Update README.md

README.md

@@ -1,19 +1,33 @@
----
-license: cc-by-4.0
----
+---
+license: cc-by-4.0
+language:
+- en
+---
 
-These are the (mostly) original files for some of the forums I scraped, repacked into ZSTD-compressed Parquet files (`compression='zstd', compression_level=7, row_group_size=500`) on a **per-thread** basis instead of a per-message basis, which should make them more convenient to handle. Unlike the other archive, they shouldn't have issues with spaces between adjacent HTML tags, as that occurred in an intermediate step where single messages were extracted from the pages. Unfortunately, I only had some of the original files for the forums scraped in 2023.
+Here are the _mostly_ original files for some of the forums I scraped in the past, repacked as HTML strings + some metadata on a **per-thread** basis instead of a per-message basis, which should make them more convenient to handle. Unlike the other archive, they shouldn't have issues with spaces between adjacent HTML tags, as that occurred by mistake in an intermediate step where single messages were extracted from the pages. Unfortunately, I do not have the original files for most of the forums scraped in 2023; those would need to be scraped again.
 
-Most forums (Elliquiy being the current exception here) were originally scraped page-by-page. Pages from the same thread were grouped together and concatenated with the string `\n<!-- [NEW PAGE STARTS BELOW] -->\n` without any other further processing, obtaining one thread per row of data.
+## Scraping details
+Most forums were scraped page-by-page using the Firefox extension [Web Scraper](https://addons.mozilla.org/en-US/firefox/addon/web-scraper/), generally only picking the top-level thread message container instead of the entire page. The scraper was configured to:
+
+1. Scan forum pages one by one, retrieve thread links and visit the links one by one;
+2. Retrieve the thread page HTML, look for the "next page" link, navigate to that;
+3. Repeat step 2 in a cyclic fashion.
+
+For the process to work properly, it was important not to scrape too many threads per run and using Firefox instead of Chrome, otherwise the process could fail (very quickly in the case of Chrome). Newer versions of Web Scraper solved some issues.
+
+After exporting the scraped pages to `.csv`, pages from the same threads were grouped together and concatenated with the string `\n<!-- [NEW PAGE STARTS BELOW] -->\n` with some Python code (involving the library Pandas), without any further processing, obtaining one row of data per thread. The records were then saved into Parquet files (`compression='zstd', compression_level=7, row_group_size=500`). It's important to make sure that the row group size is not too large, or issues can arise when loading the the files later on both with Python/pandas or in the HuggingFace dataset viewer.
+
+## Usage notes
+The data is not directly usable for finetuning as-is. You will need to split the threads into messages, extract metadata, clean/convert the HTML, etc. For the NSFW forums it will be probably best to somehow avoid using usernames directly, just to be nice(r).
 
 ## Forum listing
-Only the roleplay sections (OOC included) were scraped.
+Only the roleplay sections were scraped. Note that the SFW forums can have censored slurs, for example using asterisks like `****`.
 
 | Name | Type | Scraping period | Notes
 |------|------|-------|------
 | AllTheFallen | NSFW (Lolisho) | 2024-10 |
-| Blue Moon Roleplaying Forums | NSFW | 2023-04 |
+| Blue Moon Roleplaying Forums | NSFW | 2023-03 |
 | Elliquiy | NSFW | 2023-03 to 2023-04 | "Print" version, lacks some formatting
-| Lolicit | NSFW (Lolisho) | 2023-03 | Defunct website
 | Giant in the Playground | SFW | 2025-01
-| Menewsha | SFW | 2025-01 | No activity at the time of scraping
+| Lolicit | NSFW (Lolisho) | 2023-03 | Defunct website
+| Menewsha | SFW | 2025-01 | No activity at the time of scraping. Generally low-quality postings.