You are on page 1of 1

C.

Lu | 1036110
Web Information Retrieval & Data
End Terrorism mining | 2IMW15
D. M. Garcia Torres | 1036847
H. Qiu | 1033392
J. P. van Miltenburg | 1195537
Truth finding and rumor Analytics P. Majumder | 1036151
Prof. Dr. M. Pechenizkiy Group 20 T. Zhou | 1034799

PROBLEM - IMPACT RESULTS


National security is unfortunately a hot topic for governments all over the world Cra The crawler is API-independent and can therefore crawl tweets much
faster than the Twitter API
Terrorist organizations like Jihad and IS are finding new ways to spread uncertainty as a form of psychological impact to a wle
populace r 15 calls per 15 minutes vs 1 tweet per second
Enter will strive to limit the impact of online terrorism with the main goal:
Identify terrorist assertions and verify their truth
This way, people will not be scared by fake terrorist messages, but can be warned when real attacks happen
W The use of neural word embedding as features or as a vector
METHODOLOGY OR
representation.
Skip-Gram: In this approach the context is predicted given the
Team performance Project
D2 word. In the task of rumor identification we propose a model to
Agile work Crawl data VE perform a binary classification (rumor or not) on a single
Hackathons Word2Vec feature extraction sentence. We use word embeddings to compute an arithmetic
C mean vector for a sentence. we use this mean vector as a feature
International set-up 2 SVMs (assertion checking & rumor verification)
Data clustering vector in the learning process.
Query processing AC Input: Pool of labeled and unlabeled data, U
Initialization:
TIV Generate the small initial training dataset
E Use current labeled data set to train model M
APPLICATION FRAMEWORK LEA Repeat K-MEANS
1. Use M to label all unlabeled data in U
RNI 2. Use uncertainty sampling to select n most uncertain data and ask user for labeling
NG 3. Update training dataset
CL
US
TER
ING

RU Rumour Spreading Analysis


A rumour from a user can influence his or her follower. So for every the earliest
MO rumour spreader, we log the number of the users followers that tweet same
UR rumour later, though might not directly retweets. We believe those with a high
AN such number is the key point to spread the rumour. Or we call it rumour
spreader.
ALY Furthermore, Rumor tweets burst in the first several hours after it comes out.
SIS

CONCLUSION REFLECTION AND FUTURE WORK


Cyber terrorism is not yet that developed The data we needed was not as much available as we thought. This made developing the application more difficult.
Machines need more false data to MVP is developed, but much features can be added:
Improve performance Develop a more unified product, without missing links between different parts of the application
If social media terrorism develops: Make the program functioning on a live web pages
Develop a more UI friendly interface
Just hit Enter It is also important to note that parts of the application are not evaluated and tested for optimal results. This also needs to
be done in the next sprints.
We did handle the cultural differences quite well!

You might also like