Industry Page Classifier
Reads a short passage about a company's records and says which of 25 kinds of company it is most likely about, such as law firms, trucking and freight, manufacturers or pharma companies.
All training text is our own, from pages published at aitrainingdatabroker.com. Nothing comes from a customer.
Related tool: company-data-licensability-explorer (a Space that runs in your browser).
Why it exists
Teams that collect notes about many kinds of company need a quick way to file a passage under the right industry before a person reads it. This model does that sorting; it never decides what a company can license.
Use it
from transformers import pipeline
classify = pipeline("text-classification", model="aitrainingdatabroker/industry-page-classifier", top_k=3)
print(classify("Closing statements, escrow files and signed listing agreements from the past decade."))
How it was trained
Fine-tuned from distilroberta-base for 25 epochs on one RTX 3070. Training data: 349 passages from company-data-licensability-by-industry, company-data-licensing-faq-by-topic, which are our own published page text. It was trained on record descriptions, introductions and section paragraphs from our 25 industry pages, and tested on the questions and answers of those same pages, which it never saw in training. The scripts that built the data and trained the model are in source/; the full scores are in scores.json.
Test scores
On 87 held-out rows: accuracy 0.759, macro F1 0.746.
| Label | F1 | Count |
|---|---|---|
accounting-firms |
0.333 | 3 |
call-centers-and-bpos |
1.0 | 3 |
construction-companies |
0.8 | 3 |
consulting-firms |
0.8 | 3 |
engineering-and-architecture-firms |
0.545 | 3 |
film-and-video-production-companies |
0.75 | 4 |
financial-advisory-firms |
0.667 | 3 |
fintech-companies |
0.857 | 4 |
health-tech-companies |
0.857 | 4 |
healthcare-practices |
1.0 | 4 |
insurance-agencies |
0.667 | 3 |
investment-firms |
0.857 | 3 |
law-firms |
0.8 | 3 |
manufacturers |
0.4 | 3 |
martech-and-adtech-companies |
0.857 | 4 |
music-and-audio-companies |
1.0 | 4 |
pharma-companies |
0.667 | 4 |
property-management-companies |
0.5 | 3 |
publishing-companies |
0.75 | 4 |
real-estate-brokerages |
1.0 | 4 |
semiconductor-and-electronics-companies |
0.8 | 4 |
software-companies |
0.667 | 3 |
trucking-and-freight |
0.857 | 3 |
universities-and-research-labs |
0.889 | 4 |
warehousing-and-fulfillment-companies |
0.333 | 4 |
Limits
- 25 classes with about 14 training passages each. Related industries (fintech and financial advisory, for one) overlap, so use the top three scores, not only the first.
- Trained on one broker's page text. Passages in a very different style may be sorted worse than the score suggests.
- This model is not legal advice and not a certification. A person should review its output before anything is shared or decided, and your counsel decides what a rule requires.
Read more
- Sell company data to AI labs, by industry
- Can a law firm license data to AI?
- Can a construction company license data to AI?
License and credit
Apache 2.0. Please credit AI Training Data Broker and link to https://aitrainingdatabroker.com/ when you share or build on it.
About
AI Training Data Broker is a broker and technical advisor for companies that sell or license their data to AI labs. It guides the technical side, including export, format, de-identification and compliance, and publishes free guides, checklists and tools.
- Downloads last month
- 16
Model tree for aitrainingdatabroker/industry-page-classifier
Base model
distilbert/distilroberta-base