The E ectiveness of Internet Content Filters ff Philip

As par t of its defense of t he Child Online Protection Act, which seeks to p revent minors from viewing com mercially p ublished har mful to minors m a terial on the World Wide Web, t he U.S. Depar t men t of Jus tice com missioned a s tu dy of t he p revalence of adult m a te rials and the effectiveness of Internet conten t filters in blocking the m. As of 2005 6, abou t 1.1% of webpages indexed by Google and MSN were adult h un dreds of millions of pages. About 6% of a se t of 1.3 billion searches executed on AOL, MSN and Yahoo! in su m mer 2005 re t rieve a t leas t one adult webpage a mong the firs t ten res ul ts, and abou t 1.7% of t he firs t t he resul ts are adult webpages. These es timates are based on sim ple rando m sa m ples of webpages indexed by search engines an d a s t ra tified rando m sa m ple of searches. Webpages with sexually explicit con tent intended for adult en ter tain ment (i.e., no t in an educational, m e dical or ar tis tic con text) were u sed to tes t a variety of Internet con ten t filters for u n derblocking failing to block webpages t ha t t hey are intended to block. A rando m sa m ple of clean webpages with no sexual con ten t or reference to sex was u sed to tes t t he filters for overblocking blocking webpages t hey are not intended to block. Webpages re t rieved by the m os t po p ular searches according to Wordt racker were also categorized an d u sed to tes t filters. Generally, filters with lower ra tes of u n derblocking had higher ra tes of overblocking. If the filter m os t effective a t blocking adult m a terials were applied to search indexes, typical query resul ts or t he resul ts of pop ular queries, t he n u m ber of clean pages blocked in er ror would exceed the n u m ber of adult pages blocked correctly. Acknowledgments. This work was su p por ted by the U.S. Depar t men t of Jus tice. I te s tified on behalf of t he U.S. a t t rial in ACLU v. Gonzales, no. 98 5591 (E.D. Pa.) an d s ub mit ted exper t declara tions in t he rela ted m a t ter Gonzales v. Google, no. CV 06 8006MISC JW (N.D. Ca., San Jose Div.). Much of t he da ta collection was perfor med by a grou p led by Paul Mewett a t CRA International. I a m gra teful to David Freed man, Raphael Gome z, Theodore Hirt, Joel McElvain an d an anony mous referee for helpful conversa tions or com men ts on an earlier d raf t.