Topic Recommendation with Article Metadata using Collaborative and Content-based Filtering on Author Activeness Profiling

Topic selection is a critical step in creating scientific articles. A topic is an important basis for authors to convey their ideas in written publications. Moreover, it enables the construction of a comprehensive author profile based on their historical topic preferences. Both factors influence how an author's expertise is perceived. This research aims to build a topic selection recommendation system by introducing Author Activeness Profile (AAP), which maps relations between authors and research topics, with Collaborative Filtering (CF) and Content-Based (CB) filtering approaches. CF and CB are combined with AAP due to their advantageous characteristic of being able to study the patterns of authors and articles. In CF, authors will be grouped with other authors who have similarities with each other. Topic recommendations will be fetched from authors with the closest similarity. In CB, recommendations are generated by assessing the similarity between an author's preferred topic and other possible topics. Topics with the closest similarities will be returned as recommendation results. Evaluation is conducted by assessing the alignment between system-generated topic recommendations and the author's actual preferences, calculated with f1-score. The evaluation found that the CF method with K-NN achieved the highest f1-score of 83.44%, while the CB method with Jaccard and Cosine similarity achieved the highest f1-score of 62.9%. It concludes that the CF method with K-NN provides the best recommendations for authors' topic selection recommendation systems.