search

turtlelovesshoes / SUMOTwitterInsights

Cultures of Experimentation workshop 2018 Round 5 initiative to analyze open text of Mozilla Support @firefox customer support sentiment
0 stars 1 forks source link

Positive Negative Sentiment for Social Inbound and Outbound #8

Open turtlelovesshoes opened 6 years ago

turtlelovesshoes commented 6 years ago

One of the use cases that we could act on was to analyze the outbound responses of social support is to find what words are positive ans what are negative.

In the Firefox 63 report we did just that with these results, analyzing them will follow: Positive/Negative OUTBOUND

AoA vs Outbound @firefox Constructing Target (Dummy) Classifier... Most predictive words of sentiment in Firefox Tweets: Summary: In replies do not use ‘work’ or red words but continue to use present tense to point out actions like ‘see’ ‘check’ and ‘there’’

Most Informative Features contains(work) = True neg : pos = 70.8 : 1.0 contains(3) = True neg : pos = 69.1 : 1.0 contains(new) = True neg : pos = 36.6 : 1.0 contains(able) = True neg : pos = 36.6 : 1.0 contains(without) = True neg : pos = 36.6 : 1.0 contains(using) = True neg : pos = 35.0 : 1.0 contains(looking) = True neg : pos = 35.0 : 1.0 contains(may) = True neg : pos = 35.0 : 1.0 contains(forward) = True neg : pos = 35.0 : 1.0 contains(reply) = True neg : pos = 35.0 : 1.0 contains(experience) = True neg : pos = 35.0 : 1.0 contains(can) = True neg : pos = 35.0 : 1.0 contains(via) = True neg : pos = 35.0 : 1.0 contains(it) = True neg : pos = 35.0 : 1.0 contains(here) = True neg : pos = 31.0 : 1.0 contains(there) = True pos : neg = 29.1 : 1.0 contains(check) = True pos : neg = 24.5 : 1.0 contains(see) = True pos : neg = 23.9 : 1.0 contains(actually) = True neg : pos = 21.0 : 1.0 contains(this) = True neg : pos = 21.0 : 1.0

Positive/Negative INBOUND - users @firefox Positive/Negative ALL Constructing Target (Dummy) Classifier... Most predictive words of sentiment in Firefox Tweets: Summary: Positive when the word red shows up, negative that people think Firefox is slow, but half the time we get thank yous and greetings are positive, so we should keep that up

Most Informative Features contains(fix) = True neg : pos = 16.3 : 1.0 contains(red) = True pos : neg = 16.0 : 1.0 contains(slow) = True neg : pos = 15.7 : 1.0 contains(page) = True neg : pos = 13.3 : 1.0 contains(tried) = True neg : pos = 9.8 : 1.0 contains(why) = True neg : pos = 9.4 : 1.0 contains(2) = True neg : pos = 7.4 : 1.0 contains(times) = True neg : pos = 7.4 : 1.0 contains(0) = True neg : pos = 6.4 : 1.0 contains(without) = True neg : pos = 6.4 : 1.0 contains(thank) = True pos : neg = 6.3 : 1.0 contains(much) = True neg : pos = 5.4 : 1.0 contains(better) = True neg : pos = 5.4 : 1.0 contains(getting) = True neg : pos = 5.4 : 1.0 contains(after) = True neg : pos = 5.4 : 1.0 contains(thought) = True neg : pos = 5.4 : 1.0 contains(days) = True neg : pos = 5.4 : 1.0 contains(found) = True neg : pos = 5.4 : 1.0 contains(hi) = True pos : neg = 5.3 : 1.0 contains(watch) = True pos : neg = 4.7 : 1.0

turtlelovesshoes commented 6 years ago

Open questions here, was it effective? The report was post boned to next week and will be updated then.