|
--- |
|
language: |
|
- en |
|
pipeline_tag: token-classification |
|
tags: |
|
- legal |
|
license: mit |
|
--- |
|
|
|
# Knowledge Graph Extraction for Cyber incidents |
|
This model has been finetuned with SecureBERT (https://arxiv.org/abs/2204.02685) |
|
on the CASIE dataset (https://ieeexplore.ieee.org/document/9776031). We have implemented the approach described in the CASIE paper. |
|
|
|
# Model Description |
|
The following description is taken from the CASIE paper: |
|
|
|
- An **event nugget** is a word or phrase that most clearly expresses the event occurrence. These differ from event triggers in that they can be multi-word phrases. |
|
|
|
- An **event argument** is an event participant or property value. They can be taggable entities involved in the event, such as person or organization, or attributes that specify important information, such as time or amount. |
|
|
|
- A **role** is a semantic relation between an event nugget and an argument. Each event type specifies the roles it can have and constraints on the arguments that can fill them. |
|
|
|
- A **realis** value specifies whether or not an event occurred and can be one of the three values: Actual (event actually happened), Other (failed event, future event), or Generic (an undetermined/non-specific event, such as referring to the concept of phishing attacks). |
|
|
|
# Example |
|
 |
|
|
|
# Usage |
|
``` |
|
from transformers import AutoModelForTokenClassification |
|
model = AutoModelForTokenClassification.from_pretrained("CyberPeace-Institute/Cybersecurity-Knowledge-Graph", trust_remote_code=True) |
|
|
|
input_text = "This is a Cybersecurity-related text." |
|
|
|
output = model(input_text) |
|
``` |
|
IMPORTANT! : To get the Argument to Role coreferences, use the dedicated **space**! You can download the models under "arg_role_models/". |
