METHODS AND SYSTEMS TO FINGERPRINT TEXTUAL INFORMATION USING WORD RUNS
The present invention provides methods and systems to enable fast, efficient, and scalable means for fingerprinting textual information using word runs. The present system receives textual information and provides algorithms to convert the information into representative fingerprints. In one embodiment, the fingerprints are recorded in a repository to maintain a database of an organization's secure data. In another embodiment, textual information entered by a user is verified against the repository of fingerprints to prevent unauthorized disclosure of secure data. This invention provides approaches to allow derivative works (e.g., different ordering of words, substitution of words with synonyms, etc.) of the original information to be detected at the sentence level or even at the paragraph level. This invention also provides methods and systems for enhancing storage and resource efficiencies by providing approaches to optimize the number of fingerprints generated for the textual information.
This application claims the benefit of U.S. application Ser. No. 12/177,043, entitled “METHODS AND SYSTEMS TO FINGERPRINT TEXTUAL INFORMATION USING WORD RUNS,” filed Jul. 21, 2008, and is hereby incorporated by reference.
BACKGROUND OF THE INVENTIONWith the rapid increase and advances in digital documentation capabilities and document management systems, organizations are increasingly storing important, confidential, and secure information in the form of digital documents. Unauthorized dissemination of this information, either by accident or by wanton means, presents serious security risks to these organizations. Therefore, it is imperative for the organizations to protect such secure information and detect and react to any secure information (or derivatives thereof) from being disclosed beyond the perimeters of the organization.
Additionally, the organizations face the challenge of categorizing and maintaining the large corpus of digital information across potentially thousands of data stores, content management systems, end-user desktops, etc. It is therefore valuable to the organization to be able to identify and disregard redundant information from this vast database. At the same time, it is critical to the organization's security to be able to identify derivative forms of the secure data (e.g., changes to the sentence structure or word ordering at the sentence/paragraph level, use of comparable words in the form of synonyms/hpernyms, varied usage of punctuations, etc.) and identify any unauthorized disclosure of even such derivative forms. Therefore, any system or method built to accomplish the task of preventing unauthorized disclosure would have to address these two conflicting challenges.
One method to detect similar data is by examining the database at the file level. This can be done by comparing the file names, or by comparing the file sizes, or by doing a checksum of the contents of the file. However, even minor differences between the two files will evade a detection method.
Other prior art solutions teach partial text matching methods using various k-gram approaches. In such approaches, text-characters of a fixed length, called k-grams, are selected from the secure text. These k-grams are hashed into a number called a fingerprint. In order to increase storage and resource efficiency, the various prior art approaches propose different means by which the k-grams can be sampled so as to store only a representative subset of the k-grams. However, these prior art approaches suffer a number of disadvantages. For example, these prior systems are not robust against derivative works of the secure text. Additionally, the k-gram approaches are not suitable for use in multi-language environments (e.g., a document containing a mixture of Mandarin and English words). Also, using a character-based approach as opposed to a word-based approach does not allow for the exclusion of common or repeated words, thus resulting in overall memory and resource inefficiencies.
SUMMARY OF THE INVENTIONMethods and systems to provide fast, efficient, and scalable means to fingerprint textual information using word runs is presented. In one embodiment, the present invention provides methods and systems to efficiently fingerprint vast amounts textual information using word runs and allows these fingerprints to be recorded in a repository. This embodiment comprises a receiving module to receive textual information from a plurality of input sources. It further includes a normalization module to convert the textual information to a standardized canonical format. It then includes a word boundary detection module that detects the boundaries of words in a language-independent manner. It additionally includes a word hash list generator, where each word of the textual information is converted to a representative hash value. Several means are provided by which the word hash list can be post-processed to significantly improve memory and resource efficiencies. Examples of such post-processing include eliminating certain stop words, grouping certain categories of words and mapping them to one hash value, etc. This embodiment also includes a fingerprint generator, which generates fingerprints by applying hash functions over the elements of the word hash list. The fingerprint generator uses algorithms to generate only a representative subset of the entire word hash list, thus further enhancing the memory and resource efficiencies of the system. A repository, which can include any database or storage medium, is then used to record the fingerprints generated for the vast amounts of textual information received at the receiver module.
In another embodiment, the present invention provides methods and systems to receive any textual information entered in by a user and to match such information against a fingerprint database. This embodiment includes a receiving module to receive the user-entered information, a normalization module to convert the textual information to a standardized canonical format, a language independent word boundary detector to detect the start and end of each word, a word hash list generator to generate representative hash values to every word, and a fingerprint generator that uses a sliding window to efficiently generate a representative subset of fingerprints for the received user information. This embodiment finally matches the generated fingerprints against a previously developed fingerprint database, and provides alerts to the user in the event that any secure or protected information is indeed being disclosed.
Other embodiments of the present invention allow the fingerprints to be generated without any preference for language, and without any linguistic understanding of the underlying text, thereby allowing the invention to be applied to most languages. The present invention also provides embodiments where the fingerprints are made independent of the presence of punctuations, the ordering of words within sentences or paragraphs, and/or the presence of upper and lower case characters in the words. By doing this, the present invention allows word runs to be matched and detected both at sentence and paragraph level. Additionally, this invention allows even derivative works of the original text (e.g., changes to the sentence structure or word ordering at the sentence/paragraph level, use of comparable words in the form of synonyms/hpernyms, varied usage of punctuations, removal or addition of certain stop words, etc.) to be matched and detected.
These and other objects, features and characteristics of the present invention will become more apparent to those skilled in the art from a study of the following detailed description in conjunction with the appended claims and drawings, all of which form a part of this specification. In the drawings:
The present invention may be embodied in several forms and manners. The description provided below and the drawings show exemplary embodiments of the invention. Those of skill in the art will appreciate that the invention may be embodied in other forms and manners not shown below. It is understood that the use of relational terms, if any, such as first, second, top and bottom, and the like are used solely for distinguishing one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions.
Information may be received from several sources. In one embodiment, the source could include confidential, important, or secure information maintained by an organization, where such information needs to be recorded or registered into a database. In another embodiment, the source could include any information entered by a user having access to an organization's secure information, where such information would need to be matched and inspected against an existing database of secure information. The textual information received from either of these sources includes a plurality of words. Such words are may be present as a plurality of text-characters, with one word distinguished from another by the presence of at least one space-character. The words may also be present as plurality of text-characters, with one word separated from another by the use of punctuation marks.
The received information is first normalized to a canonical text representation 120. This can be done by converting the computer files containing the textual information into one of several raw text formats. One example of such normalization is to convert a PDF (Portable Document Format) file into a Unicode transformation format file. An example of a Unicode transformation format is UTF-16.
In one embodiment, the present invention uses a word boundary detector 125 to detect the separation of one word from a preceding or following word. The word boundary detector 125 uses a state machine and employs character-classes that dictate boundary analysis across languages. In this embodiment, the state machine utilizes mapping tables to determine what character-class a particular character belongs to. By mapping the current character and comparing that against the mapping of the previous character, the detector determines whether a word has just started or ended. Because the character-classes include generic word separators or delimiters common to most languages, this word boundary detector can be used in a language independent manner. Additionally, the characters within the words may be case-folded, such that the word-value hash assigned to a particular word does not depend upon whether the word has any upper or lower-case characters. Note that the case folding can be done at any time prior to the generation of a word hash list.
In one embodiment, the word-value hashes are computed as 32-bit unsigned integers. This is advantageous because the computation of the word-value hashes could then use 32-bit arithmetic, which would be much faster than performing 64-bit arithmetic on 32-bit architectures.
The post processing steps of
The present invention also discloses methods by which the hash function can optionally be made word-order independent.
The following description of
The received information is converted to a normalized text format within the text normalization module 920. In one embodiment, this text normalization module is any computer implemented software application that can be used to convert the data file from a non-Unicode format to a Unicode text format. A person of skill in the art can immediately appreciate the wealth of third-party software applications that are readily available to perform this normalization.
The received normalized information is then transmitted to a word detector 930. In one embodiment, the word detector could be a computer implemented software for running an algorithm to detect the boundaries of each word. In this embodiment, the word boundary detector uses a state machine and employs character-classes that dictate boundary analysis across languages. Here, the state machine utilizes mapping tables to determine what character-class a particular character belongs to. By mapping the current character and comparing that against the mapping of the previous character, the detector determines whether a word has just started or ended. Because the character-classes include generic word separators or delimiters common to most languages, this word boundary detector can be used in a language independent manner. Thus, various embodiments of this system can be developed for different languages. Additionally, a case-folding operation may be done on the words to remove any distinction between words containing upper case and lower case characters. This ensures that duplicate fingerprints are not generated for upper and lower case formats of the same word. Note that the case folding can be done at any time prior to the operation of the word hash list generation module.
The received normalized information is then used to generate a word hash list using the word hash list generation module 940. In one embodiment, this word hash list generation module is a computer implemented software that operates on every word of the received normalized textual information. In this embodiment, the module further comprises a computer implemented software to compute a hash function over all the characters of each word, resulting in a word-value hash for every word. These word-value hashes are compiled together in a list, and this list is designated as the word hash list. The word hash list can further be post-processed to exclude some word-value hashes in order to generate fingerprints that are robust and remain impervious to edits in derivative works of the original text. Examples of this include removing certain stop words that occur frequently in a language and grouping certain categories of words and mapping them to one common word-value hash. These post-processing steps can also be achieved by means of a computer implemented software.
The word hash list is finally used to generate a set of fingerprints by operation of the fingerprint generation module 950. In one embodiment, the fingerprint generation module is a computer implemented software capable of performing arithmetic and logic operations. Here, the software reads word-value hashes using a sliding window of size W, reading W number of word-value hashes at a given time. At each window instant, the software designates a distinct-valued word-value hash as an anchor, and generates a new fingerprint every time the anchor of the current window is not identical to the anchor from the immediately preceding window. The software computes the fingerprint by computing a new hash function over all word-value hashes starting from the first word-value hash of the current window up until the word-value hash corresponding to the anchor of the current window. This method of fingerprinting using wordruns is advantageous over other methods because it results in memory and resource efficiency, by reducing the total number of fingerprints that need to be stored in a fingerprint database.
The systems explained in
The algorithms and software presented herein are not inherently related to any particular computer or other apparatus. Various general purpose systems may be used with programs in accordance with the teachings herein, or it may prove convenient to construct more specialized apparatus to perform the required method steps. The required structure for a variety of these systems will appear from other portions of this description. In addition, the present invention is not described with reference to any particular programming language, and various embodiments may thus be implemented using a variety of programming languages.
In addition to the above mentioned examples, various other modifications and alterations of the invention may be made without departing from the invention. Accordingly, the above disclosure is not to be considered as limiting and the appended claims are to be interpreted as encompassing the true spirit and the entire scope of the invention.
Claims
1. A system to prevent unauthorized disclosure of secure information, the system comprising:
- a processor;
- a memory;
- a processing component configured to: receive information including a first text, wherein the first text includes a plurality of words; normalize the first text into a first canonical text expression, the first canonical text expression including a plurality of normalized words; generate a first word hash list for the first canonical text expression, wherein the first word hash list is generated at a word level; and generate one or more fingerprints for the first word hash list, wherein the generation of one or more fingerprints includes: assigning a sliding window of size W, wherein W specifies a number of word-value hashes to read from the first word hash list; using the sliding window to read the W word-value hashes from the first word hash list; designating an anchor word-value hash for the sliding window by selecting a distinct-valued word-value hash among the W word-value hashes; and applying a fingerprint hash function to all words starting from a first word-value hash to the anchor word value-hash, wherein applying the fingerprint hash function generates the one or more fingerprints.
2. The system to prevent unauthorized disclosure of secure information as recited in claim 1, wherein each of the plurality of words being a combination of one or more text characters not separated by a predefined character, and each of the plurality of words being separated from a previous word and a subsequent word by at least one of said predefined character.
3. A system to prevent unauthorized disclosure of secure information as recited in claim 2, wherein said predefined character includes more than one character type, including a space, a period, a comma, a semi-colon, a colon, an exclamation point, a dash, a parenthesis, and a quotation mark.
4. A system to prevent unauthorized disclosure of secure information as recited in claim 1, wherein the processing component is further configured to detect among the plurality of words a separation of one word from a following or preceding word using a word boundary detector.
5. A system to prevent unauthorized disclosure of secure information as recited in claim 4, wherein the word boundary detector is language independent.
6. A system to prevent unauthorized disclosure of secure information as recited in claim 1, wherein said receiving information includes receiving secure information from a local database.
7. A system to prevent unauthorized disclosure of secure information as recited in claim 1, wherein said receiving information includes receiving information entered by a user.
8. A system to prevent unauthorized disclosure of secure information as recited in claim 1, wherein said generating the first word hash list includes converting the plurality of normalized words into a plurality of word-value hashes, wherein each word-value hash of the plurality of word-value hashes represents a specific normalized word.
9. A system to prevent unauthorized disclosure of secure information as recited in claim 8, wherein said converting the plurality of normalized words includes one or more of the following:
- performing case-folding;
- removing stop words;
- mapping a predefined set of common words to a unique word-value hash;
- mapping a predefined set of synonyms to a unique word-value hash;
- mapping a predefined set of common words to a unique word-value hash; and
- mapping a predefined set of words in a particular category to a unique word-value hash.
10. A system to prevent unauthorized disclosure of secure information as recited in claim 1, wherein the fingerprint hash function includes any hash function that allows the one or more fingerprints to be independent of the order of words in the first word hash list.
11. A system to prevent unauthorized disclosure of secure information as recited in claim 1, wherein the processing component is further configured to record the one or more fingerprints from all sources in a fingerprint database.
12. A system to prevent unauthorized disclosure of secure information as recited in claim 1, wherein the processing component is further configured to monitor and detect any unauthorized disclosure of said secure information by a user, by generating the one or more fingerprints for information entered by the user.
13. A system to prevent unauthorized disclosure of secure information as recited in claim 11, wherein the processing component is further configured to monitor and detect any unauthorized disclosure of said secure information by a user, by generating the one or more fingerprints for information entered by the user and matching the one or more fingerprints against fingerprints stored in said fingerprint database.
14. A computer implemented method for preventing unauthorized disclosure of secure information, the computer implemented method comprising:
- storing a plurality of secure text fingerprints for a given organization, wherein each of the plurality of secure text fingerprints is generated using a fixed window word run hashing;
- receiving a first text that a user desires to transmit outside of the given organization;
- generating a first set of fingerprints for the first text using the fixed window word run hashing, wherein generating a first set of fingerprints includes: converting a plurality of normalized words into a plurality of word-value hashes to create an original word hash list, wherein each word-value hash represents a specific normalized word; assigning a sliding window of size W, wherein W specifies a number of word-value hashes to read from the original word hash list; using said sliding window to read said W word-value hashes from the original word hash list; designating an anchor word-value hash for the sliding window by selecting a distinct-valued word-value hash among said W word-value hashes; and applying a fingerprint hash function to all words starting from a first word-value hash to the anchor word-value hash, wherein applying the fingerprint hash function generates the first set of fingerprints;
- determining whether any of the first set of fingerprints is identical to any of the plurality of secure text fingerprints; and
- taking a security action when any of the first set of fingerprints is identical to any of the plurality of secure text fingerprints.
15. A computer implemented method for preventing unauthorized disclosure of secure information as recited in 14, wherein the fixed window word run hashing comprises:
- receiving information including an original text, the original text including a plurality of words;
- normalizing said original text into an original canonical text expression, the original canonical text expression including a plurality of normalized words;
- generating an original word hash list for the original canonical text expression, wherein the original word hash list is generated at a word level, wherein the original word hash list includes a plurality of word-value hashes; and
- generating an original set of fingerprints for the original word hash list.
16. A computer implemented method for preventing unauthorized disclosure of secure information as recited in claim 14, wherein said first text includes at least one of:
- text contained in an electronic mail;
- text contained in a file attached to an electronic mail; and
- text that is transferred using a computer's output device.
17. A computer implemented method for preventing unauthorized disclosure of secure information as recited in claim 14, wherein said security action includes at least one of:
- preventing said first text from being transferred;
- logging the event as a security violation;
- requiring a password from said user to allow said first text to be transferred;
- blocking said user's access to said first text; and
- sending out a security alert.
18. A computer implemented method for preventing unauthorized disclosure of secure information as recited in claim 14, the computer implemented method further comprising generating the plurality of secure text fingerprints for the given organization.
19. A computer implemented method for preventing unauthorized disclosure of secure information as recited in claim 14, the computer implemented method further comprising creating a fingerprint database for the given organization, wherein the fingerprint database comprises the plurality of secure text fingerprints for the given organization.
20. A computer implemented method for preventing unauthorized disclosure of secure information as recited in claim 14, wherein the fingerprint hash function includes any hash function that allows the first set of fingerprints to be independent of the order of the words in the original word hash list.
Type: Application
Filed: Sep 14, 2012
Publication Date: Mar 21, 2013
Inventors: Scott MORE (San Francisco, CA), Ilya Beyer (San Mateo, CA)
Application Number: 13/620,364
International Classification: G06F 21/00 (20060101);