JFSA Project Uses Artificial Intelligence to Examine Annual Reports

The following are the key insights from the Proof-of-concept:

• The results suggest that textual information could be widely analyzed precisely and in large quantities by artificial intelligence due to the progress of the natural language processing technology.
• It is seen in some approaches that artificial intelligence understands the sentence structure and context and tries to analyze sentences in a more human-like way. If there was sufficient time and more data to be analyzed, artificial intelligence would be able to work much better. However, since the criterion of artificial intelligence tends to be a black box, it is important to consider how to use it.
• Many companies emphasized that it was vital to develop artificial intelligence jointly with the person who judged the evaluations of descriptions. As humans and artificial intelligence have their own areas of expertise, it is important for humans and artificial intelligence to play their own roles to achieve better performance. Humans should provide their judgment and knowledge to artificial intelligence as well as interpretation of the results and feedback.

The PoC project team, with the cooperation of 20 corporate participants from the private sector, conducted a series of PoC tests by taking a variety of approaches, such as programming, machine learning, and deep learning. The test ended on June 30, 2019. After that, the PoC project team compiled test results, including respective test methods, new insights, and future possibilities.

Keywords: Asia Pacific, Japan, Banking, Annual Report, Artificial Intelligence, NLP, Fintech, Proof of Concept, JFSA