Masquerade detection based on fuzzy inclusion

Through reverse n-element syntax mapping and fuzzy inclusion detection technology, the problem that existing systems are difficult to detect advanced camouflage technology is solved, and the effective recognition of Unicode replacement and visually similar substrings is realized, which improves the efficiency and ability of phishing detection Trust.

CN112771524BActive Publication Date: 2024-11-01MICROSOFT TECHNOLOGY LICENSING LLC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201980062498.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2018-11-27
Filing Date
2019-06-26
Publication Date
2024-11-01
Estimated Expiration
2039-06-26

AI Technical Summary

Technical Problem

Existing anti-phishing systems have difficulty detecting advanced camouflage technology, especially the camouflage behavior of using Unicode character replacement and visually similar substrings, which makes it difficult to distinguish between legal and malicious text in electronic communications, and traditional methods are computationally expensive and Inefficient.

Method used

Reverse n-member syntax mapping and fuzzy inclusion detection technology are adopted to generate data structures and train neural networks to identify visually similar substrings and Unicode replacements, improve the detection ability of camouflage text and support efficient scanning in real-time applications.

Benefits of technology

Effectively identifying and classifying disguised texts reduces the risk of phishing attacks, improves the trustworthiness and efficiency of electronic communications, and is suitable for real-time application scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112771524B_ABST
    Figure CN112771524B_ABST
Patent Text Reader

Abstract

Aspects of the present disclosure include cloaking detection techniques based on fuzzy inclusion. In one embodiment, a reverse n-gram mapping is created for a list of protected entities. Suspicious strings can be divided into n-grams, and each n-gram of the suspicious string can be searched in the reverse n-gram mapping for the corresponding protected entity. Based on the protected entities found during the search, fuzzy inclusion of the protected entity may be detected in the suspicious string. Subsequently, cloaking can be identified based on the characteristics of fuzzy inclusion. In this way, the communication system is able to detect cloaking using detection of visually similar text, and can therefore take various actions to help users mitigate the risks caused by cloaking.
Need to check novelty before this filing date? Find Prior Art

Description

Background Art

[0001] Phishing is usually accomplished through email or instant messaging forgery by posing as a trustworthy entity in electronic communications. Phishing attackers often trick users into replying to emails, opening attachments, or visiting fake websites that are nearly identical to corresponding legitimate sites. Phishing attackers continue to develop new techniques to bypass existing anti-phishing systems.

[0002] Impersonation is a typical phishing technique. Email impersonation attacks attempt to disguise emails from celebrities or well-known people, trademarks, or trusted organizations by forging headers or email addresses, creating misleading URLs, or using other methods to create something that appears to be from a legitimate source but is actually malicious.

[0003] An advanced cloaking technique is to embed substrings that are visually similar to legitimate entities into email headers or email bodies. Such abuse techniques pose a risk to the public by spreading phishing, spam, and malware content. Identifying these abuses is a challenge. Traditional anti-phishing systems may not be able to detect this new cloaking technique because the substrings do not match the legitimate entity via string comparison. In addition, anti-phishing scans are computationally expensive given the exponential growth of electronic communications in our modern lives. Effective technical solutions for cloaking detection are critical for immediate or real-time applications and to maintaining the trustworthiness and efficiency of our electronic communications. Summary of the invention

[0004] The purpose of providing this Summary is to introduce a selection of concepts in a simplified form that are further described in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to determine the scope of the claimed subject matter.

[0005] Aspects of the present disclosure provide an effective method to detect types of disguises, including, for example, brand disguises or disguises of individuals or organizations. Although the disguised text may visually look like legitimate text, as if it comes from a legitimate individual, organization, or brand, the disclosed technology is still able to distinguish the disguised text from the legitimate text. In various embodiments, the disclosed technology is able to detect the following disguised texts: 1) using Unicode characters that look visually similar to the original characters but may use different language scripts; 2) inserting, removing, replacing some characters of the original string; 3) prefixing, suffixing, including, or otherwise mixing other texts with legitimate text; and / or 4) mixing the above-mentioned disguise methods.

[0006] In various aspects, systems, methods, and computer-readable storage devices are provided to improve the ability of computing devices to detect camouflage methods. One aspect of the technology described herein is to improve the ability of computing devices to generate and use various data structures for quickly and efficiently searching for visually similar substrings. Another aspect of the technology described herein is to improve the ability of computing devices to detect camouflage based on reverse n-gram mappings generated from alternative forms of protected strings. Another aspect of the technology described herein is to improve the ability of computing devices to identify various types of fuzzy inclusions. Therefore, training data for various types of fuzzy inclusions can be used to train neural networks and develop artificial intelligence to detect camouflage. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The technology described herein is illustrated by way of example and not limitation in the accompanying drawings, in which like reference numerals represent similar elements and in which:

[0008] Figure 1 is a block diagram illustrating an exemplary operating environment for implementing masquerade detection and reaction in accordance with various aspects of the techniques described herein;

[0009] Figure 2 is a block diagram illustrating a fuzzy containment based masquerade detection system in which some embodiments of the present disclosure may be employed;

[0010] Figure 3 illustrates a flow chart showing an exemplary process for detecting masquerading according to one aspect of the techniques described herein;

[0011] Figure 4 is a block diagram illustrating an exemplary data repository that may be employed with some embodiments of the present disclosure;

[0012] Figure 5 illustrates a flow chart showing an exemplary process for generating a reverse n-gram mapping according to one aspect of the technology described herein;

[0013] Figure 6 illustrates a flow chart showing an exemplary process of searching for reverse n-gram mappings according to one aspect of the technology described herein;

[0014] Figure 7 is a flowchart illustrating an exemplary process for determining fuzzy containment according to one aspect of the technology described herein;

[0015] Figure 8 is a block diagram of an exemplary computing environment suitable for implementing aspects of the techniques described herein. DETAILED DESCRIPTION

[0016] The various techniques described herein are set forth with sufficient specificity to meet statutory requirements. However, this description itself is not intended to limit the scope of the present disclosure. On the contrary, the inventors have anticipated that the claimed subject matter may also be embodied in other ways in conjunction with other current or future technologies to include steps different from the steps described herein or combinations of steps similar to the steps described herein. In addition, although the terms "step" and / or "box" may be used herein to refer to different elements of the method employed, unless and in addition to the order of the various steps being explicitly described, these terms should not be interpreted as implying any particular order among or between the various steps disclosed herein. In addition, the term "based on" generally indicates that the execution of the prior action uses a subsequent condition.

[0017] Electronic messages or documents, including short message service (SMS), multimedia message service (MMS), instant message (IM), email, social network messages, web pages, and even ordinary electronic documents, are vulnerable to phishing attacks. In phishing messages, disguisers make the message appear to come from a legitimate source, such as a well-known brand, relevant organization, important person, or close friend.

[0018] To prevent detection, the disguiser can embed a substring of a visual representation similar to a legal entity in a message. For example, in order to make the disguised text look indistinguishable from the legal entity, the disguiser can use Unicode characters visually similar to the characters in the legal entity in the disguised text. The disguiser can utilize Unicode characters to add characters to the legal entity, delete characters from the legal entity, or exchange characters with the legal entity. In addition, the disguised text can be an independent word, or can be combined with additional characters to be prefixed, suffixed, or embedded in the string of text. Unconventional symbols or separators can also be used in the disguised text to avoid anti-phishing techniques based on conventional separators. Each of these technologies can be used by the disguiser to disguise text strings, URLs, email addresses, attachments, documents, etc.

[0019] There are conventional methods for searching similar keywords in a string based on predetermined keywords, such as edit distance methods and similarity measurement techniques based on n-gram characters. In contrast, similarity measurement techniques based on n-gram characters are usually faster than edit distance methods. On the other hand, edit distance methods are usually more accurate than similarity measurements based on n-gram characters, especially for keywords with only a few characters.

[0020] However, conventional edit distance-based methods and n-gram character-based similarity measurement techniques are insufficient in handling the aforementioned attacks. For example, these conventional techniques are limited in handling advanced cloaking techniques (such as Unicode substitution). For example, a cloaking attack may replace characters in a well-known word or character string with other Unicode characters similar to the original character. These conventional methods may detect exact string matches, but lack the functionality to detect characters that have been replaced with similar Unicode characters. In particular, conventional methods cannot find Unicode substitutions in visually similar substrings embedded in a string.

[0021] In this disclosure, and at a high level, a technical solution is provided to identify substrings in a suspicious string that are visually similar to famous, well-known or otherwise legal trademarks, individuals, institutions or entities (hereinafter referred to as "protected strings") based on a dedicated data structure. The reverse n-gram map is a specific data structure created herein that can be generated from alternative forms of protected strings, combined with Figure 5 Further discussion is provided. In one embodiment, the conversion process to the alternate form is defined by a predetermined conversion protocol (such as placing each character into a common Unicode range representing a particular language) and many other rules in the conversion protocol.

[0022] A reverse n-gram mapping is created to enable fuzzy inclusion detection. There are many types of fuzzy inclusion. As mentioned above, traditional cloaking techniques, such as including legitimate entities (e.g., ) of an exact string of characters (hereinafter referred to as "exact inclusion"). However, new disguise methods based on fuzzy inclusion have emerged, such as inserting, removing or replacing some characters of a legitimate entity, for example, replacing characters of a legitimate text with Unicode characters that visually look similar to the original characters; prefixing, suffixing, including or otherwise mixing other characters with the legitimate text; and / or mixing the above-mentioned disguise methods. In order to overcome these advanced disguise methods, in various embodiments, the present disclosure extends the edit distance method from identifying fuzziness to equivalently identifying fuzzy inclusion, and utilizes reverse n-gram mapping to effectively identify inspection targets. Advantageously, the disclosed technology can effectively scan long documents or search for disguises in large quantities in targets.

[0023] For fuzzy inclusion, the present disclosure also classifies fuzzy inclusion into specific classes based on various characteristics of fuzzy inclusion, such as the start index and end index of the disguised substring, or marks the fuzzy inclusion with a class label. Subsequently, various labeled fuzzy inclusion training data can be used to train neural networks and develop artificial intelligence to detect disguises.

[0024] In one embodiment, the strings of text in the message are first quickly checked to filter out strings of no interest, such as stop words or any unimportant words specific to the domain. These strings of no interest can be stored in an efficient data structure, such as an ordered tree data structure, such as a prefix tree. This initial process advantageously reduces the computational power required and allows for faster fuzzy inclusion based visual similarity searches.

[0025] In terms of fuzzy inclusion detection, the disclosed technology can first find those strings or substrings that are exact matches to the protected string in a first pass. Then, in a second pass, the disclosed technology will find those strings that look similar to the protected string but may have characters added, deleted, or replaced. For example, in the first pass, any string that includes the protected string "Microsoft" can be found as an instance of exact inclusion. In the second pass, suspicious strings such as "Microoooosoft", or "Microsoft", as an instance of vague inclusion.

[0026] Continuing at a high level, the reverse n-gram map contains information about n-grams of interest and the protected strings they are respectively linked to. The suspicious string is converted to its alternative forms based on the same conversion protocol used when building the reverse n-gram map. Each n-gram from the alternative form of the suspicious string is then used to search the reverse n-gram map. If the protected string is found during the search, it is a hit. The number of hits can be compared to the number of n-grams in the protected string. Therefore, the disclosed technology can detect potential ambiguous inclusions of protected strings in suspicious strings based on such comparisons, which will be combined with Figure 7 Further discussion.

[0027] For the detected fuzzy inclusion, the present disclosure further classifies the fuzzy inclusion into a specific fuzzy inclusion class based on the various characteristics of the fuzzy inclusion, or marks the fuzzy inclusion with a class label. Such a label can help define the level of disguised abuse. For example, a message with multiple types of fuzzy inclusions may indicate more complex or advanced disguised abuse compared to another message with only a single type of fuzzy inclusion.

[0028] Advantageously, the various technical solutions disclosed herein improve conventional detection methods, for example, using visually similar substrings (e.g., using Unicode substitution) to detect camouflage methods. In addition, by utilizing a prefix tree data structure and a reverse n-gram mapping (combining Figure 4 Further discussion), the technical solutions disclosed in this article operate more efficiently to support real-time applications.

[0029] After briefly describing an overview of various aspects of the technology described herein, an exemplary operating environment in which various aspects of the technology described herein can be implemented is described below. Figure 1 , an exemplary operating environment for implementing masquerade detection and response is shown and generally designated as operating environment 100. Operating environment 100 is only one example of a suitable computing environment and is not intended to suggest any limitation as to the scope of use or functionality of aspects of the technology described herein. Operating environment 100 should not be interpreted as having any dependency or requirement relating to any one or combination of components illustrated.

[0030] Now turn to Figure 1 , a block diagram is provided, which shows an operating environment 100 in which some aspects of the present disclosure can be adopted. It should be understood that such arrangements and other arrangements described herein are only set forth as examples. In addition to or in place of these arrangements and elements shown, other arrangements and elements (e.g., machines, interfaces, functions, groups of sequences and functions, etc.) can be used, and for the sake of clarity, some elements can be omitted completely. In addition, many elements described herein are functional entities, which can be implemented as discrete or distributed components or combined with other components, and are implemented in any suitable combination and position. The various functions performed by the entity described herein can be implemented by hardware, firmware and / or software. For example, some functions can be implemented by a processor that executes instructions stored in a memory.

[0031] In addition to other components not shown, the operating environment 100 includes two systems, namely, a detection system 130 and a reaction system 140. It should be understood that Figure 1 The illustrated operating environment 100 is an example of a suitable operating environment. Figure 1 Each of the systems shown in the FIGURES may be implemented via any type of computing device (such as a combination of Figure 8 The components may be implemented by a computing device 800 described in the foregoing. These components may communicate with each other via a network, which may include, but is not limited to, a local area network (LAN) and / or a wide area network (WAN). In an exemplary embodiment, the WAN includes the Internet and / or a cellular network, as well as any of a variety of possible public and / or private networks.

[0032] In this embodiment, the detection system 130 receives two types of inputs, such as a document 110 and a watch list 120. The watch list 120 includes those protected strings, which can be customized for industries, fields, institutions, or even individual people or applications. For example, an organization can define a specific watch list for its email application. Document 110 is any type of document that needs to be scanned for masquerade detection. In this example, document 110 can be an email message that contains suspicious strings, such as "Micros0ftTeam", wait.

[0033] The detection system 130 will detect suspected forgery in the document 110 based on the watch list 120. In various embodiments, a reverse n-gram mapping is constructed based on the watch list 120. Figures 3 to 5 When the detection system 130 processes a string, such as "Micros0ftTeam" or It will first transform them into alternative forms based on the same transformation protocol used to build the reverse n-gram mapping, such as, if possible, converting each character in the search string into the corresponding Unicode character in English. Other rules for transformation can also be used, such as converting a character into another visually similar character. For example, "m" can be converted to "rn". As a result, "Micros0ffTeam" in document 110 can be transformed into its alternative form as "rnicros0ffteam".

[0034] The detection system 130 then searches the reverse n-gram mapping for each n-gram derived from the alternative form of the search string. Based on the discovery of such a search process, the protected string (e.g., Microsoft) in the monitoring list 120 can be determined to be fuzzily included in the search string (e.g., Micros0ffTeam). In addition, the detection system 130 will identify the type of fuzzy inclusion in this instance based on the various characteristics of such fuzzy inclusion. In this case, since the protected string "Microsoft" is a prefix of the search string "Micros0ffTeam", the label "Prefix" is assigned to "Micros0ffTeam". In summary, triple information is generated after the initial detection process. Specifically, the protected strings in the monitoring list 120 are linked to the suspicious strings in the document 110 using the labels. In addition, based on the labels, the detection system 130 can generate a disguise metric for the document 110, which will be combined with the Figure 2 Further detailed discussion.

[0035] The masquerade metric for the document 110 and other relevant information may be communicated to the reaction system 140. Accordingly, the reaction system 140 will take various responses based on the type and / or degree of masquerade detected by the detection system 130. For example, for a low level of suspected masquerade, the reaction system 140 may simply deliver a warning 152 to the recipient of the document 110. For a medium level of suspected masquerade, the reaction system 140 may additionally take the response of a mark 154, which may highlight the suspicious string, or even temporarily convert it to another form to reduce its risk. For example, a hyperlink to the suspicious string may be temporarily removed, but may be restored if the recipient decides to restore it. For a high level of suspected masquerade, the reaction system 140 may take the response of a purge 156, e.g., purge the document 110 without even delivering it to the intended recipient.

[0036] As will be appreciated by those skilled in the art, different application-specific or domain-specific reaction systems may be needed to solve application-specific or domain-specific masquerade problems. In various embodiments, both the detection system 130 and the reaction system 140 are highly customizable in solving application-specific or domain-specific masquerade problems.

[0037] Reference now Figure 2 , a block diagram is provided that illustrates aspects of an exemplary fuzzy containment-based masquerade detection system for implementing an embodiment, and is generally designated as detection system 200. Detection system 200 represents only one example of a suitable computing system architecture. Other arrangements and elements may be used in addition to or in place of the arrangements or elements shown, and some elements may be omitted entirely for clarity. In addition, as with operating environment 100, many of the elements described herein are functional entities that may be implemented as discrete or distributed components or in combination with other components, and in any suitable combination and location.

[0038] Detection system 200 is only an example of system design, and is not intended to imply any limitation on the scope of use or functionality of various aspects of the technology described herein. As shown, any component in detection system 200 should not be interpreted as having any dependency related to another component. It should be understood that such arrangement and other arrangements described herein are only set forth as examples. In addition to or in place of these arrangements and elements shown, other arrangements and elements (e.g., machines, interfaces, functions, groups of sequences and functions, etc.) can be used, and for clarity, some elements can be completely omitted. In addition, many elements described herein are functional entities, which can be implemented as discrete or distributed components or combined with other components, and are implemented in any suitable combination and position. The various functions performed by the entity described herein can be implemented by hardware, firmware and / or software. For example, some functions can be implemented by a processor that performs instructions stored in a memory.

[0039] In this embodiment, the detection system 200 is a camouflage detection system based on fuzzy inclusion, which includes a searcher 210, a classifier 220 and a detector 230, as well as Figure 2 The detection system 200 may receive a signal such as a signal from a user. Figure 1 The files or information of the watch list 120 and the document 110 discussed are used for masquerade detection. By way of example only, the input files used for masquerade detection may include but are not limited to email messages, web pages, text files, PDF files, etc.

[0040] Searcher 210 can identify fuzzy inclusion of the protected word string in the search string. Searcher 210 can be configured to compare the protected word string and the search string in their respective original forms in a first pass, e.g., for detecting exact inclusion, and then compare the protected word and the search string in their respective alternative forms in a second pass, e.g., based on a reverse n-gram mapping for detecting fuzzy inclusion. In some embodiments, to detect fuzzy inclusion, a search result such as Figure 5 In the construction process shown in FIG. 1 , the searcher 210 first constructs a reverse n-gram mapping, such as Figure 4 Next, such as in some embodiments, using Figure 6 In the search process shown, the searcher 210 can search for reverse n-gram mappings based on the search string found in the document. Based on the results of the search process, such as using Figure 7 As shown in the process, the searcher 210 can determine whether the protected string is ambiguously included in the search string.

[0041] For detected fuzzy inclusions, the classifier 220 may then classify the fuzzy inclusions into a disguise type and / or determine relevant parameters associated with the disguise type, such as shown in Table 1 below. In some embodiments, such classification is based on the position of the protected string in the search string, such as whether the protected string appears at the beginning, middle, or end of the search string. In some embodiments, such classification is based on the language used in the search string, for example, whether Unicode characters from multiple languages ​​are mixed in the search string. In some embodiments, such classification is based on other characteristics of the search string or the fuzzy inclusion, such as upper or lower case of characters, length of the string, delimiters used in the search string, whether an exact inclusion is detected in the first pass, and the like.

[0042] Additionally, in some embodiments, the detector 230 may include an artificial intelligence 232. Either supervised learning or semi-supervised learning may be used to train an artificial neural network (ANN) in the artificial intelligence 232 for various tasks, including general camouflage detection tasks or even specific classification tasks performed by the classifier 220. Supervised learning is a machine learning task that learns a function that maps an input to an output based on example input-output pairs. Semi-supervised learning uses both labeled and unlabeled data for training, such as a small amount of labeled data and a large amount of unlabeled data.

[0043] In one embodiment, the artificial intelligence 232 can use fuzzy inclusion of protected strings as features to train the ANN. Specifically, labeled data (such as the triple information discussed above, i.e., protected string, search string, and type information) can be used to train the ANN to utilize various deep learning techniques to generally identify cloaking. With proper training, the ANN can output a cloaking metric 240 for a string, paragraph, or document. In some embodiments, the cloaking metric 240 can be a general probability metric, such as a confidence score representing the likelihood of cloaking associated with a document or the severity of the cloaking. In some embodiments, the cloaking metric 240 can include, for example, category probability metrics for corresponding types of cloaking shown in Table 1, such as the severity of cloaking of a type called WeakSectionFuzzyEqual. Therefore, Figure 1 The reaction system 140 in can take appropriate measures based on the masquerade metric 240 to mitigate the risks associated with masquerade.

[0044] However, in some embodiments, the disguised metric 240 may be generated directly from the triple information, for example, based on fuzzy inclusion. For example, in a strict detection system, the disguised metric 240 may be binary, i.e., true or false. Whenever the count of a tuple of the triple information exceeds a threshold (e.g., 2), the detection system 130 may output the disguised metric 240 as true.

[0045] In some other embodiments, a hybrid disguise metric based on a probability measure from the artificial intelligence 232 and a statistical measure from the classifier 220 may also be used. For example, if the statistical measure of the important type of disguise meets certain thresholds, the detection system 130 may output the disguise metric 240 as true or 100%. Otherwise, the detection system 130 may output the disguise metric 240 as a confidence score based on the statistical measure from the classifier 220 and / or the probability measure from the artificial intelligence 232.

[0046] In addition, the detection system 130 can be implemented using components (such as hardware, software, firmware, special equipment or any combination thereof). The detection system 130 can be integrated into a single device, or can be distributed over multiple devices. The various components of the detection system 130 can be co-located or can be distributed. The detection system 130 can be formed from other systems and components thereof. In addition to or in place of the arrangements and elements shown, other arrangements and elements (e.g., machines, interfaces, functions, groups of sequences and functions, etc.) can be used, and some elements can be completely omitted. In addition, many elements described herein are functional entities, which can be implemented as discrete or distributed components or combined with other components, and are implemented in any suitable combination and position. The various functions performed by the entity described herein can be implemented by hardware, firmware and / or software. For example, a processor that executes instructions stored in a memory can perform various functions.

[0047] Reference now Figure 3 , a flow chart is provided that illustrates an exemplary process for detecting camouflage. Each block of process 300, as well as other processes described herein, comprises a computational process that can be performed using any combination of hardware, firmware, and / or software. For example, various functions can be implemented by a processor executing instructions stored in a memory. The process can also be embodied as computer usable instructions stored on a computer storage medium. The process can be provided by an application, a service, or a combination thereof.

[0048] In block 310, the Figure 2 The searcher 210 generates a reverse n-gram mapping based on the protected string. In some embodiments, the protected string is transformed into an alternate form and a reverse n-gram mapping is generated from the alternate form. The alternate form can be generated by converting characters in the protected string into a specific range of Unicode characters for a specific language, if such corresponding Unicode characters are available. In addition, some characters can be converted into corresponding visually similar characters. In various embodiments, the reverse n-gram mapping is configured for efficient fuzzy containment searching. In combination with Figure 5 Discuss this box further.

[0049] In block 320, for example, Figure 2 The searcher 210 searches the reverse n-gram map for each n-gram generated from the search string. The search string can be any string delimited by spaces in a document or message. In one embodiment, only strings that are considered suspicious are checked. For fuzzy containment searches, in order to be compatible, the search string will be transformed into its alternative form following the same transformation protocol used to build the reverse n-gram map. Combined with Figure 6 This search process in block 320 is discussed in further detail.

[0050] In block 330, the Figure 2 The searcher 210 determines fuzzy inclusion of the protected string in the search string. If one n-gram of the search string matches the protected string, it is a hit. Depending on the number of hits and the actual number of n-grams in the protected string, for example, if the ratio of the number of hits to the number of n-grams of the protected string is greater than a predetermined threshold, fuzzy inclusion of the protected string in the search string can be determined. Figure 7 The process in block 330 is discussed in further detail.

[0051] Figure 4 4 is a block diagram illustrating an exemplary data repository that can be employed with some embodiments of the present disclosure. In this embodiment, data repository 400 includes mapping 410 and prefix tree 420. Mapping 410 is a reverse n-gram mapping. An n-gram is a continuous sequence of n items from a given string. An n-gram of size 1 is called a unigram; an n-gram of size 2 is called a bigram; an n-gram of size 3 is called a trigram, and so on. In some embodiments, trigrams or quadgrams are used. Each entry in mapping 410 represents a relationship from an n-gram to a protected string. For example, n-gram 1 is linked to word_n11, word_n12, etc. In this case, when n-gram 1 is used as a key for searching mapping 410, all protected strings linked to n-gram 1 will be obtained.

[0052] Prefix tree 420 has used the data structure prefix tree, and it refers to the ordered tree data structure for storing dynamic sets or associative arrays, and wherein key is usually a string.Prefix tree 420 has many advantages that are superior to binary search tree.For example, compared with binary search tree, it is faster to find data in prefix tree in the worst case.As another example, there is no conflict of different keys in prefix tree.

[0053] Fuzzy inclusion search may be an expensive computational operation, especially in view of the high throughput required in some real-time systems. In various embodiments, prefix tree 420 can be used to exclude those strings that do not need fuzzy inclusion search. Therefore, many types of strings (e.g., strings other than protected strings) can be stored in prefix tree 420. In one example, universal stop words can be stored in prefix tree 420. In another example, strings that need exact matching can also be stored in prefix tree 420. In addition, various applications can define different application-specific categories of search strings that do not need fuzzy inclusion search, and these application-specific categories of search strings can be stored in prefix tree 420. Therefore, during the disguise search, the inexpensive search on prefix tree 420 can be performed to exclude all known strings that do not need more expensive fuzzy inclusion search. In this way, the disguise detection process disclosed herein can be performed more efficiently, and the responsiveness and throughput requirements of some real-time applications can be met.

[0054] Now turn to Figure 5 , a flowchart is provided that illustrates an example of generating a reverse grammar mapping. Each block of process 500 and other processes described herein include a computing process that can be performed using any combination of hardware, firmware, and / or software. For example, various functions can be implemented by a processor executing instructions stored in a memory. The process can also be embodied as computer usable instructions stored on a computer storage medium. In various embodiments, it can be related to Figure 3 Process 500 is performed by performing block 310 of FIG.

[0055] In block 510, an alternative form of the protected string is identified. The conversion process to the alternative form is defined by a predetermined conversion protocol. In some embodiments, only one alternative form is used. However, in some other embodiments, more than one alternative form may be used to cover additional variations of the cloaking method. For example, if the document is written in two languages, the protected string may be transformed into Unicode characters in both languages.

[0056] In terms of the conversion protocol, in some embodiments, all characters in a protected string are converted to corresponding Unicode characters in a default language. The default language can be pre-selected by the system or dynamically determined based on the specific application or user requesting the service. In one embodiment, the default language can also be dynamically selected based on the document to be scanned. Typically, there are only a limited number of protected strings. Even if the range of Unicode is dynamically selected, the process in box 510 is still computationally inexpensive.

[0057] Many other rules can be incorporated into the conversion protocol, such as converting characters to visually similar characters. Such rules can be manually configured based on specific applications. For example, a disguiser may intentionally replace "m" with "rn" to overcome conventional anti-phishing systems because "rn" is visually similar to "m". In this case, a rule for converting "m" to "rn" can be incorporated into the conversion protocol. In some other embodiments, the trained ANN in the artificial intelligence 232 can automatically convert the protected string into an alternative form without an explicit conversion protocol.

[0058] In box 520, generate n-grams based on alternative forms. Generate character n-grams based on alternative forms. In some embodiments, tri-grams are used. Using the example above, the alternative form for the protected string "Microsoft" can be "rnicrosoft". Therefore, the eight tri-grams of "rnicrosoft" include "rni", "nic", "icr", "cro", "ros", "oso", "sof", and "oft". As another example, if "mic" or "microphone" is another protected string, their tri-grams will also include "rni" and "nic", etc. In some other embodiments, other n-grams (e.g., quad-grams) can be used based on specific applications.

[0059] In block 530, the protected string is linked to the n-grams in the reverse n-gram map. Each n-gram is linked to its protected string, and such information may be stored in the reverse n-gram map, such as Figure 4 410. Continuing with the above example, the tri-gram "rni" in the reverse n-gram mapping will be linked to "Microsoft" and "microphone", as well as potentially other protected strings.

[0060] Now turn to Figure 6 , a flowchart is provided that illustrates an example of searching for a reverse n-gram mapping. Each block of process 600 and other processes described herein include a computational process that can be performed using any combination of hardware, firmware, and / or software. For example, various functions can be implemented by a processor executing instructions stored in a memory. The process can also be embodied as computer usable instructions stored on a computer storage medium. In various embodiments, it can be related to Figure 3 Process 600 is performed by referring to block 320 of the flowchart.

[0061] Initially, in box 610, the search string is transformed into its alternate form. In various embodiments, the search string is converted into its alternate form based on the same conversion protocol used in box 510. For example, the alternate form of the search string can also be generated by replacing the Unicode characters in the search string with the corresponding Unicode characters of the target language. In one embodiment, the target language can be selected based on the primary language used in the document hosting the search string. As previously mentioned, it is relatively cheap to dynamically build a reverse n-gram mapping based on the target language, especially when the number of protected strings is much smaller than the number of search strings.

[0062] In block 620, n-grams may be generated from alternative forms of the search string. With the same token, the same number of grammars should be selected. For example, if tri-grams are used to construct the reverse n-gram mapping, then for compatibility purposes, tri-grams should also be used for the search string.

[0063] In block 630, reverse n-gram mappings are searched based on the generated n-grams. Each n-gram from an alternative form of the search string is used to search for reverse n-gram mappings. In some embodiments, if the reverse n-gram mappings are constructed using the same principle, repeated n-grams are merged into one n-gram. Alternatively, if the reverse n-gram mappings are constructed using a protected string for repetitions of the same n-gram, repeated n-grams may be allowed. In general, as long as process 600 and process 500 use the same reversal protocol, the reverse n-gram mappings may be combined into one n-gram. Figure 7 The fuzzy inclusion is correctly determined in process 700.

[0064] Now turn to Figure 7 , a flowchart is provided that illustrates an example of determining fuzzy inclusion. Each block of process 700 and other processes described herein include a computational process that can be performed using any combination of hardware, firmware, and / or software. For example, various functions can be implemented by a processor executing instructions stored in a memory. The process can also be embodied as computer usable instructions stored on a computer storage medium. In various embodiments, it can be related to Figure 3 330 to perform process 700.

[0065] In block 710, a count of protected strings found in the reverse n-gram mapping during the search is determined. In one embodiment, each n-gram extracted from an alternative form of the search string is used as a key for searching the reverse n-gram mapping. If a protected string is found during the search, it is a hit. For each hit, a hit counter for the protected string is incremented by 1. Multiple hit counters can be used for multiple protected strings.

[0066] In block 720, the count of protected strings found in the reverse n-gram mapping is compared to the count of n-grams in the protected string. In the same embodiment as described above, at the end of the search process, the hit counter for the protected string can be compared to the number of n-grams in the protected string. Using the example used previously to illustrate this comparison, when the protected string is "Microsoft" and the search string is "Micros0ftTeam", there are eight tri-grams in "rnicrosoft", including "rni", "nic", "icr", "cro", "ros", "oso", "sof", and "oft". On the other hand, there are many tri-grams in "rnicros0ftteam", including "rni", "nic", "icr", "cro", "ros", "oso", "s0f", "0ft", "ftt", "tte", "tea", and "eam". After searching the reverse n-gram mapping, the hit counter for the protected string being "Microsoft" becomes 6. The number of this hit (i.e., 6) is compared to the count of n-grams in the protected string "Microsoft" (i.e., 8).

[0067] In block 730, a potential fuzzy inclusion of the protected string in the search string is determined based on the comparison. In some embodiments, the difference of the above two counters, i.e., the difference between the count of the protected string found in the reverse n-gram mapping and the count of the n-gram in the protected string, can be used. For example, if the difference is less than a threshold, it can be determined that the protected string is fuzzily included in the search string. In the above example, assuming a threshold of 3, the difference of 2 is small enough to mark the fuzzy inclusion of "Microsoft" in "Micros0ftTeam". The threshold of the difference can be dynamically adjusted based on the length of the protected string. For example, a longer protected string can have a higher threshold. In some embodiments, the ratio of the above two counters, i.e., the ratio of the count of the protected string found in the reverse n-gram mapping to the count of the n-gram in the protected string, can be used. For example, if the ratio is greater than a threshold, it can be determined that the protected string is fuzzily included in the search string. In the above example, assuming a threshold of 0.6, the ratio of 0.75 in this case is also large enough to mark the fuzzy inclusion of "Microsoft" in "Micros0ftTeam".

[0068] In box 740, the potential fuzzy inclusion is further verified, and the characteristics of the fuzzy inclusion can be determined. False positives of fuzzy inclusion can be excluded. In some cases, there will be many hits when searching for reverse n-gram mappings, but the matching n-grams are inconsistent with the order in which they are presented in the protected string. This type of false positive situation or other types of false positive situations can be distinguished from true fuzzy inclusion situations. Edit distance refers to the minimum number of operations required to convert one string into another string (for example, the insertion, deletion or replacement of a single character, or the transposition of two adjacent characters). In some embodiments, the edit distance between the disguised substrings in the protected string and the search string, or the edit distance between the protected string and the search string can be used to exclude false positive fuzzy inclusion situations. For example, the D-Levenshtein distance between the suspected substrings in the protected string and the search string can be used. In the false positive example discussed above, when the matching n-grams are inconsistent with the sequence they are presented in the protected string, the edit distance between the protected string and the search string or its substring may be greater than a predetermined threshold. The threshold for the edit distance can be preset or determined by the system or user, for example, based on the length of the protected string. In summary, at least in some embodiments, in order to confirm the fuzzy inclusion of the protected string in the search string, the above two criteria are considered, including the difference or ratio between the count of the protected string found in the reverse n-gram mapping and the count of the n-gram in the protected string, and some measures such as the edit distance measure for excluding false positive situations. In addition, for the fuzzy inclusion situation, the characteristics of the fuzzy inclusion can be determined. Various characteristics can be considered, for example, as shown in part in Table 1 below, including the edit distance discussed herein, the start index and end index of the fuzzy inclusion, the uppercase or lowercase in the search string, the (multiple) languages ​​used in the search string, the length of the search string, etc. In some embodiments, the start index and end index of the fuzzily included protected string in the search string can be determined, for example, by minimizing the edit distance between the protected string and the substring of the search string. In some other embodiments, many other characteristics of fuzzy inclusion can also be determined.

[0069] In block 750, the cloaking type based on fuzzy inclusion is classified. In various embodiments, the fuzzy inclusion of the protected string in the search string is assigned a type based on the characteristics of the fuzzy inclusion. For example, the cloaking type can be assigned based on the respective lengths or case (uppercase or lowercase) of the protected string and the search string. In addition, the cloaking type can be assigned based on which portion of the substring matches the protected string (e.g., a prefix or suffix).

[0070] Just to illustrate, if the protected string is "Microsoft" and the search string is "356MicrosoftOffice", since the protected string is included in the middle of the search string, the disguise can be marked with the type "Include". As another example, consider the search string "officemicr0soft", because the substring "micr0soft" that is fuzzily included is the suffix of the search string, the disguise can be marked with the type "Suffix".

[0071] As just another example, if the protected term is "Microsoft" and the search string is Then the type "SectionFuzzyEqual" can be used here, because the search string is partially similar to the protected string, but contains replaced characters. As another example, for the search string "Microooooosoft", the type "WeakEqual" can be used here, because the search string is largely the same as the protected string, but contains some additional characters.

[0072] The above illustrative examples may not include every type of disguise. In particular, in various embodiments, disguise types may be added or modified by a user or application based on specific needs. By way of example only, Table 1 below indicates various types or parameters that may be used in some embodiments.

[0073] Table 1: Types and parameters

[0074]

[0075]

[0076]

[0077] Thus, we have described various aspects of the technology for detecting masquerading and reducing the risk of masquerading. It should be understood that various features, sub-combinations, and modifications of the embodiments described herein are useful and can be used in other embodiments without reference to other features or sub-combinations. In addition, the order and sequence of the steps shown in the above example processes are not meant to limit the scope of the present disclosure in any way, and in fact, these steps can occur in various different sequences within its embodiments. Such variations and combinations thereof are also considered to be within the scope of the embodiments of the present disclosure.

[0078] Referring generally to the accompanying drawings, first specifically to Figure 8, an exemplary operating environment for implementing various aspects of the techniques described herein is shown and generally designated as computing device 800. Computing device 800 is only one example of a suitable computing environment and is not intended to suggest any limitation on the scope of use of the techniques described herein. Computing device 800 should not be interpreted as having any dependency or requirement relating to any one or any combination of components illustrated.

[0079] The techniques described herein may be described in the general context of computer code or machine-usable instructions (including computer executable instructions such as program components) executed by a computer or other machine. Typically, program components, including routines, programs, objects, components, data structures, etc., refer to code that performs a specific task or implements a specific abstract data type. The techniques described herein may be practiced in a variety of system configurations including handheld devices, consumer electronics, general-purpose computers, special-purpose computing devices, etc. Aspects of the techniques described herein may also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected via a communication network.

[0080] Continue to refer Figure 8 , computing device 800 includes a bus 810 that directly or indirectly couples the following devices: memory 820, processor 830, presentation component 840, input / output (I / O) port 850, I / O component 860, and illustrative power supply 870. Bus 810 may include an address bus, a data bus, or a combination thereof. Although shown with lines for clarity, Figure 8 The various boxes of FIG. 1 are not actually very clear to depict the individual components, and metaphorically, the lines would more accurately be gray and fuzzy. For example, a presentation component such as a display device may be considered an I / O component. Also, a processor has memory. The inventors of the present invention recognize that this is the nature of the art and reiterate that Figure 8 The figures are merely illustrations of exemplary computing devices that can be used in conjunction with various aspects of the techniques described herein. No distinction is made between categories such as "workstation," "server," "laptop," "handheld device," etc., as all of these are Figure 8 and means "computer" or "computing device".

[0081] The computing device 800 typically includes a variety of computer-readable media. Computer-readable media can be any available media that can be accessed by the computing device 800, and includes volatile and nonvolatile media, removable and non-removable media. By way of example, but not limitation, computer-readable media may include computer storage media and communication media. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data.

[0082] Computer storage media include RAM, ROM, EEPROM, flash memory or other storage technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices. Computer storage media do not include propagated data signals.

[0083] Communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism, and includes any information delivery media. The term "modulated data signal" refers to a signal that has its characteristics set or changed in a manner that encodes information in the signal. By way of example, but not limitation, communication media include wired media such as a wired network or a direct wired connection, and wireless media such as acoustic, RF, infrared and other wireless media. Any combination of the above should also be included within the scope of computer readable media.

[0084] Memory 820 includes computer storage media in the form of volatile and / or non-volatile memory. Memory 820 may be removable, non-removable, or a combination thereof. Exemplary memory includes solid-state memory, hard disk drives, optical drives, and the like. Computing device 800 includes a processor 830 that reads data from various entities such as bus 810, memory 820, or I / O components 860. (Multiple) presentation components 840 present data indications to a user or other device. Exemplary presentation components 840 include display devices, speakers, printing components, vibration components, and the like. I / O ports 850 allow computing device 800 to be logically coupled to other devices including I / O components 860, some of which may be built-in.

[0085] In various embodiments, memory 820 includes, among other things, temporary and persistent copies of masquerade detection logic 822. The masquerade detection logic 822 includes instructions that, when executed by processor 830, cause computing device 800 to perform masquerade detection functions, such as, but not limited to, process 300, process 500, process 600, and process 700. In various embodiments, the masquerade detection logic 822 includes instructions that, when executed by processor(s) 830, cause computing device 800 to perform masquerade detection functions in conjunction with Figure 1 The detection system 130 or the reaction system 140 or the combination Figure 2 Various functions associated with but not limited to the searcher 210, classifier 220 or detector 230.

[0086] In some embodiments, the processor 830 may be packaged together with the masquerade detection logic 822. In some embodiments, the processor 830 may be packaged together with the masquerade detection logic 822 to form a system-in-package (SiP). In some embodiments, the processor 830 may be integrated with the masquerade detection logic 822 on the same die. In some embodiments, the processor 830 may be integrated with the masquerade detection logic 822 on the same die to form a system on chip (SoC).

[0087] Illustrative I / O components include microphones, joysticks, game pads, satellite dishes, scanners, printers, display devices, wireless devices, controllers (such as stylus, keyboards, and mice), natural user interfaces (NUIs), and the like. In various aspects, a pen digitizer (not shown) and accompanying input tools (also not shown but may include, by way of example, a pen or stylus) are provided to digitally capture freehand user input. The connection between the pen digitizer and the (multiple) processors 830 may be direct, or may be coupled via the use of serial ports, parallel ports, and / or other interfaces and / or system buses known in the art. In addition, the digitizer input component may be a separate component from an output component such as a display device, or in some aspects, the available input area of ​​the digitizer may coexist with the display area of ​​the display device and be integrated with the display device, or may exist as a separate device overlaying or attached to the display device. Any and all such variations and any combination thereof are considered to be within the scope of various aspects of the technology described herein.

[0088] The computing device 800 may include a networking interface 880. The networking interface 880 includes a network interface controller (NIC) that sends and receives data. The networking interface 880 may use wired technology (e.g., coaxial cable, twisted pair, optical fiber, etc.) or wireless technology (e.g., terrestrial microwave, communication satellite, cellular, radio, and spread spectrum technology, etc.). In particular, the networking interface 880 may include a wireless terminal suitable for receiving communications and media through various wireless networks. The computing device 800 may communicate with other devices via wireless protocols such as code division multiple access (CDMA), global system for mobile (GSM), or time division multiple access (TDMA), as well as other protocols, via the networking interface 880. Radio communications may be short-range connections, long-range connections, or a combination of short-range and long-range wireless telecommunication connections. Short-range connections may include connections to devices that provide access to wireless communication networks (e.g., mobile hotspots). A connection such as a wireless local area network (WLAN) connection using the 802.11 protocol. A Bluetooth connection to another computing device is a second example of a short-range connection. A long-range connection may include a connection using any of the CDMA, GPRS, GSM, TDMA, and 802.16 protocols.

[0089] The technology described herein has been described with respect to specific aspects, which are intended in all respects to be illustrative rather than restrictive. Although the technology described herein is susceptible to various modifications and alternative constructions, certain illustrated aspects thereof are shown in the accompanying drawings and have been described in detail above. However, it should be understood that there is no intention to limit the technology described herein to the specific forms disclosed, but rather, it is intended to cover all modifications, alternative constructions, and equivalents that fall within the spirit and scope of the technology described herein.

Claims

1. A computer-implemented method comprising: accessing a search string associated with an electronic message; generating an alternative form of the search string using a transformation protocol that replaces original characters with Unicode characters or visually similar characters; identifying a plurality of n-grams of the alternative forms of the search string; accessing a reverse n-gram mapping comprising n-grams mapped to a plurality of protected strings, wherein the reverse n-gram mapping comprises n-grams of alternate forms of the plurality of protected strings, the alternate forms of the plurality of protected strings being generated based on the transformation protocol; searching the reverse n-gram mapping for the plurality of n-grams of the alternative forms of the search string; determining a count of the plurality of n-grams of the alternative forms of the search string based on searching the reverse n-gram mapping, the search string corresponding to a protected string in the plurality of protected strings associated with the reverse n-gram mapping; Based on the count, determining that the search string is an instance of a fuzzy inclusion of the protected string; as well as Based on determining that the search string is an instance of the ambiguous subsumption of the protected string, a modification operation associated with the electronic message is performed.

2. The method according to claim 1, wherein the conversion protocol comprises: Based on a visual similarity between a letter in a string and another letter, the letter is replaced with the other letter.

3. The method according to claim 1, wherein the conversion protocol comprises: Replaces letters of a string from one language to another.

4. The method according to claim 1, further comprising: The reverse n-gram mapping for the protected string in the plurality of protected strings is created by associating the protected string with each n-gram of the protected string in a data structure representing the reverse n-gram mapping.

5. The method of claim 1 , wherein determining that the search string is an instance of the fuzzy containment of the protected string is based on a ratio of n-grams of the plurality of n-grams of the alternative forms of the search string that the search string matches with n-grams of the alternative forms of the protected string.

6. The method according to claim 1, further comprising: The start index and the end index for the instance of the fuzzy containment are located by minimizing the edit distance between the protected string and a substring of the search string determined by the start index and the end index.

7. The method according to claim 1, further comprising: training a neural network using instances of the ambiguous inclusions of the protected string as features; as well as The neural network is used to detect ambiguous inclusions in electronic messages.

8. A computer storage device comprising computer-implemented instructions stored thereon, which instructions, when used by one or more hardware computer processors of a computing device, cause the computing device to: accessing a search string associated with an electronic message; generating an alternative form of the search string using a transformation protocol that replaces original characters with Unicode characters or visually similar characters; identifying a plurality of n-grams of the alternative forms of the search string; accessing a reverse n-gram mapping comprising n-grams mapped to a plurality of protected strings, wherein the reverse n-gram mapping comprises n-grams of alternate forms of the plurality of protected strings, the alternate forms of the plurality of protected strings being generated based on the transformation protocol; searching the reverse n-gram mapping for a plurality of n-grams of alternative forms of a search string associated with an electronic message; determining a count of the plurality of n-grams of the alternative forms of the search string based on searching the reverse n-gram mapping, the search string corresponding to a protected string in the plurality of protected strings associated with the reverse n-gram mapping; Based on the count, determining that the search string is an instance of a fuzzy inclusion of the protected string; as well as Based on determining that the search string is an instance of the ambiguous subsumption of the protected string, a modification operation associated with the electronic message is performed.

9. The computer storage device of claim 8, wherein the transformation protocol comprises: Based on a visual similarity between a letter in a string and another letter, the letter is replaced with the other letter.

10. The computer storage device of claim 8, wherein the instructions further cause the computing device to: Instances of the ambiguous inclusion of the protected string in the search string are determined based on ratios of n-grams in the plurality of n-grams of the alternative forms of the search string that match the n-grams of the alternative forms of the protected string.

11. The computer storage device of claim 8, wherein the instructions further cause the computing device to: The reverse n-gram mapping is created for the protected string in the plurality of protected strings by associating the protected string with each n-gram of the protected string in a data structure representing the reverse n-gram mapping.

12. The computer storage device of claim 8, wherein the instructions further cause the computing device to: The fuzzily contained instances of the protected string in the search string are assigned types based on characteristics of the fuzzily contained instances.

13. The computer storage device of claim 12, wherein the instructions further cause the computing device to: locating the start index and the end index for the instance of the fuzzy containment by minimizing the edit distance between the protected string and a substring of the search string determined by the start index and the end index; and The type is assigned based on the start index, the end index, or the edit distance.

14. The computer storage device of claim 12, wherein the type is assigned based on respective lengths or capitalization of the protected string and the search string.

15. The computer storage device of claim 12, wherein the characteristics of the instances of the fuzzy inclusion include delimiters in the search string, languages ​​used in the search string, or counts of languages ​​used in the search string.

16. The computer storage device of claim 12, wherein the instructions further cause the computing device to: training a neural network using the types of instances of the fuzzy inclusion as features; and The neural network is utilized to detect ambiguous inclusions in electronic messages.

17. An apparatus for detecting camouflage, comprising: one or more hardware computer processors; as well as a computer memory storing usable instructions that, when used by the one or more hardware computer processors, cause the one or more hardware computer processors to perform operations comprising: accessing a search string associated with an electronic message; generating an alternative form of the search string using a transformation protocol that replaces original characters with Unicode characters or visually similar characters; identifying a plurality of n-grams of the alternative forms of the search string; accessing a reverse n-gram mapping comprising n-grams mapped to a plurality of protected strings, wherein the reverse n-gram mapping comprises n-grams of alternate forms of the plurality of protected strings, the alternate forms of the plurality of protected strings being generated based on the transformation protocol; searching the reverse n-gram mapping for a plurality of n-grams of alternative forms of a search string associated with an electronic message; determining a count of the plurality of n-grams of the alternative forms of the search string based on searching the reverse n-gram mapping, the search string corresponding to a protected string in the plurality of protected strings associated with the reverse n-gram mapping; Based on the count, determining that the search string is an instance of an ambiguous inclusion of the protected string; and Based on determining that the search string is an instance of the ambiguous subsumption of the protected string, a modification operation associated with the electronic message is performed.

18. The apparatus of claim 17, wherein the transformation protocol comprises: Based on a visual similarity between a letter in a string and another letter, the letter is replaced with the other letter.

19. The apparatus of claim 17, wherein the operations further comprise: The reverse n-gram mapping is created for the protected string in the plurality of protected strings by associating the protected string with each n-gram of the protected string in a data structure representing the reverse n-gram mapping.

20. The apparatus of claim 19, wherein determining that the search string is an instance of the fuzzy containment of the protected string is based on a ratio of n-grams of the plurality of n-grams of the alternative forms of the search string that the search string matches with n-grams of the alternative forms of the protected string.