poltextlab_admin

The poltextLAB team’s submission for the SLICES-SC consortium’s “2nd SLICES-SC Open Call” has been successfully accepted. The SLICES-SC project builds a community of researchers around SLICES-RI, which offers the necessary solutions to create and manage efficiently IT-related experiments. As per the successful proposal, poltextLAB will be granted access to the imec research group iLab.t‘s GPULab infrastructure starting from April 2023. The GPULab is a testbed with 125+ GPUs with over 570.000+ cuda cores and 1.8TB+ GPU RAM for AI research and every research that needs GPUs. The poltextLAB team will exploit imec’s infrastructure to carry out a finetuning experiment of the large language models used in the CAP BABEL MACHINE project.

Nathalie Neptune – researcher at Institut de Recherche en Informatique de Toulouse (IRIT) – will be taking part of the Visiting Professorship program in April 2023. Neptune will be working in poltextLAB on the hyperparameter tuning of monolingual and bilingual Longformer models as part of the CAP BABEL MACHINE project.

On January 26-27, 2023, the Institute of Informatics of the University of Szeged hosted the Hungarian Conference on Computational Linguistics for the nineteenth time, which presented research and results in the field of language technology. At the conference, researchers from poltextLAB gave a presentation entitled “HunEmPoli: a Hungarian language, extensively annotated emotion corpus” in the Corpus and Databases section. The conference website is available here: https://rgai.inf.u-szeged.hu/mszny2023 The conference proceedings are available here: https://rgai.inf.u-szeged.hu/file/430

Orsolya Ring and István Üveges created the emotion corpus which is available on the project’s GitHub page. “The (manually annotated) HunEmPoli corpus was built using pre-agenda speeches from the 2014-2018 parliamentary term, and was created within the framework of the Hungarian Comparative Agendas Project (CAP) of the Institute of Political Science of the Centre for Social Science Research, and is freely accessible for research purposes upon registration. In the course of our research, we created an inductive emotion category system, the categories of which can later be mapped to Plutchik’s emotion category system, which distinguishes eight classes and is also convertible to the positive-negative categories used in sentiment analysis. This extended system was necessary because, in our previous experience, sentences in political texts could not be classified into one of the most commonly used Plutchik’s category systems for emotion analysis, or only with very low annotator agreement, whereas the extended system allowed the corpus to be annotated with high inter-annotator agreement. In the final annotation guide, a total of 12 so-called emotion topics (ET) were defined, each of which was accompanied by at least three call words or phrases to facilitate the annotators’ work.” You can continue on this link.

Rebeka Kiss and Miklós Sebők published a paper entitled Creating an Enhanced Infrastructure of Parliamentary Archives for Better Democratic Transparency and Legislative Research -Report on the OPTED forum in the European Parliament (Brussels, Belgium, 15 June 2022) in the latest issue of  International Journal of Parliamentary Studies. In this paper, they summarise the key findings of the forum “Data4Parliaments” Parliamentary Data for a Better Democracy, organised by the OPTED network at the European Parliament on 15 June 2022. The aim of the conference was to provide a forum for the exchange of ideas related to sharing and analysing parliamentary text data between official archives and different user groups from the social sciences and the civil society.

On the 1st of December 2022, Orsolya Ring and Ákos Máté held a beginners course of the “Text Mining and Artificial Intelligence Training Programme”.  The course introduced the theoretical and practical foundations of text mining and artificial intelligence applications in social sciences.

On Tuesday, 8 and 15 November 2022, Ákos Máté held a course entitled Data Visualization in R. The aim of the course was to provide a practical and interactive overview of data visualization using the ggplot2 package in R. We covered the basic concepts of data visualization and then put them into practice using different data sources. The presentation was in Hungarian, the slides were in English.

We are excited to announce that CompText 2023 will be held at the School of Government and Public Policy at the University of Strathclyde in Glasgow, Scotland. The meeting will be held from May 12-13, 2023 and will include a set of pre-conference training sessions covering multiple approaches to computational social science, text analysis and machine learning. See the CompText website for updates on the meeting and information on prior conferences. The official call for submissions will be announced in early November 2022 with the submission deadline in mid-December.

From 29th to 30th September, OPTED held its second annual meeting in Amsterdam. The city is home to two of our project partners, at the University of Amsterdam and the Vrije Universiteit Amsterdam, and provided a great background for fascinating conversations. Members of every Work Package came together to discuss the present and future of the project, both in-person and a hybrid format. The team of Work Package 1 and our local partners in Amsterdam organised a variety of sessions where everyone could hear about the latest work in the project, and give their input for the future direction. We talked about recent and upcoming publications, the ongoing work on the data inventories, as well as plans for the future platform.

István Üveges gave a presentation at the CPSS Workshop in Potsdam, held between 12-15 September 2022. He described a new annotation framework for the emotion analysis of political texts and the machine learning results obtained with the Hungarian BERT model. The event was part of the KONVENS 2022 Conference on the intersection between computational linguistics and political science, where participants discussed issues such as the latest challenges in the sentiment analysis of political texts, propaganda detection using machine learning, or the identification of political groups in social media. The conference proceedings are available on the following link: https://old.gscl.org/en/arbeitskreise/cpss/cpss-2022/workshop-proceedings-2022