Datasets:
Commit
·
7676290
1
Parent(s):
23b8c17
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
---
|
| 2 |
-
license:
|
| 3 |
task_categories:
|
| 4 |
- text-classification
|
| 5 |
language:
|
|
@@ -11,4 +11,65 @@ tags:
|
|
| 11 |
- url
|
| 12 |
- html
|
| 13 |
- text
|
| 14 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
task_categories:
|
| 4 |
- text-classification
|
| 5 |
language:
|
|
|
|
| 11 |
- url
|
| 12 |
- html
|
| 13 |
- text
|
| 14 |
+
---
|
| 15 |
+
# Phishing Dataset
|
| 16 |
+
|
| 17 |
+
Phishing dataset compiled from various resources for classification and phishing detection tasks.
|
| 18 |
+
|
| 19 |
+
## Dataset Details
|
| 20 |
+
|
| 21 |
+
The dataset has two columns: `text` and `label`. Text field contains samples of:
|
| 22 |
+
|
| 23 |
+
- URL
|
| 24 |
+
- SMS messages
|
| 25 |
+
- Email messages
|
| 26 |
+
- HTML code
|
| 27 |
+
|
| 28 |
+
Which are labeled as **1 (Phishing)** or **0(Benign)**.
|
| 29 |
+
|
| 30 |
+
### Source Data
|
| 31 |
+
|
| 32 |
+
This dataset is a compilation of 4 sources, which are described below:
|
| 33 |
+
|
| 34 |
+
- [Mail dataset](https://www.kaggle.com/datasets/subhajournal/phishingemails) that specifies the body text of various emails that can be used to detect phishing emails,
|
| 35 |
+
through extensive text analysis and classification with machine learning. Contains over 18,000 emails
|
| 36 |
+
generated by Enron Corporation employees.
|
| 37 |
+
|
| 38 |
+
- [SMS message dataset](https://data.mendeley.com/datasets/f45bkkt8pr/1) of more than 5,971 text messages. It includes 489 Spam messages, 638 Smishing messages
|
| 39 |
+
and 4,844 Ham messages. The dataset contains attributes extracted from malicious messages that can be used
|
| 40 |
+
to classify messages as malicious or legitimate. The data was collected by converting images obtained from
|
| 41 |
+
the Internet into text using Python code.
|
| 42 |
+
|
| 43 |
+
- [URL dataset](https://www.kaggle.com/datasets/harisudhan411/phishing-and-legitimate-urls) with more than 800,000 URLs where 52% of the domains are legitimate and the remaining 47% are
|
| 44 |
+
phishing domains. It is a collection of data samples from various sources, the URLs were collected from the
|
| 45 |
+
JPCERT website, existing Kaggle datasets, Github repositories where the URLs are updated once a year and
|
| 46 |
+
some open source databases, including Excel files.
|
| 47 |
+
|
| 48 |
+
- [Website dataset](https://data.mendeley.com/datasets/n96ncsr5g4/1) of 80,000 instances of legitimate websites (50,000) and phishing websites (30,000). Each
|
| 49 |
+
instance contains the URL and the HTML page. Legitimate data were collected from two sources: 1) A simple
|
| 50 |
+
keyword search on the Google search engine was used and the first 5 URLs of each search were collected.
|
| 51 |
+
Domain restrictions were used and a maximum of 10 collections from one domain was limited to have a diverse
|
| 52 |
+
collection at the end. 2) Almost 25,874 active URLs were collected from the Ebbu2017 Phishing Dataset
|
| 53 |
+
repository. Three sources were used for the phishing data: PhishTank, OpenPhish and PhishRepo.
|
| 54 |
+
|
| 55 |
+
#### Dataset Processing
|
| 56 |
+
|
| 57 |
+
Primarily, this dataset is intended to be used in conjunction with the BERT language model. Therefore, it has
|
| 58 |
+
not been subjected to traditional preprocessing that is usually done for NLP tasks, such as Text Classification.
|
| 59 |
+
|
| 60 |
+
_Is stemming, lemmatization, stop word removal, etc., necessary to improve the performance of BERT?_
|
| 61 |
+
|
| 62 |
+
In general, **NO**. Preprocessing will not change the output predictions. In fact, removing empty words (which
|
| 63 |
+
are considered noise in conventional text representation, such as bag-of-words or tf-idf) can and probably will
|
| 64 |
+
worsen the predictions of your BERT model. Since BERT uses the self-attenuation mechanism, these "stop words"
|
| 65 |
+
are valuable information for BERT. The same goes for punctuation: a question mark can certainly change the
|
| 66 |
+
overall meaning of a sentence. Therefore, eliminating stop words and punctuation marks would only mean
|
| 67 |
+
eliminating the context that BERT could have used to get better results.
|
| 68 |
+
|
| 69 |
+
However, if this dataset plans to be used for another type of model, perhaps preprocessing for NLP tasks should
|
| 70 |
+
be considered. That is at the discretion of whoever wishes to employ this dataset.
|
| 71 |
+
|
| 72 |
+
For more information check these links:
|
| 73 |
+
|
| 74 |
+
- https://stackoverflow.com/a/70700145
|
| 75 |
+
- https://datascience.stackexchange.com/a/113366
|