Finding More Bilingual Webpages with High Credibility via Link Analysis Chengzhi Zhang, Nanjing University of Science and Technology Xuchen Yao, Johns.

Finding More Bilingual Webpages

with High Credibility

via Link Analysis

Chengzhi Zhang , Nanjing University of Science and Technology

Xuchen Yao , Johns Hopkins University

Chunyu Kit , City University of Hong Kong

8 August 2013

BUCC2013, Sofia, Bulgaria

3 ideas

• Bilingual URL Pattern Detection

• Deep Webpage Recovery

• Incremental Bilingual Website Exploration

Bilingual URL Pattern Detection

• a URL pattern: <en, zh> (Kit and Ng, 2007)– www.legco.gov.hk/yr99-00/en/fc/esc/e0.htm– www.legco.gov.hk/yr99-00/zh/fc/esc/e0.htm

• Improvement:– pairing up speed goes up from O(|U|2) to O(|U|)

• U is the set of all URLs within a website• approach: inverted index for URLs

– token-based pair -> char-based pair• weak pairs: <1e, 1c>, <2e, 2c>, ...

– http://.../1e/i.html <-> http://.../1c/i.html• enchanced: <e,c>

– supports multiple languages• better mining multilingual websites such as EU and UN

Bilingual URL Pattern Detection

• a URL pattern: <en, zh> (Kit and Ng, 2007)– www.legco.gov.hk/yr99-00/en/fc/esc/e0.htm– www.legco.gov.hk/yr99-00/zh/fc/esc/e0.htm

• Improvement:– pairing up speed goes up from O(|U|2) to O(|U|)

• U is the set of all URLs within a website• approach: inverted index for URLs

– token-based pair -> char-based pair• weak pairs: <1e, 1c>, <2e, 2c>, ...

– http://.../1e/i.html <-> http://.../1c/i.html• enchanced: <e,c>

– supports multiple languages• better mining multilingual websites such as EU and UN

Bilingual URL Pattern Detection

• a URL pattern: <en, zh> (Kit and Ng, 2007)– www.legco.gov.hk/yr99-00/en/fc/esc/e0.htm– www.legco.gov.hk/yr99-00/zh/fc/esc/e0.htm

• Improvement:– pairing up speed goes up from O(|U|2) to O(|U|)

• U is the set of all URLs within a website• approach: inverted index for URLs

– token-based pair -> char-based pair• weak pairs: <1e, 1c>, <2e, 2c>, ...

– http://.../1e/i.html <-> http://.../1c/i.html• enchanced: <e,c>

– supports multiple languages• better mining multilingual websites such as EU and UN

Bilingual URL Pattern Detection

• a URL pattern: <en, zh> (Kit and Ng, 2007)– www.legco.gov.hk/yr99-00/en/fc/esc/e0.htm– www.legco.gov.hk/yr99-00/zh/fc/esc/e0.htm

• Improvement:– pairing up speed goes up from O(|U|2) to O(|U|)

• U is the set of all URLs within a website• approach: inverted index for URLs

– token-based pair -> char-based pair• weak pairs: <1e, 1c>, <2e, 2c>, ...

– http://.../1e/i.html <-> http://.../1c/i.html• enchanced: <e,c>

– supports multiple languages• better mining multilingual websites such as EU and UN

Top 20 Keys

Number of matched URLs (top 10)

Keys in domain “gov.hk”(rescue local weak keys if they are globally strong)

3 ideas

• Bilingual URL Pattern Detection

• Deep Webpage Recovery

• Incremental Bilingual Website Exploration

Deep Webpage Recovery

• deep webpage: pages that are not linked by any other static pages (not searchable) until created dynamically– mostly triggered by JavaScript or Flash actions

• http://www.fehd.gov.hk/tc_chi/cagenda 20070904.htm– we have discovered patterns <tc_chi, english>, <tc_chi, en>,

<tc_chi, eng>, ..., then try:– wget http://www.fehd.gov.hk/english/cagenda 20070904.htm– wget http://www.fehd.gov.hk/en/cagenda 20070904.htm– wget http://www.fehd.gov.hk/eng/cagenda 20070904.htm– ...

various structures of websites with deep pages

Deep Webpage Recovery

• deep webpage: pages that are not linked by any other static pages (not searchable) until created dynamically– mostly triggered by JavaScript or Flash actions

• http://www.fehd.gov.hk/tc_chi/cagenda 20070904.htm– we have discovered patterns <tc_chi, english>, <tc_chi, en>,

<tc_chi, eng>, ..., then try:– wget http://www.fehd.gov.hk/english/cagenda 20070904.htm– wget http://www.fehd.gov.hk/en/cagenda 20070904.htm– wget http://www.fehd.gov.hk/eng/cagenda 20070904.htm– ...

Deep Webpage Recovery

• deep webpage: pages that are not linked by any other static pages (not searchable) until created dynamically– mostly triggered by JavaScript or Flash actions

• http://www.fehd.gov.hk/tc_chi/cagenda 20070904.htm– we have discovered patterns <tc_chi, english>, <tc_chi, en>,

<tc_chi, eng>, ..., then try:– wget http://www.fehd.gov.hk/english/cagenda 20070904.htm– wget http://www.fehd.gov.hk/en/cagenda 20070904.htm– wget http://www.fehd.gov.hk/eng/cagenda 20070904.htm– ...

Deep Webpage Recovery

• deep webpage: pages that are not linked by any other static pages (not searchable) until created dynamically– mostly triggered by JavaScript or Flash actions

• http://www.fehd.gov.hk/tc_chi/cagenda 20070904.htm– we have discovered patterns <tc_chi, english>, <tc_chi, en>,

<tc_chi, eng>, ..., then try:– wget http://www.fehd.gov.hk/english/cagenda 20070904.htm– wget http://www.fehd.gov.hk/en/cagenda 20070904.htm– wget http://www.fehd.gov.hk/eng/cagenda 20070904.htm– ...

3 ideas

• Bilingual URL Pattern Detection

• Deep Webpage Recovery

• Incremental Bilingual Website Exploration

Incremental Bilingual Website Exploration

• Intuition: bilingual websites tend to link to other bilingual websites.

• Measures:– Linkout(w): total number of outgoing links from website w– PageRank(w): (Brin and Page, 1998)

– WeightedPageRank(w): weighted by "how bilingual" w is (the more bilingual URLs, the more "bilingual" w is)

Incremental Bilingual Website Exploration

• Intuition: bilingual websites tend to link to other bilingual websites.

• Measures:– Linkout(w): total number of outgoing links from website w– PageRank(w): (Brin and Page, 1998)

– WeightedPageRank(w): weighted by "how bilingual" w is (the more bilingual URLs, the more "bilingual" w is)

Incremental Bilingual Website Exploration

• Intuition: bilingual websites tend to link to other bilingual websites.

• Measures:– Linkout(w): total number of outgoing links from website w– PageRank(w): (Brin and Page, 1998)

– WeightedPageRank(w): weighted by "how bilingual" w is (the more bilingual URLs, the more "bilingual" w is)

Incremental Bilingual Website Exploration

• Intuition: bilingual websites tend to link to other bilingual websites.

• Measures:– Linkout(w): total number of outgoing links from website w– PageRank(w): (Brin and Page, 1998)

– WeightedPageRank(w): weighted by "how bilingual" w is (the more bilingual URLs, the more "bilingual" w is)

Discovering related webistes from seed websites(select the top K most related websites)

[Linkout, PageRank, WeightedPageRank]

Evaluationon number of URL pairs found and precision

total websites: 12,800

Conclusion

• Unsupervised bilingual pair detection (no heuristics)– http://mega.ctl.cityu.edu.hk/~czhang22/pupsniffer-eval/Data/

Pattern_Credibility_LargeThan100.txt

• A large collection of English-Chinese webpages