Page 120 - 臺大管理論叢第32卷第1期

Internet Celebrity Economy: Exploring the Value of Viewers’ Comment Features and Live Streamers’
Marketing Strategies in Forecasting Revenue

sending behavior. Further, a persuasive streamer has more praising (M = 1.246, SD = .608)
comments, and the more he/she shares food features (M = 1.298, SD = .636), the higher
the viewers’ gift- sending behavior. In addition, a comical streamer has more excited (M =
1.282, SD = .568), praising (M = 1.345, SD = .567), complaining (M = 1.258, SD = .549),

disappointed (M = 1.269, SD = .567), and ridiculing (M = 1.188, SD = .570) comments,
and the more she/he chats with the viewers (M = 1.282, SD = .615), the higher the viewers’
gift-sending behavior.

6. Forecasting Accuracy

We perform Leave-one-out Cross-validation (LOOCV) to evaluate the HB predic-
tion model. James, Witten, Hastie, and Tibshirani (2013) indicate that LOOCV is K-fold

cross-validation taken to its logical extreme, with K equal to N, the number of data points
in the set. This means that for N separate times, the function approximator is trained on
all data except for one point, and a prediction is made for that point. We add three new

live streams with a total of 13 data points for analysis. Therefore, this study conducts the
analysis 13 separate times, where we use 12 streams for training versus one live stream
for holdout, repeating such that each stream gets a chance to be out-of-sample once. Two
quality indices are computed for model validation: Mean Absolute Error (MAE) and the
Root Mean Squared Error (RMSE). The MAE is equal to the absolute difference between

the actual value of gift giving and posting observed at time step t, throughout the duration
of the lievestream, and the estimated gift-sending value of the HB model at that same
time step, and averaged overall values. RMSE is the standard deviation of the average of
squared differences between the estimated gift-sending value and actual gift-sending value

at time step t. We compute the MAE and RMSE across all 13 trials for the HB model and
compare the differences. The results are reported in Table 6. Details of the cross-validation
MAE and RMSE are small. As mentioned previously, we perform a log transformation for
the variable of gift giving. We use the exponential of the RMSE value to restore it to RMB.

As shown in Table 6, the RMSE value of the first trial is 3.23, so the exponential is 25.27
RMB. This shows that our prediction error of the viewers’ gift-sending value in the 13th
live stream is 25.27 RMB.

112