Data extension using machine learning and generative AI

I originally studied geometry in university. In my final year of the doctorate program, I had a hard time deciding whether or not to pursue a career in the academia. Eventually, I decided to seek practical application for research results and chose a career path in corporate research. The reason for choosing NEC was its presence in the academia. The deciding factor was that a large number of NEC’s research papers was accepted at reputed international conferences.

My current major research theme is “to solve issues in cases where data is limited.” When analyzing customer’s data, it is not uncommon that the data volume provided to NEC is small due to legal restrictions and/or corporate confidentiality. In other cases, the data may simply be insufficient. The principle of this research is to extend data using machine learning and other means to make meaningful analysis possible even in such cases. At NEC, this is established as a platform and provided as a service called NEC Data Enrichment. I worked on the research and development of the “consumer attributes extension” technology as a data extension technology more specialized in marketing. 

I also started using generative AI in addition to machine learning. While recommender systems conventionally required several weeks of data learning to produce precision reasoning, now with the application of generative AI, they can provide high-accuracy reasoning without using learning data, which I find intriguing. Another advantage is that I can present the minimum viable product to a customer and get their reaction on the spot.

Research that captures “who and how to please”

The other appeal of working at NEC is that you can work in various areas. My division primarily handles data analyses, but the range of analyses is diverse, covering audio, healthcare, intellectual properties, robots, traffic, and lots more. We also conduct research that supports these analyses. Even just me alone, I have been involved in businesses such as finance, retailing, and government services. Another key point is that there are many people who would support you with these businesses. There are people with deep expert knowledge of the industry that you want to know about in the business units. NEC is a company that has been active in business across a broad range of industries for many years. Without the need to recruit new people, NEC already has necessary people along with an accumulation of know-how that can be mobilized to provide services to a certain industry or company. This is very reassuring.

If I were to give a single piece of advice to students who are considering doing research at a company, I would say “think who and how to please.” In university, you may think submitting a paper can please a certain professor. This is also true for companies. It is important to think about who and how to please when conducting R&D of technologies and services. What is different from universities is that there is profit involved in business. You must make something that pleases customers to the point that they want to buy it. But this is also the best part of what makes corporate research exciting.

It’s the same when you submit your research paper to the academia. In 2024, my research papers were accepted at RecSys and ICDE, known as the top-class international conferences. During the peer review, some people did not give me a positive feedback, saying that the accuracy of my technology was poor. In the end, other reviewers recognized my work and thus accepted my research paper. After the publication, many corporate researchers mentioned that this technology “successfully optimizes costs without compromising accuracy.” Academic standards can sometimes be different from what corporate researchers look for. I would like to continue doing research from a perspective that there are people needing the technology or thinking that it is interesting―even when the technology does not conform with conventional standards.