Peter Dodd on Text Analytics (Complexity)
Podcasts · Technology · Source
Very similar techniques to the ones I’ve been developing for analysing political discourse, for Nalin’s book and Sanjeev’s PhD!
Hedonometer and Story Wrangler are both interesting projects to check out
StoryWrangler - Twitter Ngram Search
Some interesting ideas:
- Can look at both word usage, and how much words have been amplified over time. Difference in usage and amplification can present an interesting story
- Adding words that have suddenly emerged alongside other words can be important signals. For example, news was a very stable word for decades. Then suddenly, fake started to be associated with it
- You really care about “big movers”. Things that are a high absolute number but have suddenly increased in usage
- Word frequencies and semantic topic frequencies can be a useful predictor of geopolitical risk and macroeconomic risk
Metrics to look at:
- Story turbulence
- Narrative control
- Collective chronopathy — which stories are receiving a greater share of attention over time
Papers to look at