METHOD FOR INFORMATION RETRIEVAL IN AN ENCRYPTED CORPUS STORED ON A SERVER

Abstract

A method for information retrieval in an encrypted corpus stored on a server, from a digital request calculated on a customer device, containing a sequence of terms, includes the following steps: encryption of the request on a customer computer device and transmission of same to a database management server; and homomorphic calculation, on the server, of the encrypted response to the encrypted request recorded on the server. The method further comprises an additional requesting step performed on the customer device; and presentation of the result in an ordered form of the documents, in application of the processing of the previous step. The present disclosure also relates to a method for preparing a requestable base and to a method for information retrieval in an encrypted corpus.

Description

TECHNICAL FIELD

The field of the present disclosure relates to information retrieval from a database in a form that preserves the confidentiality of data and requests.

The present disclosure relates, in particular, to the systems for processing personal data, and, in particular, health data.

BACKGROUND

Databases are an integral part of many applications, such as financial applications and medical e-health applications. Databases can be very sensitive, containing valuable data belonging to a company or individuals. The theft of sensitive data is a growing concern for individuals, companies and governments.

Databases can be made up of collections of raw files or managed using the database management system (DBMS), such as the Oracle database, MySQL, Microsoft SQL Server, etc. A database can be deployed on a server within an organization, on a virtual server in a cloud, or on a DBMS service in a cloud. Data theft is a concern for every type of deployment.

When databases are deployed on a server in a company's premises, the server is physically under the company's control. If the server is compromised or infected by malware or viruses, hackers may be able to access the raw database data file and steal data by bypassing any company access control mechanism. On the other hand, corporate database administrators have the potential to violate privacy and data integrity intentionally or accidentally, as they can access stored data to perform database management tasks. A database system can also be deployed by a company on a virtual server that runs on a cloud such as Amazon Elastic Compute Cloud (Amazon EC2). In this case, the virtual server underlying the database is physically under the control of the cloud provider, and on the company's virtual server installs DBMS to manage their databases. As in the above case, data theft also occurs in this case, if the cloud infrastructure is compromised by attackers, infected with malware or viruses, and the company's database administrators could violate database confidentiality and integrity.

In addition, if not all cloud providers are trustworthy; they can steal database data from the virtual servers provided by them.

To address these risks, solutions using Homomorphic Encryption (HE) methods for database querying have recently been considered.

Homomorphic encryption methods have been developed for search engine applications, in particular, the user sends an encrypted request to the search engine, without the latter being aware of the request received. It applies a classic search operation to find matching documents and returns the response to the user in an encrypted form. Thus, the search engine never knows the clear content of the request.

These homomorphic encryption methods also make it possible to search among encrypted files on a remote server to retrieve files that contain a term transmitted as an encrypted request to a remote server. The server applies the search without knowing the clear terms of the query and returns the result. The server never knows the requests or documents recorded in the database other than in an encrypted form. An attack on the server therefore does not create any risk with regard to the confidentiality of information, for example, personal data or health data.

Another application relates to biometrics using a database of fingerprints of persons authorized to perform an action, for example, entering a protected building. These fingerprints are naturally encrypted, because they are personal data that cannot be revoked.

Users scan their fingerprints and these are compared with those in the database. Two fingerprints of the same person taken at two different times are never strictly identical and it is therefore not possible to make a simple comparison of the encrypted fingerprints (two encrypted ones of two different fingerprints are obviously different). Thanks to homomorphic encryption, it is possible to compare encrypted fingerprints without ever decrypting them.

STATE OF THE ART

It is known in the state of the art a document that presented the basics of homomorphic encryption that is the thesis “A FULLY HOMOMORPHIC ENCRYPTION SCHEME: A DISSERTATION SUBMITTED TO THE DEPARTMENT OF COMPUTER SCIENCE AND THE COMMITTEE ON GRADUATE STUDIES OF STANFORD UNIVERSITY” in September 2009, which describes in Chapter 7 the basic principles of the application to information retrieval.

It is also known that the article “Multi-keyword Similarity Search over Encrypted Cloud Data”; Mikhail Strizhov 1 Indrajit Ray 1 29th IFIP International Information Security Conference (SEC), June 2014, Marrakech, Morocco. Springer, IFIP. This article describes a solution based on homomorphic encryption for searching for encrypted documents on a server without requiring the documents to be decrypted before searching.

Also known is the U.S. Pat. No. 8,904,171 describing a secure search and information retrieval process that includes receiving an encrypted request, creating a swapped search tree with nodes that have been swapped and encrypted.

The search tree is encrypted with a first private encryption key. The server receives a request from a customer that includes a set of keywords, and each request term is encrypted with the first private encryption key. The search is performed using a request and evaluation at each node of the tree to determine if one or more match(es) exist(s). The response is based on the match of keywords for each document and one or more encrypted node(s) with the first private encryption key.

It is also known that the European patent E2865127 describing a homomorphic encryption for database query. Numerical values are encrypted using keys and random numbers to produce encrypted text. The encrypted text is homomorphic and consists of two or more encrypted subtexts. Queries using addition, averaging and multiplication operations can be performed without deciphering the numerical values applicable to the query. Each encrypted subtext is stored in a single record and in separate attributes. The present disclosure relates to methods for the encryption and decryption, the creation of an appropriate table, the interrogation of such a database and the updating of such a database.

In addition, US documents 2010/146299 are known.

Drawbacks of the Prior Art

One of the disadvantages of homomorphic encryption is that the size of the keys and the cost (in computing time) of the operations are much larger than traditional encryption processes.

Prior art solutions have a major disadvantage resulting from the computing power required to execute, on the server, the homomorphic encryption processing each time a new document is indexed and each time a new request is made. For this reason, prior art solutions are only applicable to very limited corpora, for example, a company directory or a small set of textual documents.

Moreover, prior art solutions are limited to searching for documents on the basis of a binary criterion of presence or absence in the document of a term of the request, without allowing to propose in an efficient way a ranking of the relevance of the documents corresponding to the request; the method according to the present disclosure proposes an efficient solution to information retrieval in a large encrypted corpus.

BRIEF SUMMARY

In order to address these disadvantages, the present disclosure relates to a first aspect of a method for information retrieval in an encrypted corpus stored on a server, from a digital request calculated on a customer device, containing a sequence of terms, comprising the following steps:

• • Encryption of the request on a customer computer device;
  • Transmission of the encrypted request to a database management server;
  • Homomorphic calculation on the server of the encrypted response to the encrypted request recorded on the server;
  • Transmission of the encrypted response to the customer device;
  • Decryption of the encrypted response 50 on the customer device and extraction of the document identifiers characterized in that it comprises:
    • a) calculation steps on the customer device, when introducing a new requestable document i, for each document i belonging to the corpus, a first table TF_i and a second table Δdf_i,
      • the first table TF_i comprising, for each indexed term w of the document i, the number of occurrences of the term w in the document i, and
      • the second table Δdf_i constituted by the index of the words w present in the document i;
    • b) the encryption of document i and the table Δdf_i, as well as the encryption by a homomorphic encryption method of the table TFi, and the transmission of these three encrypted digital information to the server for recording in a storage space dedicated to a user A or a group {A_u} of users;
    • c) a step of creating or updating an index df_A on the customer device, associated with a user A, for all the documents i accessible by the user A, with the index df_A being constituted by a table indicating for each term w the number of documents i containing the term w;
    • d) requesting steps consisting of:
      • encryption, on the customer device, of a request Q_x constituted by a succession of terms, by a homomorphic encryption belonging to the same cryptosystem as the encryption applied to TF;
      • transmission of the thus encrypted request to the server for carrying out the step of homomorphic calculation and transmission of an encrypted response to the customer and decryption by the customer device;
      • an additional step, performed on the customer device, of aggregating the identifiers of the data contained in the encrypted response and in the index df_A recorded on the customer device; and
      • the presentation of the result in an orderly form of the documents in accordance with the processing of the previous step.

According to an alternative embodiment, the method includes a step of recreating, on the customer device, the index df_A from the encrypted information {Δdf_i} stored for each document i in the dedicated space of the server assigned to the user A.

According to an alternative embodiment, the calculations performed on the server are implemented in a parallel and/or distributed manner.

According to another alternative embodiment, the server (2) is constituted by a cloud platform.

The present disclosure also relates to a method for preparing a requestable base of documents i containing a sequence of terms, characterized in that it comprises the following steps:

• • a) calculation steps on the customer device, when introducing a new requestable document i, for each document i belonging to the corpus, a first table TF_i and a second table Δdf_i,
    • the first table TF_i comprising, for each indexed term w of the document i, the number of occurrences of the term w in the document i, and
    • the second table Δdf_i constituted by the presence or not of each term w in the document i;
  • b) the encryption of the document i and the table Δdfi, as well as the encryption, by a method for the homomorphic encryption of the table TFi, and the transmission of these three encrypted digital information to the server for recording in a storage space dedicated to a user A or a group {A_u} of users;
  • c) a step of creation or updating of an index df_A on the customer device, associated with a user A, for all the documents i accessible by the user A, with the index df_A being constituted by a table indicating, for each term w, the number of documents i containing the term w.

The present disclosure also relates to a method for information retrieval in an encrypted corpus stored on a server, based on a digital request calculated on a customer device, containing a sequence of terms, characterized in that it includes the following steps:

• • Encryption of the request on a customer computer device;
  • Transmission of the encrypted request to a database management server;
  • Homomorphic calculation on the server of the encrypted response to the encrypted request recorded on the server;
  • Transmission of the encrypted response to the customer device; and
  • Decryption of the encrypted response on the customer device and extraction of the document identifiers.
  • The requesting steps consisting in:
    • encryption, on the customer device, of a request Q_x constituted by a succession of terms, by a homomorphic encryption belonging to the same cryptosystem as the encryption applied to TF,
    • transmitting the thus encrypted request to the server for carrying out the step of homomorphic calculation and transmission of an encrypted response to the customer and decryption by the customer device,
    • and includes an additional step performed on the customer device, of aggregating the identifiers of the data contained in the encrypted response and in the index df_A recorded on the customer device, and
    • and a step of presenting the result in an ordered form of the documents in accordance with the processing of the preceding step.

BRIEF DESCRIPTION OF THE DRAWINGS

The present disclosure will be best understood when reading the following description that relates to a non-restrictive exemplary embodiment, while referring to the appended drawings, wherein:

FIG. 1 is a schematic view of a computer system according to the present disclosure, and

FIG. 2 represents a schematic view of data flows between the various IT resources.

DETAILED DESCRIPTION

Hardware Architecture

FIG. 1 shows a schematic view of the hardware architecture of the present disclosure;

It includes a customer computer device (1) connected to a server (2) by a computer network, for example, the Internet.

The server (2) is associated with a memory (3) for the recording of a database. The server (2) has a processor for performing digital processing.

The server (2) and the memory devices (3) are in a particular example constituted by a set of distributed resources, for example, of the “cloud” type.

Functional Architecture

FIG. 2 shows an exemplary functional architecture;

The customer device 1 provides initial processing of a document i consisting of a digital file 9 recorded in a working memory.

Optionally, each term of the document is pre-processed by known means such as “stemming”, “stop list” (deletion of current words) and any other usual linguistic processing).

Preparation of the Requestable Encrypted Files.

The initial processing is divided into three tasks.

The first task is to apply encryption to the document i with a known cryptographic method, for example, symmetric AES encryption and records an encrypted version (10) of this document on the customer device, and optionally on the server (2) or a third-party storage service. The corpus of thus defined encrypted documents constitutes the document base (32).

A second task, performed in parallel or sequentially, consists in calculating an index of occurrences of the terms present in the file 9, and in recording a table TF_i (14) of occurrences, in the form of a list of terms wj present in the document i, each of the terms w_j in this list being associated with a number corresponding to the occurrence tfi of the term w_j in the document i.

The table TF_i (14) is therefore of the {[w_i tf_u]}_j type ; for a document i.

A third task, performed in parallel or sequentially, consists in calculating a table Δdf_i 15 corresponding, for each term wj, to the presence or not of the term in the document. This table Δdf_i (15) is therefore of the {[W_j|tf_ij>0]}_j type

The table TF_i (14) is then encrypted using a homomorphic encryption method, for example, according to a method described in article Zhou, H., & Women, G. (février 2014). Efficient homomorphic encryption on integer vectors and its applications. In Information Theory and Applications Workshop (ITA), 2014 (pp. 1-9). IEEE.

The result of this encryption of the table TF_i (14) is a set of encrypted data (11). Each set of encrypted data (11) is transmitted by the customer device (1) to the server (2).

The grouping of the encrypted data 11 constitutes an encrypted database (30) of all the {TF_i}_i.

At the same time or sequentially, the table Δdf_i (15) is encrypted using a known method, for example, AES and transmitted to the server (2) to record an encrypted file (12) on the server (2).

All the encrypted files (12) recorded on the server form a database (31).

Each encrypted file (12) recorded on the server (2) makes it possible to reconstitute a table df_A 13 by decryption with an algorithm inverse to the one used for the above-mentioned encryption.

This table df_A (13) is calculated only on the customer device 1, from:

• • Either all the encrypted tables 31 recorded on the server (2), after the transmission thereof on the customer device (1);
  • or directly by updating a table df_A (13) locally recorded on the customer device 1, the update being performed each time a table Δdf_i (15) is added.

This data preparation step leads to the recording, on the server, of data that are not directly requestable and that do not reveal meaningful information about the content or the documents, especially in the event of an attack on the server or a malicious action by a privileged user.

Requesting

Requesting is performed by sending a text request (20) formed by a combination of words from the customer device (1).

Optionally, this request (20) is pre-processed by known means of “stemming”, “stop list” (deletion of current words) and any other usual linguistic processing.

The request (20) is encrypted using the same homomorphic encryption method as that used for encrypting the table TF_i (14=to obtain an encrypted request (21).

The encrypted request (21) is transmitted to the server (2) that records to make a request (40).

By applying a homomorphic calculation on the data in the encrypted database (30) and the request (40), the server (2) calculates an encrypted response (41).

This processing consists in calculating, in the encrypted domain, the number of occurrences of each term q_k of the request (40) for each known document i.

For each of the k terms q_k and for each document i, the values tf_i,j are counted for the cases where q_k corresponds to a term wj, from the encrypted database (30) of tables {[w_j; tf_ij}_i and in the encrypted space, without decrypting the variables w_j, q_k and tf_i,j.

All these counts constitute a response (41) that is transmitted to the customer device (1) and records it locally as a response (50).

The customer is then able to decrypt the response (50) to calculate a decrypted response (51).

Finally, the customer can combine the response 51 and the table df_A (13) to calculate a score TF-IDF (52) according to a known method.

This score TF-IDF (52) constitutes a classification key for the documents i in the order of relevance to the request (20).

Optionally, the customer device 1 presents the results as a search engine and allows the user to find the corresponding record.

Claims

1. A method for information retrieval in an encrypted corpus stored on a server, from a digital request calculated on a customer computer device, containing a sequence of terms, comprising the following steps: encryption of the digital request on the customer computer device;transmission of the encrypted request to a database management server;homomorphic calculation on the server of an encrypted response to the encrypted request recorded on the database management server;transmission of the encrypted response to the customer device; anddecrypting the encrypted response on the customer device and extracting document identifiers, characterized in that it comprises: a) calculation steps on the customer device, when introducing a new requestable document i, for each document i belonging to the corpus, a first table TFi and a second table Δdfi: first table TFi comprising, for each indexed term w of the document i, the number of occurrences of the term w in the document i, andthe second table Δdfi constituted by the presence or not of each term w in the document i;b) encrypting document i and the table Δdfi, as well as encrypting, by a homomorphic encryption method, the table TFi, and transmitting these three encrypted digital information to the server for recording in a storage space, dedicated to a user A or to a group {Au} of users;c) a step of creating or updating an index df_A on the customer device, associated with a user A, for all the documents i accessible by the user A, with the index df_A being constituted by a table indicating for each term w the number of documents i containing the term w,d) requesting steps comprising: encryption, on the customer device, of a request Q_x constituted by a succession of terms, by a homomorphic encryption belonging to the same cryptosystem as the encryption applied to TF;transmission of the request thus encrypted to the server for carrying out the step of homomorphic calculation and transmission of an encrypted response to the customer and decryption by the customer device.an additional step, performed on the customer device, of aggregating the identifiers of the data contained in the encrypted response and in the index df_A recorded 011 the customer device; andthe presentation of the result in an orderly form of the documents on the customer device.

2. The method of claim 1, further comprising a step of recreating, on the customer device, the index df_A from the encrypted information {Δdfi} stored in the dedicated storage space of the server assigned to the user A. of information retrieval in an

3. The method of claim 1, wherein the calculations performed on the server are implemented in a parallel and/or distributed manner.

4. The method of claim 3, wherein the server is part of a cloud platform.

5. The method of to claim 1, further comprising the following steps: a) calculation steps on the customer device, when introducing a new requestable document i, for each document i belonging to the corpus, a first table TFi and a second table Δdfi ; the first table TFi comprising, for each indexed term w of the document i, the number of occurrences of the term w in the document i, andthe second table Δdfi constituted by the presence or not of each term w in the document i,b) the encryption of the document i and the table Δdfi as well as the encryption, by a method for the homomorphic encryption of the table TFi, and the transmission of these three encrypted digital information to the server for recording in a storage space dedicated to a user A or a group {Au} of users; andc) a step of creation or updating of an index df_A on the customer device, associated with a user A, for all the documents i accessible by the user A, with the index df_A being constituted by a table indicating, for each term w, the number of documents i containing the term w.

6. A method for information retrieval in an encrypted corpus stored on a server, from a digital request calculated on a customer device, containing a sequence of terms, comprising the following steps: encryption of a request on a customer device;transmission of an encrypted request to a database management server;homomorphic calculation on the server of an encrypted response to the encrypted request stored on the server;transmission of the encrypted response to the customer device, anddecryption on the customer device of the encrypted response and extraction of document identifiers; andwherein the requesting steps comprise: encryption, on the customer device, of a request Q_x constituted by a succession of terms, by a homomorphic encryption belonging to the same cryptosystem as the encryption applied to TF;transmission of the request thus encrypted to the server for carrying out the step of homomorphic calculation and transmission of an encrypted response to the customer and decryption by the customer device; andan additional step performed on the customer device, of aggregating the identifiers of the data contained in the encrypted response and in an index df_A recorded on the customer device, anda step of presenting the result in an ordered form of the documents on the customer device.