Live Blogging the 2020 Election with Scalable Google Trends

Updated at 5:17 pdt (at bottom)

Wow, it’s November 3rd 2020! That means election day in the US, and what an election. I’m going to eschew politics except to say GO VOTE. We have the day of at LSG, but I figured this would be a great time to deploy a new tool and to do so i’m going to live blog and tweet the election.

Specifically we hooked up Google Trends to our data analysis stack, built an elastic load balancer to be able to scale jobs with cost and now it’s just party time. And when I say party time I mean we can pass thousands of keywords into Google Trends, get them back hourly, process the keywords as NGRAMs to better understand what they are about.

Okay, first start here with these two fantastic articles by Ruth Everett:

An Introduction to Using Google Trends with Python

Visualizing Python and Google Trends

Okay, now that you have read those fantastic articles let me just show you around. First we started with a manually curated seed keyword list, then ran it through a Google scraper of ours to pull related searches. We tool that list and ran it through Google Trends and pulled back the top and rising terms as well as their scores. We then ran that through the NLTK ngram library to be able to get categories/topics/themes. Then we data warehouse and dumped it into GDS. Screens below:

This is a view of the ngrams of rising queries from our seed list visualized in Google Data Studio

Ngrams of top queries in Google Data Studio

As you can see themes and topics!

Also, we have the ability to drill down to see the terms that make up the ngrams

This screenshot below is a drill down into “results 2020”

Cool right?

Anyway, I’m going to be updating this every hour and sharing any cool info on this liveblog and on twitter.

___________________________________

5:17 PST update

Just ran the tool to capture 8am EST trends data. So far early searcher trends are looking to catch the latest poll numbers, which is unsurprising. Though I do think it curious how focused they are on candidates ages and net worth. As this beginning to develop I expect this data to be rather…interesting.