Artificial Intelligence Technology Development and Its Practical Use at NTT DATA

2.1 Reduced burden of credit card member store management operations

The use of examination support technology began two years ago, and results have already been obtained on six cases.

The credit card industry is one area where results have been obtained. This industry has seen an increase in the number of scenarios in which consumers use their credit cards with the rapid growth of e-commerce and the diversification of payment methods such as payment via smartphone. As a result, there is strong demand to develop secure environments where customers feel at ease when using their credit cards, and therefore, credit card companies are working urgently to increase the security and efficiency of managing a wide variety of retail stores (member stores) that accept credit card payments.

As part of its efforts towards increasing the security and efficiency of examination operations, NTT DATA began working with Sumitomo Mitsui Card Co., Ltd. in October 2013 to jointly initiate verification tests on the efficiency and automation of information gathering for member store examination, and in October 2014, these verification tests were implemented in Sumitomo Mitsui Card’s member store management operations. Comprehensive decisions are made after referring to and analyzing a variety of information concerning member store examination operations, but particularly in cases when examinations are conducted on non-face-to-face merchants such as e-commerce and smartphone stores.

For this reason, it is important to determine how precisely and effectively this diverse information can be analyzed within member store examination operations. Examination support technology extracts and collects information necessary for examinations from a vast amount of information existing on the Internet such as reputation information from forums, content on e-commerce sites, and third-party information on the Internet.

A dashboard function has also been implemented that displays gathered information on one screen in an easy-to-understand fashion and increases the efficiency of examination operations such as automatically highlighting and displaying areas that need attention. The AI Solution Promotion Office is focusing on R&D of knowledge acquisition technology that specifies which areas are important in examination operations by adding meanings to text information. In the example of credit card member store examinations, this technology is being utilized to not only specify corporate names, addresses, and hours of operation from e-commerce sites, but also to automatically gather the price range and price ceiling of products that those corporations sell. In addition to performing these tasks, this technology also gathers information on the reputation of the relevant companies and checks to see if there are any entries on social networking sites regarding those companies.

2.2 Semi-automatic generation of rules through the use of machine learning

Knowledge acquisition technology is a technology that estimates the structure of semi-structured and unstructured data such as sentences and lists. This is a combination of original structuring technology in line with the data format of extracted targets. To utilize this technology, rules are necessary to extract objective information, and to generate these rules, there is a need for not only business knowledge but also specialized knowledge on information extraction technology^*1, which is why specialists in the AI Solution Promotion Office have until now been receiving requests and performing operations manually. Consequently, the considerable cost and time required for this have been issues until now.

However, R&D is currently underway on technologies that semi-automatically create rules through machine learning by leveraging knowledge on rule generation that has been conducted manually up until now. For example, when creating rules that specify place names from text information, a person first confirms whether the candidate place name (written in red characters) on a screen listing example sentences is correct or not by selecting either “o” or “x” (Fig. 1). Next, the system uses machine learning based on this content to automatically generate new rules. Repeating this step increases the accuracy of the rules. This mechanism for generating rules and technology that increases the efficiency of examinations is collectively called sensu (tentative name), and verification tests on and application of sensu are currently progressing.

Fig. 1. Rules can be generated through machine learning by performing simple tasks.

3.1 Provision of advanced services at low cost

Since data collected from robots and sensor devices are handled in an integrated manner, Cloud Robotics Platform utilizes the R-env^TM technology developed by NTT Service Evolution Laboratories to control the integration of cloud support devices as well as the know-how accumulated by NTT DATA on M2M (machine to machine) and IoT (Internet of Things) (Fig. 2).

Fig. 2. Cloud Robotics Platform.

NTT DATA is already conducting verification tests that utilize sensors and Cloud Robotics Platform. Two examples are verification tests of elderly support services in nursing homes and verification tests of visitor survey collected in public facilities. These tests use Cloud Robotics Platform and also leverage technologies such as the communication robot Sota^TM*3 manufactured by Vstone Co., Ltd. and technology for speech recognition, dialogue control, and speech synthesis developed by NTT Media Intelligence Laboratories.

3.2 Support of guidance-related operations of bank clerks using sensors and robots

The number of overseas visitors to Japan is expected to increase in 2020, and efforts are therefore increasing to develop contact desk services utilizing communication robots such as those that support customers in multiple languages at reception desks in various industries. Accordingly, NTT DATA has been paying attention to customer support-related operations in financial institutions where there are great expectations for use scenarios involving communication robots. NTT DATA and Vstone have been conducting joint verification tests towards realizing customer support using communication robots in actual branches of financial institutions.

The feasibility of providing customer support services via dialogues with communication robots was evaluated in verification tests conducted at the RESONA Bank Toyosu Branch (Seven Days Plaza Toyosu), a newly established branch, from the day of opening on November 15, 2015, until the end of April 2016. During these verifications, we extracted issues in business and operations while listening to the opinions of visitors and bank clerks. For example, high-sensitivity sensors installed in the ceiling above the building entrance on the first floor would detect people entering and leaving the facility and notify the communication robot stationed at the ATM (automated teller machine) corner. Along with communicating with customers by saying “Welcome!” or “Thank you for visiting us,” the robot directed visitors to the branch services located on the second floor. High-sensitivity sensors installed at the entrance of the branch notified the communication robot placed at the reception desk when visitors came close. The robot would then start a dialogue with the visitor by greeting them with “Welcome!” and then prompt the visitor to use the self-reception tablet according to the contents of the robot’s conversation with the visitor. This endeavor achieved services that supported the guidance-related operations conventionally conducted by bank clerks (Fig. 3).

Fig. 3. Branch reception desk robot.

An increasing number of inquiries is being made regarding the utilization of communication robots in customer support-related operations. NTT DATA will continue conducting verification tests and is working to expand the range of support that communication robots provide in customer support-related operations such as customer guidance and product introductions in stores by utilizing AI technology. NTT DATA also intends to promote the early practical use of communication robots.

4.1 Large-scale traffic simulation technology based on parallel distributed processing

NTT DATA’s traffic simulation adopts a method using a multi-agent model. In this method, targets that influence cars, stoplights, roads, intersections, and other traffic elements are regarded as agents, and this method reproduces actual traffic congestion situations by operating an enormous number of these agents on a virtual network of roads generated on computers. However, conventional measures require high-spec computer resources to simulate situations where a high density of cars causes congestion in a broad area.

NTT DATA has successfully realized distributed processing by separating an original map into a mesh format and by using multiple general-purpose servers (Fig. 4). In addition, since one mesh is assigned to one core of a server and then calculated, distributed processing requires mesh partitions with equal computational complexity. The NTT Software Innovation Center has developed graph data analysis processing technology called equal granularity clustering (Fig. 5) with which NTT DATA has succeeded in conducting fast mesh partitions and conducting traffic simulation in real time.

Fig. 4. Image of the use of distributed processing.

Fig. 5. Illustration of the application of equal granularity clustering technology.

4.2 Mitigation of congestion by optimizing traffic congestion in predictions and signal controls

To verify the effectiveness of traffic control using this traffic simulation technology, NTT DATA cooperated with Jilin City in China in November and December 2014 in conducting verification tests in a city central area. During these tests, Jilin City buses equipped with on-board computer terminals were used to conduct traffic congestion prediction and stoplight control simulations in combination with statistical information such as traffic volume, routes, and information from approximately 200 probe units (the location and speed, etc., of each bus) concerning the eight routes gathered from those on-board computer terminals. Traffic was controlled by reflecting stoplight settings—which were optimized through prior assessments done in simulations—in intersection stoplight machinery found within the city, and the effects of traffic congestion mitigation and the improvement of bus service times was verified. As a result, the average speed of vehicles moving within the target area was increased, traffic congestion was reduced, and it was confirmed that the average bus service time for target bus routes was improved on average by 7% and up to 27%.

In 2015, NTT DATA cooperated with Guiyang City in China to develop technology that successfully mitigates traffic congestion by performing traffic congestion predictions and stoplight control simulations based on the data analysis results of cameras installed in Guiyang City used for managing traffic. This effort was done in collaboration with the Institute of Software, Chinese Academy of Sciences and was conducted as a horizontal deployment of know-how obtained from the verification tests conducted in Jilin City.

Initiatives promoting the practical use of traffic congestion mitigation solutions in combination with traffic simulation and stoplight control technology are progressing within NTT DATA in an effort towards successfully managing major events in 2020 and realizing smart cities in countries around the world.