應用資訊檢索技術於論文評閱者推薦模式之評估 Evaluation of Information Retrieval Based Models for Recommendation of Paper Reviewers

Abstract

隨著傳統期刊逐漸採用電子形式出刊，也帶動投稿及評閱過程愈來愈多採用電子自動化之潮流。目前一般的線上投稿暨評閱系統雖然功能逐漸齊備，但仍少有推薦評閱者功能。為評估現有可推薦評閱者技術之表現，本文分別用標題、關鍵詞、摘要、全文4種不同長度的論文表示方式，搭配7種評閱者匹配法，其中包括向量空間模式下的4種相似度匹配法，及應用於OpenConf線上投稿系統中的3種主題式匹配法，交叉組合出4×7＝28種推薦模式。測試結果顯示，向量空間模式匹配法優於主題式匹配法。又所有推薦模式中，以摘要為論文表示方式，搭配向量空間模式的餘弦相似度匹配法，其推薦效果最好。As more e-journals appear and the e-review process becomes more popular, the demand for automatic recommendation of a good peer reviewer has been ever increasing. To automate the process of paper reviewer recommendation, this work evaluates four kinds of paper representations, which include full text, abstract, title, and author defined keywords. To match reviewers with papers, this work evaluates seven scoring methods including three topic-based methods from OpenConf, a popular online submission system with source, and four similarity-based methods from the vector space model of traditional information retrieval. The results of the 28 experiments show that recommendation methods based on the vector space model are better than the three topic-based methods of OpenConf in most document representations. Among them, the abstract paper representation combined with cosine similarity matching measure has the highest average precision.

Keywords

• Vector space model
• Paper representation
• Reviewer recommendation