The Internet Encyclopedia (Volume 3)

P1: IML

Wisman WL040/Bidgoli-Vol III-Ch-59 August 14, 2003 18:3 Char Count= 0

HOWTOSEARCH—THESEARCHER’SVIEW 727

Figure 2: Retrieved versus relevant references when

searching the Web. The intersection of retrieved and rel-

evant pages will likely be relatively small in part because

only a fraction of the Web is indexed.

Several basic definitions common in the search liter-

ature, “relevancy,” “recall,” and “precision,” are also of

value to searchers. Relevancy measures the usefulness of

the references retrieved and is highly subjective. Figure 2

illustrates three useful points: (1) that relevant references

are generally a subset of all the references available;

(2) that retrieved information generally includes some

irrelevant references; and (3) that the intersection of re-

trieved and relevant references will often exclude ref-

erences that are relevant (Belew, 2000). The following

definitions ofrecallandprecisionnicely complement the

intuition illustrated in Figure 2:

Recal l=

|Retrieved∩Relevant|

|Relevant|

Precision=

|Retrieved∩Relevant|

|Retrieved|

.

Recall is the percentage of relevant references retrieved. A

recall of 50% means that only one-half of the relevant ref-

erences were retrieved; retrieving every possible reference

yields 100% recall but often includes many references that

are irrelevant. Precision is the percentage of retrieved ref-

erences that are relevant. Of the references retrieved, a

precision of 50% means that only one half of the refer-

ences were relevant; returning only one relevant refer-

ence yields 100% precision but possibly very poor recall.

A “weight loss” query can yield perfect recall by retrieving

every Web page in existence but will likely produce many

that are not relevant, so that precision is poor. Simulta-

neously achieving perfect recall and precision is nearly

impossible even when carefully designed and constructed

sets of known information are searched. Given the scope,

size, disorganization, and diversity of information sources

available on the Web, merely retrieving relevant informa-

tion is daunting.

Search engines naturally exhibit great recall when find-

ing millions of pages for a broad query, but narrowing the

query can greatly improve precision by eliminating many

pages of questionable relevancy. Fine control of a search

can yield higher quality references, is relatively easy to

use, and is common to most search engines. Table 1 lists

common search controls for popular engines with details

and examples of search refinement provided in the follow-

ing discussion (Sullivan, 2001).

Adding Search Terms:The simplest and often most ef-

fective method of improving the precision of search re-

sults is adding search terms with a more precise mean-

ing. The query “weight loss” produces millions of pages

using the Teoma search engine, each page containing

the word “weight,” “loss,” or both. Adding terms for

“vegetarian weight loss diet” produces merely 10,000

but more focused references. Each page found still

contains at least one or more of the query words. Be-

cause search engines generally rank pages higher that

match more query terms, pages with fewer discrimi-

nating terms are effectively ignored.

Searching for Phrases:Phrases of specific word group-

ings can yield more precise searches than the same

words when matched independently. Searching on the

Table 1Common Search Controls for Popular Engines

Query Match How Example Popular Search Engine Support

Any Automatic vegetarian diet Teoma, Yahoo, Inktomi

OR vegetarian OR diet AltaVista, Yahoo, MSN, Google

All Automatic vegetarian diet AltaVista, Google

AND vegetarian AND diet AltaVista, Yahoo, MSN

Include ++vegetarian diet AltaVista, Yahoo, MSN, Google, Inktomi

Exclude − vegetarian–diet AltaVista, Yahoo, MSN, Google, Inktomi

AND NOT vegetarian AND NOT diet AltaVista, MSN, Inktomi

Phrase “ “ “vegetarian diet” Google, Yahoo, Inktomi, Teoma

Wildcard ∗ veget∗diet Altavista, Yahoo, MSN

Proximity NEAR vegetable NEAR diet AltaVista

Title title: title: ”vegetarian diet” Altavista, Yahoo, Inktomi

intitle: intitle: “vegetarian diet” Google

Site Only site: site:food.com vegetarian Google, Inktomi

host: host:food.com vegetarian AltaVista

URL Only url: url:food.com AltaVista, Yahoo, Inktomi

allinurl: allinurl:food.com/diet/ Google

All References link: link:www.food.com AltaVista