Files in this item

FilesDescriptionFormat

application/pdf

application/pdfAbdelmoneim_Dareen.pdf (6MB)
(no description provided)PDF

Description

Title:Semantic deontic modeling and text classification for supporting automated environmental compliance checking in construction
Author(s):Abdelmoneim, Dareen
Advisor(s):El-Gohary, Nora
Department / Program:Civil & Environmental Eng
Discipline:Civil Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:M.S.
Genre:Thesis
Subject(s):Deontology
Deontic Logic
Text Classification
Semantic
Information Retrieval
Compliance Checking
Automated Compliance Checking
Regulatory Compliance
Construction Compliance Checking
Abstract:Compliance checking in the construction industry refers to checking the conformance of a process, plan, document, design, or action to applicable norms (regulatory norms, contractual norms, and advisory practices). Manual compliance checking has been time-intensive, resource-consuming, and error-prone. Automated compliance checking (ACC) is thus a more efficient approach to compliance assessment. However, automated compliance checking (ACC) in the construction domain continues to be a challenge. Current ACC systems do not provide the level of knowledge representation and reasoning that is needed to efficiently interpret applicable norms (laws, regulations, contractual requirements, advisory practices, etc.) and check conformance of designs and operations to those interpretations. As such, this thesis explores a new approach to automated regulatory and contractual compliance checking – applying theoretical and computational developments in the fields of deontology, deontic logic, and natural language processing (NLP) to the problem of compliance checking in construction. Deontology is a theory of rights and obligations; and deontic logic is a branch of modal logic that deals with obligations, permissions, etc. A deontology for ACC would serve as a normative model for ACC knowledge representation and reasoning. NLP is a theoretically-based computerized approach to analyzing, representing, and manipulating natural language text for the purpose of achieving human-like language processing for a range of tasks or applications. NLP is the process by which humans and computers interact using natural human language (e.g. English). It is particularly important in ACC, as all norms are documented in natural language text. As such NLP is needed to: 1) classify and retrieve applicable norms/information from large amounts of textual documents using text classification algorithms, and 2) extract and formalize natural language rules or project information expressed in textual documents using information extraction techniques. The first thesis objective is to develop an upper-level domain deontology for ACC in construction. The purpose of the deontology is to represent the laws and regulations and reason about compliance of construction operations to those laws and regulations. A deontic model deals with assessing whether a specific action or state is right or wrong, permitted or forbidden. It uses deontic logic for normative reasoning about ideal versus actual behavior or state of systems; such as formal contract representation, automated contractual analysis, violation assessment systems, etc. This deontology represents the first deontic modeling initiative in the construction domain. The deontology is composed of: 1) concepts of ACC in the construction domain, such as ‘compliance assessor’, ‘compliance agent’, ‘subject’, ‘authority’ ‘compliance checking result’, etc.; 2) inter-concept relationships, such as ‘compliance assessor assesses compliance agent’; 3) axioms, which specify the definitions of the concepts and relations in the deontology, and specify constraints on their interpretation. Axioms also represent the constraints of the ACC domain. The second thesis objective is to evaluate the deontology and demonstrate its application using real project case studies. The initial evaluation of the deontic model showed its potential in successfully addressing the needs of ACC in construction. The model was initially evaluated through: 1) answering formal competency questions that evaluate the ability of the deontology to fulfill its requirements – as set by the developer, 2) automated consistency checking that evaluates the consistency of the components of the model (concepts, relations, and axioms), 3) case studies that evaluate the applicability of the deontology to solve real project compliance checking problems (the case studies focused on environmental compliance checking, and specifically on checking storm-water pollution prevention plans with applicable norms), and 4) domain expert interviews that evaluate the deontic model from a user’s perspective. The third research objective is to develop a semantic (deontic-based) TC algorithm to classify the clauses/sub-clauses of contract general conditions as environmental and non-environmental (since the second objective focuses on environmental compliance checking). Text classification is the process of identifying the group to which a piece of text belongs. Different text classification methods such as naïve Bayes classifier (NB), support vector machines (SVM), and maximum entropy (ME), were studied and empirically evaluated in the context of construction contract text classification. Different preprocessing and feature selection methods were implemented and evaluated (in terms of recall and precision). The final classifier model implements the ‘bag of words’ feature model, stop-word removal using a standard English stop-word list, stemming, odds ratio scoring function, best 20 features, feature weighting using term frequency, and SVM algorithm for machine learning and classification. The performance of the model achieves a 100% recall and 96% precision, at 26% threshold.
Issue Date:2012-02-06
Genre:thesis
URI:http://hdl.handle.net/2142/29609
Rights Information:Copyright 2011 Dareen Abdelmoneim
Date Available in IDEALS:2012-02-06
Date Deposited:2011-12


This item appears in the following Collection(s)

Item Statistics