A sensitive text recognition method, system, electronic device and storage medium

Abstract

The present invention proposes a sensitive text recognition method, system, electronic device and storage medium, and the method and technical scheme thereof include a multi-branch tree text expansion step, pre-expanding the characters in the sensitive text, and expanding the sensitive text according to the pre-expansion result through the multi-branch tree; a DFA text expansion step, expanding the sensitive text containing invalid characters through the DFA technology; a text classification expansion step, collecting the sensitive text as a positive sample, and collecting non-sensitive text as a negative sample, and expanding the sensitive text through a text classification algorithm according to the positive sample and the negative sample; a fusion text recognition step, identifying the sensitive text according to the number of characters in the sensitive text through the multi-branch tree, the DFA and the text classification algorithm. The present application solves the problem that the existing sensitive text recognition method has a high false positive rate and unsatisfactory effect.

Description

Technical Field

[0001] The present invention belongs to the technical field of text recognition, and in particular relates to a sensitive text recognition method, system, electronic device and storage medium. Background Art

[0002] In the online world, there are various sensitive texts such as insults and pornography, which will bring a very bad experience to other users. Some people change the text slightly to avoid recognition. Therefore, sensitive recognition is of great significance to purifying the online environment. Summary of the invention

[0003] The embodiments of the present application provide a sensitive text recognition method, system, electronic device and storage medium to at least solve the problems of high false positive rate and unsatisfactory effect of existing sensitive text recognition methods.

[0004] In a first aspect, an embodiment of the present application provides a sensitive text recognition method, comprising: a multi-branch tree text expansion step, pre-expanding the characters in the sensitive text, and expanding the sensitive text according to the result of the pre-expansion through a multi-branch tree; a DFA text expansion step, expanding the sensitive text containing invalid characters through the DFA technology; a text classification expansion step, collecting the sensitive text as a positive sample, and collecting non-sensitive text as a negative sample, and expanding the sensitive text through a text classification algorithm based on the positive sample and the negative sample; a fusion text recognition step, recognizing the sensitive text through the multi-branch tree, the DFA and the text classification algorithm according to the number of characters in the sensitive text.

[0005] Preferably, the text classification expansion step further comprises: expanding the positive samples by performing the pre-expansion on the words in the sensitive text.

[0006] Preferably, the text classification expansion step further comprises: randomly segmenting the negative samples, and performing text classification on the obtained segmented blocks and the positive samples through binary classification.

[0007] Preferably, the fused text recognition step further includes: if the number of words in the sensitive text is less than a threshold, recognition is performed through the multi-branch tree and the DFA, otherwise recognition is performed through the text classification algorithm.

[0008] In a second aspect, an embodiment of the present application provides a sensitive text recognition system, which is applicable to the above-mentioned sensitive text recognition method, including: a multi-branch tree text expansion module, which pre-expands the characters in the sensitive text, and expands the sensitive text according to the result of the pre-expansion through a multi-branch tree; a DFA text expansion module, which expands the sensitive text containing invalid characters through the DFA technology; a text classification expansion module, which collects the sensitive text as a positive sample, and collects non-sensitive text as a negative sample, and expands the sensitive text through a text classification algorithm according to the positive sample and the negative sample; a fusion text recognition module, which recognizes the sensitive text according to the number of characters in the sensitive text through the multi-branch tree, the DFA and the text classification algorithm.

[0009] In some embodiments, the text classification expansion module further comprises: expanding the positive samples by performing the pre-expansion on the words in the sensitive text.

[0010] In some embodiments, the text classification expansion module further includes: randomly segmenting the negative samples, and performing text classification on the obtained segmented blocks and the positive samples through binary classification.

[0011] In some of the embodiments, the fused text recognition module further includes: if the number of words in the sensitive text is less than a threshold, then recognition is performed through the multi-branch tree and the DFA; otherwise, recognition is performed through the text classification algorithm.

[0012] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a sensitive text recognition method as described in the first aspect above is implemented.

[0013] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a sensitive text recognition method as described in the first aspect above.

[0014] This application can be applied to the field of data capability technology. Compared with related technologies, the sensitive text recognition method provided in the embodiment of this application ensures a high accuracy and recall rate by integrating multiple recognition methods to meet the needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0016] Figure 1 This is a flow chart of the sensitive text recognition method of the present invention;

[0017] Figure 2 It is a framework diagram of the sensitive text recognition system of the present invention;

[0018] Figure 3 A framework diagram of an electronic device of the present invention;

[0019] In the above picture:

[0020] 1. Multi-tree text expansion module; 2. DFA text expansion module; 3. Text classification expansion module; 4. Fusion text recognition module; 60. Bus; 61. Processor; 62. Memory; 63. Communication interface. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0022] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0023] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0024] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a kind of", "the" and the like involved in this application do not indicate a quantitative limitation and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices.

[0025] In the online environment, people's behavior cannot be controlled. Some people may post insulting, reactionary, pornographic, violent and terrorist information, which pollutes the online environment. Such texts need to be identified and removed. In order to evade identification, some people slightly change the text and escape "arrest".

[0026] For example:

[0027] Change "I am eating"

[0028] Change to:

[0029] "I am a fan."

[0030] The above is the variant of the first pinyin of the word, and there are also radical variants, for example:

[0031] "I am a rebel."

[0032] This variation can be found everywhere on the Internet, and some invalid characters are added to the text to prevent it from being recognized, for example:

[0033] "I'm &&&eating &$&*-||-meal."

[0034] and combinations of the above variants, etc.

[0035] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings:

[0036] Figure 1 For a flow chart of the sensitive text recognition method of the present invention, please refer to Figure 1 The sensitive text recognition method of the present invention comprises the following steps:

[0037] S1: Pre-expand the characters in the sensitive text, and expand the sensitive text according to the result of the pre-expansion through a multi-branch tree.

[0038] In the specific implementation, each word in the sensitive vocabulary is processed as follows

[0039] Character decomposition, for example, decomposing "饭" into "饣反"

[0040] Decomposing the first letter of the pinyin, for example, decomposing "饭" into "f"

[0041] Decomposing the full pinyin, for example, decomposing "饭" into "fan"

[0042] Decomposing the full pinyin with tone marks, for example, decomposing "饭" into "fàn"

[0043] In specific implementation, each character has 5 possibilities. For a word of length N, there are 5 N possibilities for its variants. The embodiments of this application solve this problem of construction + traversal in the form of a multi-way tree.

[0044] In specific implementation, first recursively construct a multi-way tree. This application provides a pseudo-code to illustrate the tree construction algorithm:

[0045] Input: root node (denoted as root), extended array (two-dimensional array, each dimension is the different extended results of each word, denoted as lists), depth (initially passed in as 0, denoted as depth)

[0046] Output: A multi-way tree

[0047] If the depth is equal to the length of the original text, then exit; return null

[0048] for lst in lists:

[0049] nodes.append(tree(lst[depth]))

[0050] root.set_children(nodes)

[0051] for child in root.children:

[0052] create_tree(child,lists,depth+1)

[0053] In specific implementation, then traverse the multi-way tree. Adopt the depth-first traversal method. If it is a leaf, find a correct path.

[0054] (1) Visit vertex v; visited[v]=1; / / Before the algorithm execution, visited[n]=0

[0055] (2) w = the first adjacent vertex of vertex v;

[0056] (3) while (w exists)

[0057] if(w has not been accessed)

[0058] Recursively execute the algorithm starting from vertex w;

[0059] If w is a leaf:

[0060] Find a path.

[0061] w = the next adjacent point of vertex v

[0062] S2: Expand the sensitive text containing invalid characters by using DFA technology.

[0063] In the specific implementation, the main idea of ​​DFA is to expand in sensitive words. The idea is to make empty jumps for invalid characters, thereby solving the problem of invalid characters interspersed in characters, and identifying sensitive texts with invalid characters by deterministic finite automata (DFA).

[0064] S3: Collect the sensitive text as a positive sample, and collect the non-sensitive text as a negative sample, and expand the sensitive text through a text classification algorithm according to the positive sample and the negative sample.

[0065] Optionally, the positive samples are expanded by performing the pre-expansion on the words in the sensitive text.

[0066] Optionally, the negative sample is randomly segmented, and the obtained segmented blocks and the positive sample are subjected to text classification through binary classification.

[0067] In the specific implementation, before classification, positive samples (abusive and pornographic samples) and negative samples (normal samples) are first needed. There are many negative samples, but few positive samples. Collecting positive sample data is the first step in classification.

[0068] There are automated tools on the Internet that can generate relevant swear words. However, too many negative samples will inevitably lead to sample imbalance. In order to solve this problem, the present application provides a specific embodiment. In this specific embodiment, the following expansion method is used to expand the swear words samples:

[0069]

[0070]

[0071] In specific implementation, the above-mentioned extension rules can be used to randomly change the text in the sensitive text through the above-mentioned scheme, which will generate more sensitive text data and alleviate the problem of insufficient samples to a certain extent.

[0072] In the specific implementation, the number of positive samples generated is still much smaller than the number of negative samples. The embodiment of the present application adopts the bagging method to solve the problem, that is, the negative samples are randomly divided, and a binary classification algorithm is trained with the divided blocks and all the positive samples, and then the result is obtained by voting.

[0073] Optionally, algorithms such as textcnn, textrnn, textrcnn, fasttext, etc. can be used.

[0074] S4: The sensitive text is identified according to the number of words in the sensitive text by using the multi-branch tree, the DFA and the text classification algorithm.

[0075] Optionally, if the number of words in the sensitive text is less than a threshold, the text is identified by the multi-branch tree and the DFA; otherwise, the text is identified by the text classification algorithm.

[0076] In specific implementations, multi-branch tree expansion and DFA cannot solve problems with different contexts, while text classification algorithms use CNN or RNN to extract high-order features, which may contain these semantics and can solve such problems to a certain extent.

[0077] The problem with text classification algorithms is that they are less effective for short texts due to the data itself (most of the positive samples are long, with few short texts). However, multitree expansion and DFA perform well for short texts.

[0078] Therefore, in the specific implementation, the text with less than 5 characters can be identified by using multi-tree expansion and DFA. If sensitive text is identified, it is returned; if not, or the number of characters is greater than or equal to 5, the text classification algorithm is used for identification.

[0079] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0080] The embodiments of the present application provide a sensitive text recognition system, which is applicable to the above-mentioned sensitive text recognition method. As used below, the terms "unit", "module", etc. can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0081] Figure 2 For a framework diagram of a sensitive text recognition system according to the present invention, see Figure 2 , including:

[0082] Multi - fork tree text expansion module 1: Perform pre - expansion on the characters in the sensitive text, and expand the sensitive text according to the results of the pre - expansion through a multi - fork tree.

[0083] In specific implementation, each character in the sensitive vocabulary is processed as follows

[0084] Split the character, for example, split "饭" into "饣反"

[0085] Split the first letter of the pinyin, for example, split "饭" into "f"

[0086] Split the full pinyin, for example, split "饭" into "fan"

[0087] Split the full pinyin with tones, for example, split "饭" into "fàn"

[0088] In specific implementation, each character has 5 possibilities. For a word with length N, its variants have 5 N possibilities. The embodiments of the present application use a multi - fork tree to solve this construction + traversal problem.

[0089] In specific implementation, first recursively construct a multi - fork tree. The present application provides a pseudo - code to illustrate the tree construction algorithm:

[0090] Input: Root node (denoted as root), expanded array (two - dimensional array, each dimension is different expansion results of each word, denoted as lists), depth (initially passed in as 0, denoted as depth)

[0091] Output: A multi - fork tree

[0092] If the depth is equal to the length of the original text, exit; return null

[0093] for lst in lists:

[0094] nodes.append(tree(lst[depth]))

[0095] root.set_children(nodes)

[0096] for child in root.children:

[0097] create_tree(child,lists,depth + 1)

[0098] In specific implementation, then traverse the multi - fork tree. Adopt the depth - first traversal method. If it is a leaf, find a correct path.

[0099] (1) Visit vertex v; visited[v] = 1; / / before the algorithm is executed, visited[n] = 0

[0100] (2) w = the first adjacent point of vertex v;

[0101] (3)while(w exists)

[0102] if(w has not been accessed)

[0103] Recursively execute the algorithm starting from vertex w;

[0104] If w is a leaf:

[0105] Find a path.

[0106] w = the next adjacent point of vertex v

[0107] DFA text expansion module 2: expands the sensitive text containing invalid characters by using DFA technology.

[0108] In the specific implementation, the main idea of ​​DFA is to expand in sensitive words. The idea is to make empty jumps for invalid characters, thereby solving the problem of invalid characters interspersed in characters, and identifying sensitive texts with invalid characters by deterministic finite automata (DFA).

[0109] Text classification and expansion module 3: collects the sensitive text as positive samples and collects non-sensitive text as negative samples, and expands the sensitive text through a text classification algorithm according to the positive samples and the negative samples.

[0110] Optionally, the positive samples are expanded by performing the pre-expansion on the words in the sensitive text.

[0111] Optionally, the negative sample is randomly segmented, and the obtained segmented blocks and the positive sample are subjected to text classification through binary classification.

[0112] In the specific implementation, before classification, positive samples (abusive and pornographic samples) and negative samples (normal samples) are first needed. There are many negative samples, but few positive samples. Collecting positive sample data is the first step in classification.

[0113] There are automated tools on the Internet that can generate relevant swear words. However, too many negative samples will inevitably lead to sample imbalance. In order to solve this problem, the following expansion method is used to expand the swear words samples:

[0114]

[0115]

[0116] In specific implementation, the above-mentioned extension rules can be used to randomly change the text in the sensitive text through the above-mentioned scheme, which will generate more sensitive text data and alleviate the problem of insufficient samples to a certain extent.

[0117] In the specific implementation, the number of positive samples generated is still much smaller than the number of negative samples. The embodiment of the present application adopts the bagging method to solve the problem, that is, the negative samples are randomly divided, and a binary classification algorithm is trained with the divided blocks and all the positive samples, and then the result is obtained by voting.

[0118] Optionally, algorithms such as textcnn, textrnn, textrcnn, fasttext, etc. can be used.

[0119] Fusion text recognition module 4: for the sensitive text, according to the number of words in the sensitive text, the sensitive text is recognized by using the multi-branch tree, the DFA and the text classification algorithm.

[0120] Optionally, if the number of words in the sensitive text is less than a threshold, the text is identified by using the multi-branch tree and the DFA; otherwise, the text is identified by using the text classification algorithm.

[0121] In specific implementations, multi-branch tree expansion and DFA cannot solve problems with different contexts, while text classification algorithms use CNN or RNN to extract high-order features, which may contain these semantics and can solve such problems to a certain extent.

[0122] The problem with text classification algorithms is that they are less effective for short texts due to the data itself (most of the positive samples are long, with few short texts). However, multitree expansion and DFA perform well for short texts.

[0123] Therefore, in the specific implementation, the text with less than 5 characters can be identified by using multi-tree expansion and DFA. If sensitive text is identified, it is returned; if not, or the number of characters is greater than or equal to 5, the text classification algorithm is used for identification.

[0124] In addition, combined Figure 1 A sensitive text recognition method described may be implemented by an electronic device. Figure 3 It is a framework diagram of the electronic device of the present invention.

[0125] The electronic device may include a processor 61 and a memory 62 storing computer program instructions.

[0126] Specifically, the processor 61 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0127] Among them, the memory 62 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 62 may include a hard disk drive (HDD), a floppy disk drive, a solid state drive (SSD), a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 62 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 62 may be inside or outside the data processing device. In a specific embodiment, the memory 62 is a non-volatile memory. In a specific embodiment, the memory 62 includes a read-only memory (ROM) and a random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (Programmable Read-Only Memory, PROM for short), an erasable PROM (Erasable ProgrammableRead-Only Memory, EPROM for short), an electrically erasable PROM (Electrically Erasable ProgrammableRead-Only Memory, EEPROM for short), an electrically alterable ROM (Electrically Alterable Read-Only Memory, EAROM for short) or a flash memory (FLASH) or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), wherein the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0128] The memory 62 may be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 61 .

[0129] The processor 61 implements any sensitive text recognition method in the above embodiments by reading and executing computer program instructions stored in the memory 62.

[0130] In some of the embodiments, the electronic device may further include a communication interface 63 and a bus 60. Figure 3 As shown, the processor 61, the memory 62, and the communication interface 63 are connected via a bus 60 and communicate with each other.

[0131] The communication port 63 can realize data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0132] The bus 60 includes hardware, software or both, and couples the components of the electronic device to each other. The bus 60 includes but is not limited to at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. By way of example and not limitation, bus 60 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, bus 60 may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0133] The electronic device can execute a sensitive text recognition method in an embodiment of the present application.

[0134] In addition, in combination with a sensitive text recognition method in the above embodiment, the present application embodiment may provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any sensitive text recognition method in the above embodiment is implemented.

[0135] The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program codes.

[0136] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A sensitive text recognition method, characterized in that: include: A multi-branch tree text expansion step, pre-expanding the characters in the sensitive text, and expanding the sensitive text according to the pre-expansion result through the multi-branch tree; DFA text expansion step, expanding the sensitive text containing invalid characters by using DFA technology, wherein DFA performs a blank jump for invalid characters; A text classification and expansion step, collecting the sensitive text as a positive sample and collecting the non-sensitive text as a negative sample, and expanding the sensitive text by a text classification algorithm according to the positive sample and the negative sample; Integrating the text recognition step, the sensitive text is recognized according to the number of words in the sensitive text by using the multi-branch tree, the DFA and the text classification algorithm; Among them, the text classification expansion step further includes: expanding the positive sample by pre-expanding the words in the sensitive text; randomly segmenting the negative sample, and performing text classification on the obtained segmented blocks and the positive sample through binary classification.

2. The sensitive text recognition method according to claim 1, characterized in that: The fusion text recognition step further includes: if the number of words in the sensitive text is less than a threshold, recognition is performed through the multi-branch tree and the DFA, otherwise recognition is performed through the text classification algorithm.

3. A sensitive text recognition system, characterized in that: include: A multi-branch tree text expansion module performs a pre-expansion on the characters in the sensitive text, and expands the sensitive text according to the result of the pre-expansion through the multi-branch tree; A DFA text expansion module, which expands the sensitive text containing invalid characters by using DFA technology, wherein DFA performs a null jump on the invalid characters; A text classification expansion module collects the sensitive text as a positive sample and collects the non-sensitive text as a negative sample, and expands the sensitive text by a text classification algorithm according to the positive sample and the negative sample; A fusion text recognition module is used to recognize the sensitive text according to the number of words in the sensitive text by using the multi-branch tree, the DFA and the text classification algorithm; The text classification expansion module further includes: expanding the positive sample by pre-expanding the characters in the sensitive text; randomly segmenting the negative sample, and performing text classification on the segmented blocks and the positive sample through binary classification.

4. The sensitive text recognition system according to claim 3, characterized in that: The fusion text recognition module further includes: if the number of words in the sensitive text is less than a threshold, recognition is performed through the multi-branch tree and the DFA, otherwise recognition is performed through the text classification algorithm.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the sensitive text recognition method as described in any one of claims 1 to 2 is implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the sensitive text recognition method as described in any one of claims 1 to 2 is implemented.

Images