DeepPeep was a search engine that aimed to crawl and index every database on the public Web. Unlike traditional search engines, which crawl existing webpages and their hyperlinks, DeepPeep aimed to allow access to the so-called Deep web, World Wide Web content only available via for instance typed queries into databases. The project started at the University of Utah and was overseen by Juliana Freire, an associate professor at the university's School of Computing WebDB group. The goal was to make 90% of all WWW content accessible, according to Freire. The project ran a beta search engine and was sponsored by the University of Utah and a $243,000 grant from the National Science Foundation. It generated worldwide interest.
How it works
However, what separated DeepPeep and other search engines is that DeepPeep uses the ACHE crawler, 'Hierarchical Form Identification', 'Context-Aware Form Clustering' and 'LabelEx' to locate, analyze, and organize web forms to allow easy access to users.
The ACHE Crawler is used to gather links and utilizes a learning strategy that increases the collection rate of links as these crawlers continue to search. What makes ACHE Crawler unique from other crawlers is that other crawlers are focused crawlers that gather Web pages that have specific properties or keywords. Ache Crawlers instead includes a page classifier which allows it to sort out irrelevant pages of a domain as well as a link classifier which ranks a link by its highest relevance to a topic. As a result, the ACHE Crawler first downloads web links that has the higher relevance and saves resources by not downloading irrelevant data.
Hierarchical Form Identification
In order to further eliminate irrelevant links and search results, DeepPeep uses the HIerarchical Form Identification (HIFI) framework that classifies links and search results based on the website's structure and content. Unlike other forms of classification which solely relies on the web form labels for organization, HIFI utilizes both the structure and content of the web form for classification. Utilizing these two classifiers, HIFI organizes the web forms in a hierarchical fashion which ranks the a web form's relevance to the target keyword.
When there is no domain of interest or the domain specified has multiple types of definition, DeepPeep must separate the web form and cluster them into similar domains. The search engine uses context-aware clustering to group similar links in the same domain by modeling the web form into sets of hyperlinks and using its context for comparison. Unlike other techniques that require complicated label extraction and manual pre-processing of web forms, context-aware clustering is done automatically and uses meta-data to handle web forms that are content rich and contain multiple attributes.
DeepPeep further extracts information called Meta-Data from these pages which allows for better ranking of links and databases with the use of LabelEx, an approach for automatic decomposition and extraction of meta-data. Meta-data is data from web links that give information about other domains. LabelEx identifies the element-label mapping and uses the mapping to extract meta-data with accuracy unlike conventional approaches that used manually specific extraction rules.
When the search results pop up after the user has input their keyword, DeepPeep ranks the links based on 3 features: term content, number of backlinks. and pagerank. Firstly, the term content is simply determined by the content of the web link and its relevance. Backlinks are hyperlinks or links that direct the user to a different website. Pageranks is the ranking of websites in search engine results and works by counting the amount and quality of links to website to determine its importance. Pagerank and back link information are obtained from outside sources such as Google, Yahoo, and Bing.
DeepPeep Beta was launched and only covered seven domains: auto, airfare, biology, book, hotel, job, and rental. Under these seven domains, DeepPeep offered access to 13,000 Web forms. One could access the website at deeppeep.org, but the website has been inactive after the beta version was taken down.
- Wright, Alex (2009-02-22). "Exploring a 'Deep Web' That Google Can't Grasp". The New York Times. Retrieved 2009-02-23.
- Franke, Susanne (2009-02-24). "DeepPeep: Forscher wollen verborgene Datenbanken im Web zugänglich machen" [DeepPeep: Researchers want to make hidden databases accessible on the web]. Comp. Ztg. Retrieved 2009-02-25 – via lanline.de.
- Warwick, Martyn (2009-02-25). "DeepPeep lets light in to the hidden Web". TelecomTV. Retrieved 2009-02-25.[permanent dead link]
- Sawant, Nimish (2010-03-09). "Crawling the deep web". LiveMint. Mint. Retrieved 2010-12-13.
- "Main Page". WebDB. University of Utah School of Computing. 2008-10-04. Archived from the original on 2009-02-27. Retrieved 2009-02-23.
- Pichler, Thomas (2009-02-23). "Suchansätze dringen in die Tiefen des Internets: Erforschen von Datenbanken als wichtiger Schritt" [Search phrases penetrate the depths of the Internet: Researching databases as an important step] (in German). Pressetext. Retrieved 2009-02-23.
- "Suchansätze dringen in die Tiefen des Internets" [Search phrases penetrate the depths of the Internet]. nachrichten.ch (in German). 2009-02-24. Archived from the original on 2011-07-07. Retrieved 2010-12-13.
- "Award Abstract #0713637: III-COR: Discovering and Organizing Hidden-Web Sources". NSF Award Search. National Science Foundation. Retrieved 2009-02-23.
- "Esplorando il DeepWeb, i fondali della Rete dove Google non arriva" [Exploring the DeepWeb, the depths of the Net where Google does not arrive]. Liberta di Stampa Diritto all'Informazione (This is an Italian translation of the New York Times article "Exploring a ‘Deep Web’ That Google Can’t Grasp" by Alex Wright) (in Italian). Italy. 2009-04-05. Retrieved 2009-03-05.
- Sándor, Berta (2009-02-24). "Az internet mélyét kutatja a DeepPeep" [The internet researching the depths of DeepPeep]. sg.hu (in Hungarian). SG (Hungary). Retrieved 2009-03-05.
- "Niet alles is te vinden met Google" [Not everything can be found with Google] (in Dutch). Dutch Cowboys. 2009-03-04. Retrieved 2009-03-05.
- "探索谷歌尚未把持的'深层网络'" [Explore Google's not yet dominated 'deep network'] (This is a Chinese translation of the New York Times article "Exploring a ‘Deep Web’ That Google Can’t Grasp" by Alex Wright) (in Chinese). 2006-03-03. Archived from the original on 2011-07-07. Retrieved 2009-03-05.
- "Sfida al deep web: la Kosmix prova a svelare le pagine nascoste di internet" [Challenge to the deep web: Kosmix tries to reveal the hidden pages of the internet]. Messagg. 2009-02-23. Archived from the original on 2012-08-04. Retrieved 2010-12-13.
- Barbosa, Luciano; Nguyen, Hoa; Nguyen, Thanh; Pinnamaneni, Ramesh; Freire, Juliana (2010-01-01). "Creating and Exploring Web Form Repositories". Proceedings of the 2010 ACM SIGMOD International Conference on Management of Data. SIGMOD '10. New York, NY, USA: ACM: 1175–1178. doi:10.1145/1807167.1807311. ISBN 9781450300322.
- "ViDA-NYU/ache". GitHub. Retrieved 2016-11-06.
- Duygulu, Pinar (1999-12-22). "Hierarchical representation of form documents for identification and retrieval". Proceedings of SPIE. 3967 (1). doi:10.1117/12.373486. ISSN 0277-786X.
- Beckett, Andy (2009-11-25). "The dark side of the internet". The Guardian. ISSN 0261-3077. Retrieved 2016-11-06.