README.md

---
title: TJEN
license: cc-by-4.0
language:
- da
pretty_name: Tjenesteannoncer i Enevældens Nyheder
size_categories:
- 100K<n<1M
authors:
- sofuslad
- JohanHeinsen
---

# TJEN - Tjenesteannoncer i Enevældens Nyheder / ALA - Absolutist Labour Advertisements

This dataset contains 344.000 labour advertisements from the largest town newspapers in Denmark-Norway from 1750-1850.
Labour advertisements were submitted to newspaper offices by either demand-side employers or supply-side workers looking for a job,
to be printed in the next edition of the specific newspaper.
The advertisements contains information about different aspects of buying and selling labour power, and sheds light on occupational structures and the importance of origins, experience, skills, references among other things.
The advertisements constitute a highly homogenous genre with a stable format across the entire period,
allowing us to isolate the corpus with very high precision. The dataset is a subset from the ENO dataset curated by Johan Heinsen and Camilla Bøgeskov
from Aalborg University (https://huggingface.co/datasets/JohanHeinsen/ENO). \
"Tjen" is the imperative form of "at tjene" (to serve).

# Dataset creation

The dataset was created by Sofus Landor Dam (Aalborg University) and Johan Heinsen (Aalborg University).
It was created as a part of the research project "Run Away: Coercion and Autonomy c. 1800", funded by the Independent Research Fund Denmark. \
Starting with the ENO dataset, we created a binary classifier model to isolate the labour advertisements from the rest of the material. This model found 241.000
advertisements, but the original text segmentation from the ENO dataset was geared more toward "text between rubrices" rather than individual advertisements. Therefore, we
created a bespoke RegEx-based tokenizer to segment the positively identified texts into individual labour advertisements. This process resulted in around 344.000 advertisements.
We then created two other classification models to identify the intended sex of the employee and whether an advertisement belonged to the supply-side or the demand-side,
enriching the dataset with two analytical categories.

## Models used for identification and categorization

1. Labour advertisement identifier: https://huggingface.co/JohanHeinsen/Labour_advertisement_identifier
2. Gender model: https://huggingface.co/JohanHeinsen/Labour_ads_gender
3. Demand model: https://huggingface.co/JohanHeinsen/Labour_ads_demand

# Validation

We validated the dataset using a 500-row random sample. The criterium for sucess was that one row in the dataset contained only 1 labour advertisement -
even if the text was slightly fragmented. Based on this sample, the estimated success rate is 96.12% (where failure is primarily due to the existence of more than one labour advertisement in the row).
This means the true number of labour advertisements is slightly higher than the amount presented in this dataset. 

# Dataset structure

Each entry in the dataset contains the following variables:

- 'text': The HTR-recognized text from the ENO dataset.
- 'id': A unique id for the entry from the ENO dataset. Links it to the original dataset and can be searched using keywords on the online ENO-portal at: https://hislab.quarto.pub/. Here, you can also trace the original scanned newspaper edition.
- 'date': The publication date for the advertisement.
- 'newspaper': The name of the newspaper in which the advertisement was originally printed.
- 'sex': The sex of the worker. For demand-side advertisements, the sex refers to the sex of the sought after employee, not the employer. There are also instances of "both", when, for example, employers look for both male and female employees in the same advertisement. 
- 'supply_demand': Indicator for whether the advertisement comes from the supply-side or the demand-side of the labour market.

## Data sensitivity
Due to the historical nature of the data, there is no sensitive or personal information present.

## Dataset biases
The data mirrors the hierarchichal structure of 18th and 19th century Denmark-Norway. The language reflects the idea that certain people were born to serve, whilst others were born to be masters.
Any use of the dataset should take this into consideration, as it could reproduce these inherent biases.

## More information
For any inquiries related to the dataset, please contact Sofus Landor Dam at email: sld@society.aau.dk