This analysis folder contains analysis and experimentation notebooks for 5 themes:
tk_analysis/
sentence_classifier/
skills_extraction/
skills_taxonomy/
skills_taxonomy_application/
Outputs (figures and data) from analysis are saved to a corresponding folder in the outputs/
folder.
This folder contains two notebooks:
TextKernel Data.ipynb
- Provides a summary of TextKernel dataset.TextKernel Data Sample for Skills.ipynb
- Comparison of our sample of TextKernel data to all data.
Outputs are in outputs/tk_analysis
.
This folder contains three markdown files:
Sentence Splitter Experiments.md
- Notes on experiments for the initial skills classifierskills_classifier_2.0_experiments.md
- Notes on experiments for the skills classifier after labelling more dataskills_codebook.md
- A codebook for labellers to follow when labelling sentences as either skill sentences or not.
Within notebooks
, there's a number of .ipynb
and .py
files related to experimentation:
Skills Classifier 1.0 - Analyse sentence predictions.py
- initial experiments from skills classifier 1.0Skills Classifier 1.0 - Doccano Baseline Classifier.py
- initial experiments from skills classifier 1.0Skills Classifier 1.0 - TextKernel Automated Training Data.py
- initial experiments from skills classifier 1.0Skills Classifier 1.0 - Sentence Classifier.py
- initial experiments from skills classifier 1.0Skills Classifier 2.0 - improve_classifier.ipynb
- experiments using more labelled data from skills classifier 2.0
In this folder we have two scripts for various bits of analysis and figure plotting after extracting skills:
Multi-skill sentences.pynb
- Analysis of multi-skill sentence - how long are sentences typically when they have multiple skills in.Experiment - Data reduction sample size.ipynb
- Analysis of finding a good sample size of embeddings to fit the reducer class to.Effect of sample size.ipynb
- Investigate the effect of sample size of skill sentences and how many words are in the vocab.Experiment - Clustering parameters.ipynb
- Analysis of which are the optimal clustering parameters to extract skills from.Effect of merging clusters distance threshold.ipynb
- Investigation of the distance threshold for which two skill clusters should be merged into one.Skills Extraction Analysis and Figures.ipynb
- Various analysis and figure generation of the skills extracted. Outputs are inoutputs/skills_extraction/figures/..
Extracted skills - 2021.11.05.py
In this folder we also have experimentation notebooks showing 4 approaches for skills extraction approaches, including:
- Network approach
- Transformers sentence embeddings approach - no masking
- Word2vec approach
- Transformers sentence embeddings approach - masking
The last approach was what we used in the final pipeline (refactored in skills_taxonomy_v2/pipeline/skills_extraction/
).
In this folder we perform some analysis on the skills taxonomy created when running skills_taxonomy_v2/pipeline/skills_taxonomy/build_taxonomy.py
.
Evaluate hierarchy.ipynb
- Evaluate hierarchy based on popular skill groups for job titles. Output csvs stored inoutputs/skills_taxonomy/evaluation/
.Tranversal Skills.ipynb
- Identify the most and least transversal skills and skill groups. Outputs inoutputs/skills_taxonomy/transversal/
.Renaming sample of skill groups.ipynb
- Manually creating names for some of the skill groups, outputsskills_taxonomy_v2/utils/2021.09.06_level_a_rename_dict.json
which is used in other notebooks.Skills Taxonomy Analysis and Figures.ipynb
- this notebook provides some visualisation and analysis of it. Outputs are stored inoutputs/skills_taxonomy/figures/2021.09.06/
.Plot interactive hierarchy.ipynb
- this notebook plot interactive plots of the hierarchy for interrogation.
This folder contains two notebooks to analyse the skills taxonomy in application to the job location and whether the job was advertise pre or post COVID.
Application - Geography.ipynb
See how different skill groups in the taxonomy relate with location of the job advert. Outputs inoutputs/skills_taxonomy_application/region_application/
.Application - COVID.ipynb
See how different skill groups in the taxonomy relate with whether the job advert was out pre or post COVID. Outputs inoutputs/skills_taxonomy_application/covid_application
.
This folder also contains the script locations_to_nuts2.py
to convert longitude and latitude coordinates from the job adverts to NUTS2 regional classifications - this was neccessary for use in Application - Geography.ipynb
.