EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

745. A DECISION SUPPORT FRAMEWORK FOR AUTOMATED REVIEWER ASSIGNMENT USING NLP AND OPTIMIZATION

Invited abstract in session TB-3: Machine Learning in Applied Optimization, stream Data Science Meets Optimization.

Tuesday, 10:30-12:00
Room: 1005 (building: 202)

Authors (first author is the speaker)

1. Meltem Aksoy
Research Center Trustworthy Data Science and Security
2. Seda Yanık
Industrial Engineering, Istanbul Technical University
3. Mehmet Fatih Amasyali
Computer Engineering, Yildiz Technical University

Abstract

Addressing the inefficiency of traditional manual reviewer assignments in the peer-review process, this study introduces a decision support system that automates the assignment task. Utilizing information retrieval, natural language processing (NLP) and optimization techniques, it tackles the reviewer assignment problem through a structured, three-stage approach. Initially, it gathers diverse information from various sources to build a comprehensive database of proposals and reviewers. Then, it applies word embedding techniques to convert multilingual proposal and reviewer texts into vector representations, and it uses the cosine similarity metric to determine content similarity between each proposal-reviewer pair. Concurrently, it assesses reviewer competency by analyzing their past evaluation performance and areas of expertise through predefined knowledge rules. In the final stage, a multi-objective integer linear programming model assigns reviewers to proposals, optimizing proposal-reviewer similarity and reviewer competency while avoiding conflicts of interest. Furthermore, it explores a max-min approach to improve outcomes for the least-advantaged proposals. This model is enhanced by two additional models to ensure balanced reviewer workloads. The system's efficiency is tested with a real-world dataset from the project proposal evaluation process of a regional development agency. The results show that the proposed system significantly outperforms traditional methods.

Keywords

Status: accepted

Back to the list of papers