Brave Men and Emotional Women: Analyzing Gender Bias in Bollywood Songs

Stereotypes exist in several sections of society including various means of popular entertainment. We believe that Bollywood songs are no exception as there has been a certain change in the characteristics of the songs’ lyrics over the past few years. Hence, to computationally study Bollywood’s songs lyrics from the Hindi movie industry, in this paper, we examine their style of writing and the presence of any biases. We analyze the changes in the songs’ lyrics over time and quantitatively show the change in the pattern by evaluating the rank of certain sensitive words in the songs. We calculate embeddings for the lyrical vocabulary using Word2Vec, FastText, and GloVe algorithms and use the WEAT similarity score to show that Hindi songs indeed suffer from racial and gender bias. The metrics we obtain can be further used for more formal problems of music recommendation, lyrics generations and popularity prediction.