Keywords: keyword pattern, procurement, information search, machine learning, e-commerce, artificial intelligence.

As more and more procurement websites become available on the Internet, obtaining information from websites has become an essential part of a contractor's procurement undertaking. Several e-markets, specifically for construction, have also been established, including bLiquid.com [1] and ProcureZone [2]. However, most websites provide only two primary ways of searching for information, namely by index-menu or by keyword. The keyword search method is probably most often preferred by users not familiar with the content or indexing scheme of a website. Nevertheless, it may still be necessary for most buyers to input keywords using a trial-and-error procedure, in order to narrow down the search to find the desired information.

Finding information related to construction procurement is a more complex process than for consumer goods or general-purpose documents, which was the primary focus of the aforementioned research works. The higher complexity and scale of construction projects, emerging construction materials and technologies, and changes in building codes and regulations make the preparation of accurate tendering specifications a challenge for contractors. For example, different specifications are necessary for different types of construction materials: to order steel, accepted industry standard specifications may be required (e.g. the American Society for Testing and Materials (ASTM)), for the grade, diameter, strength and weight of the steel; to order ceiling boards for a building project, it may be necessary to specify length, width, thickness, material type, fireproof certification, texture and colour. A pre-determined or fixed search guide, therefore, seems impractical.

Our research recognized that professional procurement experience helped users to more effectively carry out website information searches, by using fewer keywords. We planned to capture such experience in order to guide inexperienced users in their search. The proposed learning framework required experienced users to annotate the relationships among the series of keywords entered during their web search, and then corrected the user's keyword input or prompted subsequent keyword candidates in order to help inexperienced users to reduce the number of keywords required during their search.

The research goal was to improve search effectiveness by guiding the user's search using three approaches; namely correction, specification and extension. The correction guide corrects misspelled or misused keywords. The specification guide constrains the search space by adding more "AND" words to a keyword phrase or by replacing the keyword with a more specific term. The extension guide extends the search space by adding more "OR" keywords, by suggesting keywords the user may need for subsequent searches. Based on these three approaches, this research applied the following guides: correction; specification-by-equivalence; specification-by-detail; extension-by-time; extension-by-location; extension-by-team; and extension-by-component.

For example, the E-location guide suggests keywords for procured items whose construction normally occurs adjacent to the item to which the input keyword relates. For example, the guide may suggest "reverse circular concrete pile" for "diaphragm wall" because the construction activities requiring these items often occur at adjacent locations. The E-team guide suggests keywords for procured labor or other resources relating to the type of labour the entered keyword represents. For example, the guide may suggest "high-rise welding labour" for "steel erection licensed labour" because the two types of labour are often utilized simultaneously, as a team, in building projects. The E-component guide suggests keywords for procured items that are normally embedded within the item to which the entered keyword relates. For example, the guide may suggest "#4 deformed rebar" for "type C reinforced concrete pipe" because the former item is often embedded in the latter.

The paper will describe how users are classified for learning credibility, and the learning framework for recording expert users' search patterns. The learning framework is the foundation for suggesting prospective keywords when a user inputs a keyword. It calculates the likelihood of each prospective keyword becoming the next keyword the user needs based on the frequency of occurrence, type of user and date for inputting the keyword, and the user's desired search guide direction.

Twelve professionals, using 14 procurement packages, with 64 items in total, evaluated the proposed framework. It will be demonstrated that the proposed learning keyword guide facilitated a dynamic, customized menu and indexing system, and reduced the number of keywords required for the professionals to find the information they required.

1

bLiquid.com, http://bliquid.com(last updated, 2004) (last visited, June 21, 2004).

2

ProcureZone, http://www.procurezone.com (last updated, 2002) (last visited, June 21, 2004).

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £135 +P&P)