Please enable JavaScript in your browser.

fltech - 富士通研究所の技術ブログ

富士通研究所の研究員がさまざまなテーマで語る技術ブログ

Experience ‘Trustable Internet’ a technology for fighting disinformation

Hello. I'm Kouichi from the Data & Security Research Laboratory. I'm writing to introduce a new demo in which users can experience our anti-disinformation technology powered by Trustable Internet, which was published on the Fujitsu Research Portal in December. The Fujitsu Research Portal, which enables the immediate free trial of Fujitsu's advanced technologies in a variety of formats, is available to the global public at the following URL.

https://en-portal.research.global.fujitsu.com/

The concept of Trustable Internet was established in partnership with the Keio University SFC Research Institute in October 2022. In March 2023 we announced Endorsement Graph, a primary function for realizing Trustable Internet, exhibiting it in a demo at IETF.

https://www.fujitsu.com/global/about/resources/news/press-releases/2022/1013-01.html

To enable easy trial of the Endorsement Graph function, we have made it available as a Chrome extension that can be installed on your browser. In this article, I will explain what users can experience once the extension is installed.

Please note this article uses the English version of the portal. For the Japanese version, please see the post linked below.

Trustable Internetを用いた偽情報対策技術が体験できます - fltech - 富士通研究所の技術ブログ

What you can experience with the Chrome extension

1. Visualize evidence information related to an article

This Chrome extension retrieves a variety of evidence information related to an article, managed by Trustable Internet, using the article's URL (*1). It then considers the connections between the retrieved information and structures it as a graph.

Below is an example of a generated graph, called an Endorsement Graph. The graph grows each time a third party contributes information about the published article, such as evidence information for an image used in the post or a formal announcement from a local government. Thus, the contents of an Endorsement Graph differ dramatically depending on the date and time it is retrieved.

(*1) In this demo, the evidence information has already been cached in the Chrome extension.

2. Analyze an Endorsement Graph to evaluate authenticity

As the Endorsement Graph grows, it becomes more complex and difficult for human users to interpret. The Chrome extension deciphers this complicated graph on behalf of the user and explains it in natural language that is easy to understand.

In particular, the Chrome extension reads out evidence information connected in the Endorsement Graph and uses a large language model (LLM) to calculate whether there are any inconsistencies between them (*2). It then extracts some consistent evidence information and uses it to generate an explanation, enabling easier understanding.

(*2) In this demo, because LLM processing takes time, the result has already been calculated.

How to experience the Chrome extension

1. Create an account on the Fujitsu Research Portal (FRP)

For details on how to create an account, please see the following.

https://blog.fltech.dev/entry/2023/11/13/announcing-fujitsu-research-portal-en

2. Install the Chrome extension

Access the Trustable Internet page on the FRP site and follow the instructions to install the Chrome extension on a Chrome browser.

3. Press the TI button to begin the experience

With this page open, press the TI button displayed on the toolbar. The program code associated with the TI button will find the sample article on the page, retrieve evidence information related to the article, and display analysis results for the article based on the evidence information. You can click the analysis results to view details.

Sample Article

The following is a sample article posted on a fictional social networking site for the purpose of this demo.

"Shizuoka flood damage" article posted on a fictional SNS on September XX, 2022

Future Direction

Through continued collaboration with our customers, we aim to improve the accuracy and processing time of our anti-disinformation technology for use in various fields. Please look forward to future releases, including demos that acquire and analyze the evidence information of articles in real time, and APIs supporting Trustable Internet.