Model Comparison
Models
Model Comparison
RQ2 & RQ3: How well do baseline and advanced representations detect loneliness self-disclosure?
linguistic_only
Sparse
Handcrafted psycholinguistic features
13-d · Classical
tfidf
Sparse
TF-IDF unigram + bigram bag-of-words
15,000-d · Classical
tfidf_ling
Sparse
TF-IDF combined with 13 linguistic features
15,013-d · Classical
word2vec
Dense
Averaged Word2Vec token embeddings
200-d · Classical
sbert
Dense
Sentence-BERT contextual embeddings (all-MiniLM-L6-v2)
384-d · Classical
distilbert
Neural
End-to-end DistilBERT fine-tuning
— · Fine-tuned
Show metrics:
Best Model per Representation
F1, Precision, Recall, and ROC-AUC for the best-performing model per representation (validation set)
