Page 113 - 臺大管理論叢第32卷第1期
P. 113
NTU Management Review Vol. 32 No. 1 Apr. 2022
streamer behavior occurs. Moreover, DouYu offers several types of gifts as donations, and
the value of each gift is different. However, Zhou et al. (2019) only calculate the number
of gifts, which might have produced significant measurement errors. This study revises the
coding of the gift values according to the gift values provided by the DouYu website (See
the appendix).
This study extracts the following attributes for each live-stream show on food
products: (1) reviewer ID, (2) time at which viewers post a comment, (3) time at which
the streamer receives a gift, (4) content of the comments, and (5) the characteristics
and behaviors of the streamers. Information extracted from viewers’ comments can be
classified in terms of volume and valence. Besides, discrete emotion is extracted from the
content of viewers’ comments.
3.2 Sentiment Analysis
Given the wide variations in the features of various social media platforms, this study
performs a sentiment analysis of the emotional features of comments on live-streaming
platforms. We divide the viewers’ comments into eight categories as follows: volume,
positive, negative, excited, amused, ridiculing, complaining, and disappointed. Because
the danmaku appeared in the 10 streaming channels are written in Chinese, this study
6
adopts the package jiebaR to perform word segmentation. To identify the valence (i.e.,
positive, negative, and neutral) and emotion (i.e., excited, amused, praising, complaining,
disappointed, and ridiculing) of a word, we adopt the text segmentation technique. First,
we use an emotion dictionary code called the “National Taiwan University Semantic
Dictionary in tmcn (Li, 2019)” to infer the valence and emotion of a word. Second, we
7
mark the words with the valence or the emotion. This study calculates the volumes of
the total comments, positive comments, negative comments, and comments with various
discrete emotions every second. At the end of this procedure, a final coding is assigned to
6 Package “jiebaR” is a vetted package that provides Chinese text segmentation, keyword extraction,
and speech tagging. Proir studies have adopted this package (e.g., Zhou et al., 2019).
7 Package “tmcn” is also a vetted package that provides the text mining toolkit for simplified Chinese,
which includes facilities for Chinese string processing, Chinese NLP supporting, encoding detecting
and converting. Moreover, it provides some functions to support the ‘tm’ package in Chinese. This
package has been widely used in practice.
105