(To)pic (Mo)delling (Co)mparison

A project aimed at comparing the results of topic modelling tasks performed with various techniques:

• gensim implementation of LDA (Latent Dirichlet Aloocation).
• prodLDA: implemented using Pyro.
• LLM_TopicModel (implemented): a generative AI powered topic model, directly outputting labels from input texts .
• BERTopic: a transformer-based topic model using BERT embeddings and clustering to extract topics from text.

The dataset used for this experiment is a corpus of the UN debates held throughout 1946-2023.

Requirements

• Ollama installation + downloading the DeepSeek model used in the code via Ollama CLI: ollama pull deepseek-r1:8b.
• packages in requirements.txt.

Having a GPU is reccomended to run LLM_TopicModel