My non-mathematician friends often look at me strangely when I talk about equations and models as if they are people.
Yesterday was another classic example, in answer to the standard, “how was your day” I replied ‘Not good, my model isn’t cooperating and I can’t figure out what to train it with or how to get it learn’. Cue bemused expressions and an awkward silence.
Well today the team at Black Swan are working on the answers.
The model we have has so far been limited when co-operating with predicting Cannes Lion winners. We can see a ‘general trend’ in how our odds match up to the awards, we are indeed giving higher odds in advance to those that are winning. We can see that if you enter more categories you tend to win more awards. And for some of the categories we see that the most tweeted entrants win the Grand Prix prize. Excellent! But, sadly useless, as we can still not pinpoint the exact winners.
Overnight, Tim (out in the LA office) has got the model to learn from the week’s results and integrated this with the current model . This has improved the odds, meaning we can see a bigger division for those that win an award and those that don’t. Still, historical data isn’t giving us enough information to make clear predictions. Check back on our blog at 5.30 to find out our predictions!
Also, Twitter just isn’t twittering enough!
We have decided (contrary to stereotypes) that advertisers just don’t talk enough. The volume of tweets with #CannesLions is worryingly low.
Is Twitter just not cool anymore, if so where have all the cool kids taken their chat?
We need some new data sources.
Today we are looking to YouTube views and Facebook shares to get ‘crowd sourced’ information on what people liked. We are hoping popular shares amongst the general public are reflected or correlated with what the judges think. We don’t know if this is true, pulling in all this data and finding out is an experiment. As with all science, you start with an idea of what may be true and then test it.
A better source of data would be non-anonymised and specific to those in the industry. The judges come from within the industry, so anything we can find to infer their preferences will help us, so today we will also try and hunt that down.
It may seem like we are grasping out into the ether ad hoc, however we are actually working through the timescales to find one or a combination that is predictive. Let’s consider how does this fit to the maths? It’s all to do with timescales – something incredibly difficult and important for predictions. The historical data gives us the background trends over a long time period, on the characteristics of nominees and their success rates (historic), the tweets (if there were more) could give us a ‘now-cast’ of what’s hot at Cannes (current). YouTube and Facebook will fill in the mid term time scale.
Think of it like a weather prediction, based on three components.
Climate is something we can project well, it changes slowly (relative to a human’s lifetime) and we can use all historic data to follow long-term trends and see how they change. This is like our historic model, it gives us our baselines.
Weather changes quickly (not just every day but intraday), there are some season changes that we can see, hence we can know which month it is and narrow our choices of weather. This is like the YouTube and Facebook data (we hope)! It gives us some context for our choices.
Then there are the current conditions. In the case of the UK there are up to 5 weather systems battling it out above our heads. How do you choose one? Maybe say the one that’s winning right now is likely still to be winning tomorrow? We were hoping tweets would tell is this, but sadly this is not so.
Follow Black Swan’s predictions, insights and other activity around Cannes Lions 2015 on our Twitter feed:@BlackSwanData