Research On Song Emotion Classification Methods Combining Music Content And Lyrics

In today's era of digital information, the Internet impacts greatly on people's lives. A large number of text documents, audio, images appear in the Internet. How to seek the magnanimous information effectively is a hot research in the field of information processing. The song emotion classification methods combining music content and lyrics can solve these problems well. However, the methods still exist the problem about song emotion multiplicity and the effective integration of classification results.In this thesis, the methods about combining music content and lyrics for song emotion classification were researched deeply, the following work were done:1. Analyzed and summarized the advantages and disadvantages of song emotion classification based on music content and lyrics respectively;2. Provided a method combiningθ-MLkNN and Term Frequency*Inverse Document Frequency (TFIDF) rules , used the lyrics'right labels to correct the music content's wrong labels, in order to improve the accuracy of song emotion classification;3. Established a method combiningθ-MLkNN and lyric-based song emotion detection. In this combined approach, clustering analysed the lyrics'results, calculated the emotional threshold value, determined the lyrics, music content classification results. And optimized the classification results by linear fusion method to determine the importance of emotional categories in each song. So improved the accuracy of song emotion classification;4. Tested proposed song emotion classification methods on the dataset of 396 English songs, compared the results to the MLkNN's results, verified the effectiveness of the proposed methods. Finally, the research work involved in the thesis was summarized and the future developments were forecasted.