The original vision of Nutch, 14 years later: Building an open source search engine
- 1. The original vision of Nutch, 14 years later:
Building an open source search engine
Apache Big Data Europe 2016
sylvain@sylvainzimmer.com
@sylvinus
- 2. /usr/bin/whoami
• Jamendo (Founder & CTO, 2004-2011)
• TEDxParis (Co-founder, 2009-2012)
• dotConferences (Founder, 2012-)
• Pricing Assistant (Co-founder & CTO, 2012-)
- 3. "The original motivation for the Nutch project was to provide
a transparent alternative to the growing power of a
handful of private search services over most users’ view of
the Web.
CommerceNet Labs Technical Report, Nov 2004
However, as Nutch has been adopted with greater
enthusiasm by smaller organizations, the Nutch
Organization has de-emphasized operating a multi-
billion-page index in the public interest."
- 14. Agenda
• Values & tech choices
• Search engine components
• Challenges
• Opportunities
- 18. Privacy
• Results can be tailored by language/country, but
NOT by user/cookie/sessionid
• o/ Cache everything!
• Tor service: http://comsearchl2zlnre.onion
- 19. Participation & Pragmatism
• Use high-level languages as much as possible
(Python, Go)
• Embrace active communities (Spark, Elasticsearch)
• Use mainstream participation platforms, even if they
are nonfree (GitHub, Slack)
- 28. Specs
• HTML parsing & analysis
• Tokenization / NLP
• Static rankings
• Language detection
• I/O from crawls to databases
- 30. Common Search Pipeline
Doc sources
Common Crawl,
WARC files,
URLs ...
Filter
plugins
Document
parsing
Output
plugins
Data output
Database, file,
HDFS, S3, ...
- 33. Gumbocy
• Use Cython instead of ctypes
• Smaller API
• Tree traversal on the Cython side with basic
boilerplate/visibility support
https://github.com/commonsearch/gumbocy
- 40. Ranking formula
rank = f( static_score , dynamic_score( query ) )
Alexa
DMOZ
Blacklists
PageRank
...
ElasticSearch & Lucene
TF-IDF
BM25
...
- 45. Specs
• Send user query to databases
• Search-as-you-type
• HTML & JSON endpoints
• High performance
- 50. Funding / Scale
• Frugalism
• Caching
• In-kind services
• Individual donations / Foundation grants
• General economic incentives
- 51. Spam
• Email spam
• Wikipedia vandalism
• Algorithm complexity & scale
• Given enough eyeballs, all spam is shallow?
- 58. Research
• More people should know how to build search
engines
• Spam, Relevance, Large-scale data processing
• We need more open datasets!
- 60. Make the Web a better place!
• SEO
• Transparency
• Influence of money
• Public service