Presentation of the research by Masafumi Iwanaga at ACM Hypertext 2025

• #research publication
• #social network analysis
• #Chicago
• #USA

We presented our research findings at the ACM Conference on Hypertext and Social Media 2025. This research focuses on investigating whether friend recommendation in social media accelerates the formation of echo chambers. In a simulation starting from a subgraph of the actual X (formerly Twitter) follow graph, we compared four types of friend recommendation methods:

1. Friend recommendation based on the similarity of followees’ topics (CM)
2. Friend recommendation based on Most-Common-Neighbors (MCN)
3. Friend recommendation based on the Adamic-Adar index (AA)
4. Friend recommendation based on pSalsa (pSALSA)

Only the first one, CM, is content-based recommendation, while the remaining MCN, AA, and pSALSA are link-structure-based recommendations. Under the conditions assumed in our simulation, we obtained the following results:

1. All recommendation methods strengthen the modularity of the entire graph (the degree to which each user has many links within a specific group).
2. Specifically, MCN and AA show an **accelerating effect**, where the rate of increase further increases over time.
3. On the other hand, CM, which is content-based recommendation, increases modularity but does not show an accelerating effect.
4. Furthermore, despite being a link-structure-based recommendation, pSALSA does not exhibit the accelerating effect seen in MCN and AA.
5. When looking at individual users rather than the entire graph, specifically the diversity of communities from which the recommended friends are selected, pSALSA is the most diverse—even more diverse than the content-based CM.

The venue for this year’s conference was the Illinois Institute of Technology in Chicago, USA.
Right outside the train station nearest to the venue was the baseball stadium that is home to the Chicago White Sox.

This is from another person’s presentation, and you can see the frequency of the word “delve” rapidly increasing in the graph on the top left.

Regarding “delve,” it was briefly a well-known phenomenon that when you have ChatGPT write an English paper, it excessively uses “delve into.”

This is from yet another presentation, where the speaker, for some reason, was talking about “zakkyo”-type buildings in Tokyo. I (Japanese) didn’t know that it is special compared to buildings in other countries.

There was this object in a park near the hotel. It seems to be called The Bean.

When you go directly underneath and look up, you can see yourself reflected in various parts of the complex curved surface, which turns into a “Where’s Wally?” (or “Where’s Waldo?”) game of trying to find yourself.

This is the south staircase of Union Station—the one from this.

There are two very similar staircases, but it seems the south staircase was the one used for the filming. The south end of the hall is visible partway through the scene. (Reference 1, Reference 2).

Chicago is apparently famous for having many interesting buildings.

There is a building with fragments of famous structures from around the world embedded in its walls, and the representative from Japan was the “golden castle of Osaka,” as shown in the picture below. I wondered what that was, but it seems to be Osaka Castle.

The picture below shows the section with a fragment of the Egyptian pyramids.