Evaluating automatic annotation of lexicon-based models for stance detection of M-pox tweets from May 1st to Sep 5th, 2022
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Manually labeling data for supervised learning is time and energy consuming; therefore, lexicon-based models such as VADER and TextBlob are used to automatically label data. However, it is argued that automated labels do not have the accuracy required for training an efficient model. Although automated labeling is frequently used for stance detection, automated stance labels have not been properly evaluated, in the previous works. In this work, to assess the accuracy of VADER and TextBlob automated labels for stance analysis, we first manually label a Twitter, now X, dataset related to M-pox stance detection. We then fine-tune different transformer-based models on the hand-labeled M-pox dataset, and compare their accuracy before and after fine-tuning, with the accuracy of automated labeled data. Our results indicated that the fine-tuned models surpassed the accuracy of VADER and TextBlob automated labels by up to 38% and 72.5%, respectively. Topic modeling further shows that fine-tuning diminished the scope of misclassified tweets to specific sub-topics. We conclude that fine-tuning transformer models on hand-labeled data for stance detection, elevates the accuracy to a superior level that is significantly higher than automated stance detection labels. This study verifies that automated stance detection labels are not reliable for sensitive use-cases such as health-related purposes. Manually labeled data is more convenient for developing Natural Language Processing (NLP) models that study and analyze mass opinions and conversations on social media platforms, during crises such as pandemics and epidemics.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it