Vincent Verger : Couverture médiatique des demandeurs d’emploi : étude d’un corpus de texte

Vincent Verger : Couverture médiatique des demandeurs d’emploi : étude d’un corpus de texte

Vincent Verger: Media coverage of job seekers: study of a text corpus

Title history
  • Vincent Verger : Couverture médiatique des demandeurs d’emploi : étude d’un corpus de texte — observed on 20261117-0330 UTC (YouTube)
🎙 Vincent Verger 👥 2K 📅 November 16, 2026 ⏱ 42 min 👁 8 📄 original study 🧭 2026-11-17
Available in: English (current) Français

Keywords

stigmajob seekersmedia coverageNLPaudiovisual

Summary

This seminar presentation by Vincent Verger, a PhD student at CREST, introduces an ongoing research project on the media coverage of job seekers in France. Using transcripts from France 2’s evening news (2012-2024), the study applies advanced NLP methods to measure three theoretical concepts: salience (how often unemployment is mentioned), framing (whether coverage is neutral, supportive, or stigmatizing), and thematic classification. The methodology involves building a keyword dictionary, expanding it via word embeddings, filtering with a large language model (Llama 4), and classifying frames and topics. Preliminary findings show that media salience fluctuates and correlates with the unemployment rate, and that periods of high coverage are associated with increased stigmatizing frames. The presentation also reveals that themes like conflict with unions and allocations are more stigmatizing, while growth and social measures are less so. The next steps include obtaining France Travail data on job search behavior and exploiting local media variations for causal identification. The talk emphasizes methodological innovation and calls for feedback on the approach.

166 words

Critical Evaluation

Value of the Information & Strength of the Argument

The value of the information is high, as it addresses a significant social issue with modern computational methods. The argumentation is well-structured: the presenter clearly defines the research question, situates it in existing literature, and outlines the methodological choices with justifications. The use of LLMs for annotation is particularly novel in this context, and the validation steps (e.g., linking stigma to internal locus of control) strengthen the credibility of the measures. However, the causal claims are prospective and not yet established. The presentation is honest about limitations, such as the need for human annotation validation and the lack of causal evidence so far.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is generally strong: the speaker cites relevant literature in psychology, sociology, and economics, and builds on established theories like framing and salience. The sources mentioned in the description (INAthèque and INA le lab) are institutional and relevant. The title accurately reflects the content, which focuses on media coverage and corpus analysis. The methodology is detailed, and the presenter openly discusses potential weaknesses (e.g., reliance on LLM annotations). The talk is appropriate for an academic seminar, and the presenter demonstrates deep engagement with the subject.

205 words

Title / Content Match

The title accurately reflects the content: the talk is about analyzing media coverage of job seekers using a text corpus. It is clear and matches the presentation's scope.

Quality & Reliability

8/10

The research is methodologically advanced, using state-of-the-art NLP tools (LLMs, word embeddings, topic modeling) on a large corpus. The presentation demonstrates careful validation (e.g., checking framing against psychological theories), but the work is ongoing and not yet peer-reviewed. The speaker is transparent about limitations and open to feedback.

Key Moments

Cited Sources

  • INAthèque — Mentioned as the source of the France 2 transcripts used in the project.
  • INA le lab hypotheses — Seminar series and research hub; provided in the video description.

Contribution & Novelties

The original contribution lies in applying modern NLP techniques (LLMs, word embeddings, and automated topic modeling) to an audiovisual corpus to study media stigmatization of job seekers, which is a novel approach compared to previous keyword-based methods on print media. The presentation also provides a detailed methodology for combining theoretical frameworks with automated text analysis, and it opens the door for causal inference in future work.

Pour aller plus loin :

115 words

Radar Profile

The radar profile shows a balanced and high performance across all dimensions, with a slight emphasis on technical level and quantity of information, reflecting the methodological depth and comprehensive presentation. Quality and reliability are also strong, though the ongoing nature of the research prevents a perfect score on reliability.

Reliability 8/10