User:AlgoAccountabilityBot/Enwiki Damaging Model Card
Enwiki Damaging Model Card edit
 | This model card was created and written entirely by the Algorithmic Accountability Bot account. The bot is operated by Hal Triedman. It regularly creates and updates pages about the provenance and statistical performance of machine learning models and datasets owned by the Wikimedia Foundation. Note: Any and all edits to this page will be overwritten the next time it is updated. Please put all questions and discussion of this algorithmic component in the talk page, or else contact Hal or the WMF ML team directly. |
 | This is a Tier 1 model card. That means it is generated by testing the model against a more recent, independent test set, which provides the deepest possible dive into how the model is working on MediaWiki right now. This tier also includes: - Detailed statistics of model performance at training time - Full description of model architecture - An in-depth explanation of the model rationale, owners, creators, provenance, etc. |
Qualitative Analysis edit
What is the motivation behind creating this model? edit
To prioritize review of potentially damaging edits or vandalism. The model provides a guess at whether or not a given revision is damaging, and provides some probabilities to serve as a measure of its confidence level.
Who created this model? edit
Aaron Halfaker (aaron.halfaker@gmail.com) and Amir Sarabadani (amir.sarabadani@wikimedia.de).
Who currently owns/is responsible for this model? edit
WMF Machine Learning Team (ml@wikimediafoundation.org)
Who are the intended users of this model? edit
English Wikipedia uses the model as a service for facilitating efficient edit reviews. On an individual basis, anyone can submit a properly-formatted API call to ORES for a given revision and get back the result of this model.
What should this model be used for? edit
This model should be used for prioritizing the review and potential reversion of damaging edits and vandalism on English Wikipedia.
What should this model not be used for? edit
This model should not be used as an ultimate arbiter of whether or not an edit ought to be considered damaging. It should not be used for any other English- language wiki besides English Wikipedia, and shouldn't be used for other languages.
What community approval processes has this model gone through? edit
English Wikipedia decided (note: don't know where/when this decision was made, would love to find a link to that discussion) to use this model. Over time, the model has been validated through use in the community. The links below are just an example to show what this product might look like.
Dates of consideration forums edit
What internal or external changes could make this model deprecated or no longer usable? edit
- Data drift means training data for the model is no longer usable.
- Doesn't meet desired performance metrics in production.
- English Wikipedia community decides to not use this model anymore.
How should this model be licensed? edit
Creative Commons Attribution ShareAlike 3.0
If this model is retrained, can we see how it has changed over time? edit
To my knowledge, this model has not been retrained over time — it still uses the original dataset from 2014-2015.
How does this model mitigate data drift? edit
This model does not mitigate data drift.
Which service(s) rely on this model? edit
This model is one of many models that powers ORES, the Wikimedia Foundation's machine machine learning API.
Learn more about ORES here
Which dataset(s) does this model rely on? edit
This model was trained using hand-labeled training data from 2014-2015. More details are available in the makefile of the editquality github repository.
Train dataset is available for download here
Test dataset is available for download here
Quantitative Analysis edit
How did the model perform on training data? edit
counts (n=19230):
label n ~True ~False
------- ----- --- ------- --------
True 745 --> 420 325
False 18485 --> 657 17828
rates:
True False
---------- ------ -------
sample 0.039 0.961
population 0.034 0.966
match_rate (micro=0.916, macro=0.5):
True False
------ -------
0.054 0.946
filter_rate (micro=0.084, macro=0.5):
True False
------ -------
0.946 0.054
recall (micro=0.951, macro=0.764):
True False
------ -------
0.564 0.964
!recall (micro=0.577, macro=0.764):
True False
------ -------
0.964 0.564
precision (micro=0.963, macro=0.672):
True False
------ -------
0.359 0.984
!precision (micro=0.381, macro=0.672):
True False
------ -------
0.984 0.359
f1 (micro=0.956, macro=0.707):
True False
------ -------
0.439 0.974
!f1 (micro=0.457, macro=0.707):
True False
------ -------
0.974 0.439
accuracy (micro=0.951, macro=0.951):
True False
------ -------
0.951 0.951
fpr (micro=0.423, macro=0.236):
True False
------ -------
0.036 0.436
roc_auc (micro=0.922, macro=0.922):
True False
------ -------
0.922 0.922
pr_auc (micro=0.978, macro=0.723):
True False
------ -------
0.449 0.997
How does the model perform on test/real world data across different geographies, different devices, etc.? edit
| AUC Score | Overall accuracy | Negative sample precision | Negative sample recall | Negative sample f1-score | Negative sample support | Positive sample precision | Positive sample recall | Positive sample f1-score | Positive sample support | True Positives | True Negatives | False Positives | False Negatives | True Positive Rate (Sensitivity) | True Negative Rate (Specificity) | False Positive Rate | False Negative Rate | Positive Predictive Value | Negative Predictive Value | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| All data | 0.679 | 0.735 | 0.823 | 0.843 | 0.833 | 83 | 0.380 | 0.347 | 0.363 | 23 | 8 | 70 | 13 | 15 | 0.347 | 0.843 | 0.156 | 0.652 | 0.380 | 0.823 |
| New editors (<1 year) | 0.646 | 0.697 | 0.8 | 0.8 | 0.800 | 65 | 0.380 | 0.380 | 0.380 | 21 | 8 | 52 | 13 | 13 | 0.380 | 0.8 | 0.2 | 0.619 | 0.380 | 0.8 |
| Experienced editors (>=1 year) | 0.722 | 0.9 | 0.9 | 1.0 | 0.947 | 18 | 0.0 | 0.0 | 0.0 | 2 | 0 | 18 | 0 | 2 | 0.0 | 1.0 | 0.0 | 1.0 | nan | 0.9 |
| Anonymous editors | 0.671 | 0.691 | 0.795 | 0.78 | 0.787 | 50 | 0.421 | 0.444 | 0.432 | 18 | 8 | 39 | 11 | 10 | 0.444 | 0.78 | 0.22 | 0.555 | 0.421 | 0.795 |
| Named editors | 0.606 | 0.815 | 0.861 | 0.939 | 0.898 | 33 | 0.0 | 0.0 | 0.0 | 5 | 0 | 31 | 2 | 5 | 0.0 | 0.939 | 0.060 | 1.0 | 0.0 | 0.861 |
| Mobile editors | 0.828 | 0.75 | 0.75 | 0.818 | 0.782 | 11 | 0.75 | 0.666 | 0.705 | 9 | 6 | 9 | 2 | 3 | 0.666 | 0.818 | 0.181 | 0.333 | 0.75 | 0.75 |
| Desktop editors | 0.602 | 0.732 | 0.835 | 0.847 | 0.841 | 72 | 0.153 | 0.142 | 0.148 | 14 | 2 | 61 | 11 | 12 | 0.142 | 0.847 | 0.152 | 0.857 | 0.153 | 0.835 |
Model Information edit
What is the architecture of this model? edit
{
"type": "GradientBoosting",
"version": "0.5.1",
"params": {
"scale": true,
"center": true,
"labels": [
true,
false
],
"multilabel": false,
"population_rates": null,
"ccp_alpha": 0.0,
"criterion": "friedman_mse",
"init": null,
"learning_rate": 0.01,
"loss": "deviance",
"max_depth": 7,
"max_features": "log2",
"max_leaf_nodes": null,
"min_impurity_decrease": 0.0,
"min_impurity_split": null,
"min_samples_leaf": 1,
"min_samples_split": 2,
"min_weight_fraction_leaf": 0.0,
"n_estimators": 700,
"n_iter_no_change": null,
"presort": "deprecated",
"random_state": null,
"subsample": 1.0,
"tol": 0.0001,
"validation_fraction": 0.1,
"verbose": 0,
"warm_start": false
}
}
What is the score schema this model returns? edit
{
"title": "Scikit learn-based classifier score with probability",
"type": "object",
"properties": {
"prediction": {
"description": "The most likely label predicted by the estimator",
"type": "boolean"
},
"probability": {
"description": "A mapping of probabilities onto each of the potential output labels",
"type": "object",
"properties": {
"true": {
"type": "number"
},
"false": {
"type": "number"
}
}
}
}
}