Automated Identification of Microaggressions
Bibliographic record
Abstract
Building a successful community requires people with diverse backgrounds to feel included. United Nations’ Sustainable Development Goals for promoting peaceful and inclusive societies (SDG 16), emphasize the importance of this matter to our world. One element of human communication which makes marginalized groups feel excluded are Microaggressions (MAs). MAs are linguistically subtle statements that uphold stereotypes and contain offensive content. Often such comments are not said with the intent to cause harm but come from an individual’s biases. They can make the recipients of the statements feel excluded, with the offender typically unaware of the impacts on the recipient. Increasing awareness of the harms of MAs and flagging cases of MA statements when they are made can decrease these negative experiences for marginalized group members. This can educate people by examples and help them stop the unintended negative impacts of their statements.To this end, we are building an automated solution for detecting MAs. Our proposed solution uses Natural Language Processing (NLP) and machine learning (ML) models for MA detection on text data. Moreover, to better educate people and contextualize the statement, our proposed solution has the ability to classify MAs based on the aspect of one’s identity that is being targeted. For both tasks, we have built ML classifiers using NLP tokenizers. More specifically, we have built a binary classifier to detect MAs and a multi-class classifier to identify the category of MAs. We have trained several ML models for each task and compared their performance. Our results show a high accuracy of 94.21% and 89.37% for our binary and multi-class classifiers for detecting and categorizing MAs, respectively. These models outperform the state of the art models in this area. We have built tools around these models that can be used by others and our future work discusses other potential tools that can be built around these models to assist people.
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.
How this classification was reachedexpand
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 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.001 |
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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".