Depression detection from social media texts such as Tweets or Facebook comments could be very beneficial as early detection of depression may even avoid extreme consequences of long-term depression i.e. suicide. In this study, depression intensity classification is performed using a labeled Twitter dataset. Further, this study makes a detailed performance evaluation of four transformer-based pre-trained small language models, particularly those having less than 15 million tunable parameters i.e. Electra Small Generator (ESG), Electra Small Discriminator (ESD), XtremeDistil-L6 (XDL) and Albert Base V2 (ABV) for classification of depression intensity using Tweets. The models are fine-tuned to get the best performance by applying different hyperparameters. The models are tested by classification of depression intensity of labeled tweets for three label classes i.e. 'severe', 'moderate', and 'mild' by downstream fine-tuning the parameters. Evaluation metrics such as accuracy, F1, precision, recall, and specificity are calculated to evaluate the performance of the models. Comparative analysis of these models is also done with a moderately larger model i.e. DistilBert which has 67 million tunable parameters for the same task with the same experimental settings. Results indicate that ESG outperforms all other models including DistilBert due to its better deep contextualized text representation as it gets the best F1 score of 89% with comparatively less training time. This study helps to achieve better classification performance of depression detection as well as to choose the best language model in terms of performance and less training time for Twitter-related downstream NLP tasks.INDEX TERMS Depression classification; transfer learning; transformer language models; public health

The study focuses on news category prediction and investigates the performance of sentence embedding of four transformer models (BERT, RoBERTa, MPNet, and T5) and their variants as feature vectors when combined with Softmax and Random Forest using two accessible news datasets from Kaggle. The data are stratified into train and test sets to ensure equal representation of each category. Word embeddings are generated using transformer models, with the last hidden layer selected as the embedding. Mean pooling calculates a single vector representation called sentence embedding, capturing the overall meaning of the news article. The performance of Softmax and Random Forest, as well as the soft voting of both, is evaluated using evaluation measures such as accuracy, F1 score, precision, and recall. The study also contributes by evaluating the performance of Softmax and Random Forest individually. The macro-average F1 score is calculated to compare the performance of different transformer embeddings in the same experimental settings. The experiments reveal that MPNet versions v1 and v3 achieve the highest F1 score of 97.7% when combined with Random Forest, while T5 Large embedding achieves the highest F1 score of 98.2% when used with Softmax regression. MPNet v1 performs exceptionally well when used in the voting classifier, obtaining an impressive F1 score of 98.6%. In conclusion, the experiments validate the superiority of certain transformer models, such as MPNet v1, MPNet v3, and DistilRoBERTa, when used to calculate sentence embeddings within the Random Forest framework. The results also highlight the promising performance of T5 Large and RoBERTa Large in voting of Softmax regression and Random Forest. The voting classifier, employing transformer embeddings and ensemble learning techniques, consistently outperforms other baselines and individual algorithms. These findings emphasize the effectiveness of the voting classifier with transformer embeddings in achieving accurate and reliable predictions for news category classification tasks.