(c) Maria Indrawan Distributed Information Retrieval
(c) Maria Indrawan Challenges in Managing Distributed Information No topology of the data organisation. Dynamic data. The size of the collection. No control over quality of the data. Multimedia data.
(c) Maria Indrawan Challenges-Human Factor Diversity of users –Expert to novice Ill-formed queries. Specific behaviour –Favour precision over recall (85% users only look at the first screen – Lan Huang A survey on Web Information Technology)
(c) Maria Indrawan Types of Distributed IR Directory –Yahoo Search Engine –Google, AskJeeves, Yahoo, Teoma Meta Search –Metacrawler, Dogpile Distributed Broker –Harvest
(c) Maria Indrawan Directory Listing Manually created –Yahoo, Google, MSN –Open Directory Project
(c) Maria Indrawan Directory Listing Automatic classification TERENA. – seac.htmlhttp:// seac.html Scorpion –
(c) Maria Indrawan Search Engine Architecture Crawler (robots) –Collecting the pages from the WEB. Indexer –Indexing pages collected by the crawler and represent them in an efficient data structure. Query Server –Accepting, process and return the results of the query from the user.
(c) Maria Indrawan Crawler – Design Considerations Crawling algorithm –Breadth-first vs Depth first How do we handle URL-aliases? How do we reduce server load? How do we detect a duplicate page or a mirror- site? How often we need to revisit a site?
(c) Maria Indrawan Update Rate (May 2003) Search EngineNewest page Found Rough AverageOldest Page Found Google2 days1 month165 days MSN (Ink)1 day4 weeks51 days HotBot (Ink)1 day4 weeks51 days AlltheWeb1 day1 month599 days Gigablast45 days7 months381 days Teoma41 days2.5 months81 days WiseNut133 days6 months183 days
(c) Maria Indrawan Indexer - Design Considerations How do we handle typing mistakes? Do we use stop list and stemming algorithm? How much do we want to index in a given web page? –Google index only the first 101 KB of a web page and 120 KB of PDF file. How big do we want the database indexed to be? –response time vs coverage Do we want to index PDF, PS files?
(c) Maria Indrawan Size Growth
(c) Maria Indrawan Estimated Size Dec 31,
(c) Maria Indrawan Query Server- Design Considerations Retrieval model. Complexity of the query syntax. HCI – human computer interface. Output display.
(c) Maria Indrawan Retrieval Model Traditional approach: –Keywords matching returns to many low quality matches – low precision. Search engines need a VERY high precision output – even on the expense of RECALL. How can we achieve this?
(c) Maria Indrawan Google Retrieval Model Utilise the popularity of a page –If a page has many other pages pointed to this page, the page must be very important. We can assign a high weight to this page during search. –If a page is pointed by a popular page, this page can be considered as important because it is referred by a reputable source (a popular page). –PageRank Function.
(c) Maria Indrawan PageRank Example
(c) Maria Indrawan Google Retrieval Model Utilise the anchor text. –Anchors often provide more accurate descriptions of web pages than the pages themselves. –Anchors may exist for documents which cannot be indexed by a text-based search engine. Utilise the appearance of the text. –Larger and bolder font text are weighted higher than other words.
(c) Maria Indrawan Results Overlap
(c) Maria Indrawan Metasearch Meta searches do not build their own index. They use the index of the existing search engines. When user posted a query to a meta search, the meta search sends the query to a number of search engines and collates the results. A list of metacrawler: – http://
(c) Maria Indrawan Meta Search metacrawler, –uses google, yahoo,askJeeves, About, Looksmart, Teoma, Overture, FindWhat. dogpile, –uses google, yahoo,askJeeves, About, Looksmart, Teoma, Overture, FindWhat
(c) Maria Indrawan Metasearch Design Issue Potential problems: –Translating the user query into a different query in a different search engine. –Query time is bounded by the least powerful (slowest) underlying system. –Combining results into a single ranked list is difficult. Effectiveness depend on heuristics and information passed back from underlying search engines. detecting overlap in the query results different scoring schemes (some do not use)
(c) Maria Indrawan Distributed Broker Information is indexed locally by geographical locations or institutional boundaries. –Suitable for supporting community that to have a common search database. Local indexes are combined to provide wider coverage. Document scoring is performed locally by each index server.
(c) Maria Indrawan Distributed Broker broker CSSE broker SIMS broker ACC broker MGM broker FIT broker F. Bussiness broker Monash
(c) Maria Indrawan Distributed Broker Example: Harvest – hing/schwartz.harvest/schwartz.harvest.htmlhttp:// hing/schwartz.harvest/schwartz.harvest.html
(c) Maria Indrawan General architecture Hierarchical vs Flat Hierarchical: underlying index servers are connected through a hierarchy of brokers. –broker hierarchy provides efficient and global coverage. –brokers can be geographical, institutional or subject based. broker query broker query broker index server...
(c) Maria Indrawan Flat Graph Model broker index server broker index server broker index server broker index server... query
(c) Maria Indrawan Useful site –Provides links to most of the information discovery tools.
(c) Maria Indrawan Summary Type of Distributed Information Discovery –Directory Listing yahoo –Search Engines. Google, AskJeeves, Teoma –Metasearch metacrawler, dogpile –Distributed Broker Harvest