'You can improve yourself especially in the area that you want to grow most'
'Predicting tune-in, a Bayesian multi-level model to explain and predict tune-in to premieres of US TV series”. This is the title of my thesis that I wrote at Pointlogic last year. To explain and predict tune-in percentages of new TV series, I had information on some show-characteristics, such as the genre of the series, and the channel it was broadcast on, but I also had a data set of respondent-level characteristics available (by show).
The respondent-level data contained information on what media advertisements a respondent has seen, and on what his or hers personal TV viewing preferences were. Using this information, I designed a multi-level model that explained which people tuned-in for a certain type of show. Such a model can be used to optimize media campaigns to maximize tune-in for new shows, but it also gives general insights in the most important drivers of tune-in for premieres of series.
Great place to learn
An important advantage of using a Bayesian multi-level model in this case is that such models are very suitable when one wants to obtain out-of-sample predictions. In my thesis, I want to be able to provide accurate tune-in predictions for premieres of series that are broadcast on (for example) different channels than that are used in the modelling data. A Bayesian multi-level model could indeed provide such predictions, but there was a problem... My data set on multiple premieres of series was quite big, and estimating a relatively simple model already took several days. Therefore, I have used the much faster and relatively new Variational Bayes method to approximate the complex Bayesian multi-level models.
Good place to grow
You’re working in a very motivated team, that is eager to learn and to solve the challenges that come with every new project. It seems like none of the projects is standard, which means there’s always something new to learn. With every new project, we’re also trying to improve methodology. I am focusing on integrating more Machine Learning techniques in the project workflow, which should help us in understanding the effect of the variables on the dependent variable. By better understanding the effects, we should be able to develop better models in a smaller amount of time.
Every member of the team has its own focus points and responsibilities. This means that you can improve yourself especially in the area that you want to grow most.
The right working place for writing a thesis
What I mostly liked about writing my thesis at Pointlogic was the relevance of my thesis topic to the company. I worked with real data, and the results of my thesis provided useful insights for the company that can actually be used for future projects. Furthermore, I did not have much experience with Bayesian modeling before I started writing my thesis, nor did I have any experience with multi-level modeling and Variational Bayes. This was no problem at all, because an excellent thesis supervisor at Pointlogic guided me, and I feel like I’ve really learned a lot here.
Why working at Pointlogic?
I think the two most important factors to make a job a nice one are diversity and challenges. Both of these together describe my work experience at Pointlogic so far very well. I have been working on various projects, and all of these projects had their own challenges that needed to be solved. I got a lot of responsibility in these projects already when I just started working, and it’s very rewarding to deliver results for a project that you’re really proud of. Despite this, I am also working in a team with very motivated people. If I am not able to solve a challenge by myself, there’s always someone willing to help out.
Interested in our Intern Quantitative Marketing position? Apply here.