Enhanced context-based document relevance assessment and ranking for improved information retrieval to support environmental decision making. Issue 4 (October 2016)

• Record Type: Journal Article
• Authors: Lv, Xuan; El-Gohary, Nora M.
• Is Part Of: Advanced engineering informatics. Volume 30:Issue 4(2016:Oct.)
• Page Start: 737
• Page End: 750
• Publication Date: 2016-10
• Subjects: Information retrieval -- Context-based relevance assessment -- Context-enhanced document ranking -- Vector space model -- Statistical language model -- Project environmental review; Computer-aided engineering -- Periodicals; Engineering -- Data processing -- Periodicals; 620.00285
• DOI: 10.1016/j.aei.2016.08.004
• Languages: English
• ISSNs: 1474-0346

Abstract

Abstract: There is a need for enhanced context-based document relevance assessment and ranking to facilitate the retrieval of more relevant information for supporting environmental decision making. This paper proposes a new context-based relevance assessment method, which allows for enhanced context representation and context-based document relevance recognition through: (1) a context-aware and deep semantic concept indexing approach, and (2) a deep and semantically-sensitive relevance estimation approach. The proposed relevance assessment method was integrated into two widely-used document ranking models [vector space model (VSM) and statistical language model (SLM)], resulting in two improved ranking methods: (1) a context-enhanced VSM-based method, and (2) a context-enhanced SLM-based method. The two context-enhanced document ranking methods were evaluated in retrieving webpages that are relevant to transportation project environmental review. The two context-enhanced methods were compared with each other and with their provenance methods (i.e., original VSM and SLM) in terms of mean precision (MP) and mean average precision (MAP). The context-enhanced VSM-based method outperformed the context-enhanced SLM-based method on every metric. It achieved 48% MAP, 79% MP at the top 10 retrieved documents, and over 65% MP at the top 50 retrieved documents, on the testing data. It also showed significant improvement over the state-of-the-art keyword-based VSM method.