Users must be cautioned that THE TOOLS ARE STILL PRONE TO ERROR due largely to ambiguities in natural-language

processing and high computational costs.

 

subject and the number of all results found by the same subject or by an expert. Domain experts provided answers for judging the subjects’ browse-task performance. After using a system, a subject filled out a questionnaire with comments and satisfaction ratings on a seven-point Likert scale.

In the CBizPort experiment with 30 Chinese sub-
jects from mainland China, Hong Kong, and Taiwan
and with three Chinese business academics and prac-
titioners serving as experts, the results show that the
effectiveness of existing Chinese search engines can be
improved significantly by adding
CBizPort. However, no significant ^Portal
difference was found in the effec- ^Research
questions

tiveness of the summarizer and categorizer in the Chinese portal and in their user-satisfaction rat- ^Languages ings. Despite this, 11 subjects commented that the summarizer and categorizer facilitated their understanding and searching. ^Regions These results indicate that CBizPort could augment these search engines in searching and browsing Chinese business information but also that the summarizer and cat- _Result egorizer need further improve- ^highlights ment in precision and browse support.

In the SBizPort experiment with 19 Spanish subjects from six countries—Colombia, Mexico, Panama, Peru, Puerto Rico, and the U.S.—and a veteran Spanish business consultant serving as the expert, the SOM visualizer achieved significantly better browse effectiveness than BIWE, the benchmark search engine, showing that the tool helps alleviate information overload and supports browsing effectively. The use of a domain-specific collection achieved higher mean accuracy and search efficiency than not using it, though the differences were not significant. Subjects rated SBizPort significantly better

than BIWE, citing the precision and relevance of returned results. These results further indicate that the information-visualization tool can be an alternative to presenting search results as text in a list.

In the AMedPort experiment with 11 Arab subjects from five countries—Iraq, Jordan, Lebanon, Mauritania, and Morocco—and an Arab microbiology researcher serving as the expert, the AMedPort achieved a significantly higher mean accuracy, efficiency, and satisfaction rating than (and comparable browse effectiveness to) Ayna, the benchmark search

 

CBizPort SBizPort AMedPort

• How does the framework support Web searching in a multilingual world?

• How can Web searching and browsing be made more effective by using the portals developed by the framework?

• What are the lessons learned and implications for non-English Web searching?

Traditional Chinese is used in Hong Spanish is used in many Arabic is used in more than Kong and Taiwan, while simplified regions and countries in 22 countries in the Middle Chinese is used in mainland China. North and South Americas East and North Africa. It is They have different encoding and Europe; Catalan (another written from right to left. formats. Word segmentation is version of Spanish) is widely Web content in Arabic is a problem. used as well. generally not rich.

Mainland China, Hong Kong, Mexico, Honduras, Costa Lebanon, Saudi Arabia, Tai wan Rica, Panama, Colombia, Bahrain, Canada, Tunisia,

Venezuela, Ecuador, Peru, Kuwait, Egypt, Switzerland, Chile, Argentina, Spain, United Arab Emirates, Bolivia, Paraguay, Uruguay, Russia, U.K., U.S. Guatemala, Nicaragua, U.S.

BIWE.com Ayna.com

Benchmarks Sina.com.cn, Openfind. com.t w, HK.yahoo.com

CBizPort’s meta-searching, summarization, and categorization are helpful for searching and browsing Chinese Web pages in the three regions. CBizPort needs improvement on search precision and information quality.

SBizPort achieved a better browse performance and satisfaction rating than the benchmark but needs to be improved on its domain-specific collection.

AMedPort achieved better search accuracy than comparable browse performance in Ayna. AMedPort obtains better user ratings in user satisfaction.

Table 2. Experimental studies on developing

Web search portals in various languages.

engine. Nine subjects said AMedPort was useful and provided more topics and information than the benchmark. The portal provided high-quality information from many sources but needs improvement in both its summarizer and categorizer.

There was a probability of 0.05 (or less) that the confirmed results were actually not statistically valid in the experiments in which the best search engines in the respective languages were used as benchmarks.

 

LESSONS LEARNED