歌詞情緒標註及辨識

Lyrics Sentimental Annotation and Detection

郭冠呈(Kuan-Cheng Kuo)

指導教授 ： 林守德

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.00822

摘要

近年來，找出音樂的特徵及建立音樂與情緒間的關聯性，並判斷音樂所表達的情緒成了很重要的議題。本論文致力於收集歌詞的情緒標籤並且建立分類模型來偵測歌詞中的情緒。為了收集歌詞的情緒標籤，我們製作了兩個人類計算( Human Computation) 的遊戲並結合了Facebook的社群功能。在第一個遊戲中，我們讓玩家回答一些標記問題，並且在玩家作答時收集情緒標籤。在第二個遊戲中，我們透過收集複數個標籤來驗證標籤的正確性。在收集完標籤後，我們利用歌詞中語意及文法的特徵並使用半監督式學習來訓練分類模型。相較於音樂訊號，歌詞較缺乏旋律的資訊，因此我們也提出一些新的特徵來從歌詞中取得節奏的資訊。

關鍵字

標記 ； 人類計算 ； 歌詞 ； 半監督式學習 ； 情緒分類

並列摘要

This thesis aims at collecting lyrics sentiment labels for creating a classification model to detect the emotion of songs. To collect the labels of the lyrics, we propose two human computation games. In the first game, the emotion labels are collected when players answer some annotation questions about Chinese lyric. In the second game, multiple labels of a song are collected for verification purpose. The games are implemented with Facebook API. The second goal of this thesis is to construct a model for lyric sentiment detection. We propose several syntactic and semantic features and exploit the semi-supervised learning approaches to learn the classification model. Furthermore, some heuristics are proposed to extract information about the tempo of songs from text to compensate for the insufficient melodic information in lyrics.

並列關鍵字

Annotation ； Human computation ； Lyrics ； Semi-supervised learning ； Emotion classification

參考文獻

[1] Von Ahn, L. (2006). Games with a purpose. Computer, 39(6): 92-94.

[2] Von Ahn, L., Maurer, B., McMillen, C., Abraham, D., & Blum, M. (2008). reCAPTCHA: Human-based character recognition via web security measures. Science, 321(5895): 1465-1468.

[6] Russell, J. A. (1980). A circumplex model of affect. Journal of personality and social psychology, 39(6): 1161.

[8] Mandel, M. I., & Ellis, D. P. (2008). A web-based game for col-lecting music metadata. Journal of New Music Research, 37(2): 151-165.

[12] Scharl, A., & Weichselbraun, A. (2008). An automated approach to investigating the online media coverage of US presidential elections. Journal of Information Technology & Politics, 5(1), 121-132.