Skip to content

statarczuk16/Panacirce

Repository files navigation

Copyright Notice:

© 2021 statarczuk16 The files within this repository are copyrighted by statarczuk16 unless otherwise noted and may not be redistributed without written permission.

About

This is a Python program that uses natural language processing and a Latent Dirichlet allocation model to scrape data from a set of Reddit subreddits and then find ten topics that summarize what people on those subreddits talk about. The output is presented as a wordcloud representing the words that make up each topic.

Install/Use

  • Run the python scripts in the following order:
  • panacirce_scraper.py - scrapes data from Reddit and outputs to "scraped_data" folder
    • Scrapes data from minTimeOfInterest to maxTimeOfInterest
    • Scrapes data from assigned subreddits (default: subreddits = [['Chicago','flair:"Ask CHI"'],['Boston','flair:tourism'],['NYC',""],['LosAngeles',""],['Seattle',""]])
      • Second argument is a flair filter, though it is not currently used by the scraper
      • Submissions that are not "self" (text-only submissions as opposed to links) and auto-moderators posts are filtered out
      • Hyperlinks are extracted, though not used for anything yet.
  • panacirce_cleaner.py - cleans data created by panacirce_cleaner
    • Goes through each .csv file in scraped_data and consolidates each reddit submission into one row of data containing
      • phrased self text submissions
      • permalink to subsmission
      • a column for each most common wordcloud
  • panacirce_lda.py - uses LDA to create a word cloud from all cleaned_data and displays

Output Example

Future

  • panacirce_model_builder
    • uses cleaned_data which has been manually identified with a topic to build a predictive model
      • To create training data:
        • Set panacirce_cleaner.py :: mode to "1", which will output occurrences for each most common word
        • eg "these are most common words" -> "these 5 are 3 most 2 common 1 words 1"
        • replace left most column with 1 or 0 indicating whether or not the list of words matches the topic you want to detect.
        • eg, if you see "restaurant" or "burger" or whatever, a 1 would indicate this list of words indicates the topic of food
        • place resulting file in "training_data" folder and run panacirce_model_builder.py

About

Scrapes data from input Reddit subreddits and uses natural language process and an LDA model to determine what topics those subreddits are concerned with

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages