Good morning. This is Noriko Arai speaking from Tokyo. Before starting my talk, let me first apologize for not being able to attend a conference in person, probably next time. Some of you may know my name through the Todai Robert Project,

making an AI, passing the entrance exams of the University of Tokyo, the top university in Japan. Today, I'd like to discuss my another project called Research Map, Japan's national platform for science for science policy, operated by the Japan Science Technology Agency, JST.

So let me share the slides.

Okay. Science, a field marked by costly endeavors and uncertain outcomes, demands significant resources and time before yielding tangible successes.

Historically, developed countries as principal patrons of scientific research have acknowledged its important role in driving progress and innovation. However, recent times have seen these nations grappling with economic stagnation and pressing fiscal challenges,

affecting their capacity to fund scientific endeavors consistently. Japan exemplifies this trend, of course. Known for its technological advancement and a strong emphasis on education and science,

Japan faces economic slowdown and budget constraints. The science of science policy is an emerging interdisciplinary field, aiming to develop theoretical and empirical models to understand the scientific enterprise better.

This research area is vital because it helps frame the value of science in a way that resonates with the public and policymakers. The key here is to show that investment in science is not just a cost, but a crucial step towards long term prosperity.

When the nation channels funds into science, it's essentially using the public's money, right? Making every citizen a stakeholder in scientific progress.

This makes it essential for a government to be as accountable for these investments as companies are to their investors. The challenge lies in effectively communicating why investing in science today

can bring significant benefits even if they are not immediately visible. SUSP, Science for Science Policy, also focuses on the optimization of science investments. Every nation compiles macro data on their science research and development efforts, of course.

This includes financial investments, the volume of papers published in leading journals, and the number of researchers employed by universities. This macro data, of course, is important as it paints a broad picture of countries' commitment to science.

However, it is not enough, I believe. Actual depth of research activities is revealed through micro data. Micro data offers granular insights essential for crafting targeted and effective L&D strategies.

The input to scientific investment can be enumerated as follows. The education received by researchers, of course, their positions at universities, their salaries, research hours,

fundamental research funding, competitive research grants, and, of course, the research environment, and so on. While the output of science is the fruit of scientific investment complex and cannot

be simplistically qualified by measures such as impact factor or the number of patents alone. However, important parameters do include papers and books written by researchers, the patents they secure, the

hours they receive, the education they provide, and the research groups they organize, and so on. Of course, we need to understand the relationship between the input and output as network.

The importance of this micro data cannot be overstated. It forms a complex network that is vast and intricate, reflecting the interconnected nature of scientific research.

But how do we capture this level of detail in a meaningful way? No nation has fully mastered the collection and processing of micro data into a machine readable format that can efficiently handle this volume.

Yet, this is the frontier we must conquer. The introduction of ResearchMap in 2008 marked, yeah, it's already 15 years ago, marked notable shift in how researchers can manage and share their professional accomplishments.

Its integration with established database like PubMed and DBLP through APIs was particular significant advancement. This feature simplified one's tedious task of maintaining a record of one's achievement substantially reduced,

I would say, one tenth the time required for researchers to manage their professional profiles. We designed the ResearchMap not with a top-down manner, but from the bottom up.

It was conceived as a stylish website for researchers to obtain their ideas and express their professional selves, much like a social network service. This approach was key to our success.

In the recent landscape of academia was a significant rise in young researchers holding temporary positions only. ResearchMap has become indispensable tool for them. They have little incentive to enhance universities' websites because these pages are often removed once they move to another institution.

As a result, ResearchMap has become a vital platform for these scholars to create and showcase their accomplishments, bolstering their visibility in the competitive job market.

By 2013, ResearchMap had further evolved, uploading its utility. It began offering an API for institutions, allowing them to access and present detailed profiles of their academic personnel easily, including temporary staff.

This service gave universities a cost-effective way to showcase the breadth and depth of their faculty's expertise and output to the society, too.

This dynamic has led to richer data aggression within ResearchMap, as researchers diligently update their profiles with new publications, grants, and hours to attract future opportunities. For universities, the API access provided by ResearchMap has become a viable asset, enabling

them to maintain an up-to-date overview of the contributions made by their staff. This symbiotic relationship has fostered a win-win scenario.

Researchers are given a stable and enduring platform to highlight their work, while universities benefit from a comprehensive and easily accessible record of their diverse scholarly activities. After launch of ResearchMap, various international organizations began to release APIs for academic information.

By utilizing APIs provided by entities like ORCIP and DOI, ResearchMap has continuously improved year by year.

Researchers no longer needed to manually input data in ResearchMap except for their career history and hours. By 2018, the number of registers on ResearchMap had surpassed 300,000.

Utilizing over 10 million achievement data entries they had registered, we embarked on the development of AI. By 2020, we had successfully integrated AI into ResearchMap.

Harnessing this wealth of information as training data to enhance the platform's capabilities. Before unveiling the pioneering solution we in Japan had forged, let's take a look to grasp the issue's complexity.

Let us consider a scenario to search for the author one way on Google Scholar. It is a perplexing mix of results. The first entry is a collection of poems and the second is a chemical review paper.

Intuitively, we, and also probably machines, can understand these works cannot stem from the same one way. But can't drum deepens when another chemistry paper, the third one, appears, authored, but yet another one way.

One was affiliated with the University of California, another with the University of Illinois. Indicating different individuals behind a familiar name.

This one-way problem with John Smith's problem has been a well-known global challenge in academic attribution. And it is still hard to solve fully automatically with machines. That's the point. Just like we need to identify researchers' accuracy accurately, we also face a similar challenge with their works.

Take, for instance, two of my papers, Tractability of Cat-free Genshin-type Propositional Calculus with Permutation Influence and its sequel,

Tractability of Cat-free Genshin-type Propositional Calculus with Permutation Influence II, both published in the Journal of Theoretical Computer Science. To the human eye, especially to us experts, these papers clearly distinct continuation of one another however.

The difference is not obvious to computer scanning through the titles. The vector similarity in titles and the subject matter might lead a machine to consider them identical. And this isn't an isolated example.

Consider how many important articles a title just preface or introduction. So that is why the unique identifying system distinguish these works becomes manageable. So that is why every paper, just like every book has an ISBN, should have a unique identifier like DOI.

But it is not enough. Because DOI is not assigned to research funding provided by each country, each institution, which is an essential input for source,

making it impossible to construct a network with input and output. To truly follow the science of science policy, we must weave together a comprehensive tapestry of the researcher's journey. We can craft a microdata-driven source that captures the essence of

the scientific pursuit only by integrating this full spectrum of input and output. ResearchMap fills this gap by providing researchers with functionality to interlink microdata within the platform.

For the first time, this enables the feasibility of science of science policy based on microdata, something we are confident in achieving. Let me take a moment to explain the AI assistance system in ResearchMap on this slide.

It is not fully automatic. It is assistance in the AI and human collaboration. To begin your journey with ResearchMap, you will start by securing your researcher's ID from Ministry of Education, Science and Technology of Japan.

Admittedly, this initial step might seem tedious, but it is not as bad as any initial step signing up for social media like Facebook.

Your registration requires inputting your official name, in kanji, probably, and any alphabetical, ripple-z dishes that might appear in your published works. You will detail your affiliations, the roles you hold within each, and the proportion of your efforts dedicated to these positions.

Personal details such as birthdate, gender, and nationality also play a part in creating a holistic profile of researchers. What comes next is the showcase of ResearchMap's ingenuity.

The AI assistance steps in. This AI shifts through the extensive data network within ResearchMap to suggest possible co-authors you've previously collaborated with. Engaging with this step is very critical. By confirming or denying these suggestions, you are training

the AI, enhancing its ability to accurately identify your contribution to papers, patents, books, and much more. The interactive process doesn't end here. As you affirm your co-authors, the

AI extends its suggestions to your work, offering potential matches for you to verify. This collaborative dance with AI fine-tunes the system with each interaction. The more

you engage, the sharper the AI becomes, rewarding your diligence with ever-increasing precision rate. Your task is to simply confirm the correct associations and dismiss inaccuracies. With

a simple yes or no, you are crafting a network of your professional footprint. The impact of this partnership with AI and researchers is profound. The accuracy of the

AI's recommendations, the precision, recall, and F1 score have surpassed 99% in a year. It's a testament to the power of cooperative human-AI interaction, and it's reshaping how we catalog and recognize academic

efforts. Imagine a system that grows smarter with each contribution you make. That is a model we have achieved with ResearchMap. With the integration of AI into ResearchMap in February 2020, we witnessed a remarkable surge in the platform's utility.

Astonishingly, the number of papers registered on ResearchMap doubled in the year following the year's introduction. This increase

isn't a reflection of Japanese researchers working twice as hard, especially during the challenging times of COVID-19. But rather, it's a testament to the hard work of AI working behind the scenes.

The AI's contribution has been invaluable, sifting through data, identifying connections, and populating records, thereby liberating researchers from the clerical burdens of data entry. The academic community has welcomed this advancement, and the sense of satisfaction amongst researchers is palpable.

Today, I'd like to give you a glimpse into my own ResearchMap profile. The sheer volume and complexity of the data it holds are far too expansive to be captured on a single slide.

For those curious to see the full extent of its capabilities, please visit ResearchMap directly and explore. My commitment to enhancing ResearchMap is relentless. I am constantly fine-tuning it, adding new features

that bring joy and ease to its researchers, and sprinkling the elements designed to enrich the data. Much like a shop perfecting their signature dishes. So please keep your eyes on our advances.

Thank you very much for your kind attention.