Martin is a brilliant network scientist and professor of physics with a focus on computational science at Umeå University, where he heads the Integrated Science Lab (IceLab) and its excellence center on modeling adaptive mechanisms in living systems under stress. His research develops mathematical models, algorithms, and visualizations that reveal structure and dynamics in complex systems—from information flow and citation networks to ecology and spreading processes.

And Martin is awesome in many dimensions: he’s a deep thinker, he’s someone who carefully crafts meticulous and mindblowing science communication. But he is probably best known (with Carl T. Bergstrom) for introducing the map equation framework for flow-based community detection in networks, and for the widely used Infomap approach built on those ideas. His work has been highly influential across network and data science, with publications and tools used broadly by researchers in many disciplines. Do not miss his talk!

The talk details are

• Time: March 10th, 14:00
• Place: DTU Compute Building 324, room 240
• Title: Predicting interactions in dynamic networks 
• Abstract: Predicting future interactions or novel links in networks is an indispensable tool with many applications across diverse domains, including drug repurposing based on genetic networks, money laundering detection in financial systems, and recommendation systems using transactional data. Among the many techniques developed for link prediction, those leveraging the networks’ community structure have proven highly effective. For example, the recently proposed MapSim predicts links based on a similarity measure derived from the code structure of the map equation, an information-theoretic community-detection objective function that operates on network flows. The map equation benefits from Infomap, its fast optimization method widely regarded as one of the best network-clustering algorithms. However, we developed Infomap for static networks. While its stochastic greedy search algorithm excels at identifying reliable communities in network snapshots, Infomap cannot effectively integrate new data or adapt to smooth transitions over time in continuously evolving relational networks. This shortcoming raises a computational challenge: How can we equip the map equation framework with a computationally efficient optimization method to enable adaptive analysis of dynamic networks, leveraging evolving relational data to predict future interactions and uncover novel links in real time?

This is a great opportunity, don’t miss it!

References
[1] Data from https://valgdatabase.dst.dk/
[2] Note that parishes don’t align perfectly with constituencies, so some values are interpolations.

And if you’re really interested there’s a nice PDF with “infinite” zoom that I can send upon request.

Our first speaker is the amazing Cesar Hidalgo. Did I mention that Cesar is amazing? Let me give you some reasons. He’s the the Director of the Center for Collective Learning which spans the Universities of Toulouse, Manchester, and Harvard. He’s a highly cited scientist, famous for (among other things) the notion of economic complexity. He has written several bestselling books. There’s a TV show about him. He started a successful company. The list goes on.

But all this worldly success is not why you should care about coming to hear Cesar speak. You should come because Cesar is a fountain of creativity. He has more unusual and inventive ideas before breakfast than most people have in months. And he’s brimming with enough energy and excitement to make the energizer bunny envious. In summary, there’s no excuse not to stop by.

Title: How time, technology, and language impact collective memory and attention

Time: February 9, 10am

Place: Det Samfundsfaglige Fakultetsbibliotek. Gothersgade 140. Auditorium 1.

Abstract: From writing to the web, humans have used communication technologies to enhance our collective memory. Yet, much of what was once popular is now forgotten. In this talk, I will present research exploring the roles played by time, language, and technologies on the dynamics of collective memory and attention. Using data on the attention received by biographies, scientific papers, songs, and movies, we will explore the universal decay of collective memory, the role played by languages in global fame, and the biases in attention and collective memory introduced by changes in technology.

Official bio: César A. Hidalgo is a Chilean-Spanish-American scholar known for his many contributions to economic complexity, data visualization, and applied artificial intelligence. Hidalgo leads the Center for Collective Learning at the Artificial and Natural Intelligence Institute (ANITI) of the University of Toulouse. He is also an Honorary Professor at the University of Manchester and a Visiting Professor at Harvard’s School of Engineering and Applied Sciences. Between 2010 and 2019 Hidalgo led MIT’s Collective Learning group. Prior to working at MIT, Hidalgo was a research fellow at Harvard’s Kennedy School of Government. Hidalgo is also a founder of Datawheel, an award-winning company specializing in the creation of data distribution and visualization systems. He holds a Ph.D. in Physics from the University of Notre Dame and a Bachelor’s in Physics from Universidad Católica de Chile. His contributions have been recognized with numerous awards, including the 2018 Lagrange Prize and three Webby Awards. He is also the author of three books: Why Information Grows (Basic Books, 2015),  The Atlas of Economic Complexity (MIT Press, 2014), and How Humans Judge Machines (MIT Press, 2021).

Cesar is also an old friend from my time as a postdoc in Laszlo Barabasi’s lab. To prove this connection is a BarabasiLab halloween-party photo from 2007 (from the author’s personal collection) which shows the young Cesar (with Nick Blumm to his left and Pu Wang to his right).

If the talk topic intrigues you, you can check out these papers if you’re interested in hearing more

• Candia, Cristian, et al. “The universal decay of collective memory and attention.” Nature human behaviour  (2019)
• Ronen, Shahar, et al. “Links that speak: The global language network and its association with global fame.” Proceedings of the National Academy of Sciences  (2014)
• Jara-Figueroa, C., Amy Z. Yu, and César A. Hidalgo. “How the medium shapes the message: Printing and the rise of the arts and sciences.” PloS one (2019)
• Yu, Amy Zhao, et al. “Pantheon 1.0, a manually verified dataset of globally famous biographies.” Scientific data 3.1 (2016)

• So many clustering algorithms create division wherever they’re applied.
• Since stopping Twitter, I’m using RSS feeds instead. Kottke and Daring Fireball is all I’ve added so far and it works AMAZING.
• Blown away by Messi yesterday, I was reminded of this glorious #fivethirtyeight data science piece from 2014 “Lionel Messi is Impossible” https://fivethirtyeight.com/features/lionel-messi-is-impossible/

The two camps: When I look at the literature, there seems to be two camps.

• On one hand, there is amazing research [1,2,3] (explained with exemplary clarity in my recent interview with Piotr Sapiezynsky) which a) clearly reveals the mechanisms behind algorithmically sorted content on Facebook and b) explains how this sorted content leads to filter bubbles along partisan/racial/etc lines.
• On the other hand there is another stream of research stating that the issue with social media is a kind of breaking of filter-bubbles (bubbles that exist in the first place). This one is illustrated below

(Actually it was the tweet above from my friend and collaborator Michael, that made me think about this whole thing.)

The paradox. So what’s going on!?! It’s a paradox: There’s incontrovertible evidence of filter bubbles. Yet the problem is that social media exposes us to opposing viewpoints (from outside the bubbles). That made me think: Is there a way that these viewpoints can both be true simultaneously?

And upon reflection, I think there is.

One of the things that came up in my podcast talk with Piotr is that it’s difficult for democrats to target content to republicans for a surprising reason. It is difficult because it’s highly painful to republicans to see Biden portrayed in a positive light (e.g. in political ads) that republican users actually stop browsing Facebook sooner when they see such ads (too much cognitive dissonance, perhaps). This leads to lost revenue from other advertisers. That’s the essence of how the filter bubble works, in fact. (And it’s the same for democrats and Trump, obviously.)

But the research in Michael’s tweet above suggests that there’s a different kind “democrat content” that does not have that effect on republicans (and vice versa).

In fact my hypothesis is that there is content from the “other side” which does not compel people to leave the platform. Stuff that does not result in cognitive dissonance. One candidate for such content could be statements that are particularly outrage-generating (and conforming with our negative image of the other side, hence no cognitive dissonance).

Stated differently, perhaps there’s not just partisan and non-partisan content. It’s highly plausible to me that there are some types of partisan content that drive people away from their social platform of choice (displaying reasonable aspects of political opponents, discussing them in a positive light) and other types of partisan content (the more outrageous stuff) that keep people engaged.

Resolving the paradox. This explains the apparent paradox. There is filtering, but the filter bubbles do not filter all partisan content. It just separates out the reasonable stuff and only sends the crazy stuff across political divides.

Hence the title of this piece, post versus person. This sorting of content is only possible because social media is about posts rather than people. Posts are bite-sized opinions, factoids, etc. People are complex, nuanced, good & bad.

In the off-line world, I am presented with the full person, not just highly selected parts of them. And (in my experience) even people I strongly disagree with politically are usually pretty reasonable in most of their opinions and behavior. Maybe the person I’ve know as a fun and friendly coffee-machine acquaintance for years will say something pretty outrageous one day, and I’ll think “wow, that was weird and unexpected” … and move on.

But in the online world, I don’t get the whole person. I don’t get the reasonable and boring stuff. I don’t get the history and context of the person. I just get the single crazy utterance, just the disembodied single post, because that’s what’ll keep me staring at the screen longer [4].

What do you say? My hypothesis around which types of content makes it across the partisan divide is totally testable. I hope someone will go and investigate!

(Also, I haven’t really done a literature study, so perhaps this idea isn’t new at all. If it isn’t, let me know!)

We’ve long known that social media only rewards only extreme snippets with attention, but I think this filter bubble aspect is new: There is some partisan content that reinforces existing views in a righteous not-cognitive-dissonance-inducing way (that’s the stuff that makes it across the divide), then there’s the more nuanced stuff that causes us to have so see the other side in a different light – a list of the good deeds done by my enemy. That stuff we never see.

Further, by only exposing us to the most extreme views of our neighbors, social media is making us forget all the mundane stuff we have in common with all people.

By the way, realizing this issue even shows the outline of a solution: As we connect people online, on path towards a better system is one that establishes mechanisms that help us remember to see other people as whole beings – not just hot-takes.

Let me know what you think

References

1. https://dl.acm.org/doi/abs/10.1145/3359301
2. https://www.ieee-security.org/TC/SPW2022/ConPro/papers/ali-conpro22.pdf
3. https://dl.acm.org/doi/abs/10.1145/3514094.3534135
4. This view also explains why the old non-algorithmic Twitter somehow felt less insane. You were exposed to all tweets from entire people. There was still a selection, people want likes resulting in more extreme content, but at least you got the full person. With algorithmic sorting of maximally engaging content, we get rid of all the reasonable stuff that doesn’t reinforce our existing beliefs.

Christoph is an Associate Professor of Social Networks at ETH Zürich. His work investigates how social networks evolve over time and how individuals are affected by the emerging social structures that they are embedded in. The Swiss StudentLife Study that he conducted with his team from 2016 to 2020, for example, studies how friendship networks and social groups evolve among previously unacquainted university students, and how the emerging levels of social integration affect their well-being and academic success. He also develops statistical network models and computational methods to study social networks dynamics.

And Christoph delivers truly award winning science. He received the Raymond Boudon Award of the European Academy of Sociology in 2018 and the Freeman Award of the International Network of Social Network Analysts in 2021. (full bio below)

Title: The emergence of social networks and how they matter for individuals

• Date: Thursday, April 28th, 2022.
• Time: 13:00
• Location: Technical University of Denmark, Building 321, third floor lab-space (at the end of the hall).

Abstract: Individuals have the deep psychological need to form ties with others. The relationships they build – who they become friends with, who they spend time with, who they talk to, who they approach for support – follow certain micro-level patterns. These are partly rooted in individuals’ preferences and partly in the opportunities that come with their existing social networks and environments. Individuals are in turn affected by their position in social networks, for example, through processes of social influence, or the access to social support and resources. Two important questions in social network research are thus, first, to explain how macro-level structures of social networks emerge from micro-level processes and, second, how these emerging, complex social networks affect the individuals they connect. In this talk I will discuss empirical research strategies that can help getting insights into these questions. From a methodological point of view I will discuss the utility of different data collection strategies, statistical data analysis, and agent-based simulations. The talk will further present empirical insights on the emergence of friendship ties and groups in social networks, and on the well-being and academic success of individuals.

Bio: Christoph Stadtfeld is an Associate Professor of Social Networks at ETH Zürich in Switzerland. His empirical work investigates how social networks evolve over time and how individuals are affected by the emerging social structures that they are embedded in. The Swiss StudentLife Study that he conducted with his team from 2016 to 2020, for example, studies how friendship networks and social groups evolve among previously unacquainted university students, and how the emerging levels of social integration affect their well-being and academic success. He also develops statistical network models and computational methods to study social networks dynamics. Probably best-known in recent years is his work on actor-oriented models for relational event data (DyNAM) and corresponding open source software (the goldfish package in R). Christoph’s work appeared in sociological, methodological and interdisciplinary journals. He received the Raymond Boudon Award of the European Academy of Sociology in 2018 and the Freeman Award of the International Network of Social Network Analysts in 2021.

There is also a video of me doing “research-y” stuff.

And there’s even a little written “portrait” of me (in Danish).

I can’t really put this stuff in my normal photo library (don’t want to do that to the rest of the family), So now, I’m doing a little blog-post to have it all collected in a single one place.

The true highlight of the day, however, was when I had to go on stage. My last name is first alphabetically, so I was the first person on stage. Prior to the ceremony we had been carefully briefed about where to stand, members of the Danish Royal Family weren’t shaking hands yet and much more. But what I hadn’t anticipated is that the minister and the crown princess had not received that briefing. So when I got on stage and walked to my designated spot (where the Crown Princess gave me a warm handshake), there was a moment of confusion.

This confusion was picked up by Billedbladet, a Danish gossip magazine which subsequently ran a story entitled (my translation): Confusion on stage: Crown Princess Mary helped get things under control. If you click through, there’s even a video of the “incident”.

So that’s perhaps the real breakthrough: I made it to the tabloids

A few days ago, I was scrolling on Twitter when I came across a Twitter thread by Dr. Matthew Sweet (@drmatthewsweet) about Johann Hari’s new book Stolen Focus. I haven’t read Stolen Focus, only the parts about our work. Those parts were completely fine, see full backstory below.

I don’t know if it counts as “viral”, but the thread has 1.9K retweets and almost 5K likes at the time of writing.

As I read the thread, I have to admit that I grew somewhat uneasy.

Disturbingly (from my perspective), the thread uses our paper (with Philipp Lorenz-Spreen, Philipp Hövel, and Bjarke Mønsted) Accelerating Dynamics of Collective Attention as a main example of a problem with Hari’s book. Here’s Dr. Sweet getting into it:

He goes on:

This is where it got even more troublesome.

Sweet doesn’t seem to understand the context of the quote and ends up completely misrepresenting the point we’re making in the paper.

We do say that the “phenomenon lacks a strong empirical foundation”. But the point of that statment (which is from the abstract) is to suggest that the research we present in the paper, provides that very missing foundation.

(I’m also a little bit offended that Sweet takes a potshot at us by stating that the paper “didn’t make a big impact” [1].)

So what’s the criticism of our work?

Ok. The next tweets contain the actual criticism of our paper:

That’s it?!

The argument seems to be that if you’re a “fellow historian”, you won’t like our approach. Perhaps, between the lines, there is also a claim that the datasets we study are not representative enough.

Importantly, from substantive perspective, Sweet doesn’t engage with what we’re trying to do in the paper at all. And there are zero attempts to address the empirical behaviour that we are observing in the paper.

Our careful, peer reviewed work documents systematic changes across time in a range of datasets. Somehow that doesn’t matter.

Collective versus Individual

To be fair, in one subsequent tweet Sweet goes beyond criticising our work and points to a problem he has with the book.

I fully agree that these two phenomena should not be conflated. For the record, we are very careful not to do that in the paper [2]. (And in the excerpt I read from the book, that was not a problem either. )

The point about Chaucer

But regarding the part of the tweet above which mentions Chaucer’s 9-day wonder, it once again is clear that Sweet hasn’t understood what we’re doing in the paper.

The point of our paper is that this “9-day wonder” is a period of time that we show to be shrinking over time. And it looks like it has been shrinking for a long time. So whatever period Chaucer was talking about is much shorter today.

Through careful empirical work, our paper documents precisely that evolution.

So what now?

The big question now is what we, as authors, should do in a situation like this?

On one hand, I am proud of our paper. As explained above, even with my most generous reading, I find Sweet’s criticism of our work to be loud, full of mistakes, and without almost any substance (good old argumentum ad lapidem, cf [1]). And since thousands of people have read his criticism on Twitter, I’d like to publicly call bullshit on the whole thing.

On the other hand, my intuition is that it’s best just to stay out of Twitter fights [3]. Twitter is not the place for nuanced discourse.

For now, I’ve chosen the option of writing this blog post.

I think the post is long enough and navel-gazing enough that no-one will read it. (Sorry, for making so much out of a small thing, folks, but it is my blog after all.)

It’s also an opportunity to add some nuance to the discussion that’s just not possible on Twitter. For now, this is my solution to have a defence of our work on the record, while not getting in Twitter fights

Addendum: Backstory

This whole thing started back in late 2019 I was contacted by Johann Hari who wanted to talk about a paper that I’d recently written in collaboration with Philipp Lorenz-Spreen, Philipp Hövel, and Bjarke Mønsted: Accelerating Dynamics of Collective Attention.

The paper had come out earlier that year, and we’d all been doing a lot of press for it, so I didn’t think much of it. Hari was in Copenhagen, he stopped by my office and we had a long, nice chat about the paper.

Then, in January 2021, I heard from Hari again. He was hard at work on his book, and sent me the text of the part about our work (and the conversation we had) to read through. His email specifically noted “It’s very important to me to make sure every aspect of the book is entirely factually accurate so I wanted to send it to you so you could read it through.”

I read the excerpt, and I found it to be a careful description of our work, with more detail than many other journalists would care to add. There were also a few sections of me speculating about reasons and mechanisms, but I found those sections to be clearly labeled as such. So I sent it back with a couple of small technical notes.

Full disclosure: I was also a little bit flattered because Hari had turned me into a kind of living character in the book – a kind of hero, which was new and exciting.

Addendum: I’m not alone in thinking that @drmatthewsweet is being disingenuous

After writing the above, I came across a Twitter thread by Erik Hoel making very similar points. I’ve unrolled it below for the reader’s convenience

Originally tweeted by Erik Hoel (@erikphoel) on January 7, 2022.

1/ How to grind an axe with a pop-sci book. You might have seen the viral thread of @DrMatthewSweet implying that the author @johannhari101 cannot “be trusted to handle data.”

2/ This is a great way to get some attention for yourself: pick a book, then find a way to say the author is misusing sources (one of the gravest authorial sins) when really you simply disagree with the original sources (a commonplace, boring objection that gets you no likes)

3/ @DrMatthewSweet does precisely this switcheroo in his recent viral thread. Let’s first look at what he says is the most “serious claim,” the strongest evidence of @johannhari101‘s inability to “handle data” (the second objection)

4/ @DrMatthewSweet implies the authors of the paper @johannhari101 cites *concede* themselves that the hypothesis of shrinking cultural attention spans “lacks a strong empirical foundation.” The implication is that Hari misuses the citation.

5/ Except the authors don’t concede anything like what @DrMatthewSweet is saying! That’s a quote from the Introduction of the paper, when the authors talk about *previous research* and say “but so far” – implying their very paper *is* the empirical research that’s needed.

6/ Does @DrMatthewSweet quote that? No. So in accusing another author of cherry-picking, he cherry picks, and not just some data, but a wild out-of-context quote that makes it sound like the conclusion of the study. It’s not! It’s the *previous* research. Here’s the abstract:

7/ It’s obvious that the paper says what @johannhari101 says it says. That, of course, doesn’t mean it’s correct! It could be wrong. I haven’t read @johannhari101‘s book, I don’t know him personally, and I’m pretty skeptical of pop-sci books lately:

https://erikhoel.substack.com/p/why-do-most-popular-science-books

8/ But this was @DrMatthewSweet‘s “most serious claim” and it’s really just that he doesn’t agree with the paper, NOT that @johannhari101 misused it. To make make this point, @DrMatthewSweet took a pull quote totally out-of-context. The Discussion is clear

https://www.nature.com/articles/s41467-019-09311-w

9/ The other claims are much smaller, like that @johannhari101 cited a non-peer-reviewed study to make a point, and that he doesn’t know that Facebook let’s you find nearby friends. Frankly, they seem like smaller sins than @DrMatthewSweet‘s misrepresentation of that pull quote.

10/ So it looks like rather than @johannhari101 misusing sources, @DrMatthewSweet directly does so in his thread. Overall, it’s just a substitution of a boring disagreement (“I don’t like this book & the research behind it”) with the exciting implication of authorial misconduct.

Originally tweeted by Erik Hoel (@erikphoel) on January 7, 2022.

Notes

[1] For fun, I clicked on the paper’s “metrics” page – which provides a measure of online impact. Here’s what it states:

I don’t know what you have to do to “make a big impact” in Sweet’s world, but I am personally of the opinion that being in the 99th percentile of all tracked articles of a similar age in all journals is pretty decent.

And while we’re nitpicking and speculating wildly. Half the author team is from Germany, so it’s also factually incorrect to state that the researchers are “in Denmark”. Probably it’s my paranoia here, but how is it relevant where we’re from!? Am I detecting a hint that research from Denmark should be taken less seriously?

[2] As detailed in the “Backstory” addendum, I got to read the part of the book about our work, and there Hari is careful to talk about collective attention and not individual attention.

[3] Also, it turns out that Hari is a controversial figure in the UK. While my personal interactions with him have been great, a Twitter search reveals that a lot of people have an axe to grind with him. So I also worry that if I jump in to defend our work that somehow it will be confused with a defence of the book as a whole (which I still haven’t read in full and thus don’t have an opinion about), and which might direct the intense vitriol of the Hari-haters my way … something that I am also not really interested in.

To this day I still have fond memories of that talk. It was a packed room, in a stylish old derelict cinema, where – incidentally – the video for this pop hit (which I kind of dig, btw), was shot. And that night people actually liked my jokes. It’s not so often you get laugh-out-loud reactions to a science talk. This one was my favorite experience of giving a talk ever.

The talk itself is entirely non-technical, full of silliness, and contains a kind of history of Physics-y Network Science, plus in-the-thick-of-it results from our (then on-going) project Sensible DTU.

Now, a few days ago in early October 2021 [1], for reasons unknown to me, someone put the video of that night on YouTube. I haven’t watched it (I worry a little bit that it won’t live up to my recollection of it), but I’m posting it here so I have a way of finding it again.

Notes

[1] Interestingly, it’s been 7 years since I gave the talk. And the pop video I linked in the post was the song “7 years” by Lucas Graham. Coincidence? Or proof that the Illuminati are running everything? We will never know.