數據科學與工程研究所「數據科學系列演講」
題 目： Music Recommendation Based on Multiple Contextual Similarity Information
主講人： Prof. Ming-Feng Tsai (Department of Computer Science, National Chengchi University)
時 間：2018.10.17(三)15:30-17:20
地 點：本校工程三館115室
摘 要：In this talk, I will introduce a music recommendation approach by using various contextual similarity information based on the framework of Factorization Machine (FM). In specific, we will go through the FM approach, the idea of feature similarity, and the incorporation of multiple feature similarities into the FM framework. By integrating different feature similarities, the approach enables users to discover diverse items that they never listened before. In addition, in order to avoid the high computational cost and noise within the large number of similarity features, we also present a grouping FM technique to alleviate the problems. In the experiments, a real-world dataset is used to assess the performance of the proposed method. The dataset is collected from an online blogging website (LiveJournal), which includes user listening history, user profiles, social information, and listened music information. Our experimental results show that, with the multiple feature similarities based on the FM framework, the proposed method improves the recommendation performance significantly. Furthermore, with the proposed grouping technique, the efficiency of the method also gets improved significantly. In addition to the FM approach, in this talk I will also share the experience and lessons we learned from the collaboration with industry, and will briefly introduce the further advanced methods published in recent years based on the network embedding techniques.
題 目： Music Recommendation Based on Multiple Contextual Similarity Information
主講人： Prof. Ming-Feng Tsai (Department of Computer Science, National Chengchi University)
時 間：2018.10.17(三)15:30-17:20
地 點：本校工程三館115室
摘 要：In this talk, I will introduce a music recommendation approach by using various contextual similarity information based on the framework of Factorization Machine (FM). In specific, we will go through the FM approach, the idea of feature similarity, and the incorporation of multiple feature similarities into the FM framework. By integrating different feature similarities, the approach enables users to discover diverse items that they never listened before. In addition, in order to avoid the high computational cost and noise within the large number of similarity features, we also present a grouping FM technique to alleviate the problems. In the experiments, a real-world dataset is used to assess the performance of the proposed method. The dataset is collected from an online blogging website (LiveJournal), which includes user listening history, user profiles, social information, and listened music information. Our experimental results show that, with the multiple feature similarities based on the FM framework, the proposed method improves the recommendation performance significantly. Furthermore, with the proposed grouping technique, the efficiency of the method also gets improved significantly. In addition to the FM approach, in this talk I will also share the experience and lessons we learned from the collaboration with industry, and will briefly introduce the further advanced methods published in recent years based on the network embedding techniques.
