JobTrends Dataset – Online Job Postings Across Platforms

This repository contains the JobTrends corpus, a collection of online job postings gathered from multiple platforms, along with cleaned, labeled, and subsetted versions of the data. The dataset is intended for labor market analysis, NLP on job descriptions, and related research tasks. Contents Research paper / primary documentation describing the corpus, collection methodology, and (in detail) the label schema. - `data/` - `raw/` – Unprocessed, scraped job postings per source platform. - `hahu_raw.json` - `linkedin_raw.json` - `telegram_raw.json` - `processed/` – Cleaned and (where applicable) labeled job postings. - `all_clean_labeled.json` – Combined cleaned + labeled data from all platforms. - `by_source/` – Cleaned + labeled data split by platform. - `hahu_clean_labeled.json` - `linkedin_clean_labeled.json` - `telegram_clean_labeled.json` - `category_subsets/` – Filtered subsets for specific categories (e.g., computer science only). - `cs_only_clean_labeled.json` - `cs_only_hahu.json` - `cs_only_linkedin.json` - `cs_only_telegram.json` File naming conventions - `*_raw.json` – Raw, unprocessed job postings scraped from a specific source. - `*_clean_labeled.json` – Cleaned postings with labels (e.g., categories, skills, or other attributes), as defined in the paper. - `cs_only_*` – Subsets restricted to a particular category (likely computer-science-related jobs). How to use the dataset 1. Exploratory analysis - Start with `data/processed/all_clean_labeled.json` for a unified view. - Use `data/processed/by_source` if you need to compare platforms. 2. Category-specific experiments - Use `data/processed/category_subsets` (e.g., `cs_only_*` files) for focused modeling or analysis. 3. Rebuilding your own preprocessing - Use `data/raw/*.json` as input and reproduce/modify preprocessing steps, following the methodology described in the PDF.