1. médialab Sciences Po
  2. News
  3. MetAt - May 9, 2023 logbook

MetAt - May 9, 2023 logbook

Share our methodological expertise and skills.

Event, Workshop

What is METAT? 

METAT is a research methods support workshop: every month, a three-hour slot to help you resolve the methodological difficulties you encounter in the course of a scientific project.

Who is METAT for?

METAT is aimed at anyone needing occasional support in using a research tool or method. All profiles are welcome: students, doctoral students, researchers, research engineering professionals or others, inside and outside Sciences Po, without restriction of status or affiliation.

How to register?

Registration is compulsory via the form available on the METAT page.

Session of 09/05/2023

Place: Sciences Po

Number of participants: 7

Number of supervisors: 14

Non-medialab supervisors: Blazej Palat (CDSP, Sciences Po), Yuma Ando (CEE, Sciences Po), Cyril Heude (DRIS, Sciences Po), Guillaume Levrier (LISIS, Université Gustave Eiffel)

Bibliometrics in an emerging field of research

First support for a doctoral student in history who came to do bibliometrics on the subject of rewilding. The coaching took the form of a review of available data sources (Scopus, HAL, Google scholar, Ulrich web (at BnF), FACTIVA, Base search engine) and methodological advice on comparing the coverage of these databases with each other. After the first tests carried out during the session, it became clear that the field of research on rewilding is emerging and therefore not very structured at the moment. Participants were encouraged to diversify their search terms, in particular by looking at the keywords provided by the authors of the articles they found.

Recovering Twitter user metadata

Second accompaniment of a political theory student returning to discuss the future of Twitter data collection in light of E. Musk's impactful politics, and to get help doing network analysis on Twitter data already collected on journalists. The session was used to update Minet, deduplicate lists of Twitter follower IDs and use the API to retrieve metadata from these users.

Automatic classification of vimeo videos

Helped a research engineer write a Javascript program using Artoo.js to extend the video library in the backoffice of his website. Addition of a function to automatically select the checkboxes on each line corresponding to a video whose identifier is in the desired list. 

3D visual representation

First coaching of a doctoral student in law who needed to quickly enter a large amount of information and situate it in a 3D representation tool created prior to the session. The coaching took the form of a brainstorming session on the different methods of data collection according to the student's problematic. The supervisors helped to code the data according to established criteria, and to work out the implications between these codes, their justifications and possible visual representations (heat matrices, scatterplots, graphs, etc.). These elements were then put into practice in a LibreOffice spreadsheet.

Classification model training on manifesto dataset

Support for a PhD student in political science in machine learning. The work involved training a classification model to detect certain parameters in multilingual sentences belonging to the European dataset manifestos. Using The Augmented Social Scientist notebook, pre-parameterized models for English (BERT) and other languages (XLM-RoBERTa) were used to train them on the manifestos data previously annotated by the PhD student. The participant was then coached to divide datasets by uploading them to the notebook and then reading them with pandas.  

Processing textual data from online forums and exporting them in csv format

Second support for an HDR dissertation. The first part of the work involved understanding a database containing extracts from online forum discussions. The supervisors then helped create csv files according to criteria defined during the session, and began merging these four files in R.

Data analysis with R 

Second support session for a Master 1 student's dissertation. Firstly, the supervisors gave advice on how to access the database of interest to the participant. Secondly, the supervisors helped to use the R code to classify individuals according to socio-demographic variables. Results were visualized and interpreted using the FactoMineR package and explor. Finally, the output of the Imer function was used to interpret the fixed and random effects of the variables.