Network Analytics for Neural Network Optimization

Your brain is constantly building connections and mental relationships in order to make sense of the world.

This is the equivalent of 86 billion neurons forming a neural network to make associations and predictions based on information and knowledge you’ve gained. Over time your brain takes shortcuts, seeking the most likely relationship or association based on an input or past experience.

A team of data scientists at O2 Czech Republic are building advanced neural networks to gain accurate context of its mobile network. Mobile network analytics creates better context into network usage and subscribers, creating a better prediction of customer behavior, more personalized services, and scaling superior customer experiences from thousands to millions.

Translating this to a mathematical equation and words that grab the attention of a data scientist, an algorithm can determine the vector, or relationship distance between an input and its context.

For example, if your colleague asked, “How much money you have in your pocket,” you may think they are asking to understand if you can afford the lunch you are eating together. But rather they were asking to understand how much economic impact you have.

This is an example of “Word2Vec,” a deep learning algorithm used to understand a group of words. Why are you saying what you are saying in an effort to understand context. While this is a linguistic mathematical equation, the same principle can be applied to a telecommunications company with network analytics to better understand the context of its network to improve customer experience and reduce customer churn.

O2 Czech Republic by the numbers

mobile subscribers

96%

4G LTE coverage

335K

O2-TV subscribers

Jan Romportl Chief Data Scientist

Jan Romportl is chief data scientist at O2 Czech Republic where he helped build the data science team with a strong focus on machine learning from telco big data. He is also involved in the startup scene as a chief science officer. Jan has more than 10 years of academic research and teaching experience in AI, man-machine interaction, speech technologies and philosophy.

Deep learning on telco data.

“We do deep learning on telco data. We are able to cluster customers. We want the corporation to know the customers and for that you need the brain of the corporation.”

Jan Romportl, Chief Data Scientist

O2 Czech has created an advanced neural network, applying deep learning techniques to personalize the customer experience, scaling to 8M mobile subscribers across the Czech Republic and Slovakia.

Every phone emits two types of distinct data patterns which can be used to better understand the context of its users – Cell_IDs and SIM Card_IDs.

Cell_IDs are cell tower IDs and call routes that emit geolocation data which can determine the optimal sequence of calls in an effort to reduce dropped calls, busy signals, and other call quality concerns. These Cell_IDs are dynamic, changing based on a given location.

A SIM Card_ID is unique to the individual SIM card within the device. Think of your SIM Card_ID as a uniquely identifiable feature to you. While Cell_IDs are dynamic, a SIM Card_ID is permanent. You only get one. This ID, when paired with your customer account profile data, depicts your relevant data such as cellphone type, mobile usage, call patterns, gender, demographics, and other account-related data points.

By combining Cell_IDs, SIM Card_IDs, and customer data, O2 Czech gains a more accurate representation of who their customers are and the type of experiences they may receive at certain locations.

Applying a Word2Vec algorithm to cell data (Cell2Vec), O2 Czech visualizes the context of its subscribers related to its network. These visuals depict the Czech Republic, user behaviors, network performance, and more.

Network analytics gives O2 Czech the ability to understand how to optimize its mobile network. By relating cell tower and SIM card data, O2 Czech better predicts customer experiences and paths to churn. This scales personalization of its services to millions of subscribers.

For example, an inner city with a high concentration of tech-savvy individuals who primarily text and stream video may require different network requirements than a low-density rural area where users are older and primarily place voice calls. Matching network optimization with relevant, highly contextual personalized offers drives customer loyalty and increases in O2 Czech Republic’s mobile, fixed, and internet-based offerings.

O2 Czech recently participated in a Teradata Business Value Realization engagement.

Together, O2 Czech and Teradata identified 1.6B Kč ($70M USD) of revenue benefit for their customer experience efforts using Teradata for their data analytics. The four-year, fully realized Teradata solution applies data analytics for customer churn, new customer acquisition, campaign effectiveness, and upselling. In 2019 alone, 654M Kč ($28M USD) in revenues were identified as business value realized from data analytics on Teradata.

Cutting-edge technology for cutting-edge data science

Building cutting-edge deep neural networks requires a platform and experts spanning across multiple technologies in order to perform the data science methodologies needed to scale personalization from thousands to millions. O2 Czech turns to Teradata to make it possible.

Learn more about Vantage

O2 Czech Republic's answer:

O2 Czech Republic