NEC Develops AI-based Customer Profile Estimation Technology

NEC
Develops AI-based Customer Profile Estimation Technology

Tokyo, September 2,
2016 – NEC Corporation (NEC; TSE: 6701) today
announced the development of a Customer Profile Estimation Technology that
automatically estimates detailed customer profiles on individual customers,
including their interests and preferences to a high level of precision, without
the involvement of marketing experts.

Based on NEC’s unique
relationship mining technologies, the newly-developed technology estimates
detailed and hard-to-obtain profiles on individual customers completely
automatically and to a high degree of precision from basic profile information
that is relatively easy to obtain, such as age and gender, combined with
purchase histories.

NEC has verified the
effectiveness of the technology using public data. As a result, NEC has
confirmed that an analysis that would take conventional experts three months to
complete can be carried out in three days, and to a degree of precision
that surpasses those experts (*).

“With this technology, we
can respond to lifestyles that change from moment to moment, quickly discover
true individualized needs that may have been overlooked, and devise appropriate
measures. In addition to estimating detailed customer profiles, the technology
can also be applied to estimating product attributes,” said Akio Yamada,
General Manager, Data Science Research Laboratories, NEC Corporation.

Moving forward, NEC will
continue to pursue research and development of the technology, aiming to provide
it in the retail and distribution sectors, including department stores,
supermarkets, convenience stores, e-commerce sites and point card systems.

Main features of the new
technology include the following:

1. Automatically
estimates detailed profiles based on basic profile data and purchase histories

Through NEC’s unique
relationship mining technologies, detailed profile information for each
customer can be estimated by simply inputting their basic profile information
and purchase histories. As the fully automated process completely eliminates
tasks such as manual product labeling, detailed profiles on each individual
customer can be estimated in a short time, thereby cutting three months down to
three days (*).

2. A process of
hypothesis generation and verification is repeated to achieve an estimation
precision that surpasses an expert

An artificial
intelligence (AI) component references product purchase histories to produce a
hypothetical detailed profile for each customer. This is interconnected with
another AI that references questionnaire results provided by a subset of
customers to verify the accuracy of the hypotheses. This cycle of hypothesis
generation, verification and feedback is repeated. The use of AI allows
objective detailed profile estimation untainted by the subjective views of an
analyst. By repeating the cycle many times at high speed, the estimation of
detailed customer profiles to a degree of precision that surpasses the analysis
results of an expert is achieved.

***

Note:

* Tested with a public dataset from a movie review website.
Estimated profiles (movie preferences) of each user from thousands of basic
profiles (age and gender) and 1 million movie ratings.