Compliance short message auditing scheme applied to dispute case mediation

Abstract

The invention relates to the technical field of short message auditing, and discloses a compliance short message auditing scheme applied to dispute case mediation, which comprises the following steps: step S101, constructing a keyword library which comprises a plurality of sensitive words; step S102, acquiring a mediation short message text made by an endorsement provider, and judging whether sensitive words appear in the mediation short message text or not; s103, a character length library is constructed, the character length library comprises m common character lengths, and the character lengths represent the number of characters in each row of the short message text. According to the method, when the mediation short message is processed, whether sensitive words exist in the short message text or not is considered, misunderstanding of dispute parties due to the typesetting problem is also considered, and various short message reading typesetting modes of different mobile phones are simulated in the process of analyzing the mediation short message text, so that the short message text is analyzed, and the short message reading efficiency is improved. And each typesetting mode is analyzed, so that misunderstanding of dispute parties is avoided, and the credibility of endorsement providers is prevented from being influenced.

Description

Technical Field

[0001] The present invention relates to the technical field of SMS auditing, and more specifically, to a compliant SMS auditing solution applied to mediation dispute cases. Background Art

[0002] Contacting the parties to a mediation case by sending text messages is a basic means of initiating effective mediation and facilitating the parties to the dispute to reach a settlement consensus. However, under the current situation, it is impossible to completely filter out non-compliant text messages by relying solely on text message service providers. Incorrect text message content may damage the credibility of the mediation endorsement provider and cause deception and distress to the parties involved in the mediation.

[0003] The existing technology is to connect to multiple SMS provider channels, ignore the provider category when sending notification SMS, and send through multiple channels. The control of SMS signature and SMS content is in the hands of the developer / user of the dispute resolution system, and the pressure of content filtering is on the SMS provider.

[0004] However, since different SMS providers have different filtering strategies, the compliance of the content sent cannot be guaranteed. It is easy for non-compliant SMS content to be sent to the parties in the dispute, damaging the judicial credibility of the endorsement provider. Summary of the invention

[0005] The present invention provides a compliant SMS review solution for use in mediation dispute cases, which solves the technical problems in the above-mentioned background technology.

[0006] The present invention provides a compliant SMS review solution for mediation dispute cases, comprising the following steps:

[0007] Step S101, constructing a keyword library, the keyword library includes a number of sensitive words;

[0008] Step S102, obtaining the mediation text message prepared by the endorsement provider, and determining whether the mediation text message contains sensitive words;

[0009] If yes, the mediation SMS text will not be sent and sensitive words will be marked;

[0010] If not, proceed to step S103;

[0011] Step S103, constructing a character length library, the character length library contains m commonly used character lengths, and the character length represents the number of characters in each line of the SMS text;

[0012] Step S104, arranging and adjusting the SMS texts according to the m types of character lengths in the character length library to generate m detection SMS texts;

[0013] Step S105, determining whether there is an arrangement of the m test text messages that misleads the parties into mistaking a combination of normal characters for a sensitive word;

[0014] If yes, adjust the text of the detected SMS and go to step S104;

[0015] If not, the detection SMS text is sent to the parties to the dispute as a mediation SMS text;

[0016] m is a custom parameter.

[0017] Furthermore, the specific steps for constructing a keyword library are as follows:

[0018] Step S110, inputting a number of words as first sensitive words into the keyword library;

[0019] Step S120, generating derivative words according to the vocabulary in the keyword library, and listing the derivative words as first sensitive words in the keyword library, wherein the derivative words include words that have the same pronunciation as the first sensitive words but different characters and have no meaning of their own;

[0020] Step S130, obtaining a word that has the same pronunciation as the first sensitive word and has its own meaning, and inputting it into the keyword library as the second sensitive word.

[0021] Furthermore, the specific steps for determining whether sensitive words appear in the mediation text are as follows:

[0022] Step S201, searching the vocabulary in the keyword library to determine whether a first sensitive word appears in the text of the mediation text message;

[0023] If yes, the mediation SMS text will not be sent and sensitive words will be marked;

[0024] If not, proceed to step S202;

[0025] Step S202, searching the words in the mediation text message to see if a second sensitive word appears;

[0026] If yes, proceed to step S203;

[0027] If not, proceed to step S103;

[0028] Step S203, constructing a semantic analysis model, inputting the sentence containing the second sensitive word in the mediation SMS text into the semantic analysis model, outputting the meaning of the second sensitive word, and determining whether it is the same as the first sensitive word;

[0029] If yes, the mediation SMS text will not be sent and sensitive words will be marked;

[0030] If not, go to step S103.

[0031] Furthermore, the calculation formula of the semantic analysis model is as follows:

[0032]

[0033] Q=W Q ×X;

[0034] K=W K ×X;

[0035] V=W V ×X;

[0036] In the formula, Attention(Q, K, V) represents the meaning of the second sensitive word output by the semantic analysis model; Q, K and V represent the first, second and third conversion features of the semantic analysis model respectively; X represents the sentence where the second sensitive word in the mediation text message is located; T represents the transposition operation of the matrix; W1 and W2 represent the first and second weight parameters of the semantic analysis model respectively; W Q , W K and W V denote the first, second, and third weight matrices respectively; softmax denotes the softmax activation function.

[0037] Furthermore, the specific steps of determining whether there is an arrangement pattern in the m test text messages that misleads the parties into mistaking a combination of normal characters for a sensitive word are as follows:

[0038] Step S301, constructing a detection window set, where the detection window set includes h detection windows;

[0039] Step S302, the detection window slides in the m adjustment text messages respectively;

[0040] Step S303: Each time the detection window slides, a detection word is generated, and whether the detection word belongs to the first sensitive word is determined.

[0041] Furthermore, the specific steps of constructing the detection window set are as follows:

[0042] Step S401, preset a piece of experimental text, and divide each character of the experimental text into a cell;

[0043] Step S402, randomly selecting a first sensitive word with a character length of not less than 2, breaking up the first sensitive word and randomly inserting it into the experimental text;

[0044] Step S403, select an experimenter to read the experiment text. If the experimenter recognizes the sensitive word, the arrangement of the cells of the characters of the sensitive word is used as a detection window.

[0045] Furthermore, the specific steps of determining whether there is an arrangement pattern in the m test text messages that misleads the parties into mistaking a combination of normal characters for a sensitive word are as follows:

[0046] Step S501, constructing graph network data according to m detected SMS texts;

[0047] The graph network data includes a first node and an edge between the first nodes;

[0048] The first node is generated according to the data of the character map;

[0049] The existence of an edge between the first nodes indicates that: in one of the mediation SMS texts, two characters are adjacent;

[0050] Step S502, input the graph network data into the sensitive word analysis model, and output 1 or 0, 1 means that there are m detected SMS texts with arrangements that mislead the parties into mistaking normal character combinations for sensitive words; 0 means there are none.

[0051] Furthermore, the sensitive word analysis model is built based on the graph neural network model.

[0052] Furthermore, the calculation formula of the sensitive word analysis model is as follows:

[0053]

[0054] In the formula, represents the feature vector of node v at the kth layer; N(v) represents the set of nodes that have edges with node v; Indicates the features of the nodes in the k-1 layer that have edges with node v; Represents the feature vector of node v at the k-1th layer; the AGGREGATE function is used to aggregate all UPDATE function is used for aggregation and

[0055] Furthermore, the training samples used to train the sensitive word analysis model include:

[0056] Randomly select a piece of text as training text, select a character length for arrangement, randomly select a first sensitive word with a character length of not less than 2, split the first sensitive word and randomly insert it into the training text, which serves as sample data of a training sample;

[0057] The sample labels of training samples are manually annotated.

[0058] The beneficial effect of the present invention is that when processing a mediation text message, the present invention not only takes into account whether there are sensitive words in the text of the text message, but also takes into account the misunderstanding of the parties to the dispute due to typesetting problems. In the process of analyzing the text of the mediation text message, the present invention simulates various typesetting methods of reading text messages on different mobile phones, and analyzes each typesetting method to avoid misunderstandings among the parties to the dispute and affect the credibility of the endorsement provider. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is the overall flow chart of the present invention. DETAILED DESCRIPTION

[0060] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0061] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0062] like Figure 1 As shown, the purpose of the first embodiment of the present invention is that due to different filtering strategies of SMS providers, it is impossible to ensure that the SMS sent to the parties to the dispute are legal and compliant, which will damage the judicial credibility of the endorsement provider. In addition, due to the problem of text typesetting, when sending in the form of SMS, it is possible to mislead the parties to the dispute. Therefore, a compliant SMS review scheme applied to mediation dispute cases is proposed, including the following steps:

[0063] Step S101, constructing a keyword library, the keyword library includes a number of sensitive words;

[0064] The specific steps to build a keyword library are as follows:

[0065] Step S110, inputting a number of words as first sensitive words into the keyword library;

[0066] Step S120, generating derivative words according to the vocabulary in the keyword library, and listing the derivative words as first sensitive words in the keyword library, wherein the derivative words include words that have the same pronunciation as the first sensitive words but different characters and have no meaning of their own;

[0067] Step S130, obtaining a word that has the same pronunciation as the first sensitive word and has its own meaning, and inputting it into the keyword library as the second sensitive word.

[0068] It is worth noting that when the first sensitive word appears in the mediation SMS text, it means that the mediation SMS text is illegal / compliant; when only the second sensitive word appears in the mediation SMS text, it means that the mediation SMS text may be illegal / compliant and needs to be analyzed based on the context.

[0069] Step S102, obtaining the mediation text message prepared by the endorsement provider, and determining whether the mediation text message contains sensitive words;

[0070] If yes, the mediation SMS text will not be sent and sensitive words will be marked;

[0071] If not, proceed to step S103;

[0072] The specific steps to determine whether sensitive words appear in the mediation text are as follows:

[0073] Step S201, searching the vocabulary in the keyword library to determine whether a first sensitive word appears in the text of the mediation text message;

[0074] If yes, the mediation SMS text will not be sent and sensitive words will be marked;

[0075] If not, proceed to step S202;

[0076] Step S202, searching the words in the mediation text message to see if a second sensitive word appears;

[0077] If yes, proceed to step S203;

[0078] If not, proceed to step S103;

[0079] Step S203, constructing a semantic analysis model, inputting the sentence containing the second sensitive word in the mediation SMS text into the semantic analysis model, outputting the meaning of the second sensitive word, and determining whether it is the same as the first sensitive word;

[0080] If yes, the mediation SMS text will not be sent and sensitive words will be marked;

[0081] If not, go to step S103.

[0082] The semantic analysis model is built based on the attention mechanism.

[0083] The calculation formula of the semantic analysis model is as follows:

[0084]

[0085] Q=W Q ×X;

[0086] K=W K ×X;

[0087] V=W V ×X;

[0088] In the formula, Attention(Q, K, V) represents the meaning of the second sensitive word output by the semantic analysis model; Q, K and V represent the first, second and third conversion features of the semantic analysis model respectively; X represents the sentence where the second sensitive word in the mediation text message is located; T represents the transposition operation of the matrix; W1 and W2 represent the first and second weight parameters of the semantic analysis model respectively; W Q , W K and W V denote the first, second, and third weight matrices respectively; softmax denotes the softmax activation function.

[0089] It should be noted that the training of semantic analysis models is a relatively mature existing technology and will not be elaborated here.

[0090] It should be noted that due to differences in screen size, resolution, font size type, operating system, UI design and reading mode settings of SMS applications, the length of characters displayed in each line when reading SMS on a mobile phone will be different, which will lead to differences in typesetting. Therefore, after searching for sensitive words in the content of the mediation SMS text, due to typesetting issues, sensitive words may appear in the vertical or tilted direction of the SMS, which can easily cause misunderstandings among the parties to the dispute.

[0091] Therefore, in step S103, a character length library is constructed, which contains m commonly used character lengths, and the character length represents the number of characters in each line of the SMS text;

[0092] The length of characters displayed on a mobile phone screen is limited by the size of the screen, generally between 5 and 15, so the character length library generally contains 11 character lengths. In addition, the character length library can be changed accordingly according to market changes, and the character length in the character length library can be appropriately increased or decreased.

[0093] Step S104, arranging and adjusting the SMS texts according to the m types of character lengths in the character length library to generate m detection SMS texts;

[0094] Step S105, determining whether there is an arrangement of the m test text messages that misleads the parties into mistaking a combination of normal characters for a sensitive word;

[0095] If yes, adjust the text of the detected SMS and go to step S104;

[0096] If not, the detection SMS text is sent to the parties to the dispute as a mediation SMS text;

[0097] m is a custom parameter, preferably, m is 11.

[0098] The specific steps of determining whether there is an arrangement pattern in the m test text messages that misleads the parties into mistaking a combination of normal characters for a sensitive word are as follows:

[0099] Step S301, constructing a detection window set, where the detection window set includes h detection windows;

[0100] Step S302, the detection window slides in the m adjustment text messages respectively;

[0101] Step S303: Each time the detection window slides, a detection word is generated, and whether the detection word belongs to the first sensitive word is determined.

[0102] The specific steps to construct a detection window set are as follows:

[0103] Step S401, preset a piece of experimental text, and divide each character of the experimental text into a cell;

[0104] Step S402, randomly selecting a first sensitive word with a character length of not less than 2, breaking up the first sensitive word and randomly inserting it into the experimental text;

[0105] Step S403, select an experimenter to read the experiment text. If the experimenter recognizes the sensitive word, the arrangement of the cells of the characters of the sensitive word is used as a detection window.

[0106] It should be noted that the purpose of adjusting the detection SMS text is to avoid the content therein being easily misunderstood by the parties to the dispute due to the arrangement of the mediation SMS text. Therefore, the arrangement can be adjusted appropriately, and operations such as adding space characters or irrelevant symbols can be adopted. For example, adding or deleting book title marks in a file name can appropriately adjust the arrangement of the SMS text to avoid the above situation.

[0107] The purpose of the second embodiment of the present invention is that in addition to setting up the detection mine mouth, a neural network model can also be used for analysis. Therefore, the specific steps of judging whether there is an arrangement pattern in the m detection text messages that misleads the parties to mistake the combination of normal characters for sensitive words are as follows:

[0108] Step S501, constructing graph network data according to m detected SMS texts;

[0109] The graph network data includes a first node and an edge between the first nodes;

[0110] The first node is generated according to the data of the character map;

[0111] The existence of an edge between the first nodes indicates that: in one of the mediation SMS texts, two characters are adjacent;

[0112] Step S502, input the graph network data into the sensitive word analysis model, and output 1 or 0, 1 means that there are m detected SMS texts with arrangements that mislead the parties into mistaking normal character combinations for sensitive words; 0 means there are none.

[0113] The sensitive word analysis model is built based on the graph neural network model.

[0114] The calculation formula of the sensitive word analysis model is as follows:

[0115]

[0116] In the formula, represents the feature vector of node v at the kth layer; N(v) represents the set of nodes that have edges with node v; Indicates the features of the nodes in the k-1 layer that have edges with node v; Represents the feature vector of node v at the k-1th layer; the AGGREGATE function is used to aggregate all UPDATE function is used for aggregation and

[0117] The training samples used to train the sensitive word analysis model include:

[0118] Randomly select a piece of text as training text, select a character length for arrangement, randomly select a first sensitive word with a character length of not less than 2, split the first sensitive word and randomly insert it into the training text, which serves as sample data of a training sample;

[0119] The sample labels of training samples are manually annotated.

[0120] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A compliant SMS review solution for mediation dispute cases. It is characterized in that The steps include: Step S101, constructing a keyword library, the keyword library includes a number of sensitive words; Step S102, obtaining the mediation text message prepared by the endorsement provider, and determining whether the mediation text message contains sensitive words; If yes, the mediation SMS text will not be sent and sensitive words will be marked; If not, proceed to step S103; Step S103, constructing a character length library, the character length library contains m commonly used character lengths, and the character length represents the number of characters in each line of the SMS text; Step S104, arranging and adjusting the SMS texts according to the m types of character lengths in the character length library to generate m detection SMS texts; Step S105, determining whether there is an arrangement pattern in the m test text messages that misleads the parties into mistaking a combination of normal characters for a sensitive word; If yes, adjust the text of the detected SMS and go to step S104; If not, the detection SMS text is sent to the parties to the dispute as a mediation SMS text; m is a custom parameter.

2. According to claim 1, a compliant SMS review solution for mediation dispute cases, It is characterized in that The specific steps to build a keyword library are as follows: Step S110, inputting a number of words as first sensitive words into the keyword library; Step S120, generating derivative words according to the vocabulary in the keyword library, and listing the derivative words as first sensitive words in the keyword library, wherein the derivative words include words that have the same pronunciation as the first sensitive words but different characters and have no meaning of their own; Step S130, obtaining a word that has the same pronunciation as the first sensitive word and has its own meaning, and inputting it into the keyword library as the second sensitive word.

3. According to claim 2, a compliant SMS review solution for mediation dispute cases, It is characterized in that The specific steps to determine whether sensitive words appear in the mediation text are as follows: Step S201, searching the vocabulary in the keyword library to determine whether a first sensitive word appears in the text of the mediation text message; If yes, the mediation SMS text will not be sent and sensitive words will be marked; If not, proceed to step S202; Step S202, searching the words in the mediation text message to see if a second sensitive word appears; If yes, proceed to step S203; If not, proceed to step S103; Step S203, constructing a semantic analysis model, inputting the sentence containing the second sensitive word in the mediation SMS text into the semantic analysis model, outputting the meaning of the second sensitive word, and determining whether it is the same as the first sensitive word; If yes, the mediation SMS text will not be sent and sensitive words will be marked; If not, go to step S103.

4. According to claim 3, a compliant SMS review solution for mediation dispute cases, It is characterized in that The calculation formula of the semantic analysis model is as follows: Q=W Q ×X; K=W K ×X; V=W V ×X; In the formula, Attention(Q, K, V) represents the meaning of the second sensitive word output by the semantic analysis model; Q, K, and V represent the first, second, and third conversion features of the semantic analysis model, respectively; X represents the sentence where the second sensitive word in the mediation text is located; T represents the transposition operation of the matrix; W 1 and W 2 Respectively represent the first and second weight parameters of the semantic analysis model; W Q , W K and W V denote the first, second, and third weight matrices respectively; softmax denotes the softmax activation function.

5. According to claim 2, a compliant SMS review solution for mediation dispute cases, It is characterized in that The specific steps of determining whether there is an arrangement pattern in the m test text messages that misleads the parties into mistaking a combination of normal characters for a sensitive word are as follows: Step S301, constructing a detection window set, where the detection window set includes h detection windows; Step S302, the detection window slides in the m adjustment text messages respectively; Step S303: Each time the detection window slides, a detection word is generated, and whether the detection word belongs to the first sensitive word is determined.

6. According to claim 5, a compliant SMS review solution for mediation of dispute cases, It is characterized in that The specific steps to construct a detection window set are as follows: Step S401, preset a piece of experimental text, and divide each character of the experimental text into a cell; Step S402, randomly selecting a first sensitive word with a character length of not less than 2, breaking up the first sensitive word and randomly inserting it into the experimental text; Step S403, select an experimenter to read the experiment text. If the experimenter recognizes the sensitive word, the arrangement of the cells of the characters of the sensitive word is used as a detection window.

7. According to claim 1, a compliant SMS review solution for mediation dispute cases, It is characterized in that The specific steps of determining whether there is an arrangement pattern in the m test text messages that misleads the parties into mistaking a combination of normal characters for a sensitive word are as follows: Step S501, constructing graph network data according to m detected SMS texts; The graph network data includes a first node and an edge between the first nodes; The first node is generated according to the data of the character map; The existence of an edge between the first nodes indicates that: in one of the mediation SMS texts, two characters are adjacent; Step S502, input the graph network data into the sensitive word analysis model, and output 1 or 0, 1 means that there are m detected SMS texts with arrangements that mislead the parties into mistaking normal character combinations for sensitive words; 0 means there are none.

8. According to claim 7, a compliant SMS review solution for mediation of dispute cases, It is characterized in that The sensitive word analysis model is built based on the graph neural network model.

9. According to claim 7, a compliant SMS review solution for mediation of dispute cases, It is characterized in that The calculation formula of the sensitive word analysis model is as follows: In the formula, represents the feature vector of node v at the kth layer; N(v) represents the set of nodes that have edges with node v; Indicates the features of the nodes in the k-1 layer that have edges with node v; Represents the feature vector of node v at the k-1th layer; the AGGREGATE function is used to aggregate all UPDATE function is used for aggregation and 10. A compliant SMS review solution for mediation of dispute cases according to claim 9, It is characterized in that The training samples used to train the sensitive word analysis model include: Randomly select a piece of text as training text, select a character length for arrangement, randomly select a first sensitive word with a character length of not less than 2, split the first sensitive word and randomly insert it into the training text, which serves as sample data of a training sample; The sample labels of training samples are manually annotated.

Images