Keywords: information retrieval, project extranet, knowledge-based systems, frame-based knowledge representation, information overload, classification standards, information technology, construction project management, artificial intelligence.

The growth of the Internet and the development of its associated technologies spurred the creation of web-based project management systems for the AEC (Architecture / Engineering / Construction) industry. These systems, usually offered as an ASP service, are also known as project extranets or Internet-based project management systems. They typically store thousands of documents generated during the building design and construction processes. Most of project extranets have standard document retrieval engines, based on keyword search, which do not provide adequate precision and recall to user queries. Searching in these systems is like using a search engine on the web and the results are alike: tens to hundreds of documents are retrieved for any single search, leading to the information overload problem.

Enhancing an information retrieval system effectiveness implies improving its precision and recall parameters [1]. Precision (the proportion of relevant documents out of those returned) can be increased if the system is aware of the user query context, discarding false hits. On the other hand, improving recall (the proportion of returned documents out of the relevant ones) requires the system to detect non- explicit document relevancy. For both cases, domain knowledge is useful.

This paper discuss a new knowledge-based information retrieval system designed for project extranets that integrates building design and construction knowledge. The retrieving performance of this system is expected to be superior as the use of AEC knowledge improves both recall and precision.

System recall is improved through query-term expansion based on equivalence and hierarchical relations on the concepts in the knowledge base. Precision is improved by enhanced filtering.

Although an ontology would be the ideal mechanism for precise and complete representation of the building construction domain knowledge, its creation is, as for any other domain, a complex and laborious task. Instead, the approach taken in the system described here is to use expert knowledge, thesauri and classification systems information - like OCCS (OmniClassConstruction Classification System) - as well as standards like ISO 13567 (Organization and naming of layers for CAD) to create a custom knowledge base. Enterprise specific data like time schedule, personnel names and functions, special tools and materials, names of places, etc. are also input.

The knowledge representation scheme used is frame-based [2] and adopts a hierarchical structure. Each information item (thesaurus terms, classification items, etc.) is represented as a frame with a set of slots corresponding to its class. Associated to each frame slot are values or other frames. All search terms are weighted according to procedural rules. These weights are used to select terms for actual document base search and for ranking result listings presented to the user.

The proposed system analyses information from the user query in order to infer what is the required information. Then it uses domain knowledge to discover which kinds of construction documents are the best candidates to fulfil that need. These document types supply information for defining values of filtering parameters like file type, date and author. Finally, a ranking order for retrieved documents is computed for presentation to the user.

The user interface of this new information retrieval system is the same as the traditional keyword search in place today, as well as is the output format. Further implementation is needed to assess the complete system performance, but it is expected to greatly improve search mechanisms available today on commercial project extranets, enhancing both precision and recall of such systems and reducing information overload on users.

1

Baeza-Yates, R., Ribeiro Neto, B., "Modern Information Retrieval", Addison Wesley, Harlow, 1999.

2

Way, E.C., "Knowledge Representation and Metaphor", Kluver, Dordrecht, 1991.

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 £82 +P&P)