Here, BIrefers to the product of acquisition, interpretation, collation, assessment, and exploitation of information in the business domain [ 4]. CBizPort includes two versions of its user interface—one for simplified Chinese, one for traditional Chinese—each with the same look and feel. Relying on a ^{Virtual Arabic} _keyboard conversion dictionary with 6,737 Chinese characters in each of the two encodings (Big5 and GB2312), the encoding converter converted all Chinese characters into the encoding of the interface version. The eight information sources used in the portal’s meta-searching are major Chinese search engines and Click to summarize in business-related portals from the ^{3 or 5 sentences.} three regions. Relying on two Chi-nese-phrase lexicons to extract phrases, the portal’s categorizer organizes retrieved Web pages into various folders labeled with the key phrases in the page summaries and titles.

The Spanish Web portal ( SBizPort) supports searching and browsing of business information from 22 Spanish-speaking regions. In addition to keyword searching, summarization, and categorization, like those in CBizPort, SBizPort provides a comprehensive collection of business Web pages for searching and supports visualization of retrieved pages (see Figure 2). Users visualize Web pages by clicking on a region to see a list of pages on the right and open pages by clicking the link-embedded titles.

The Arabic Web portal, AMedPort (see Figure 3) focuses on the medical domain in some 22 Arab regions and supports all search and browse functions available in SBizPort. AMedPort includes a customized user interface with a right-to-left text display and a virtual keyboard to facilitate Arabic input.

Retrieved pages titles and abstracts are listed.

Figure 3. Screenshots from

AMedPort.

 

EXPERIMENTAL FINDINGS Sixty native speakers participated in the experiments (detailed in [ 3, 6]) of the three Web portals to evaluate the framework’s usability in supporting Web searching in a multilingual world (see Table 2). In each experiment (about an hour), a Web portal was

The user types “excretion” to search for medical information about the topic in Arab regions.

Retrieved pages are categorized into folders labeled with key phrases.

The summary is listed on the left, while he original page is on the right.

The SOM visualizer categories 30 Web pages onto six regions and displays hyperlinks on the right.

 

compared with a benchmark search engine in each of the three languages. Each subject was introduced to the portal and the benchmark search engine and was randomly assigned different task scenarios (one scenario per system). Each scenario contained three or four search and browse tasks based on standards developed through the National Institute of Standards and Technology Text Retrieval Conference ( trec.nist.gov). All questionnaires used in the experiment were administered in the subjects’ native language. Subjects spent an average of three minutes to finish a search task and eight minutes to finish a browse task. The order in which the systems were used was randomly assigned to avoid bias due to the sequence of use.

Several information-retrieval measures—precision, recall, and F value—revealed the search and browse effectiveness of the system by computing the ratios between the number of relevant results found by a