A simple yet powerful library to predict toxicity/profanity of a review/comment or list of reviews/comments.
cuss_inspect is a logistic regression based model trained on 180K+ reviews and tested on 24K+ reviews. The library does not uses any specific wordlist/swear-words-list but is able to detected most of the swear words easily.
| 1 Prediction (ms) | 10 Predictions (ms) | 100 Predictions (ms) | 1000 Predictions (ms) | 10000 Predictions (ms) | |
|---|---|---|---|---|---|
| cuss_inspect | 0.2 | 0.3 | 0.8 | 4.3 | 24.7 |
The accuracy,precision and recall are quite impressive as compared to other models. Logistic regression for text classification outperforms many other classifcation algorithms such as SVC,Decision Tree and Naive Bayes.
| Precision | Recall | F1 Score | |
|---|---|---|---|
| 0 | 0.83 | 0.94 | 0.88 |
| 1 | 0.99 | 0.96 | 0.97 |
| Accuracy | 0.95 | ||
| macro avg | 0.91 | 0.95 | 0.93 |
| weighted avg | 0.96 | 0.96 | 0.96 |
$ pip install cuss_inspect
from cuss_inspect import predict, predict_prob
# for simple string
text_0 = "this is simple review. you have done a good job"
print(predict(text_0))
# [0]
print(predict_prob(text_0)
# [0.05]
text_1 = "son of a bitch"
print(predict(text_1))
# [1]
print(predict_prob(text_1)
# [1.]
# for list of inputs
test = ['who are you?' , 'what do you want?' , 'son of a dog' , 'how the hell can you say that' , 'fuck it']
print(predict(test))
# [0 0 1 1 1]
print(predict_prob(test))
# [0.12 0.22 0.55 0.96 1.]*predict() and predict_prob return numpy arrays.