Unsupervised [randomly responding] survey bot detection: In search of high classification accuracy
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
While online survey data collection has become popular in the social sciences, there is a risk of data contamination by computer-generated random responses (i.e., bots). Bot prevalence poses a significant threat to data quality. If deterrence efforts fail or were not set up in advance, researchers can still attempt to detect bots already present in the data. In this research, we study a recently developed algorithm to detect survey bots. The algorithm requires neither a measurement model nor a sample of known humans and bots; thus, it is model agnostic and unsupervised. It involves a permutation test under the assumption that Likert-type items are exchangeable for bots, but not humans. While the algorithm maintains a desired sensitivity for detecting bots (e.g., 95%), its classification accuracy may depend on other inventory-specific or demographic factors. Generating hypothetical human responses from a well-known item response theory model, we use simulations to understand how classification accuracy is affected by item properties, the number of items, the number of latent factors, and factor correlations. In an additional study, we simulate bots to contaminate real human data from 36 publicly available datasets to understand the algorithm’s classification accuracy under a variety of real measurement instruments. Through this work, we identify conditions under which classification accuracy is around 95% or above, but also conditions under which accuracy is quite low. In brief, performance is better with more items, more categories per item, and a variety in the difficulty or means of the survey items.
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 | 0.008 | 0.005 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| 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