Get inside the DOCOMO integrated data analysis platform
Manager of Big Data Group at NTT DOCOMO, Hiroyuki Ishii gives us a glimpse into his cloud-based DWH driving business value creation from raw data.
After receiving his B.S. and M.S. degrees from the University of Tokyo, Hiroyuki spent the bulk of his career working on HSDPA and LTE commercial services including 3GPP standardisation as a Delegate of 3GPP RAN4. More recently he was assigned to DOCOMO Innovations in California to work as a project manager in the Big Data Group with the Service Innovation Department.
We sat down with Hiroyuki to gain some insight into his integrated data analysis platform which delve into big data analysis and soon to become 5G network.
What is the DOCOMO integrated data analysis platform?
We are using cloud-based DWH. More specifically speaking, we are using Amazon Redshift.
Some features are shown below:
- We are using two sets of “ds2.8xlarge * 125”. The total data size is 4 PB.
- 10 TB data is coming to our platform per day.
- The number of users, who analyse the data using the platform, is more than 1000.
- We develop and operate the platform by ourselves.
Tell us your thoughts on Amazon's AWS Redshfift
Our views on using AWS redshift are as follows:
- Easy setup
- Functionality (at the beginning)
- Early adapters burden (at the beginning)
How do you drive business value creation from DOCOMO integrated data analysis platform?
We have all kinds of data, such as network data, device data, and service data. And there are many kinds of problems to be solved, and how to derive the business value creation would change problem-by-problem.
What is Service-quality-based silent failure detection?
In NW operations, the performance is sometimes degraded without any alarms. This is a so-called silent failure problem. The customers are unsatisfied with the quality of services, but the service operators cannot notice it.
Especially, in case that the throughput performance is slightly degraded, it would be difficult to distinguish bad cells from good cells because the throughput behaviour is inherently dynamic.
How did you solve this problem?
What we did to solve the problem is as follows:
- We gathered various traffic data such as data rate, call drop rate, UL interference, and so on. The total number of data is 900.
- We did a binary classification, in which these traffic data were used as Explanatory Variables and the results of actual failure were used as Objective Variable.
- Then, we achieved very good detection performance, such as 98% detection rate and 0.03% false alarm rate.
How can we use AI to forecast taxi demand?
We forecasted the taxi demand using the population statistics, the taxi data, such as location data and status, and the weather data. The population statistics were derived from the mobile network.
We used two approaches, such as multivariable auto-regression model and deep learning. In high demand area, the multivariable auto-regression model was better than the deep learning approach, but in medium demand area, the deep learning approach was better than the multivariable auto-regression model. That is why we combined the two approaches.
NTT DOCOMO, Inc. is the predominant mobile phone operator in Japan. The name is officially an abbreviation of the phrase, "do communications over the mobile network", and is also from a compound word dokomo, meaning "everywhere" in Japanese. Wikipedia
Explain how banner ads are served using bandit model?
We optimised the banner ads using multi-armed bandit model. The bandit model decides the sequence of arm pulls to maximise the sum of rewards. So, in this case, the arms correspond to the advertisements, and the rewards correspond to the clicks.
Will 5G bring better data analytics?
The answer is YES, because more and more data will be obtained in 5G NW and the services using 5G NW. We cannot perfectly understand what 5G is though.
Where is data analytics moving over the next few years?
To be honest, I am not sure. In any case, it would be definitely true that data analytics would be more and more important.
What can the audience take away from your presentation later this month?
We would like to discuss with other mobile operators what mobile operators should do in terms of data analytics now and in the near future.
Speaker at Telco Data Analytics Europe
Hiroyuki Ishii will be arriving in Madrid later this month to share more wisdom around Business Value Creation with an Integrated Data Analysis Platform.
He will be covering:
- Service-quality-based silent failure detection
- Business value creation from DOCOMO integrated data analysis platform
- Forecasting taxis demand with AI
- Serving banner ads using bandit model