My undergraduate honour’s dissertation was a Pure Language Processing (NLP) analysis mission. It centered on multilingual textual content technology in under-represented languages. As a result of current metrics carried out very poorly on evaluating outputs of fashions educated on the dataset I used to be utilizing, I wanted to coach a realized regression metric.
Regression could be helpful for a lot of textual duties, comparable to:
- Sentiment evaluation: Predict the energy of optimistic or unfavourable sentiment as a substitute of straightforward binary classification.
- Writing high quality estimation: Predict how excessive the standard of an editorial is.
For my use case, I wanted the mannequin to attain how good one other mannequin’s prediction was for a given process. My dataset’s rows consisted of the textual enter and a label, 0 (unhealthy prediction) or 1 (good prediction).
- Enter: Textual content
- Label: 0 or 1
- The duty: Predict a numerical likelihood between 0 and 1
However transformer-based fashions are often used for technology duties. Why would you utilize a pre-trained LM for…
