Emotional targeting

Why do moods matter politically? Think first about economics. If you can predict the mood of a group of people then you might be able to predict how likely they are to buy (and sell) and how much they will pay (and sell for). So investors who speculate on the stock market have a lot to gain by accurately assessing and predicting the moods of markets.

Here “mood” is a handy catchall for an assessment of whether potential customers and investors en mass are optimistic or pessimistic, take a long term or short term view, or are prepared to take risk or are risk averse in relation to particular commodities. Such reductive mood assessment and prediction is important for hedge fund managers.

Hedge funds 101

Hedge funds are investment “vehicles” for speculating on markets (as far as I understand them). So a company managing a hedge fund invests your money to buy shares on the stock market, with the expectation that they (the management company) will later sell the shares at a profit. The return on the resale allows the company to pay back your investment with interest and make a profit. (In fact, individuals can’t invest directly in hedge funds, but your bank or other portfolio manager might — and I’m simplifying.)

There are risks, as the value of the shares the investment manager buys may drop. In this case the company loses on that transaction. But if, on average, the hedge fund company sells the shares it acquires at a price sufficient to exceed what it paid plus the costs of the operation and any interest owing, then it wins.

Hedge fund managers that speculate in this way have to be big and robust enough to secure borrowing, to absorb the risks, and be agile and good enough at making predictions.

Algorithms and emotions

That’s where sophisticated computation comes in. Hedge fund managers use algorithms to buy and sell shares and transact funds very quickly, even over minutes and seconds. They also look increasingly to all the data out there to gauge mood shifts in markets — including social media data.

In February this year, feature writer in the Guardian Carole Cadwalladr delivered an engaging and disturbing article about how some of the smarts deployed to predict market behaviour are being deployed in political campaigns. There are reports online about billionaire Robert Leroy Mercer, joint CEO of Renaissance Technologies, an investment management firm deploying sophisticated software to manage hedge funds.

Mercer is also a computer science PhD with academic credentials. He co-authored articles on automated language translation algorithms as deployed in Google Translate. Such systems infer translation mappings from vast repositories of sentences for which the translation is already known, e.g. English and their equivalent French sentences.

The algorithms use statistical methods for effecting translation between these languages, bypassing lexicons and grammar rules. The programmers don’t have to know much about the languages they are designing for. They just need an extremely large (big data) set of sentences and their translations into other languages.

Data politics

The political punch line to this narrative is that Mercer was one of the major funders for both the Trump and the Brexit campaigns. He also has a large stake in a company called Cambridge Analytica (worth a look) that was employed by both of these campaigns. Apparently, “trackers from sites like Breitbart could also be used by companies like Cambridge Analytica to follow people around the web and then, via Facebook, target them with ads,” according to Cadwalladr.

She reports, “On its website, Cambridge Analytica makes the astonishing boast that it has psychological profiles based on 5,000 separate pieces of data on 220 million American voters – its USP is to use this data to understand people’s deepest emotions and then target them accordingly.” She adds, “The system … amounted to a ‘propaganda machine.'”

It’s all about mood manipulation. From then on the story gets complicated …


  • Brown, Peter F., Vincent J. Della Pietra, Stephen A. Della Pietra, and Robert L. Mercer. 1993. The mathematics of statistical machine translation: parameter estimation. Comput. Linguist., (19) 2, 263-311.
  • Levitt, Steven, and Stephen J. Dubner. 2005. Freakonomics: A Rogue Economist Explores the Hidden Side of Everything. London: Penguin http://freakonomics.com/
  • Cadwalladr, Carole. 2017. Robert Mercer: The big data billionaire waging war on mainstream media. Link.



  1. Daniel Bridgman says:

    Along a related vector … struck by purveyors/hawkers of OpenCV, Deep Learning skill-building that commoditize facial-expression recognition and “mood” detection (for subscription or publication fees). A late-breaker: “Compete in Kaggle’s Facial Expression Recognition Challenge and train a CNN (from scratch) capable of recognizing emotions/facial expressions in real-time.” . This conjures scenarios of travelers, crowds at public events, being mined en masse for affect, mood data via facial “expression” algorithms utilizing deep neural networks , results would then be auctioned (as per Cambridge Analytica) to highest bidders in whatever political arena’s relevant/awash in big money. You don’t have to be clinically paranoid to get a whiff of techno-rational malice seeking baser nature. But as you demonstrate via the CP Centre tool, accuracy, reliability’s currently dubious. Nonetheless, as with Google Translate, leaps of improved accuracy do happen, especially when efforts and resources are unrelenting. Just channeling the mood from above …

    1. Thanks for this addition Daniel. If the goal is achieved of automated recognition of emotions from photographs of faces then the next challenge will be to identify emotions from moving images. It is possible to capture a split second expression that gives the face whatever emotion the photographer wants. In any case, the cues to reading someone’s emotions are presumably highly dynamic. But this is perhaps an argument about the difficulty of automated emotional recognition, whereas you are referring to the social and political implications, which are indeed a challenge.

  2. Daniel Bridgman says:

    Perhaps this belongs under “joys and sorrows of automation” … for although facial-expression mining for clues to mood, emotional disposition, is at present, naively obsessed with key frames–weighted still images in captured flows–the ultimate goal remains stark: extraction of marketable significance from copious strands of video. The auctioning of results to whomever’s willing to pay is triply ominous. At the moment the butt of these efforts seem to be neo-liberals, who were ascendant under previous administrations both in the US and UK; while 2017’s beneficiaries appear increasingly to be right-leaning political PACs with more money than god. Nonetheless deep learning, Open Computer Vision, etc., are mostly open-source distributions that aim to emancipate cold-war (covert) military-industrial surveillance legacies from outmoded automation techniques and submit them to newer, fleeter, ones. Even though the code’s rarely copied verbatim, project objectives, notorious debacles, catastrophic time-n-money sumps (of previous eras) tune development and tools of choice. But as you (and others) have observed, to and fro swings of fortune can alter the direction of benefit without notice. This may be why anxiety runs so high.

Leave a Reply