• Commonly held view; that while current search engines have limitations (too slow, find too much) future versions will find desired information in response to a few keywords. 85% of users use search engines (Georgia Tech)

  ---

Subject directories (aka subject gateways, jumpstations, linkfarms…)

• Human beings select and classify sites – a form of cataloguing
• E.g’s Yahoo! (general), Te Puna Web Directory (localised), SOSIG (specialised)

• Software "robot" or "spider" traverses web, adds URL’s to database, which can be searched.
• e.g.’s Alta Vista, Northern Light, ANZWERS.

• Need to admit that modern search engines are powerful, and surprisingly effective for specific classes of searches.
• Could these ideas be used in library catalogues?
• Some ways in which Search Engines are making effective searches easier:

• Make judgements of which items are most relevant to query, based on frequency, position, etc of search terms. Means we are less concerned with how many items found, than what’s in the first 20 ("first 20 precision").
• Often little information on how relevancy is arrived at. Concerns, e.g. advertising bias (GoTo - "Promote your site").

• Searches for items similar (eg. terms used, domain, etc) to one that searcher has deemed relevant. E.G. AltaVista

• Sort search on common characteristics, e.g. Northern Light. Enables a kind of sub-search.

• Synonyms automatically translated to search terms, e.g. "elderly" also searches "senior citizens" - used by Excite

• Attempts to automatically bias ranking to most "popular" sites: problems – may be self fulfilling; time lag, "popular" site not necessarily relevant to specific search.
• Hits, e.g. Direct Hit
• Boosts relevancy according to whether other searches followed link
• Links, e.g. Google
• Boosts relevancy according to number of links to site

• Really a form of cataloguing

• Precision usually good (so users feel satisfied); but recall (overlap) low.
• Bar-Ilan 1998 – low recall (max 50%), high precision, little overlap
• Clarke 1997 – relative recall 50-60%
• Hawking 1999 – commercial search engines sacrifice effectiveness for speed
• Schwartz 1998 – performance difficult to measure
• Pollock 1997 - users need understanding of Internet knowledge structures to carry out effective searches.

"indexable web"

• Lawrence99 estimate 16M web servers, only 2.8M on indexable web.

• Pages not linked to others on Web
• Databases
• Frames
• Javascript
• Image based information Not possible to find many organisations directly, since their pages only include name as graphic logo.
• Subscription only sites
• Robot exclusion sites

• Of indexable web, main search engines index 42%, best (Northern Light) only covers 16%. (Lawrence99)
• Search engines going for speed, profitablity rather than coverage. (AltaVista times out during busy periods)
• Little overlap between search engines, e.g. Notess00

Many different ways even a simple query can be expressed, e.g.

• +"native people" +internet
• +"indigenous culture" +web
• Users will give very different search statements for the same query (Bates98/Saracevic: only 1.5% of sample of users used same term for same query)
• Most people only use 2-3 terms, single word searches common. Examples from MetaSpy
• People think they get higher recall than they actually do (Blair and Maron 1985)
• Size of vocabulary: typical vocab is 50K words; therefore average hit on single word search of 800M pages of 500words each is 800M x 500 / 50K = 8M hits!
• Mis-spellings (mis-spelt words will retrieve something.) Example from AltaVista
• Search engines open to bias, e.g. Some conservative mid-west sites removed word "Gay" from Veronica indexes (Poulter97)

• If authors add metadata, surely search engines will be able to utilise this, so having external cataloguers isn't necessary.

• Metadata frequently doesn’t relate to content of site
• 34% use any meaningful metadata
• 0.3% use Dublin Core

• Therefore many search engines ignore – "search engines do not trust metadata. It's fine to talk about how nice it would be if all web pages were categorized, but the search engines know from experience that people will lie, mislead or do whatever they can to get on top". (Sullivan 1997)
• Ignored by Excite, Google, Lycos, Nlight
• Do not boost ranking on AltaVista, Excite, Google, Lycos, NLight

Known item

• E.g. Title, Author
• Likely that many "known item" searches can be handled by search engines, IF in "indexable" web, and search terms sufficiently unique.

• Challenge: Need to find terms that define our topic in such a way that search engine can present a managable list of sites closely matching our topic.

Uniform vocabulary access

• Can potentially map topic to search terms
• Provide structured access to names of people and organisations

• If we try to catalogue everything, we’ll run out of index terms; but cataloguers have created MARC records for ~40M print items (Weinberg 99), so cataloguing significant web sources should be achievable.
• Cataloguing "significant" sites may be a form of selection.

• If we catalogue items, may be easier to find/recover if they disappear or change location

• Bar-Ilan,J (1998): On the overlap, the precision and estimated recall of search engines. A case study of the query "ERDOS". Scientometrics 42(2), 207-228.
• Bates,MJ (1998): Indexing and access for digital libraries and the Internet: human, database, and domain factors. Journal of the American Society for Information Science 49(13), 1185-1205.
• Clarke,SJ; Willett,P (1997): Esimating the recall performance of Web search engines. Aslib Proceedings 49(7, July/August), 184-189.
• Hawking,D; Craswell,N; Thistlewaite,P; Harman,D (1999): Results and Challenges in Web Search Evaluation. In: WWW8 Proceedings. http://www8.org/w8-papers/2c-search-discover/results/results.html
• Lawrence,S; Giles,CL (1999): Accessibility of information on the Web. Nature 400(8 July), 107-109.
• Pollock,A; Hockley,A (1997): What's Wrong with Internet Searching. D-Lib Magazine (March)

(http://www.dlib.org/dlib/march97/bt/03 pollock.html)
• Poulter,A (1997): The design of World Wide Web search engines: a critical review. Program 31(2, April), 131-145.
• Schwartz,C (1998): Web Search Engines. Journal of the American Society for Information Science 49(11), 973-982.
• Weinberg,BH (1999): Improved Internet access: guidance from research on indexing and classification. Bulletin of the American Society for Information Science 25(2, December/January), 26-29. (http://www.asis.org/Bulletin/Jan-99/weinberg.html)