Even though COMPTEXT2025 closed last weekend, its spirit will live on in the splendid photography of our dedicated group of colleagues from the University of Vienna: @jamoeberl.bsky.social, @amplanitzer.bsky.social, @apeksha.bsky.social! We thank and appreciate their work!

Thank you, #COMPTEXT2025! ❤️ It was amazing to see familiar faces, meet new ones, and have so many insightful conversations and discussions. Thank you for the great feedback and inspiring presentations! I'm excited about where the field of computational social science is heading!

Thanks @comptext.bsky.social for a great conference in #Vienna! I presented (1) a new method -- "MQ-Class" -- for LLM-classifying the full spectrum of media ideology and (2) my project mapping the 2D structure of global media ideology. Great people, panels, and most importantly food! #COMPTEXT2025

Thanks also to @comptext.bsky.social for keeping the ball rolling! I couldn't be happier this beautiful community exists 😁

The validation panels were epspecially popular at #COMPTEXT2025 To continue the discussion in June:👇

Off to Vienna for this year’s #COMPTEXT2025 conference. Really looking forward to seeing everyone again!

Absolutely excited to have @ahrabhikat.bsky.social in Vienna - for COMPTEXT2025 and also for coffee ☕💚

I absolutely loved @holnburger.bsky.social‘s presentation of his tool crowdbucket.io! What a well-designed service to have LLM topic modeling results validated by humans in a convenient and fun way…like swiping on Tinder. Will definitely use it for my historical gossip! #COMPTEXT2025

🔥second-to-last panel at #COMPTEXT2025 on Hate, Counterspeech & Algorithmic Influence! @simonkarg.bsky.social on tracking hate/counterspeech Thomas Kirchmair on cross-platform variations of hate Chico Camargo on (in)effective user agency Chaired/discussed by Travis Coan & @jadevrielink.bsky.social 👏

Such a great & needed roundtable - for #COMPTEXT2025 and beyond! 👏👏

Where does the bias hide? 🔎 Let's discuss the applicability of multimodal models for clustering multilingual SM data w @daweg.bsky.social @nadjaoz.bsky.social @hdnrch.bsky.social Sunny Yang @tristancann.bsky.social Chair @hajoboo.bsky.social An awesome panel to end #COMPTEXT2025 ! ✍️

Had an amazing time at my first #COMPTEXT2025 in Vienna. Happy to see so much cool research, get the chance present our paper and represent a growing team of researchers from the Institute of Political Studies @fsv.charlesuni.cuni.cz

My personal highlight of #COMPTEXT2025 😊

Forgot to mention: there are now also other #rstats packages that connect to Ollama like ellmar, tidyllm, ollamar. I still think ours works best for annotation and other tasks, but you might like the syntax or approach of something else more...

Post-Conference-Schnitzel. #COMPTEXT2025

#COMPTEXT was great, thanks to the organizers for putting together such a strong interdisciplinary conference!

As this amazingly successful #COMPTEXT2025 comes to an end, it’s time for a bit of self-reflection 🔥

Started at COMPTEXT two years ago, I was very happy to present today the complete models for detecting social group appeals at #COMPTEXT2025 with @lenamariahuber.bsky.social and @rwillh11.bsky.social Watch this space for use and access details in the coming weeks!

Some statements from the last two days of #COMPTEXT2025 worth discussing…🧐 🤓 come to the TOP cinema bar tonight to keep the conversation going and to 🥳 party with us after an inspiring conference! 🥂After Drinks & After Thoughts starting at 7pm🥂 @compcommlab.bsky.social @comptext.bsky.social

After 3 packed days of research, workshops, and networking, #COMPTEXT2025 has officially wrapped up! In 48+ panels, 220+ presentations, and a full day of tutorials we received a great review of the state-of-the-art in our field. Thanks everyone and see you next year!

How can we better capture marginalised voices through computational text analysis? An inspiring roundtable at #COMPTEXT2025 sparked important reflections. Huge thanks to @ahrabhikat.bsky.social, @krangelie.bsky.social, @indiiigo.bsky.social for guiding this vital conversation!

The success of COMPTEXT 2025 was made possible by the indefatigable and amazing work of the local organising team led by @fabiennelind.bsky.social + @hajoboo.bsky.social @univie.ac.at @compcommlab.bsky.social. The COMPTEXT Association is grateful for their work and to our sponsors (bit.ly/430VUye)!

We are happy to announce the recipients of the COMPTEXT Association travel grants. Grants were based on priorities such as ensuring global attendance and diversity. 2025 recipients are: Junaid Ahmed, Jiyoung Lee, Larissa Miller, Naama Rivlin, Ren Tao, Zeynep Zulfikar

A lively late morning at #COMPTEXT2025 brought stimulating sessions on narrative and framing, LLM performance, affective polarisation, political language in crisis, media bias, and protest in authoritarian regimes.

Happening now at #COMPTEXT2025 🤩🤩🤩 @ahrabhikat.bsky.social @krangelie.bsky.social @indiiigo.bsky.social

Absolutely fascinating and packed panel on enhancing and (critically) evaluating LLM performance at #COMPTEXT2025 this morning. w/ great work by @zacdgreene.bsky.social, Laura Liebig, Lukas Erhardt, and @christianpipal.bsky.social. Excellently chaired by @amplanitzer.bsky.social! 👏

@ahrabhikat.bsky.social is presenting her fantastic work 👀 #COMPTEXT2025

Loved the roundtable led by @ahrabhikat.bsky.social "Marginalised Voices and Perspectives in Computational Text Analysis" /w @krangelie.bsky.social & @indiiigo.bsky.social Everyone was taking notes! ✍️ #COMPTEXT2025

I had a great time presenting my work for the first time at #COMPTEXT25 - excited to keep coming back to this excellent conference!

Another fascinating morning at #comptext25 I was happy to join a panel on data and themes from the Horizon project #ActEU with @matildecer.bsky.social on gender and polarisation, @giucarny.bsky.social on gender and climate change. Looking forward to more from ActEU and more panels this afternoon!

Important topic! On after lunch in BIG Hörsaal at #COMPTEXT2025 with @indiiigo.bsky.social @krangelie.bsky.social @ahrabhikat.bsky.social

What a great panel on media bias & misinformation at #COMPTEXT2025 Investigating political sentiment, ownership influence & public engagement with @lucas-p-dasilva.bsky.social @janabernhard.bsky.social @ahrabhikat.bsky.social @hajoboo.bsky.social Patrick Walkowiak @saminenno.bsky.social Jiyoung Lee

Some advice: Don't be a jerk. Avoid superiority. Understand trauma and retraumatization. Don't take shortcuts. More human intervention. Automation might not help. Get domain + interdisciplinary expertise. Always ask: for what purpose? Is curiosity enough? Ethical considerations? #COMPTEXT2025

Finally Sarah Lynn Jewett speaking about "Another Problem from Hell: Navigating an Archive of Trauma" 🔥 She discusses the behind-the-scenes challenges of working with extremely sensitive data. She studies perpetrators of mass atrocities and genocides, focused on court records. #COMPTEXT2025

The Arkadenhof: Recommended spot to check out for a break at #COMPTEXT2025

Extensive evaluation of Whisper models for transcription, some discussion of the openness of this tool. Many visualizations of frames and ideologies, such as how frames have changed over time (e.g., pre- and post-covid). I believe this all assumes a list of ideologies as input (?). #COMPTEXT2025

Prior work has mostly relied on manual identification of political channels. Given voice transcriptions and image descriptions, we can automatically get issue-based frames (TopicGPT) and setting types (clustering). #COMPTEXT2025

Next is @cklamm.bsky.social to talk about "Pixels, Texts and Politics: Decoding Ideological Framing on YouTube with a Multimodal Approach." Youtube has become a platform for both mainstream and alternative political ideologies. How can multimodal content on YouTube reveal framing? #COMPTEXT2025

How the dynamics of media coverage change as a result of the interplay between issues and actors? @janabernhard.bsky.social shows in an excellent comp text analysis study on media bias in 🇦🇹 news before the 2024 Austrian elections. /w @ahrabhikat.bsky.social & @hajoboo.bsky.social #COMPTEXT2025

He closes with an example of a meme (image and text). In the "old days" we would reduce this to a sentiment score. Now we can extract rich qualitative descriptions using LLMs. Exciting! #COMPTEXT2025

Current challenges: 1. Speaker diarisation (multiple viewpoints are hard) 2. Scene detection (thresholds are hard, data dependent) 3. Community detection is static #COMPTEXT2025

At the end is an visual, interactive tool with the various extracted information and summaries. Should allow qualitative researchers to examine dominant characters, main (hero, villain) nodes in the network, connections between events, timelines. Goal is to bring in theory. #COMPTEXT2025

Input: 1 million Tiktok videos Text analysis of Whisper transcripts: topics, entities, events, framing Image analysis of automatically segmented scenes: face and object recognition #COMPTEXT2025

Operationalizing narratives requires detecting narrative elements (entities, characters, plot, relationships). Images can reinforce textual content and are important parts of the narrative in the text. They focus on short videos but the pipeline should work for texts and images. #COMPTEXT2025

Next up is Chico Camargo to discuss his work on "The Narrative Navigator: A Tool for Visualising Culture War Narratives in Shortform Video Content." He covers some background theory about levels (micro, meso, macro) of narratives, and different types of narrative strategies. #COMPTEXT2025

This is all followed by a network analysis for community detection and qualitative analysis of those communities. The network analysis uses the extracted structure to build edges + nodes. #COMPTEXT2025

Thanks! Slides are here, btw: tinyurl.com/comptextcost #comptext2025

Excellent talk by @damiantrilling.net about his framework that helps you decide which approach to use for computational text analysis. A really helpful decision aid - might save you lot of headaches! #COMPTEXT2025

The study includes multiple classification steps, using an OpenAI model (?) to classify URLs as political and the Claude Sonnet 3.7 to extract structured information: what (action, event, behavior, topic), where, who, how (sentiment), why (apparent objectives), when. #COMPTEXT2025