Live Inference Demo

Classify Reddit posts for loneliness self-disclosure using the best-trained model

Predictions run on the HuggingFace Spaces CPU instance using the best-performing model from the pipeline. The decision threshold is ROC-optimised (Youden J statistic) rather than the default 0.5.

Single Post Classification

Enter one Reddit post to classify

0/2000

Try an example:

How it works

1.Text is cleaned: URL removal, lowercasing, emoji removal, normalisation
2.Cleaned text is tokenised and lemmatised with spaCy (en_core_web_sm)
3.The best-performing representation is applied (e.g. SBERT encoding or TF-IDF)
4.The trained classifier produces a probability score for the Lonely class
5.The ROC-derived optimal threshold is applied instead of the default 0.5
6.Result: label (0/1), confidence, and threshold are returned