rosehelfrich / nlpJobPosts

Data Science Trends Analysis Project

Overview

This project is designed to categorize current trends in the data science field by analyzing mid-level job postings related to data analytics, data science, and machine learning. The process involves two main parts: web scraping to collect data, and natural language processing to categorize the information.

Part A: Data Collection with Web Scraping

Objective: Collect and compile data from job postings to create a comprehensive dataset for analysis.

1_get_urls

• Description: Retrieves URLs from job postings under "Data + Analytics" for mid-level positions from BuiltIn.
• Output: url_list.json and url_df.csv

2_get_info

• Description: Gathers detailed information from the retrieved URLs, including job title, description, employment type, company, salary, and location.
• Output: raw_info_df.csv containing 507 unique job posts from January 2024.

Technology:

• Beautiful Soup
• Python
• Pandas
• Zyte (formerly known as Scrapy Cloud)

Part B: NLP Machine Learning Model

Objective: Clean and analyze the collected job descriptions to identify prevalent data science technologies and applications.

3_cleanMetadata

• Description: Cleans metadata for consistency and enhanced readability.
• Output: metadataCleaned.csv

4_deepClean

• Description: Performs deeper cleaning on job titles and descriptions in preparation for NLP.
• Output: dfCleaned.csv

5_jobSummary

• Description: Summarizes each job description using the facebook/bart-large-cnn model.
• Output: jobSummaries.pkl
• Technology: Transformers, BartForConditionalGeneration

6_labelPreprocess

• Description: Lemmatizes and filters tokens, identifies frequently used words, and creates additional columns for detailed analysis:
  • Tech Stack: Derived from cross-referencing techList.csv with job descriptions.
  • Applications: Originates from a list in the LinkedIn group "Artificial Intelligence, Machine Learning, Data Science & Robotics," updated with model iterations.
  • Bag of Words: Compares job descriptions against frequently occurring tokens.
• Outputs: dfPreprocessed.csv, visTokenFreq.png

7_modelCategories

• Description: Utilizes an unsupervised BERTopic model to process job descriptions based on the "Applications" column and clusters them into relevant topics.
• Outputs:
  • dfCategorized.csv: Jobs assigned a topic or labeled "general" if insufficient data.
  • visCategory.png: Visualization of data science topics.
  • Additional visualizations based on the model performance.
• Technology: BERTopic

Technology

This project utilizes the following programming languages and libraries:

• BERTopic
• ClassTfidfTransformer
• DataMapPlot: 0.2.2
• HDBSCAN
• Matplotlib: 3.8
• MaximalMarginalRelevance
• NLTK: 3.8.1
• NumPy: 1.26
• Pandas: 1.5.3
• Python: 3.11
• PyTorch: 2.2.1
• Scikit-learn (Sklearn): 1.2.2
• SciPy: 1.11
• Seaborn: 0.12.2
• SentenceTransformers: 2.3.1
• Spacy: 3.7.2
• Transformers: 4.36.2
• UMAP