We often hear about data used in many tech-driven companies, but how does it work for a media business?

With his three years of experience working in Astro as a data scientist, Ahfaz Khan mainly focuses on customer acquisition analytics such as acquisition, upselling strategies, retention, etc.

Here, he shares with us his experience and insights that will give data science students a better understanding of the technical roadmap in this industry. 

Astro Malaysia Holdings Berhad is Malaysia’s leading content and consumer company in the TV, over-the-top (“OTT”), radio, digital and commerce space

Supervised vs Unsupervised Learning for Media Companies

supervised vs unsupervised

When it comes to these two models, it’s difficult to determine which is used more often as there is no stand-alone usage.

For example, at Astro, when we build a Predictive Churn Model, we will also be required to further build segments within the predictive list in order to customize the offerings.

Hence, if you are looking at data science from a business perspective, it is hard to separate both supervised and unsupervised learning. There will always be a balance of both in order to come out with result-oriented outcomes.

Application of Deep Learning & Machine Learning in Media

With media, there are so many different forms of content, such as images, videos, text etc; hence, how we apply deep learning is very different from other industries.

There is still the existence of structural data, but it is only one component of the entire data ecosystem. If you restrict yourself with the traditional formats of data analysis, your results become very limited.

Since the media industry has become so dynamic, the focus of many organizations is individual-centric, with personalization and hyper-targeting efforts.

Businesses are looking to understand individual customers, such as their likes and dislikes, their preferences and behaviors, and tailor products and services to customers accordingly.

In order to get that level of information, there is a critical need to dig deep into data. With that, data will progressively play a very big role in media organizations.

Watch Ahfaaz’s presentation on deep learning in D/M Summit 2019

Main Challenge of Data Scientists in the Media

In general, no matter what industry you are in, you will always struggle with the quality of data.

However, for the media industry, there is a whole different set of input as compared to just your structural data.

One of the biggest challenges is to access all these different forms of data and harnessing information to produce action-oriented results.

Nevertheless, data scientists in these industries have been exploring various platforms to harness these data, such as looking into customer interactions, social media behaviors and even call center recordings.

It is safe to say about 70% – 80% of the time and effort in the media industry goes into data preparation and filtering out quality data before going through analysis.

Recommended Deep Learning Framework

Ahfaz personally started with Keras when he embarked in this data science journey.

As many beginners venture into data science learning Python, Keras that is built on Python would ease the learning process. Also, it is a high-level API for neural networks and it allows instant prototyping which is a much simpler approach as compared to other frameworks.

Keras is good if you just want to plug and play, or if you have a data set that you know would definitely work with deep learning.

When it comes to Deep Learning Frameworks, which Neural Network should you start with?

This depends on what you want to do.

If you want to get into text processing and mining, you should pick up LSTM networks, as it is one of the most preferred frameworks when it comes to text processing. On the other hand, if you are looking into image processing, CNN might be more suitable for your needs.

When it comes to choosing which neural network you want to learn, it’s important to first figure out your purpose.

Corporate Enterprise Software Platforms (eg: SAS) vs Open Source Platforms (eg: TensorFlow):

Ultimately, when choosing a software platform, it is important to pick one that has strong technical support.

It’s crucial to ensure that you always have an easily accessible technical support that can guide you through the problems you face.

If you are looking for open-source, select those that have a strong community support, as this would show it is preferred by many; while at the same time, you would be able to debug or find solutions to your problems just by searching it online as many people would have probably experience the same problems before.

Advice to data scientists looking to venture into media organizations like Astro?

In general, there are 4 critical components of an outstanding data scientist:

Technical Skills:

  • Programming Skills
  • Statistics & Machine Learning Skills

Soft Skills

  • Business Understanding
  • Presentation & Communication Skills

In order to grow in the data science industry, there is no way of avoiding these components. If you are a complete beginner looking to enter this industry, start with picking up a programming language such as Python, and then slowly moving into Machine Learning skills and you can decide where to specialise your skills in.

On top of good technical skills, business understanding together with your communication skills will make you a better contributor to grow your company’s business. As much as it comes with experience, you should also take the initiative to read up on business reviews and articles, be aware of business news and trends etc, so that you are able to gain better insights.

You can learn more about how Ahfaaz, utilize machine learning in the media industry in his workshop at D/M Summit, titled “Deep Learning Frameworks in Media and Entertainment” plus learn how he transitioned from a medical student to being a senior data scientist, in a fireside chat we had with him.

Get your all-access pass to learn from Ahfaaz and many other industry experts on data science and marketing subjects.