This is a fork from Genetic-Algorithm-on-K-Means-Clustering and was developed as a task for the Evolutive Systems class of 2019 for the Computer Science Masters Course at UFPel.

The resulting paper is available here (in portuguese).

PHD program clustering using Genetic K-Means and BM25 scores

• Corpus of each PHD program made of the title and resume of all their indexed production at the official brazilian plataform Sucupira -- articles, books, presentations, thesis, and etc, with resume being available mostly for dissertation and thesis only
• Using ElasticSearch and the generated dataset, extracted the top 50 scoring keywords for each program with Okapi BM-25
• For each unique term (around 11k), gets the score of that term for each program, generating a matrix
• Minmax normalization for standardization
• Davies–Bouldin index for evaluation of each cluster
• In Genetic
  • Rank based selection
  • One point crossover

• panda
• numpy

python __main__.py

• config.txt contain control parameters
  • kmax : maximum number of clusters
  • budget : budget of how many times run GA
  • numOInd : number of Individual
  • Ps : probability of ranking Selection
  • Pc : probability of crossover
  • Pm : probability of mutation

• norm_data.csv is normalization data
• cluster_json is centroid of each cluster
• result.csv is data with labeled to each cluster