Fig 1.
Typical positive and negative emoticons in Weibo.
The tweets containing both the positive and negative emoticons will be identified as ambivalent ones.
Fig 2.
Distribution of tweets posted at different time.
(a) The hourly pattern. (b) The weakly pattern. In both (a) and (b), insets show the absolute fraction of each type of tweets at different time, from which it can be seen that the ambivalent tweets only occupy a small fraction in Weibo.
Fig 3.
Comparison of topic preferences between ambivalent and ordinary users.
(a) Variation of positive emotion for topic detection. (b) Variation of negative emotion for topic detection. Peaks with clear event semantics are selected from the positive or negative emotional line of the ambivalent users and the points with the same time stamp are also highlighted for ordinary users.
Table 1.
Positive Topics Discussed by Ambivalent Users.
Table 2.
Nagative Topics Discussed by Ambivalent Users.
Fig 4.
Fraction of tweets mentioning others at different hours.
The ambivalent tweets contain significantly more mentions, especially at 20:00 and 22:00.
Fig 5.
Fractions of tweets posted in three days before and after shopping.
For x axis, the unit is a day and 0 stands for the shopping date. (a) Positive tweets. (b) Negative tweets.