TL;DR
This paper introduces an ontology-driven semantic retrieval system to improve how construction projects learn from past experiences, specifically for risk management. By building a structured knowledge base (ontology) of construction risks and using it to enhance search queries, the system aims to make retrieving relevant information from project documents more efficient and accurate.
What Problem Does This Paper Address?
Construction projects are inherently complex and prone to risks. Learning from the experiences of previous projects is crucial for effective risk management, but this information is often buried within vast amounts of unstructured project documents. Existing methods struggle to efficiently extract and utilize this knowledge due to a lack of a common vocabulary and the difficulty of semantic understanding. This paper tackles the challenge of bridging the gap between unstructured project data and the need for intelligent, context-aware risk management by proposing an ontology-based semantic retrieval approach.
Methods
The core of this work is the development and application of a risk domain ontology for construction projects. The ontology construction follows a five-step process: standardization, conceptualization, formalization, implementation, and evaluation. The paper focuses on the risk of "cost overrun" for its ontology.
For the retrieval system, the authors employ several key techniques:
- Document Processing: Construction project documents are processed using text segmentation, with a custom vocabulary incorporating the risk domain ontology to improve accuracy.
- Eigenvalue Extraction and Weight Calculation: The TF-IDF algorithm is used to calculate the importance (weight) of terms (verbs and nouns) within documents.
- Query Expansion: This is a crucial novel aspect. When a user's query is too short or ambiguous, the system leverages the risk domain ontology to expand the query. This involves using synonym expansion and generalization extension based on the ontology's hierarchical structure. If a query term isn't directly in the ontology, context analysis is used to find related concepts.
- Similarity Calculation: Cosine similarity is used to measure the relevance between the expanded query vector and document vectors.
Datasets
The paper does not explicitly mention specific datasets used for training or evaluation. Instead, it refers to "previous projects" and "numerous project documents" as the source of information. The experimental results are based on a set of seven sample documents (D1-D7) and two sample queries (Q1-Q2) provided within the paper for demonstration purposes.
Key Insights and Ideas
- Ontology as a Knowledge Backbone: The paper highlights the power of ontologies in providing a structured, shared vocabulary for a complex domain like construction project risk management. This structured knowledge is essential for moving beyond simple keyword matching.
- Intelligent Query Expansion is Key: The most significant insight is the effectiveness of ontology-driven query expansion. By intelligently broadening queries based on conceptual relationships within the ontology, the system can significantly improve retrieval accuracy, especially for short or ambiguous user inputs.
- Bridging Unstructured and Structured Data: The research demonstrates a practical method for correlating the conceptual knowledge encoded in an ontology with the content of unstructured documents, enabling more meaningful information retrieval.
- Focus on "Cost Overrun" as a Starting Point: While the ontology is currently focused on "cost overrun," the methodology is presented as a foundation that can be extended to other risk categories, suggesting a scalable approach.
Key Results
The experimental results, though based on a limited set of sample documents and queries, demonstrate the efficacy of the proposed method.
- Improved Retrieval Accuracy: When the query "unexpected events" was expanded using the ontology (Q2), the system was able to filter out less relevant documents and identify document D1 as the most relevant, achieving a higher similarity score (0.829) compared to the unexpanded query (Q1, max similarity 0.042). Similarly, for Q2, document D7 also showed a significant improvement in similarity (0.509) compared to Q1.
- Enhanced Efficiency: The ability to filter out less relevant documents through similarity calculation and query expansion leads to improved retrieval efficiency.
Limitations and Future Work
- Limited Ontology Scope: The current risk domain ontology is focused solely on "cost overrun." The authors acknowledge that other critical project objectives like quality, duration, and customer satisfaction are not covered.
- Future Work: The authors intend to expand the ontology to include other risk categories to provide comprehensive support for construction project risk management. They also aim to further study domain ontologies related to these other risks.
Summary updated