Icacci presentation-isi-text categorization
0 likes28 views
This document presents a methodology for deep encrypted text categorization using recurrent neural networks and long short-term memory networks. The proposed approach avoids feature engineering by using word embeddings with deep learning. It evaluates the approach on a dataset of news articles from six sources across five categories. The results show that an LSTM network performed better than an RNN at encrypted text categorization, achieving accuracy above 80% on the test set using a 3-layer stacked LSTM model with 32 memory blocks. Future work is suggested to evaluate more complex architectures on encrypted text.
![Description of the data set and Results
The dataset is sourced from 6 online news media: The New
Zealand Herald [1], Reuters [2], The Times [3] , Yahoo News [4],
BBC [5] and The Press [6]. Business, entertainment, sport,
technology, and travel are the selected five news categories.
Table 1. Description of Data set
6](https://image.slidesharecdn.com/icacci-presentation-isi-textcategorization-180519172549/85/Icacci-presentation-isi-text-categorization-6-320.jpg)
![References
[1] www.nzherald.co.nz
[2] www.reuters.com
[3] www.timesonline.co.uk
[4] news.yahoo.com
[5] www.bbc.co.uk
[6] www.stuff.co.nz
11](https://image.slidesharecdn.com/icacci-presentation-isi-textcategorization-180519172549/85/Icacci-presentation-isi-text-categorization-11-320.jpg)
Icacci presentation-isi-text categorization
- 1. Deep Encrypted Text Categorization Vinayakumar R1, K.P Soman1 and Prabaharan Poornachandran2 1Centre for Computational Engineering and Networking (CEN), Amrita School of Engineering, Coimbatore, Amrita Vishwa Vidyapeetham, Amrita University, India. 2Center for Cyber Security Systems and Networks, Amrita School of Engineering, Amritapuri, Amrita Vishwa Vidyapeetham, Amrita University, India.
- 2. Outline • Introduction • Methodology • Description of the data set and Results • Summary • Future Work • References 2
- 3. Introduction • Text categorization focuses on classifying text to its categories and it has roots in many natural language processing (NLP) applications, mainly in content security. • Content security is an approach used by network administrator to safeguard internet security from malicious attacks with the text available in online. 3
- 4. Methodology • Text categorization has been existing as a difficult task mainly due to the traditional machine learning approaches relies on bag-of-words (BoW) model or bag-of-n-gram vectors, where unigrams, bigrams, n-grams, punctuation, stop words, emoticons, negation words, lexicons, elongated words and other delineated patterns are considered as features. • To alleviate, we have used word embedding with deep learning specifically recurrent neural network and long short-term memory. 4
- 5. Contd. Figure 1. proposed deep learning architecture for encrypted text categorization 5
- 6. Description of the data set and Results The dataset is sourced from 6 online news media: The New Zealand Herald [1], Reuters [2], The Times [3] , Yahoo News [4], BBC [5] and The Press [6]. Business, entertainment, sport, technology, and travel are the selected five news categories. Table 1. Description of Data set 6
- 7. Contd. Table 2. 5-fold cross-validation results of RNN and LSTM networks 7
- 8. Contd. Table 3. Summary of test results using 3 layer stacked LSTM network with 32 memory blocks 8
- 9. Summary • The proposed deep learning architecture for encrypted text categorization avoids the feature engineering method, thereby itself serves as a robust in handling drifting of encrypted texts in the scenario of adversarial machine learning setting. • LSTM network performed well in comparison to the RNN. 9
- 10. Future Work • we are lack behind in showing the experimental results for encrypted text categorization with more complex architecture. The reported results can be further enhanced by using more complex architecture by using an advanced hardware and training in a distributed environment. 10
- 11. References [1] www.nzherald.co.nz [2] www.reuters.com [3] www.timesonline.co.uk [4] news.yahoo.com [5] www.bbc.co.uk [6] www.stuff.co.nz 11