2016.8.30 Time series analysis - continual session

Situation

• Today, we are calculating whether continual post behavior of the fanpage owners would have an impact on the like it gets.

First thought

• We first took out the posts that were post continually and boxplot their like, then compare it to the boxplot of all posts. We found it stupid at last becuz we can never define appropriately ‘Post Continually’

Solution

• Later, our supervisor told us that we could draw it in two smart ways
  • 1st, y:like, x:time interval to the previous post.
  • 2nd, y:like, x:number of posts in the last x hours.

Analysis

• The 1st method is gorgeous that we do not need to add any of our subjective thought (we do not need to decide a ‘threshold’)
• The 2nd method is also good because compare to our orignal thought, it can include all posts on the plot (while our original one can only include ‘continual’ ones)
• It’s always better that we include as little as we can our subjective thoughts in data analysis while having no prior knowledge about the data.