=0.4 <=0.5 <=0.6 <=0.7 <=0.8 <=0.9 <=1.0 Novelty measure 20,000,000 20,000,000 2004-01 delta=0.2 18,000,000 2004-05 delta=0.2 18,000,000 2004-01 delta=0.1 2004-05 delta=0.1 2004-01 delta=0.0 16,000,000 16,000,000 2004-05 delta=0.0 14,000,000 14,000,000 Number of pages Number of pages 12,000,000 12,000,000 10,000,000 10,000,000 8,000,000 8,000,000 6,000,000 6,000,000 4,000,000 4,000,000 2,000,000 2,000,000 0 0 =0 <=0.1 <=0.2 <=0.3 <=0.4 <=0.5 <=0.6 <=0.7 <=0.8 <=0.9 <=1.0 =0 <=0.1 <=0.2 <=0.3 <=0.4 <=0.5 <=0.6 <=0.7 <=0.8 <=0.9 <=1.0 Novelty measure Novelty measure

  21. Precision and Recall • Given threshold , p is judged to be novel when < N(p N(p) ) p – Precision: #(correctly judged) / #(judged to be novel) – Recall: #(correctly judged) / #(all novel pages) • Use URLs including dates as a golden set – Assume that they appeared at their including time – E.g. http://foo.com/2004/05 – Patterns: YYYYMM, YYYY/MM, YYYY-DD Jul 2003 Jan 2004 May 2004 With old date (before t-1) 299,591 (33%) 87,878 (24%) 402,365 (33%) With new date (t-1 to t) 593,317 (65%) 270,355 (74%) 776,360 (64%) With future date (after t) 24,286 (2%) 7,679 (2%) 36,476 (3%) 917,194 (100%) 365,912 (100%) 1,215,201 (100%) Total

  22. Precision and Recall (1/2) 1 • Positive gives 0.9 80% to 90% precision in all 0.8 snapshots 0.7 Precision / Recall 0.6 • Precision jumps from the 0.5 baseline when becomes 0.4 positive, then gradually 0.3 2003-07 Precision delta=0.2 increases 0.2 2003-07 Precision delta=0.1 0.1 • Positive delta values give 2003-07 Precision delta=0.0 0 slightly better precision 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Novelty measure min. threshold 1 1 0.9 0.9 0.8 0.8 0.7 0.7 Precision / Recall Precision / Recall 0.6 0.6 0.5 0.5 0.4 0.4 2004-01 Precision delta=0.2 0.3 0.3 2004-05 Precision delta=0.2 0.2 2004-01 Precision delta=0.1 0.2 2004-05 Precision delta=0.1 0.1 0.1 2004-01 Precision delta=0.0 2004-05 Precision delta=0.0 0 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Novelty measure min. threshold Novelty measure min. threshold

  23. Precision and Recall (2/2) 1 • Recall drops according 0.9 0.8 to the distribution of 0.7 Precision / Recall novelty measure 0.6 0.5 • Positive delta values 0.4 decrease the recall 0.3 2003-07 Precision delta=0.2 2003-07 Precision delta=0.1 0.2 2003-07 Precision delta=0.0 2003-07 Racall delta=0.0 0.1 2003-07 Racall delta=0.1 2003-07 Recall delta=0.2 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Novelty measure min. threshold 1 1 0.9 0.9 0.8 0.8 0.7 0.7 Precision / Recall Precision / Recall 0.6 0.6 0.5 0.5 0.4 0.4 2004-01 Recall delta=0.0 0.3 0.3 2004-05 Precision delta=0.2 2004-01 Recall delta=0.1 2004-05 Precision delta=0.1 2004-01 Recall delta=0.2 0.2 0.2 2004-05 Precision delta=0.0 2004-01 Precision delta=0.2 2004-05 Recall delta=0.0 2004-01 Precision delta=0.1 0.1 0.1 2004-05 Recall delta=0.1 2004-01 Precision delta=0.0 2004-05 Recall delta=0.2 0 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Novelty measure min. threshold Novelty measure min. threshold

  24. Guideline for Selecting Parameters • When higher precision is required – 0 < < 0.2 – Higher • When higher recall is required – = 0 – Small positive

  25. Miss Rate • Fraction of pages miss-judged to be novel – Use a set of old pages as a golden set • Last-Modified time < t -1 – Check how many pages are assigned positive N values N Time # old pages in U(t) |U(t)| Jul 2003 4.8M 25M Jan 2004 0.17M 6M May 2004 3.8M 30M

Recommend

More recommend