結合社會性標籤及文獻內容於個人化學術文章推薦

Combining Social Tagging and Reference Content for Personalized Academic Document Recommendation

胡雅涵(Ya-Han Hu) ； 李彥賢(Yen-Hsien Lee) ； 林正賢(Jheng-Hsien Lin)

《資訊管理學報》 22卷2期 (2015/04) Pp. 171-197

摘要

學術文章推薦是近年來熱門的研究議題，過往針對學術文章推薦研究上，普遍利用學術文章內的屬性資料，如：標題、摘要、關鍵字、作者名稱以及參考文獻標題等進行推薦。然而除了上述的「內部資訊」外，學術文章中亦包含其他與該研究相關的「外部資訊」，像是參考文獻摘要及具相同社會性標籤（Social tagging）之文件等。藉由分析外部資訊，應能有助於取得與原始學術文章相關之其他研究主題的關鍵字詞，進而推薦更符合學術研究需求的文章。本研究同時考量使用者喜好文章之內、外部資訊，包括標題、摘要、關鍵字、參考文獻標題、參考文獻文章內容、社會性標籤、以及具相同社會性標籤文章內容，並以內容導向式推薦方法為基礎來建構推薦系統。此外由於不同文章屬性應具有不同程度的重要性，本研究進一步運用層級分析法（Analytic Hierarchy Process）制定出各個文章屬性之權重值，用以對文章之各個屬性相似度進行加權運算，並產生最終之推薦清單。本研究最後以實驗方式進行推薦效能評估，並以不同屬性組合之推薦方法做為評估比較基準。實驗結果顯示，在本研究採用之成對比較法（Pair Match）及命中率（Hit Rate）兩個評量指標下，本研究提出的推薦方法相較於傳統僅考慮內部資訊之推薦方法，能有較高之推薦命中率，且在文章推薦排序上能有更顯著地改善，亦即將使用者喜好程度較高之文章給予優先的推薦順位，說明本研究之學術文章推薦方法具有較佳的推薦效果。

關鍵字

推薦系統 ； 內容導向式推薦方法 ； 學術文章推薦 ； 社會性標籤 ； 文字探勘

並列摘要

Purpose-This study aims at developing a novel academic article recommender system. The abstract of the articles that share similar social tags with and that are referred by the preference articles of the targeted user are usedto improve the effectiveness of the recommender system. Design/methodology/approach-This study adopts the content-based method to determine the similarity between two academic articles and make recommendations. Seven attributes relevant to academic article, including title, abstract, keyword, reference, reference article, social tag, and article with similar social tag, are used to extend the original preference document vector. The analytic hierarchy process method is applied to determine the weights among the seven attributes by their degree of importance. Aweb-based recommender system was developed for the evaluation purpose. We invited over 90 subjects and adopted Pair Match and Hit Rate as performance criteria in the evaluation experiment. Findings-Theevaluation results show our recommendation method outperforms the five benchmarks. External information such as articles with similar social tags and reference articles used in the system can contribute to the better ranking list of recommending articles, in terms of the Pair Match and the Hit Rate. Research limitations/implications-This study only considers a limited set of academic articles as the investigated corpus. Future research shall expand the array of articles and considerco-authorship and co-citation relationships as important features. Practical implications-Though the rise of digital libraries makes the acquisition of academic resources easier, a sharp increase of academic articles leadsthe search for relevant articles ineffective. The proposed method can facilitate researchers in searching and retrieving relevant academic articles based on their individual preference profile. Originality/value-This paper is the first that investigates the influence of both internal and external information onmaking recommendation of academic article. It advances literature in determining valuable article features for optimizing the performance of recommender system.

並列關鍵字

recommender systems ； content-based recommendation ； academicarticle recommendation ； social tagging ； text mining

參考文獻

Ahlgren, P.,Colliander, C.(2009).Document document similarity approaches and science mapping: Experimental comparison of five approaches.Journal of Informetrics.3(1),49-63.

Basilico, J.,Hofmann, T.(2004).Unifying collaborative and content-based filtering.Proceedings of the 21st International Conference on Machine Learning.(Proceedings of the 21st International Conference on Machine Learning).:

Cao, Y.,Li, Y.(2007).An intelligent fuzzy-based recommendation system for consumer electronic products.Expert Systems with Applications.33(1),230-240.

Chen, Y.L.,Wei, J.J.,Wu, S.Y.,Hu, Y.H.(2006).A similarity-based method for retrieving documents from the SCI/SSCI database.Journal of Information Science.32(5),449-464.

Chi, E.H.,Mytkowicz, T.(2007).Understanding navigability of social tagging systems.Proceedings of the SIGCHI Conference on Human Factors in Computing Systems.(Proceedings of the SIGCHI Conference on Human Factors in Computing Systems).: