TY - GEN
T1 - Similarity-based dependency parsing for extracting dependency relations from bridge inspection reports
AU - Liu, Kaijian
AU - El-Gohary, Nora
N1 - Publisher Copyright:
© 2017 American Society of Civil Engineers.
PY - 2017
Y1 - 2017
N2 - Bridge inspection reports provide a large amount of technically-detailed bridge condition and maintenance history data/information that could be utilized for advancing the understanding of bridge deterioration for supporting improved maintenance of bridges. However, the valuable information buried in these inspection reports remains unexploited. There is, thus, a need for information extraction (IE) methods to automatically extract information about existing bridge deficiencies and performed maintenance actions from these reports for further analysis. However, existing IE methods are limited in their ability to extract information from such highly-technical text, with varying levels of technical detail, text patterns, and quality. To address this gap, this paper proposes an ontology-based IE framework that extracts entities and relations (i.e., dependency relations) about bridge deficiencies and maintenance actions, and represents the extracted information in a structured way. The proposed IE framework is composed of two primary components: (1) a name entity recognizer for term identification, and (2) a relation extractor for relationship association. This paper focuses on presenting the proposed similarity-based dependency parsing (DP) methodology for automated relation extraction. The proposed DP methodology utilizes a transition-based DP model to capture sentence-level dependency configurations, and represents each configuration with a distributed representation. Each unlabeled configuration is compared to labeled configurations in the distributed representation space by a cosine-similarity measure, and is then labeled with its most similar labeled configuration's transition. The proposed DP methodology achieved a configuration-based accuracy of 78.0%.
AB - Bridge inspection reports provide a large amount of technically-detailed bridge condition and maintenance history data/information that could be utilized for advancing the understanding of bridge deterioration for supporting improved maintenance of bridges. However, the valuable information buried in these inspection reports remains unexploited. There is, thus, a need for information extraction (IE) methods to automatically extract information about existing bridge deficiencies and performed maintenance actions from these reports for further analysis. However, existing IE methods are limited in their ability to extract information from such highly-technical text, with varying levels of technical detail, text patterns, and quality. To address this gap, this paper proposes an ontology-based IE framework that extracts entities and relations (i.e., dependency relations) about bridge deficiencies and maintenance actions, and represents the extracted information in a structured way. The proposed IE framework is composed of two primary components: (1) a name entity recognizer for term identification, and (2) a relation extractor for relationship association. This paper focuses on presenting the proposed similarity-based dependency parsing (DP) methodology for automated relation extraction. The proposed DP methodology utilizes a transition-based DP model to capture sentence-level dependency configurations, and represents each configuration with a distributed representation. Each unlabeled configuration is compared to labeled configurations in the distributed representation space by a cosine-similarity measure, and is then labeled with its most similar labeled configuration's transition. The proposed DP methodology achieved a configuration-based accuracy of 78.0%.
UR - http://www.scopus.com/inward/record.url?scp=85021753829&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85021753829&partnerID=8YFLogxK
U2 - 10.1061/9780784480823.038
DO - 10.1061/9780784480823.038
M3 - Conference contribution
AN - SCOPUS:85021753829
SN - 9780784480823
T3 - Congress on Computing in Civil Engineering, Proceedings
SP - 316
EP - 323
BT - Computing in Civil Engineering 2017
A2 - Lin, Ken-Yu
A2 - Lin, Ken-Yu
A2 - El-Gohary, Nora
A2 - El-Gohary, Nora
A2 - Tang, Pingbo
A2 - Tang, Pingbo
T2 - 2017 ASCE International Workshop on Computing in Civil Engineering, IWCCE 2017
Y2 - 25 June 2017 through 27 June 2017
ER -