Dialogue with the corporate strategy executives of Osaka Gas USA Corporation: Using AI to understand the momentum of decarbonisation — the potential of AI-driven management (Part 2 of 2)

Public opinion can be quantified.

Edamoto:
Now let's talk about the dashboard design and implementation phase. Our goal was to collect, analyse and visualise data on the momentum of decarbonisation in the world in a short period of 11 weeks. Generally speaking, quantifying unstructured data, such as data on social media, and identifying indicators of momentum in numerical terms is very challenging. While it makes sense that if text on social media can be quantified, trends can be identified, I think you had some concerns and worries about whether it was really possible to analyse such vast amounts of information.

Hayashi:
You’re right. The question of whether unstructured data could really be quantitatively evaluated was our main interest, and at the same time, our main concern. We knew that it was necessary to understand changes in the attitudes of the younger generations as social trends, but when Ms. Edamoto and the other members of the team suggested that we should try to quantify these changes, we honestly wondered whether it would be an expedient way of acquiring and analysing the data and whether the data contents might not be arbitrary. However, we decided to take on the challenge.

Miyoshi:
Using conventional operational methods such as human visual observation of social media and word-of-mouth information to understand public opinion inevitably results in localised sampling. It’s also undeniable that the specific data that gets picked up is influenced by the individual person in charge. The good thing about AI is that it can comprehensively and objectively categorise topics and their composition ratios. Even so, due attention should be paid to the population of social media users as a data source and the keywords used to extract the comments to be analysed.

Edamoto:
In our project, we spent a lot of time determining the keywords to be monitored when analysing posts on social media.

Kagawa:
That’s right. We considered various keywords related to environmental issues and climate change, such as ‘fossil fuels’, ‘Paris Agreement’ and ‘decarbonisation’, but in the end we chose the hashtag #GreenNewDeal, because the Green New Deal is an environmental policy of great interest in the US. We thought that we could get a lot of data on the number of people posting their opinions on social media, as well as comments for and against the policy. One of the objectives of this project was to first experience AI analysis, so we chose the keyword without thinking too deeply.

Edamoto:
This was a kind of trial, and it was very important to have a ‘just do it’ attitude. There were approximately 175,000 posts using the hashtag #GreenNew Deal from 2015 to 2019 in the US alone. Mr. Miyoshi took the lead in quantifying this huge amount of information.

Miyoshi:
Yes. We used two approaches to natural language processing in this project. One was sentiment analysis to determine whether each statement was positive or negative, using VADER, a specialised algorithm for social media in English, and the other was the development of a categorisation method from the perspective of STEEP*, which is unique to our project. The statements about the Green New Deal were diverse and included topics related to politics, human rights, well-being, social infrastructure, the environment and many others. To classify these correctly, we needed an understanding of US society and industry knowledge in addition to technology, so the algorithm is full of our expertise.
*Society (including culture and lifestyle), technology, economics, environment and politics

Edamoto:
Mr. Tsuzaki, you’ve studied physical statistics and have a deep knowledge of technology. What did you think when you saw the results of the natural language processing?

Tsuzaki:
My own past experience was limited to analysing quantified structured data, so I found this method of analysing unstructured data fresh and new. In the US, we frequently see examples of instant analysis of voice and chat utterances in various aspects of daily life for use in various services. Therefore, I had the impression that advanced natural language processing technology had been established. I was genuinely impressed that such technology had reached a level where it could be applied to the analysis of people's emotions on social media.

Edamoto:
In the US, we see some trends on social media that are not so common in Japan. People of all generations tend to express their emotions in a straightforward manner; for example, candidates’ comments in televised presidential debates ‘go viral’, announcements of events organised by environmental groups spread around the world, and users who have installed solar power systems post the benefits of it on the Internet every day. Images and videos are often posted without text, and evoke our emotions in a variety of ways.

Miyoshi:
Yes. And in response to such situations, analysis methods are also becoming more diverse. For example, image analysis enables us not only to identify the subject of images (dogs, cats, etc.), but also to determine the type of setting and estimate attributes (such as gender or age) and emotions from people's faces. For voices, technology is also being developed to not only transcribe voice data into text, but also to estimate the speaker’s emotions based on voice colour and wavelength. Although only textual data was analysed in this project, we hope to make further use of images and video in the future to better understand the opinions of the world.

Building a corporate culture of bold and enterprising data analysis to interpret social changes

Edamoto:
So, what are your final thoughts looking back on the project?

Hayashi:
Looking at the dashboard we used to visualise social momentum regarding decarbonisation, we felt that it could contribute to the objective of ‘keeping up with social change’, which was the starting point of our efforts. We also received positive feedback from the company president. It would be nice if we could target specific keywords that could increase the momentum of decarbonisation, such as ‘methanation’, ‘carbon dioxide capture, utilisation and storage’ (CCUS), ‘hydrogen’ and ‘renewables’ and use them to identify business opportunities, but through this project, we have realised that the right answer cannot be found directly and that we need to be aware of a broad range of latent possibilities.

    We were also able to experience the effectiveness of the latest technology. In the future, we believe it will be important for our organisation as a whole to foresee signs of change by increasing accuracy, trying new approaches and inviting unique ideas through discussions within the company.

Kagawa:
The design and implementation of a dashboard was a pilot initiative carried out under time and budget constraints, so I feel that there is still room for further ingenuity and effort in terms of scrutinising the analysis targets and the scope of data to be correlated. I believe that people’s attitudes towards environmental policy have changed dramatically around the world during the one year we have been working on the project. The social environment and awareness seem to have changed to such an extent that now we cannot go one day without hearing about decarbonisation or climate change issues. As our company president said, ‘The world is dynamically changing and what is considered good now may soon become obsolete.’ With this statement in mind, we would like to be even more vigilant.

Tsuzaki:
After participating in this project, I realised that data analysis methods have developed dramatically. If machines could take on the role of supplementing the analysis and scenario studies conducted by humans, I think even more effective approaches would be possible. I think this technology can be applied in a wide range of fields, so I hope we can consider using it in other businesses within the group. I believe that we can get closer to a better vision of the future by using AI and other digital technologies.

Miyoshi:
AI and big data analysis methods and business applications are evolving every day, and while we have only covered the basics of social media analysis, including the extraction of trends over time from people’s actual statements, we can also extract highly suggestive insights by using applied analysis that focuses more on users. For example, by analysing the poster’s own attributes, hobbies and preferences, trends in follows and followers, detailed inflow routes and events that trigger emotional shifts (positive or negative), I believe it will be possible not only to understand decarbonisation momentum, but also to achieve more in-depth prediction of future signs from a human perspective.

AI is not only good at social media analysis, but also at combining different data sources and data types to discover multifaceted relationships. Therefore, by accumulating large amounts of data and improving analysis models, it should be possible to improve the accuracy and speed of decision-making. PwC Consulting’s Data & Analytics team will continue to provide support for business issues in a variety of areas, including the collection of necessary data, the design and implementation of appropriate AI utilisation approaches, the development of IT infrastructure such as data platforms that support these approaches, and organisational and human resource development.

Edamoto:
There is no indication that COVID-19 will be contained, and the future of the business environment is likely to remain unclear and uncertain. However, we believe that continuing and developing initiatives such as this project will bring about behavioural change in businesses and will be positive for future operations. What kind of future awaits us once the cycle of accumulating and analysing fresh information by using flexible thinking has become established in society? I’m really looking forward to seeing it. Thank you very much for this valuable dialogue.

Click here for Part 1