58 lines
4.2 KiB
Markdown
58 lines
4.2 KiB
Markdown
# 4 Chan Webscraper, Version 2
|
|
|
|
Consider doing your own data analysis. If you save your CSV, and make a pull request, I can add it to this repository for plotting word usage changes over time.
|
|
|
|
**Highlights:**
|
|
|
|
- Written in R.
|
|
- Objective: Datamining text
|
|
- Uses the following libraries: rvest, tidyverse, tidytext, ggplot2, wordcloud, tinytex, syuzhet, scales, reshape2, and dplyr.
|
|
- If you don't have these installed in your RStudio software, then install them.
|
|
- After installing, and running this script into your RStudio IDE, you can download all posts.
|
|
- Downloaded posts are then manipulated to show word frequencies.
|
|
- Differs from V1 by scraping all replies to OP, and has a much larger noise filter.
|
|
- Sentiment analysis is also performed.
|
|
- **X number of "posts by this ID" with graphical representation.**
|
|
|
|
**Limitations**
|
|
- 4chan will recycle poster IDs, so they are not unique identifiers. Therefore, the data mining on n-pbtid isn't fully accurate for the upper-bound, but it should be closer to representative at the lower-bound.
|
|
- Time of day scrapes based on time-zone of interest (e.g. New York posting hours) hasn't been implemented, but this can be easily solved by scraping the threads from 0900h - 1700h by intervals of 3 hours to allow old threads to die, for your target local time.
|
|
- Some synonyms will not be counted under a single word (e.g. glowie, glow, glows, glower), so manual input will need to be implemented in the v2 script. The same idea goes for plural words, **The package is trash**~~but there is another package that will take root words, and remove word modifiers (e.g. -ing, -ed, -s, etc)~~
|
|
- **It has been completed** ~~n-gram analysis has not been completed, but a simple edit to the v2 script will accomodate this. n-gram means the number of paired words (e.g. 2gram = he is, she is, it is; 3gram = i am going, she is going, he is going; etc). This can provide more context of rhetoric within the entire board scrape.~~
|
|
|
|
# html_text vs html_text2 from rvest
|
|
|
|
**I changed my mind. Using html_text2 will correct run-on words that don't naturally belong together. (e.g. webmhello)**
|
|
|
|
I did an experiment comparing the `tidy_pol_fixed2` output of text.
|
|
|
|
html_text = 21776 observations
|
|
|
|
html_text2 = 20004 observations
|
|
|
|
~~I will continue using html_text because it contains more observations, which I can later filter out the noise as needed.
|
|
There were no substantial differences that I noticed in the graphs, so retaining a greater number of observations seems better than less.~~
|
|
|
|
From the rvest::html_text website:
|
|
|
|
There are two ways to retrieve text from a element: html_text() and html_text2(). html_text() is a thin wrapper around xml2::xml_text() which returns just the raw underlying text. html_text2() simulates how text looks in a browser, using an approach inspired by JavaScript's innerText(). Roughly speaking, it converts <br /> to "\n", adds blank lines around `<p>` tags, and lightly formats tabular data.
|
|
|
|
html_text2() is usually what you want, but it is much slower than html_text() so for simple applications where performance is important you may want to use html_text() instead.
|
|
|
|
# Calculating, and Displaying Post Frequency Differences
|
|
|
|
I added an extra R script, with an example PDF output. In these files it takes the CSV the scraper auto-saves to your statistics directory, and imports the data set. Then, the top 20 positive, and top 20 negative numbers are taken from subtracting day 2 from day 1. So, if you see a negative number in the bar graph, then there were more mentions of the key word made on day 1, than on day 2. The opposite is also true: if you see a positive numbr in the bar graph, then there were less mentions made on day 1, than on day 2.
|
|
|
|
I hope you find this additional script helpful.
|
|
|
|
Please also consider sending me CSV files of your scrapes to this GitHub repository.
|
|
|
|
# Sentiment Analysis
|
|
|
|
I added a sentiment analysis which will categorize used words into several columns and count their occurances. I am using the NCR library.
|
|
|
|
|
|
# X Posts by This ID
|
|
|
|
I wrote a script for finding all unique IDs from all threads, and performed analysis on the number of posts per unique ID (e.g. 2 posts by this ID) and plotted the frequency of the number of posts across all threads. See the example for a more in-depth explanation.
|