Research activities
Contents of research activities
The goal for our research undertakings is that of “creating a comprehensive research center that aims to deepen data science for business.” For this, we apply the data science techniques that have been cultivated thus far to a variety of business sectors, and build a research structure for comprehensively and seamlessly carrying out a series of processes that includes developing basic techniques and applications for various business sectors, developing models, and verifying these by putting them into practice. We have also created a research center that is one of the best in the world.
Specifically, we use basic techniques related to various types of data like those below in working to implement processes in application areas.
Streaming Data Mining Specialized to Customer Paths
Clarifying the series of processes by which consumers purchase products and which lead to sales is crucial for managers in order for them to consider optimal sales strategies. We strive to elucidate consumer behavior for customers in stores by analyzing not only sales history data such as POS data, but also customer path data obtained from RFID technology. Combining these two types of data lets us understand which customers bought how many of which products and the sorts of routes they took through the store. To give one example, it lets us understand that in retail stores like supermarkets, most customers trace a route along the outer perimeter of the store. Expressing and analyzing this as time-series data lets us see which areas customers stayed in and for how long, and in what sequence they purchased goods.
For example, what would happen if we were to visualize the purchase routes of every customer in a single day? Such a perspective is necessary for analyzing consumer behavior over a single day. Therefore, by taking the general concept of customer paths one step further and calculating the probability densities for where customers will be exist within a store, we are advancing research to find places that customers tend to visit. By comparing and analyzing the probability of customer existence and sales history data for each area, we can link the characteristic of customer path data with purchase information.
The above is an example of the research on customer path data that DS Lab performs, but we are also advancing various different types of research on modeling consumer behavior by means of customer flow estimates, licensing, and hierarchical Bayesian models. In tackling our daily research, we do not just focus on understanding consumer behavior through our research on customer path data, as we also focus on verifying the utility of our approach in actual stores and on giving back to society.
Understanding Consumer Behavior Using Eye Tracking and Creation of New Sales Department
Recently, we has been able to obtain visual attention data of customer within stores by using eye-tracking technology, in addition to customer shopping path data with RFID technology. Conventionally, such studies have primarily consisted of laboratory experiments and experiments that fixed the subject’s face in place, so analyses using eye-tracking were carried out for online shopping sites and online advertisements. In response to this, the development of eyeglass-style recording devices has recently made it possible to perform field experiments in actual stores. This has allowed us to clarify behavior related to decision-making at sales departments, such as whether or not they pay attention to products and advertisements, the number of times they look at them, the time they spend looking, and the order in which they look at them.
DS Lab has created a mechanism for acquiring not only the POS data with IDs and customer shopping path data within stores that has traditionally been accumulated, but also visual attention data. The transaction data and shopping path data have made it possible to identify which customers bought how many of which products, when, and the sorts of route they took through the store. Using eye tracking here allows us to clarify questions like what interests and excites customers and to what extent product placements and advertisements entice customers based on how people look around. However, since not much mining using conventional customer visual attention data has been carried out, numerous challenges for this remain, such as developing analytical techniques and generalizing indicators.
DS Lab applies the analytical techniques and data science techniques that have been cultivated to date in an attempt to quantify the way consumers view points of sale from a variety of different angles. Therefore, we will clarify the causal relationship between quantified viewpoints on sales departments and consumers’ purchasing behavior in a scientific manner. We aspire to make it possible to create effective sales departments in the future based on new knowledge by establishing this sort of basic theory.
