資訊管理學報
胡志堅;陳昱安;
頁: 209-237
日期: 2024/04
摘要: 資通訊科技的蓬勃發展，驅使眾多音樂愛好者透過音樂串流服務平台聆聽和分享音樂創作。然而，隨著音樂作品數量的增加，有效管理這些作品並提升音樂檢索效能，成為音樂數位典藏的重要課題。目前的音樂素材檢索和歸類，缺乏同時考量音樂風格及其創作之文化背景。因此，本研究嘗試運用深度學習、以及文字探勘等資訊科技，來分析歌曲風格和歌詞之間的關聯。首先，採用卷積神經網路模型(CNN)進行歌曲風格分類，接著結合組合式主題模型(CombindTM)分析歌詞主題傾向。研究結果顯示，所建立的歌曲風格分類模型能準確分類音樂風格並解析其音樂的組成成份。透過歌詞主題傾向雷達圖分析歌詞內容，也能解釋不同類別歌詞的意涵。整合這兩種機制能更深入地檢索音樂作品，並連接歌曲風格與歌詞意境，以及相關背景知識。建議將來可將歌曲風格音樂成分分析與歌詞主題傾向雷達圖分析整合到音樂數位典藏系統和網路音樂串流平台，提升音樂作品的檢索效能，並建立音樂知識脈絡。
關鍵字: 音樂風格分類;卷積神經網路;音樂資訊檢索;文字探勘;主題模型;
Abstract: The development of information and communication technology has led to many music enthusiasts using music streaming platforms to enjoy and share music. However, with the increasing number of music works, effectively managing these works and improving music retrieval efficiency have become important issues in digital music preservation. The current classification and analysis of music materials often overlook the music genres and cultural backgrounds. Therefore, this study utilizes information technology to analyze the relationship between songs and lyrics. Firstly, a convolutional neural network (CNN) is used for song genre classification, followed by the combination of a composite topic model (CombindTM) to analyze the thematic tendencies of lyrics. The research results show that the established song genre classification model can accurately classify music genres and analyze the compositional elements of music styles. By analyzing the lyrics' thematic tendencies using a radar chart, the textual meaning of lyrics in different categories can also be interpreted. Integrating these two mechanisms allows for a deeper retrieval of music works and the synthesis of song styles, lyrics' moods, and relevant knowledge. It is suggested to integrate both mechanisms into digital music preservation systems and online music streaming platforms in the future to enhance the retrieval efficiency of music works and establish the context of music knowledge.
Keywords: Music genre classification, Convolutional neural network (CNN), Music information retrieval (MIR), Text mining, Topic model;
瀏覽次數: 1648 下載次數: 120
引用 導入Endnote
頁: 209-237
日期: 2024/04
摘要: 資通訊科技的蓬勃發展，驅使眾多音樂愛好者透過音樂串流服務平台聆聽和分享音樂創作。然而，隨著音樂作品數量的增加，有效管理這些作品並提升音樂檢索效能，成為音樂數位典藏的重要課題。目前的音樂素材檢索和歸類，缺乏同時考量音樂風格及其創作之文化背景。因此，本研究嘗試運用深度學習、以及文字探勘等資訊科技，來分析歌曲風格和歌詞之間的關聯。首先，採用卷積神經網路模型(CNN)進行歌曲風格分類，接著結合組合式主題模型(CombindTM)分析歌詞主題傾向。研究結果顯示，所建立的歌曲風格分類模型能準確分類音樂風格並解析其音樂的組成成份。透過歌詞主題傾向雷達圖分析歌詞內容，也能解釋不同類別歌詞的意涵。整合這兩種機制能更深入地檢索音樂作品，並連接歌曲風格與歌詞意境，以及相關背景知識。建議將來可將歌曲風格音樂成分分析與歌詞主題傾向雷達圖分析整合到音樂數位典藏系統和網路音樂串流平台，提升音樂作品的檢索效能，並建立音樂知識脈絡。
關鍵字: 音樂風格分類;卷積神經網路;音樂資訊檢索;文字探勘;主題模型;
Constructing an Integrated Analysis Mechanism for Music Genre and Lyric Semantics using Deep Learning and Topic Modeling
Abstract: The development of information and communication technology has led to many music enthusiasts using music streaming platforms to enjoy and share music. However, with the increasing number of music works, effectively managing these works and improving music retrieval efficiency have become important issues in digital music preservation. The current classification and analysis of music materials often overlook the music genres and cultural backgrounds. Therefore, this study utilizes information technology to analyze the relationship between songs and lyrics. Firstly, a convolutional neural network (CNN) is used for song genre classification, followed by the combination of a composite topic model (CombindTM) to analyze the thematic tendencies of lyrics. The research results show that the established song genre classification model can accurately classify music genres and analyze the compositional elements of music styles. By analyzing the lyrics' thematic tendencies using a radar chart, the textual meaning of lyrics in different categories can also be interpreted. Integrating these two mechanisms allows for a deeper retrieval of music works and the synthesis of song styles, lyrics' moods, and relevant knowledge. It is suggested to integrate both mechanisms into digital music preservation systems and online music streaming platforms in the future to enhance the retrieval efficiency of music works and establish the context of music knowledge.
Keywords: Music genre classification, Convolutional neural network (CNN), Music information retrieval (MIR), Text mining, Topic model;
瀏覽次數: 1648 下載次數: 120
引用 導入Endnote