Method and electronic device for generating semantic representation of document to determine data security risk

1. A method for generating semantic representation of a document using an electronic device ( 100 ) to determine data security risk associated with the document, the method comprising:

receiving, by a document semantics controller ( 160 ) of the electronic device ( 100 ), a document in an electronic form, wherein the document comprises a plurality of content;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), raw text from the plurality of content;

generating, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a plurality of sentence blocks of a predefined size using the raw text;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), at least one embeddings for each of the plurality of sentence blocks;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks;

generating, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document to determine the data security risk associated with the document;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), at least one attribute of a plurality of attributes associated with a user requesting access to the document, wherein at least one attribute indicates a user security risk profile;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a document security risk profile based on the semantic representation of the document and semantic representation of neighboring documents; and

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), whether the user security risk profile matches the document security risk profile to determine access to the document.

2. The method as claimed in claim 1 , further comprising:

receiving, by the document semantics controller ( 160 ) of the electronic device ( 100 ), at least one request from at least one user to access the document in the electronic form; and

performing, by the document semantics controller ( 160 ) of the electronic device ( 100 ), one of:

allow the user to access the document, in response to determining that the user security risk profile matches the document security risk profile, and

deny the user access to the document, in response to determining that the user security risk profile does not match the document security risk profile.

3. The method as claimed in claim 1 , wherein determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the raw text from the plurality of content comprises:

extracting, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the raw text from the plurality of content;

performing, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a character filtration on the raw text to eliminate predefined types of characters from the raw text;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), at least one word and at least one sentence in the raw text extracted from the document which is not corresponding to the semantic representation of the document;

eliminating, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the at least one word and the at least one sentence not corresponding to the semantic representation of the document; and

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the raw text from the plurality of content.

4. The method as claimed in claim 1 , wherein generating, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the plurality of sentences blocks using the raw text comprises:

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), that a number of extracted raw text meets a predefined threshold for the raw text;

generating, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a plurality of tokens by combining a set of the raw text extracted from the document; and

generating, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the plurality of sentence blocks based on sentence splicing of the plurality of tokens, wherein each sentence block of the plurality of sentence blocks comprises a predefined number of tokens.

5. The method as claimed in claim 1 , wherein the at least one embeddings for each of the plurality of sentence blocks is determined using a deep-learning language model.

6. The method as claimed in claim 1 , wherein determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks comprises:

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), an embedding matrix comprising the at least one embeddings for each of the plurality of sentence blocks;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a document embedding by aggregating the at least one embeddings for each of the plurality of sentence blocks in the embedding matrix; and

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document based on the document embedding.

7. The method as claimed in claim 1 , wherein determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks comprises:

generating, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a similarity graph using the plurality of sentence blocks;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a page-rank score for the each of the sentence blocks of the plurality of sentence blocks;

performing, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a maximum margin relevance scoring on the page-rank score determined for each of the sentence blocks;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a set of sentence blocks with highest maximum margin relevance scoring;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a document embedding by aggregating the set of sentence blocks with the highest maximum margin relevance scoring; and

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document based on the document embedding.

8. The method as claimed in claim 1 , wherein determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks comprises:

identifying, by the document semantics controller ( 160 ) of the electronic device ( 100 ), embeddings of a set of sentence blocks of the plurality of sentence blocks that are outlying;

filtering, by the document semantics controller ( 160 ) of the electronic device ( 100 ), to exclude the set of sentence blocks that are outlying;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a document embedding by aggregating the embeddings of remaining sentence blocks after filtering the set of sentence blocks that are outlying; and

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document based on the document embedding.

9. The method as claimed in claim 1 , wherein determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks comprises:

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a plurality of clusters comprising each of the embeddings of each of the plurality of sentence blocks;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), at least one cluster of the plurality of clusters that is a semantic noise;

filtering, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the at least one cluster of the plurality of clusters that is the semantic noise;

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), a document embedding by aggregating the embeddings of remaining sentence blocks after filtering the semantic noise; and

determining, by the document semantics controller ( 160 ) of the electronic device ( 100 ), the semantic representation of the document based on the document embedding.

10. An electronic device ( 100 ) for generating semantic representation of a document to determine data security risk associated with the document, the electronic device ( 100 ) comprising:

a memory ( 120 );

a processor ( 140 ) coupled to the memory ( 120 );

a document semantics controller ( 160 ) coupled to the memory ( 120 ) and the processor ( 140 ), and configured to:

receive a document in an electronic form, wherein the document comprises a plurality of content;

determine raw text from the plurality of content;

generate a plurality of sentence blocks of a predefined size using the raw text;

determine at least one embeddings for each of the plurality of sentence blocks;

determine the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks;

generate the semantic representation of the document to determine the data security risk associated with the document;

determining at least one attribute of a plurality of attributes associated with a user

requesting access to the document, wherein at least one attribute indicates a user

security risk profile;

determining a document security risk profile based on the semantic representation of the document and semantic representation of neighboring documents; and

determining whether the user security risk profile matches the document security risk profile to determine access to the document.

11. The electronic device ( 100 ) as claimed in claim 10 , wherein the document semantics controller ( 160 ) is further configured to:

receive at least one request from at least one user to access the document in the electronic form; and

perform one of:

allow the user to access the document, in response to determining that the user security risk profile matches the document security risk profile, and

deny the user access to the document, in response to determining that the user security risk profile does not match the document security risk profile.

12. The electronic device ( 100 ) as claimed in claim 10 , wherein the document semantics controller ( 160 ) is configured to determine the raw text from the plurality of content by:

extract the raw text from the plurality of content;

perform a character filtration on the raw text to eliminate predefined types of characters from the raw text;

determine at least one word and at least one sentence in the raw text extracted from the document which is not corresponding to the semantic representation of the document;

eliminate the at least one word and the at least one sentence not corresponding to the semantic representation of the document; and

determine the raw text from the plurality of content.

13. The electronic device ( 100 ) as claimed in claim 10 , wherein the document semantics controller ( 160 ) is configured to generate the plurality of sentences blocks using the raw text by:

determine that a number of extracted raw text meets a predefined threshold for the raw text;

generate a plurality of tokens by combining a set of the raw text extracted from the document; and

generate the plurality of sentence blocks based on sentence splicing of the plurality of tokens, wherein each sentence block of the plurality of sentence blocks comprises a predefined number of tokens.

14. The electronic device ( 100 ) as claimed in claim 10 , wherein the at least one embeddings for each of the plurality of sentence blocks is determined using a deep-learning language model.

15. The electronic device ( 100 ) as claimed in claim 10 , wherein the document semantics controller ( 160 ) is configured to determine the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks by:

determine an embedding matrix comprising the at least one embeddings for each of the plurality of sentence blocks;

determine a document embedding by aggregating the at least one embeddings for each of the plurality of sentence blocks in the embedding matrix; and

determine the semantic representation of the document based on the document embedding.

16. The electronic device ( 100 ) as claimed in claim 10 , wherein the document semantics controller ( 160 ) is configured to determine the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks by:

generate a similarity graph using the plurality of sentence blocks;

determine a page-rank score for the each of the sentence blocks of the plurality of sentence blocks;

perform a maximum margin relevance scoring on the page-rank score determined for each of the sentence blocks;

determine a set of sentence blocks with highest maximum margin relevance scoring;

determine a document embedding by aggregating the set of sentence blocks with the highest maximum margin relevance scoring; and

determine the semantic representation of the document based on the document embedding.

17. The electronic device ( 100 ) as claimed in claim 10 , wherein the document semantics controller ( 160 ) is configured to determine the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks by:

identify embeddings of a set of sentence blocks of the plurality of sentence blocks that are outlying;

filter to exclude the set of sentence blocks that are outlying;

determine a document embedding by aggregating the embeddings of remaining sentence blocks after filtering the set of sentence blocks that are outlying; and

determine the semantic representation of the document based on the document embedding.

18. The electronic device ( 100 ) as claimed in claim 10 , wherein the document semantics controller ( 160 ) is configured to determine the semantic representation of the document based on the at least one embeddings for each of the plurality of sentence blocks by:

determine a plurality of clusters comprising each of the embeddings of each of the plurality of sentence blocks;

determine at least one cluster of the plurality of clusters that is a semantic noise;

filter the at least one cluster of the plurality of clusters that is the semantic noise;

determine a document embedding by aggregating the embeddings of remaining sentence blocks after filtering the semantic noise; and

determine the semantic representation of the document based on the document embedding.