Providing Answers to Questions from Automatically Collected Web Pages for Intelligent Decision Making in the Construction Sector

Abstract

The construction sector is notorious for the dichotomy between (a) its intensive use of information in its decision-making processes and (b) its limited access to, and insufficient use of, the pertinent information that is potentially available. In the context of the potential availability of valid information on the Web, we have developed a question-and-answer system, which enables construction practitioners to seek for information by posing questions in English or French, instead of entering a list of relevant keywords. Based on a modular systems approach using natural language, relevant answers to questions are selected and presented in a convivial way, thus improving and speeding up the classical Web search procedure. Our system consists of two main components: An intelligent robot that traverses a Web space and decides whether a page is construction oriented or not, and a question-answer (Q-A) component (comprising three modules) that uses a domain-specific thesaurus to extract meaningful parts of a question and to detect, process, and present paragraph passages extracted from relevant Web pages. The robot is trained on positive and unlabeled examples using a machine learning approach, while the Q-A component uses natural language processing techniques. In our experiments, we show that our automatically collected database consists of approximately 16% of noise, while the performance of the Q-A component is 65.45% in mean reciprocal rank.