Analyze your twitter favorites over time

twitter-glm.r

library(tidyverse)
library(lubridate)
library(rtweet)

user.name = 'PhDemetri'
#Get user's most recent tweets
user = get_timeline(user = user.name, n = 5000)

#Clean the data.

#Remove outliers and get a reasonable time.
#This part will depend on the user
user.data = user %>% 
  filter(created_at<round_date(today(), unit ='month'),
         favorite_count<300, #change this if you are routinely wracking up numbers like this
         year(created_at)>2016) 

#Count number of tweets and sum up favorites.
#Create a variable for months since begining fo time period
user.data  = user.data %>% 
  mutate(month = round_date(created_at, unit = 'month')) %>% 
  group_by(month) %>% 
  summarise(N = n(), fc = sum(favorite_count)) %>% 
  ungroup %>% 
  mutate(t = interval(min(month),month)%/%months(1))


#Get an idea of the trend
user.data %>% 
  ggplot(aes(month,fc/N))+
  geom_line()+
  geom_smooth()+
  theme_minimal()+
  theme(aspect.ratio = 1/2)


#Here is the model
#Use negative binomial.  You could use a poisson glm,
#but my data seems to be overdispersed.
model = MASS::glm.nb(fc ~ t + offset(log(N)), data = user.data)

#summary
summary(model)