Artificial intelligence (AI) speakers, autonomous vehicles, humanoid robots and unmanned stores tell us that AI is already serving a role in many aspects of our lives.
AI speakers based on speech recognition technology has been popularized to the extent that many tasks in demand at home can be processed with a simple voice command, connected with video or music channels, cable TVs, food delivery or online shopping services. The same is happening for autonomous vehicles. Autonomous vehicles for research run regularly in restricted environments like universities and research institutes, driver assistance AI technology detecting lane changes and obstacles are being developed by a number of car manufacturers, and autonomous driving technology with no driver intervention is waiting for commercialization.
Adopting machine learning in mobile services is not that easy
Just like manufacturing and other traditional industries dream of a technology based innovation, mobile services strive to become smart too.
The field that most mobile services adopt machine learning at first is recommendation algorithm. The algorithm acts based on the assumption that user groups with similarity will also show a similar purchasing pattern. Contrary to the businesses’ beliefs, however, the field does not give an instant boost to customer satisfaction or business results, as recommendation algorithm is a somewhat general technology with a wide range of application. Therefore it is better to focus on the imminent issues and pursue a practical solution, instead of trying to take care of the big things.
A Practical way using machine learning algorithms – Customer services
We take increasing operation efficiency of a customer service center with the new technology for example.
In the Indian market, we have to bear with the specificity of processing a variety of languages simultaneously, including 7 of the official languages. As customer base grows, the number of inquries and complaints the customer care center receives increases. If we assign these tasks to CS staffs, we must recruit additional staff for additional customers. This translates into a linear increase in operation cost with respect to the increase in customers. Moreover, managing CS staff training, overall service quality, response time and customer satisfaction is something more than a matter of cost.
The following processes can be applied to streamline CS work and manage quality.
Customers’ voices sent through CS channels are classified by the computer, based on the rules it picked up by itself. Natural language processing is applied in data pre-processing where the language characteristics are important, and converting the language into a type of signal or network makes it possible to process different languages collectively. When a simple response is available for the classified voice, the computer sends an immediate response based on a pre-set pattern. This is a technology nearly identical to those of chatbots or robot writers producing press release data. In this way, about 70% of the CS can be processed with a computer.
It is also crucial to monitor if a new group can be found in the voices received. The computer regularly modifies and updates the classification system, and forming new categories not found in previous classifications from customers’ complaints and inquiries ensures a swift adaptation to user response. This is a great advancement from the traditional rule based approach of call centers with CS staff, which have a hard time detecting the customer response because new types of issue are just classified into ‘etc.’.
Another application enhancing operational efficiency – the mobile security
For mobile services, users’ fraud or abusing is always a major issue. Such users intercept the benefits provided to users for example in a marketing promotion, compromising service integrity and giving negative experiences to normal users. Companies frequently suffer losses as cost because of them.
The fradulent users usually switch to new tactics when caught, so supervised learning, in which the classification is based on pre-defined properties or patterns, is inadequate. Traditional human-powered analysis methods also have its limits because their terminal or user information is altered or forged unexpectedly.
Outlier analysis is commonly used in this case to scan for suspicious behaviors or patterns. Sometimes, a network analysis between terminals or accounts showing suspicious patterns reveal groups of fraudsters who have been forging information to avoid getting caught.
Authentication technology confirms the legitimacy of terminal users using information about mobile terminal usage pattern, call patterns, location data and social relationships, and it has been developed and applied to actual services in response to various needs.
Data infra is an essential prerequisite for machine learning systems
A well-established data infrastructure is necessary in order to apply a machine learning-based algorithm to the services. The result of the learning depends on the type, quantity and quality of data to be learned.
We first need a data loader to confirm the source of usable data, collect them at regular intervals, pre-process to a optimal form for learning, then finally load the data. We should also get an algorithm library to store the learning algorithm of the computer and the revisions by version. The last ingredient is a feedback process to apply the learning results to the actual service, and record and reflect user responses recursively.
With a user count of 1.2 billion, the Indian communications market is also a major challenge in data analysis. Counting the digits of number of active users amounts to eight or nine, and the amount of collected data for a basic level of analysis needed for mobile app service operation is out of the operation limits for a typical system. For this reason, a good data analysis in the Indian market requires not only the capacity to analyze, but also the capacity to manage data infrastructure.
In summary, I recommend a small and practical field when you apply machine learning algorithm to a mobile service for the first time. The process of solving a small problem using new technologies will teach you how to use those technologies. We should also note that the quantity and quality of data and its update frequency is at the very core of machine learning, and a data infrastructure should be built to facilitate the process.
It is difficult in this era for businesses to expand without using data and machine learning to analyze patterns and networks. Machine learning and artificial intelligence is not a futuristic science fiction anymore.