Tag: classical-ml
All the articles with the tag "classical-ml".
- NLPClassical MLText classification
Classifying user feedback when 97% of it is one class
An honest look at categorising free-text user feedback with classical ML: upsampling, two text views, two feature sets, five model families, and a 2,000-item human review that showed why accuracy was the wrong number to watch.
- NLPClassical MLAnalytics
Summarising thousands of support chats before LLMs: what extractive NLP got right
How we turned a year of advertiser support chats into readable summaries, themes and topics for monthly business reviews, using cleaning, TF-IDF sentence scoring and topic modelling, and saved about $300K.
- ForecastingClassical MLTime series
Forecasting support demand for markets with almost no history
How we forecast monthly support volumes by language and product area to plan vendor capacity for new markets: comparable markets, a 14-model tournament per series, ensembles, sparse-series rules and a pandemic year to deal with. Saved about $70K.
- Causal inferenceClassical MLAnalytics
Did our advice actually help? Measuring customer impact with synthetic controls
How we measured whether proactive advice from account and support teams changed advertisers' results, using synthetic controls to build a fair comparison for every engaged customer, automated end to end.