Algorithmic performativity in universities’ engagement analytics

Replication package for the article in algorithmic performativity in universities’ engagement analytics.

WHAT THIS DEPOSIT CONTAINS
– data.zip
  A PDF corpus of public university governance documents about engagement / learning analytics (35 publicly available PDF documents from 24 universities in 11 countries)

– document_index.csv
  Metadata for each document: country, institution, title, document type, language, source URL, retrieval date,

– run_analysis_v18.py + requirements.txt
  Reproducible pipeline that extracts text from the PDFs, computes multilingual lexicon indicators, applies a
  polysemy safeguard for English ‘withdraw*’ (programme withdrawal vs consent withdrawal), aggregates to the
  HEI policy-stack level, computes indices and archetypes, and writes the results tables used in the manuscript. 

 And txt file with Python environment requirements.

– lexicons_v18.json
  Lexicons used for EN/NL/DE/FI quantitative analysis.

HOW TO REPRODUCE RESULTS
1) Download and unzip this deposit.
2) Install Python 3.10+ and the dependencies:
     pip install -r requirements.txt
3) Run the pipeline from the repo root:
     run_analysis_v18.py
4) Outputs will be (re)written into the analysis/ folder:
   – doc_level_metrics_v18.csv
   – hei_level_metrics_v18.csv
   – archetype_assignments_v18.csv
   – stats_summary_v18.json

NOTES ON SCOPE AND LANGUAGES
– Quantitative lexicon metrics are computed for Latin-script languages included in the lexicon file (EN/NL/DE/FI).
– Japanese/Korean documents are retained for qualitative contextual reading only and flagged in the metadata.