How machine learning predicted traitors on Game of Thrones

How machine learning predicted traitors on Game of Thrones

Using predictive models for marketing

The key challenge in the predictive models our team develops for marketing is to predict over a specific time period. We are trying to find the precise time points behind the data that determine when consumers are willing to buy. We took inspiration from Airbnb and developed a machine learning model for the fictional characters in Game of Thrones, adding coefficients that reflect the difficulty of selling in the real world, so that we can grasp the precise time when consumers have the need to buy and respond in time. If you assume that a potential buyer is actually a resident of Westeros, and you blur the line between "good" and "evil", then you have to think carefully - everyone may be a traitor lurking around you at any time (you can also understand that he will buy your product).

So how can you predict when they are going to defect (or buy out)? Our first challenge was to transform the data we needed for training - a list of various behaviors of people with different personality traits - into personality trait data that could be processed by the model. Then, we used these activity data that could represent their personal personality traits to make predictions about them.

Behavioral scoring method

There is a way to count the number of actions associated with a persona, which can be used to train our predictive model (similar to how marketing automation systems score leads). Unfortunately, this method does not help us distinguish whether an action happened in the past or in the present. This ability to distinguish is critical if we want to predict actions that will happen in the near future.

From another perspective, we might only focus on recent activities. This does help us keep our data up to date and address the negative impact of old data on our evaluation. But what if a character hasn't done anything recently? We'll still evaluate him based on his past actions. And we'll still keep some of his recent historical activity data, because some seemingly one-off activities in the past can turn into a significant pattern and affect future decision making.

We can benefit from a hybrid approach. For example, we can combine the past activities of the person in the model with the recent activities. In addition, we can use a series of different interface windows to treat recent activities differently. This way we can remember the actions that happened three weeks ago, but we can use different standards to measure the actions that happened yesterday.

Tracking behavioral targets on the move

Please remember that the emotional state behind the character will change over time. The following chart reflects the historical changes in personality to see how the emotions behind the character's personality traits affect our prediction goals:

You can see that in August, based on his recent behavior, our model thought he would defect (buy the product) during this period. Although this was expected, he remained loyal for several months. Of course, his defection did happen subconsciously. Since people's emotional state (whether they are ready to defect) changes over time, our model also needs to know whether someone is about to defect, so that we can know the exact time to strike back.

Model evaluation thinking: scoring and re-scoring within a time series

In order to understand whether our model can accurately reflect the motivations of characters, each character should be attached with a reference score - we need to assess how trustworthy they are, and this reference score should be adjusted over time. But this will make our evaluation system very complicated, because our views on the "good" or "evil" of a character will change over time, just as the motivations of these characters' own behavior will also change over time.

Another problem will arise when the scores stay at a peak before returning to a level. The spread of misleading predictions may make us temporarily skeptical of some extremely loyal characters, so we must ensure that the model's evaluation function is based on all scores over the entire period. When we retrain the model, we must penalize these incorrect scores and compare them to see which model is relatively better.

To evaluate whether a model is performing well, we simply need to consider the score we get each time we evaluate a character (daily or weekly), and observe how well the model predicts their behavior the following week. If at the beginning of a week, we assert that a character is very likely to defect, and they defect on Thursday of that week, then our model is working well and is giving positive and correct recommendations. However, if this character still does not defect in the predicted week, but does not defect until Thursday of the following week, then our model's recommendation is wrong. In this case, we have to consider how to score this character the following week.

Conclusion: What do we learn from this inspiration?

From this hypothetical case, we can see that if you want to build an accurate behavior assessment model, you need a lot of sufficient thinking and experience, and you need to use the correct measurement method to determine the accuracy of the score. When the time span of machine learning is very long, it becomes particularly important to monitor the model and be careful about possible deviations. Always remember that if your model makes the same wrong judgment about the same person over and over again, then your model will face the end, and at the same time, it also indicates that your model needs to be updated.

If you follow all the tips above, your behavioral scoring model will bring valuable machine learning to your business needs. Knowing what people do and when is the key to predictive success.

<<:  Apple launches new iPod Touch: starting at 1,495 yuan, is it worth buying?

>>:  SEMI: Global semiconductor materials market output value reached US$43.4 billion in 2015

Recommend

Spring Festival holiday arrangements for 12 app stores

360 Mobile Assistant Application & Game Revie...

Channel pitfalls: 8 hidden rules of App operation

First, let’s deconstruct the position of App Oper...

3 key points for copywriting, verified by 127 advertisers

In information flow advertising consultation, the...

Google brings Android Live Captions to Chrome

Live Caption is a feature released with the launc...

Talk about the three methods of gaining users through paid social networks

In recent years, various paid communities are sti...

Broken bones by a leaf? Leaves in the south are really hard to deal with...

recently Mr. Liu from Jieyang, Guangdong By a fal...

Event Operation: How to achieve high conversion rate through poster copywriting?

Operations staff often come into contact with eve...