Language Model Text Classification

A language model pretrained on large amounts of text and fine-tuned on labelled examples so it can assign a category or a probability to a new free-text document.

Uses AIemerginglanguage-model

Language model text classification is worth considering when the information needed for a decision lives in free text rather than in structured fields.

It earns its place only after a model on structured data has been tried, and when the organisation has the text volume, the labels, and the team to train and maintain it.

General strengths

• Uses the information in free text that structured fields leave out.
• Pretraining on an organisation's own documents lets it learn that organisation's vocabulary and writing habits.
• Can be fine-tuned for several different predictions from the same pretrained model.

General limitations

• Needs a large body of text and labelled outcomes, plus the computing and skills to pretrain or fine-tune a model.
• The reasons behind a prediction are hard to explain, because they are spread across the wording of a long document.
• Performance can fall when writing practices, templates, or the population of documents change.
• The documents it reads are often the most sensitive data an organisation holds, which limits where it can run.

Applied to business problems

Each of these is written for its own problem. The technique is shared, the solution is not.

• Denial prediction from admission notesA language model trained on a health system's own clinical notes that reads each inpatient admission's history and physical note and flags encounters whose claims are likely to be denied, so utilisation review and clinical documentation staff can check them during the stay.Claims Denied by Insurers
• Contact-reason analysis for product fixesClassifying every ticket by the reason the customer had to contact support, such as a defect, a confusing setting, a missing feature, or a billing question, and sending the largest and fastest-growing reasons to the product and engineering teams that can remove them.Support Queues Growing Faster Than the Team