Contextual Encryption with Natural Language Processing

Contextual Encryption Using Natural Language Processing

Various symmetric and asymmetric cryptography algorithms ensure the security of millions of documents being transferred from man in the middle attacks. Most of these algorithms contain a computationally infeasible task for any hacker trying to gain access to some information. However, with the advent of post quantum cryptography, existing security algorithms may turn to be breakable. A prominent mechanism of hiding a message from an attacker in post quantum cryptography, would be to prevent him from knowing that the message is encrypted.

Principles

A way to contextually encrypt sentences is to alter the semantics of the sentence, without bringing a change in the sentence structure. This is usually followed by a Pretty Good Privacy(PGP) algorithm encryption.
We have a precomputed database of public domain articles such as news, blogs, columns and magazines and store them in a online database. We can then replace sentences in the original message with a sentence from the public domain which has a similar structure.


Consider the sentence:
Raphael has lost his consignment in Ecuador.



The following sentence can be replaced with a sentence that has the same typed dependencies, picked from the public domain.

Example:

'Obama has delivered his speech in Seoul’

The sentence must be cohesive enough for a encrypted text reader to believe the text is not encrypted. For this, sentences and paragraphs must contain a flow of ideas that bind together a story. To establish such a relation, we use Text Tiling as described in [2].

It is also necessary for the original message and the encrypted message to be mutually exclusive. The encryption scheme would not be secure if looking at the encrypted message gives away some idea of the original message.

Contexts

A context is a hashmap that summarizes the main words of a sentence. These are usually nouns and verbs. Prepositions,adjectives and adverbs are usually not included in contexts. Once a message is to be encrypted, we compute it’s context and store it in a context map. A context map is a mapping of the word and it’s index in the sentence.

Consider the sentence:
Raphael has lost his consignment in Ecuador.

Context Map for above sentence:
Raphael: 1, Lost: 3, Consignment: 5, Ecuador: 7

Encryption Decryption Algorithm

Encryption Mechanism  

Encryption

We construct a context map(as shown above) and send it encrypted by existing encryption algorithms. Then we send the contextually encrypted message, also encrypted with a PGP algorithm. Refer to Encryption Diagram above.

Decryption Mechanism  

Decryption

The private context is decrypted first. Then the contextually encrypted message is passed to a stanford NLP parser. This generates the dependency tree for the contextually encrypted message. In this tree, we replace words, with indices in the private context, thus reconstructing the original message.

References

1. https://babutrack.files.wordpress.com/2014/12/paper-context-encryption-using-nlp.pdf
2. Text Encryption Algorithm Based on Natural Language Processing by Xianghe Jing, Yu Hao, Huaping Fei Zhijun Li, 2012 Fourth International Conference on Multimedia Information Networking and Security, IEEE 
3. Hearst, M. TextTiling: Segmenting Text into Multi-Paragraph Subtopic Passages, Computational Linguistics , 23 (1), pp. 33-64, March 1997. 
4. Stanford NLP parser, http://nlp.stanford.edu:8080/parser/