- 學位論文
歌詞情感分析於中文流行音樂情緒判斷之研究
Affective Lyrics Analysis for Mood Estimation of Chinese Pop Music
摘要
本研究為基於歌詞內容分析以協助判斷歌曲之情緒。此研究可應用於以情緒為主的歌曲推薦系統。由於現今音樂情緒分析研究多以分析樂曲的數位訊號為主，但缺乏與歌詞情緒相關之研究，因此，探索歌詞如何影響歌曲情緒為本研究之重點。 有鑑於目前音樂情緒分析研究缺乏公開的資料庫，因此，我們建立iPop資料庫，包含分別對樂曲、歌詞、及兩者兼具之資料內容，共三類情緒標記，並使用統計方法分析樂曲與歌詞如何影響音樂之情緒。 我們使用愉悅(valence)及激動程度(arousal)做為iPop資料庫之情緒標記。統計分析結果顯示歌詞內容為影響音樂情緒之重要因素，並可幫助傳統以樂曲為主的音樂情緒分析研究。分析標記的結果更顯示歌詞主導音樂之愉悅程度，而樂曲主導音樂之激動程度。此外，根據上述之分析結果，我們藉由計算歌詞文字的情感分數，來擷取歌詞中的情感文字(affective words)，以協助預測音樂之愉悅程度。實驗中，我們分別比較以歌詞及樂曲為主的情緒分析結果，結果顯示情感文字能提昇音樂情緒判斷之效能，其精確率可達80%。
並列摘要
Music mood estimation (MME) is a key technology in mood-based music recommendation.While mainstream MME research nowadays relies on audio music analysis, exploring the significance of lyrics text in predicting song emotion is gaining attention in recent years. One major impediment to MME research is the lack of a clearly labeled and publicly available dataset annotating the emotion ratings of lyrics text and audio separately. In light of this, we compiled a dataset of 600 pop songs (iPop) from the mood ratings of 246 participants who experienced three different song sessions, lyrics text (L), audio music track (M), and lyrics text plus audio music track (LM). We then applied statistical analysis to estimate how lyrics text and audio contribute to a song's overall valence-arousal (V-A) mood ratings. Our results show that lyrics text are not only a valid measure for estimating a song's mood ratings but also provide supplementary information that can improve audioonly MME systems. Furthermore, the lyrics text is more dominant at deciding the valence value of music than audio music track and audio music track is more dominant at deciding the arousal value of music than lyrics text. To improve the performance of MME system, we proposed sentiment score approach which extracts affective words to be lyrics text feature. The approach first calculates sentiment score base on the words' average term frequencies in positive music and negative music, then extracts affective words according to their sentiment scores. We also construct an MME system system which uses affective words as lyrics text feature. In estimating mood of music, the model with affective words as lyrics text feature performs better than the model with lyrics text feature including both affective words and non affective words or the model with only audio track features.