Commit
•
fa8fb73
1
Parent(s):
99f75a7
Update README.md
Browse files
README.md
CHANGED
@@ -40,9 +40,10 @@ pipeline_tag: text-classification
|
|
40 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
41 |
should probably proofread and complete it, then remove this comment. -->
|
42 |
|
43 |
-
#
|
44 |
|
45 |
-
This model is a fine-tuned version of [bert-large-uncased](https://huggingface.co/bert-large-uncased) on an [phishing dataset](https://huggingface.co/datasets/ealvaradob/phishing-dataset)
|
|
|
46 |
|
47 |
It achieves the following results on the evaluation set:
|
48 |
|
@@ -57,21 +58,7 @@ It achieves the following results on the evaluation set:
|
|
57 |
BERT is a transformers model pretrained on a large corpus of English data in a self-supervised fashion.
|
58 |
This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why
|
59 |
it can use lots of publicly available data) with an automatic process to generate inputs and labels from
|
60 |
-
those texts.
|
61 |
-
|
62 |
-
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input
|
63 |
-
then run the entire masked sentence through the model and has to predict the masked words. This is different
|
64 |
-
from traditional recurrent neural networks (RNNs) that usually see the words one after the other, or from
|
65 |
-
autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a
|
66 |
-
bidirectional representation of the sentence.
|
67 |
-
|
68 |
-
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining.
|
69 |
-
Sometimes they correspond to sentences that were next to each other in the original text, sometimes not. The
|
70 |
-
model then has to predict if the two sentences were following each other or not.
|
71 |
-
|
72 |
-
This way, the model learns an inner representation of the English language that can then be used to extract
|
73 |
-
features useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a
|
74 |
-
standard classifier using the features produced by the BERT model as inputs.
|
75 |
|
76 |
This model has the following configuration:
|
77 |
|
@@ -80,15 +67,13 @@ This model has the following configuration:
|
|
80 |
- 16 attention heads
|
81 |
- 336M parameters
|
82 |
|
83 |
-
##
|
84 |
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
and HTML code. This sample variability broadens the detection range of the model and allows it to be used in
|
91 |
-
various contexts.
|
92 |
|
93 |
### Training hyperparameters
|
94 |
|
|
|
40 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
41 |
should probably proofread and complete it, then remove this comment. -->
|
42 |
|
43 |
+
# BERT FINETUNED ON PHISHING DETECTION
|
44 |
|
45 |
+
This model is a fine-tuned version of [bert-large-uncased](https://huggingface.co/bert-large-uncased) on an [phishing dataset](https://huggingface.co/datasets/ealvaradob/phishing-dataset),
|
46 |
+
capable of detecting phishing in its four most common forms: URLs, Emails, SMS messages and even websites.
|
47 |
|
48 |
It achieves the following results on the evaluation set:
|
49 |
|
|
|
58 |
BERT is a transformers model pretrained on a large corpus of English data in a self-supervised fashion.
|
59 |
This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why
|
60 |
it can use lots of publicly available data) with an automatic process to generate inputs and labels from
|
61 |
+
those texts.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
This model has the following configuration:
|
64 |
|
|
|
67 |
- 16 attention heads
|
68 |
- 336M parameters
|
69 |
|
70 |
+
## Motivation and Purpose
|
71 |
|
72 |
+
Phishing is one of the most frequent and most expensive cyber-attacks according to several security reports.
|
73 |
+
This model aims to efficiently and accurately prevent phishing attacks against individuals and organizations.
|
74 |
+
To achieve it, BERT was trained on a diverse and robust dataset containing: URLs, SMS Messages, Emails and
|
75 |
+
Websites, which allows the model to extend its detection capability beyond the usual and to be used in various
|
76 |
+
contexts.
|
|
|
|
|
77 |
|
78 |
### Training hyperparameters
|
79 |
|