- 學位論文
基於特徵標籤之飛輪運動音樂推薦研究與實作
Research on the Recommendation of Flywheel Sports Music Based on Feature Tags
摘要
近年來人們的生活瑣事都會搭配著音樂進行，運動也不例外，其中飛輪運動更是注重與音樂的協調性。而飛輪運動因為疫情影響導致從傳統的實體教室授課轉變為線上影音教學，但就目前兩種形式的訓練都是由教練安排音樂及動作，缺少考量學員的偏好制訂。 在過往研究中我們已建立完成飛輪課程平台，但其中對於課程音樂的適性化推薦尚未有效達成，而許多研究指出藉由聆聽喜好的音樂與節奏有助於延長訓練時間以及大幅提升運動的成效，所以本研究以改良推薦機制為目的，修正過往研究中找到的問題，再藉由蒐集音樂特徵標籤計算歌曲之間的相似度，以及取得使用者所給予的偏好評分，分析使用者可能喜好的歌曲加以推薦，並且本研究加上考量與飛輪動作搭配的合適度，過濾其標記不合適的歌曲以優化推薦成效。 本研究之實驗將受試者分為兩組，將收集到的偏好評分進行平均計算，並且與過往研究的結果一同作為比對，經實驗結果發現基於音樂特徵標籤推薦方法明顯效果較好，平均分數高於過往數據約0.5分左右。我們修正了特徵標籤的蒐集步驟和音樂評分的尺度等等，並且發現在有加上合適度詢問的機制的實驗組二當中更是有效過濾歌曲，選擇使用者更喜愛的音樂作推薦，使得平均分數趨勢向上，以此表示達成我們優化之目的，有效提升推薦程成效。
並列摘要
In recent years, people's daily life will be accompanied by music, and sports are no exception. Among them, flywheel sports pay more attention to the coordination with music. However, due to the impact of the epidemic, flywheel sports have changed from traditional physical teaching to audio-visual teaching. However, the current two forms of training are arranged by the coaches to arrange music and movements, and lack of consideration of the preferences of the students. In the past research, we have established a platform for completing the flywheel course, but the adaptive recommendation for music has not been effectively achieved. Many studies have pointed out that listening to your favorite music and rhythm can help prolong training time and greatly improve the effectiveness of exercise. Therefore, this study aims to improve the recommendation mechanism and correct the problems found in previous studies. Calculate the similarity of all songs by collecting music feature tags, obtain the user's preference score, and then recommend the user's favorite songs. In this study, in addition to considering the suitability of music and flywheel action, inappropriate songs were marked and filtered to optimize the recommendation effect. We divided the subjects into two groups for the experiment, averaged the collected preference scores, and compared them with the results of previous studies. The experimental results show that the recommendation based on music feature tags is obviously effective, and the average score is higher than the previous data. We have revised the collection steps of feature labels and the scale of music scores, etc. In addition, the mechanism containing the suitability query is more effective in filtering songs, which can select the music that users prefer, and make the average score trend upward. This means that we achieve our optimization goal and effectively improve the effectiveness of the recommendation process.