NLP_Project

Folders/Content:

• NLP_Project (Home)
  • NLP
    • RecommEng
      • NLP_RecommEng.ipynb --> Jupyter Notebook showing text analytics (NLP preprocessing, LDA topic modelling, etc.) + recommender engine designs
      • NLP_Repord.pdf --> Report explaining rationale and results of recommender engines explored.
    • Web Scraping
      • Selenium_Webscrape.pdf --> Report explaining rationale for scraping White House Press Briefings + webscraper design spec.
      • Selenium_Webscrape.ipynb --> Jupyter Notebook for webscraper design
      • white_house_news.xlsx --> Text Data Webscraped from White House Press Briefings

Recommender Engines Folder

I designed and assessed the efficacy of two types of recommender engines (collaborative filtering (CF) and content-based). Of each type, I explored three variations to see which would best recommend university courses and reading titles. To do this, NLP techniques were applied, including stop-word removal, stemming, lemmatisation, and n-gram dependency grammar, and topic modelling using Latent Dirichlet Allocation (LDA). The recommender systems were quantitatively assessed using descriptive statistics of recommendations' cosine similarity scores above a 0.4 threshold.

Web Scraping Folder

A selenium webdriver was designed to scrape text from the official US White House Press Briefings website. This was motivated to conduct future text analytics and/or topic modelling that could serve to evaluate the administrations' priorities and agenda.