An intelligent short message interception prevention method based on AI deep learning

Abstract

The application discloses an intelligent short message anti-blocking method and device based on AI deep learning and a storage medium, and comprises the following steps: setting a short message sending demand acquisition module to acquire short message content required to be sent by each customer; setting an information classification module to detect and automatically classify the short message content information required to be sent; setting an intelligent short message reconstruction module to regenerate intelligent short messages and short message sending schemes by using the short message content information provided by the customer; setting a short message blocking simulation module to simulate short message blocking, evaluate a short message sending success rate and a click rate; when the sending success rate is lower than a threshold value, the short message blocking simulation module feeds back reasons that may cause blocking to the intelligent short message generation module, the intelligent short message generation module improves the short message content and regenerates intelligent short messages and short message sending schemes until the evaluation sending success rate is higher than the threshold value; and a user confirmation module is arranged, and when the sending success rate is higher than the threshold value, the intelligent short messages and the sending schemes can be fed back to the customer.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of short message interception, and particularly relates to an intelligent short message anti-interception method based on AI deep learning. BACKGROUND

[0002] With the rapid progress of information technology, mobile phone short messages have become an indispensable way of communication and information acquisition in people's daily life. However, the disorderly expansion of various commercial promotion activities and the rampant spread of fraudulent information have caused a large number of spam messages to continuously flow into users' mobile phones like a flood. These spam messages not only waste users' valuable time and effort to view and clean up one by one, but more seriously, they greatly interfere with the normal short message communication order. On the one hand, normal commercial promotion short messages are easily submerged in the vast amount of spam messages, making it difficult for target users to detect them; on the other hand, due to the rampant spread of spam messages, the existing short message interception mechanism often adopts a more stringent strategy to deal with this situation, which may cause normal commercial promotion short messages to be misjudged as spam messages and be intercepted. SUMMARY

[0003] In order to overcome the problems of false interception and inability to effectively intercept spam messages mentioned in the prior art, the application designs an intelligent short message anti-interception method based on AI deep learning, which can effectively solve the problems of false interception and inability to intercept new types of spam messages in the prior art through multi-dimensional label classification, user feedback adjustment, multi-stage interception and continuous learning, and provide a safe and convenient short message use environment for users.

[0004] To solve the above technical problems, the technical scheme adopted by the application is as follows: an intelligent short message anti-interception method based on AI deep learning, comprising the following steps:

[0005] Step S1: setting a short message sending demand acquisition module for acquiring short message content to be sent by each customer;

[0006] Step S2: setting an information classification module to detect and automatically classify the short message content information to be sent; information is divided into text, pictures and videos according to carriers, and text information is divided into greetings, main text, marketing promotion information and links according to content;

[0007] Step S3: setting an intelligent short message reconstruction module to regenerate intelligent short messages and short message sending schemes using the short message content information provided by the customer; the short message sending scheme includes sending number, sending target, sending time and sending interval;

[0008] Step S4: Set up a short message interception simulation module for simulating short message interception, and input the intelligent short message generated by the intelligent short message generation module and its short message sending scheme into the short message interception simulation module. The short message interception simulation module detects the content of the intelligent short message, evaluates the short message sending success rate and click rate in combination with the short message sending scheme, and marks possible interception reasons;

[0009] Step S5: When the sending success rate is lower than the threshold, the short message interception simulation module feeds back the reasons that may cause interception to the intelligent short message generation module. The intelligent short message generation module improves the short message content and regenerates the intelligent short message and the short message sending scheme, and inputs them into the short message interception simulation module again until the evaluation sending success rate is higher than the threshold;

[0010] Step S6: A user confirmation module is provided. When the sending success rate is higher than the threshold, the user confirmation module can feed back the intelligent short message and its sending scheme to the customer for confirmation of whether to send the short message.

[0011] Further, the step S3 includes the following sub-steps:

[0012] Step S3-1: Set up a short message reconstruction module, which accesses the short message content classified by the information classification module, and processes the text, picture and video information respectively according to the different information carriers;

[0013] Step S3-2: The short message reconstruction module detects whether the text information is compliant. If sensitive words or extreme words are detected, a report is generated to prompt the customer that there are non-compliant words, and a text scheme is generated to replace the non-compliant words with compliant words and recommended to the user for selection;

[0014] Step S3-3: The short message reconstruction module detects the picture information and detects whether the picture contains illegal content related to yellow gambling, rumor, fraud, etc. When the above content is detected, a report is generated to prompt the customer that there are non-compliant pictures, and the customer re-uploads compliant pictures;

[0015] Step S3-4: The short message reconstruction module detects whether the audio, subtitles and pictures of the video contain illegal information. When illegal information is detected, a report is generated to prompt the customer that the video contains illegal information, and the customer re-uploads compliant videos;

[0016] Step S3-5: The short message reconstruction module converts the compliant video to ensure that the video can be normally played on various mainstream mobile phones;

[0017] Step S3-6: Obtain user demand from the short message sending demand acquisition module, confirm the specific sending object group of the short message, such as users of specific age, geographical range, consumption level, etc., and accurately position the target audience;

[0018] Step S3-7: According to the preferences of the target audience, the short message reconstruction module uses natural language processing technology to optimize the sentence, adjust the fluency of the sentence, the accuracy of the word, and the size of the font, so that the short message content is more in line with the target user's daily reading habits and the meaning is clear;

[0019] Step S3-8: The short message reconstruction module detects the marketing promotion information contained in the text information, such as product introduction, contact information, and promotion link, and converts the marketing promotion information contained in the text information into pictures or video display;

[0020] Step S3-9: The processed text, picture and video information are integrated, and a complete intelligent short message is generated according to the preset short message template and layout rule, and the corresponding short message sending mode is generated according to the preferences of the target sending object group.

[0021] Further, the step S3-8 includes a sub-step S3-8-1: when converting the marketing promotion information contained in the text information into pictures or video display, the text is split into different pictures and videos, and the pictures and videos are closely arranged. The complete marketing promotion information is spliced by the incomplete text appearing in different pictures or video pictures, and the complete continuous picture and text information are presented in the short message.

[0022] Further, the step S3-8 further includes a sub-step S3-8-2: when converting the marketing promotion information contained in the text information into pictures or video display, interference noise points are added in the picture or video picture, which is used to interfere with the text recognition of the short message interception program on the short message picture, and does not affect the human eye reading information.

[0023] Further, the step S4 includes the following sub-steps:

[0024] Step S4-1: Establish a deep learning model training unit, and train a short message classification model by using short message samples with labeled labels through deep learning;

[0025] Step S4-2: Set a short message analysis unit, input the short message and its sending mode input into the trained short message classification model to analyze the type of the short message;

[0026] Step S4-3: Set an interception probability evaluation unit, according to the judgment result of the short message classification model, evaluate the short message interception probability, and judge the intelligent short message with the interception probability higher than the set threshold as intercepted, and mark the reason for being intercepted.

[0027] Further, the step S4-1 includes the following sub-steps:

[0028] Step S4-1-1: Collecting short message samples of various types, including normal short messages, harassment spam short messages, fraud spam short messages, promotion spam short messages, etc., and accurately labeling the labels, which can be divided into content theme labels, behavior feature labels, sending source labels, and language style labels according to the types;

[0029] Step S4-1-2: Selecting a deep learning algorithm to build an initial short message classification model;

[0030] Step S4-1-3: Inputting the short message samples with labeled labels into the short message classification model for training.

[0031] Further, the step S4-2 includes the following sub-steps:

[0032] Step S4-2-1: Preprocessing the input short message, including removing special symbols from the text, word segmentation operation, video caption recognition and audio recognition, and image content recognition and text recognition;

[0033] Step S4-2-2: Inputting the preprocessed short message data into the trained short message classification model to obtain a feature vector of the short message;

[0034] Step S4-2-3: Adding labels to the input short message according to the feature vector, judging the type of the short message according to the labels, and determining whether it is a harassment spam short message, a fraud spam short message, or a promotion spam short message.

[0035] Further, the step S4-3 includes the following sub-steps:

[0036] Step S4-3-1: The interception probability evaluation module first reads the sending number, sending time, and key information of the text content of the short message;

[0037] Step S4-3-2: According to the type of the short message, the corresponding interception rule is called for preliminary judgment, if it is a known harassment spam short message sending number or there is a fraud keyword feature, the corresponding interception rule is called for judgment;

[0038] Step S4-3-3: For new short messages, further analyze the text content, and combine the deep learning model to judge whether it meets the characteristics of harassment, fraud, or promotion spam short messages;

[0039] Step S4-3-4: According to the classification result, the short message interception probability is evaluated, and when the interception probability is higher than the set threshold, the reason for the short message being intercepted is marked and fed back to the short message reconstruction model for optimizing the short message reconstruction method;

[0040] Step S4-3-5: When the interception probability is lower than the preset value, the premium short message is determined, the features of the premium short message are extracted, including but not limited to language style, word characteristics, content structure, theme relevance and other features, and the features of the premium short message are fed back to the short message reconstruction model for optimizing the short message reconstruction method, so that the subsequent generated short message is closer to the standard of the premium short message, further reducing the possibility of interception and improving the success rate and effectiveness of the short message.

[0041] An AI deep learning-based intelligent short message anti-interception device, comprising a memory and a processor; the memory is used to store a short message reconstruction model and a short message classification model; the processor executes the AI deep learning-based intelligent short message anti-interception method.

[0042] A storage medium storing a computer program, which implements the steps of the AI deep learning-based intelligent short message anti-interception method when executed by the processor.

[0043] The AI deep learning-based intelligent short message anti-interception method has the following advantages: when obtaining the short message content from the customer, the text, pictures and videos contained in the short message are detected for violation, and the short message is deleted or replaced in advance to avoid being intercepted after being sent; a short message interception simulation module is set, which can simulate short message interception before sending, evaluate the short message interception probability, and further predict the sending success rate and click rate, and mark the possible interception reason, when the sending success rate is lower than the threshold, the intelligent short message generation model can improve the short message content and sending scheme, and continuously optimize until the sending success rate is higher than the threshold; the specific sending object group of the short message is confirmed, which can accurately locate the users such as specific age group, regional range and consumption level. According to the preferences of the target audience, the sentence optimization and short message sending mode are generated, so that the short message content is more in line with the needs and interests of the target users, and the relevance and affinity between the short message and the users are improved. This accurate positioning and personalized pushing method can make the customer's promotion information more effectively reach the target audience, enhance the attention and response rate of the users to the short message, and thus improve the promotion effect and return on investment; the marketing promotion script that may cause the short message to be intercepted is converted into a picture or video form to avoid being detected by the interception mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 It is a flowchart of the AI deep learning-based intelligent short message anti-interception method;

[0045] Figure 2 It is a specific step diagram of step S3 of the AI deep learning-based intelligent short message anti-interception method;

[0046] Figure 3The specific step diagram of step S3-8 of the intelligent short message anti-blocking method based on AI deep learning of the present application is as follows:

[0047] Figure 4 The specific step diagram of step S4 of the intelligent short message anti-blocking method based on AI deep learning of the present application is as follows:

[0048] Figure 5 The specific step diagram of step S4-1 of the intelligent short message anti-blocking method based on AI deep learning of the present application is as follows:

[0049] Figure 6 The specific step diagram of step S4-2 of the intelligent short message anti-blocking method based on AI deep learning of the present application is as follows:

[0050] Figure 7 The specific step diagram of step S4-3 of the intelligent short message anti-blocking method based on AI deep learning of the present application is as follows. DETAILED DESCRIPTION

[0051] The technical solutions of the present application will be described clearly and completely in combination with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0052] As shown in the drawings, the present application is an intelligent short message anti-blocking method based on AI deep learning, which comprises the following steps: Figures 1-7

[0053] Step S1: setting a short message sending demand acquisition module for acquiring the short message content to be sent by each customer;

[0054] Step S2: setting an information classification module for detecting and automatically classifying the short message content information to be sent; the information is classified into text, picture and video according to the carrier, and the text information is classified into greeting, text, marketing promotion information and link according to the content;

[0055] Step S3: setting an intelligent short message reconstruction module for regenerating intelligent short messages and short message sending schemes using the short message content information provided by the customer; the short message sending scheme comprises a sending number, a sending target, a sending time and a sending interval;

[0056] Step S4: setting a short message blocking simulation module for simulating short message blocking; the intelligent short message generated by the intelligent short message generation module and the short message sending scheme thereof are input into the short message blocking simulation module, the short message blocking simulation module detects the content of the intelligent short message, evaluates the short message sending success rate and click rate in combination with the short message sending scheme, and marks the possible blocking reasons; ​

[0057] Step S5: When the success rate of sending is lower than the threshold, the short message interception simulation module feeds back the reason that may cause interception to the intelligent short message generation module, the intelligent short message generation model improves the short message content and regenerates the intelligent short message and the short message sending scheme, and inputs the short message interception simulation module again until the evaluation sending success rate is higher than the threshold;

[0058] Step S6: A user confirmation module is provided, when the success rate of sending is higher than the threshold, the user confirmation module can feed back the intelligent short message and its sending scheme to the customer, and confirm whether to send the short message by the customer.

[0059] Further, the step S3 includes the following sub-steps:

[0060] Step S3-1: A short message reconstruction module is set, which accesses the short message content classified by the information classification module, and processes the text, picture and video information respectively according to the difference of information carriers;

[0061] Step S3-2: The short message reconstruction module detects whether the text information is in compliance, if sensitive words and extreme words are detected, a report is generated to prompt the customer that there are non-compliant words, and a text scheme is generated to replace the non-compliant words with compliant words and recommended to the user for selection;

[0062] Step S3-3: The short message reconstruction module detects the picture information, detects whether the picture contains illegal content related to yellow gambling, rumor, fraud, generates a report when the above content is detected, and prompts the customer that there are non-compliant pictures, and the customer re-uploads the compliant pictures;

[0063] Step S3-4: The short message reconstruction module detects whether the audio, subtitle and picture of the video contain illegal information; when illegal information is detected, a report is generated to prompt the customer that the video contains illegal information, and the customer re-uploads the compliant video;

[0064] Step S3-5: The short message reconstruction module converts the compliant video by video coding to ensure that the video can be normally played on various mainstream mobile phones;

[0065] Step S3-6: The user demand is obtained from the short message sending demand acquisition module, and the specific sending object group of the short message is confirmed, such as users of specific age, geographical range, consumption level, etc., to accurately position the target audience;

[0066] Step S3-7: According to the preference of the target audience, the short message reconstruction module uses natural language processing technology to optimize the sentence, adjusts the fluency of the sentence, the accuracy of the word, and the size of the font, so that the short message content is more in line with the target user's daily reading habits and the meaning is clear;

[0067] Step S3-8: The short message reconstruction module detects marketing promotion information contained in the text information, such as product introduction, contact information, and promotion link, and converts the marketing promotion information contained in the text information into a picture or a video for display.

[0068] Step S3-9: The processed text, picture, and video information are integrated, a complete intelligent short message is generated according to a preset short message template and layout rule, and a corresponding short message sending mode is generated according to the preference of a target sending object group.

[0069] Further, the step S3-8 includes a sub-step S3-8-1: when the marketing promotion information contained in the text information is converted into a picture or a video for display, the text is split into different pictures and videos, the pictures and videos are closely arranged, and the complete marketing promotion information is spliced from the incomplete text appearing in different pictures or video pictures to present complete and continuous picture and text information in the short message.

[0070] Further, the step S3-8 further includes a sub-step S3-8-2: when the marketing promotion information contained in the text information is converted into a picture or a video for display, interference noise points are added in the picture or video picture to interfere with the text recognition of the short message interception program on the short message picture, and the information is not affected by the human eye reading.

[0071] Further, the step S4 includes the following sub-steps:

[0072] Step S4-1: Establishing a deep learning model training unit, using short message samples with labeled labels to train a short message classification model through deep learning;

[0073] Step S4-2: Setting a short message analysis unit, inputting the short message and its sending mode input into the trained short message classification model to analyze the type of the short message;

[0074] Step S4-3: Setting an interception probability evaluation unit, evaluating the short message interception probability according to the judgment result of the short message classification model, and determining the intelligent short message with an interception probability higher than a set threshold as intercepted, and marking the reason for being intercepted.

[0075] Further, the step S4-1 includes the following sub-steps:

[0076] Step S4-1-1: Collecting short message samples of various types, including normal short messages, harassment type spam short messages, fraud type spam short messages, promotion type spam short messages, etc., and accurately labeling the labels, the labels can be divided into content theme labels, behavior feature labels, sending source labels, and language style labels according to the types;

[0077] Step S4-1-2: Selecting a deep learning algorithm to construct an initial short message classification model;

[0078] Step S4-1-3: input the labeled short message sample into the short message classification model for training.

[0079] Further, the step S4-2 includes the following sub-steps:

[0080] Step S4-2-1: pre-process the input short message, including removing special symbols from the text, word segmentation operation, caption recognition and audio recognition for the video, content recognition and text recognition for the picture;

[0081] Step S4-2-2: input the pre-processed short message data into the trained short message classification model to obtain the feature vector of the short message;

[0082] Step S4-2-3: add a label to the input short message according to the feature vector, judge the type of the short message according to the label, and determine whether it is a harassment type spam short message, a fraud type spam short message or a promotion type spam short message.

[0083] Further, the step S4-3 includes the following sub-steps:

[0084] Step S4-3-1: the interception probability evaluation module first reads the sending number, sending time and key information of the text content of the short message;

[0085] Step S4-3-2: call the corresponding interception rule according to the preliminary judgment of the short message type, if it is a known harassment type spam short message sending number or there is a fraud type keyword feature, then the corresponding interception rule is called for judgment;

[0086] Step S4-3-3: for a new short message, further analyze the text content, and judge whether it meets the characteristics of harassment, fraud or promotion type spam short message in combination with the deep learning model;

[0087] Step S4-3-4: according to the classification result, evaluate the short message interception probability, when the interception probability is higher than the set threshold, mark the reason for the short message being intercepted, and feed back to the short message reconstruction model for optimizing the short message reconstruction method;

[0088] Step S4-3-5: when the interception probability is lower than the preset value, it is determined as a high-quality short message, the features of the high-quality short message are extracted, including but not limited to language style, word characteristics, content structure, theme relevance and other aspects, the features of these high-quality short messages are fed back to the short message reconstruction model for optimizing the short message reconstruction method, so that the subsequent generated short messages tend to be closer to the standard of high-quality short messages, further reducing the possibility of being intercepted, and improving the success rate and effectiveness of the short message sending.

[0089] An AI deep learning-based intelligent short message anti-blocking device, comprising a memory and a processor; the memory is used to store a short message reconstruction model and a short message classification model; the processor executes the above-mentioned AI deep learning-based intelligent short message anti-blocking method.

[0090] A storage medium stores a computer program, which implements the steps of the above-mentioned AI deep learning-based intelligent short message anti-blocking method when executed by the processor.

[0091] The beneficial effects of the AI deep learning-based intelligent short message anti-blocking method of the present application are as follows: when obtaining the short message content from the customer, the text, picture and video contained in the short message are detected for violation, and if there is any violation information, it is deleted or replaced in advance to avoid the short message being blocked after being sent; a short message blocking simulation module is set, which can simulate short message blocking before sending the short message, evaluate the probability of short message being blocked, and further predict the success rate of sending and the click rate, and mark the possible blocking reason, when the sending success rate is lower than the threshold value, the intelligent short message generation model can improve the short message content and the sending scheme, and continuously optimize until the sending success rate is higher than the threshold value; confirming the specific sending object group of the short message can accurately locate the users such as specific age group, regional range, consumption level, etc. According to the preferences of the target audience, the sentence optimization and the generation of the short message sending mode are carried out, so that the short message content is more in line with the needs and interests of the target users, and the relevance and affinity between the short message and the users are improved. This accurate positioning and personalized pushing method can make the customer's promotion information more effectively reach the target audience, enhance the attention and response rate of the users to the short message, and thus improve the promotion effect and the return on investment; the marketing promotion script that may cause the short message to be blocked is converted into a picture or video form to avoid being detected by the blocking mechanism.

[0092] The above describes the present application and its embodiments, which are not limited, and the drawings only show one of the embodiments of the present application, and the actual content is not limited thereto. In general, if a person skilled in the art is inspired by it, without departing from the purpose of the present application, without creative design, similar structural methods and embodiments of the technical solution can be designed, which should belong to the protection scope of the present application.

Claims

1. A smart SMS anti-interception method based on AI deep learning, characterized in that, Includes the following steps: Step S1: Set up the SMS sending requirement acquisition module to obtain the SMS content that each customer needs to send; Step S2: Set up the information classification module to detect and automatically classify the SMS content to be sent; information is classified by carrier into text, images and videos, and text information is classified by content into greetings, body text, marketing promotion information and links; Step S3: Set up the smart SMS reconstruction module to regenerate smart SMS messages and SMS sending plans using the SMS content information provided by the customer; the SMS sending plan includes the sending number, sending target, sending time, and sending interval; Step S4: Set up the SMS interception simulation module to simulate SMS interception. The smart SMS generated by the smart SMS generation module and its SMS sending scheme are input into the SMS interception simulation module. The SMS interception simulation module detects the content of the smart SMS, evaluates the SMS sending success rate and click rate in combination with the SMS sending scheme, and marks possible reasons for interception. Step S5: When the sending success rate is lower than the threshold, the SMS interception simulation module reports the possible reasons for the interception to the intelligent SMS generation module. The intelligent SMS generation model improves the SMS content and regenerates the intelligent SMS and SMS sending plan. It is then input into the SMS interception simulation module again until the evaluation of the sending success rate is higher than the threshold. Step S6: A user confirmation module is provided. When the sending success rate is higher than the threshold, the user confirmation module will send the smart SMS and its sending plan back to the customer, who will then confirm whether to send the SMS. Step S4 includes the following sub-steps: Step S4-1: Establish a deep learning model training unit and train a text message classification model using labeled text message samples through deep learning. Step S4-2: Set up the SMS analysis unit, input the SMS messages and their sending methods from the SMS interception simulation module into the trained SMS classification model for analysis, and determine the type of SMS message; Step S4-3: Set up an interception probability assessment unit to assess the probability of SMS being intercepted based on the judgment results of the SMS classification model, and determine that smart SMS messages with an interception probability higher than a set threshold are intercepted, and mark the reasons for the interception.

2. The intelligent SMS anti-interception method based on AI deep learning according to claim 1, characterized in that, Step S3 includes the following sub-steps: Step S3-1: Set up the SMS reconstruction module. The SMS reconstruction module receives SMS content classified by the information classification module and processes text, image and video information separately according to the different information carriers. Step S3-2: The SMS reconstruction module detects whether the text information is compliant. If sensitive words or extreme words are detected, a report is generated to prompt the customer that there are non-compliant words, and a text scheme to replace the non-compliant words with compliant words is generated and recommended to the user for selection. Step S3-3: The SMS reconstruction module detects image information and whether the image contains illegal content related to pornography, gambling, drugs, rumors, or fraud. If the above content is detected, a report is generated to remind the customer that there is a non-compliant image and the customer is required to re-upload a compliant image. Step S3-4: The SMS reconstruction module detects whether the audio, subtitles, and video of the video contain illegal information; when illegal information is detected, a report is generated, prompting the customer to re-upload the video segment containing illegal information and to re-upload a compliant video. Step S3-5: The SMS reconstruction module performs video encoding conversion on compliant videos to ensure that the videos can be played normally on various mainstream mobile phones; Step S3-6: Obtain user needs from the SMS sending needs acquisition module, confirm the specific target group of the SMS, and accurately locate the target audience; Step S3-7: Based on the preferences of the target audience, the SMS reconstruction module uses natural language processing technology to optimize sentences, adjust the fluency of sentences, the accuracy of word choice, and the size of font, so that the SMS content is more in line with the daily reading habits of the target users and is clear in meaning; Step S3-8: The SMS reconstruction module detects the marketing and promotional information contained in the text information and converts the marketing and promotional information contained in the text information into images or videos for display. Step S3-9: Integrate the processed text, image and video information, generate a complete smart SMS according to the preset SMS template and layout rules, and generate the corresponding SMS sending method according to the preferences of the target audience.

3. The intelligent SMS anti-interception method based on AI deep learning according to claim 2, characterized in that, Step S3-8 includes sub-step S3-8-1: When converting the marketing promotion information contained in the text information into images or videos, the text is split into different images and videos, the images and videos are arranged closely together, and the incomplete text appearing in different images or videos is pieced together to form complete marketing promotion information, presenting complete and continuous images and text information in the text message.

4. The intelligent SMS anti-interception method based on AI deep learning according to claim 3, characterized in that, Step S3-8 further includes sub-step S3-8-2: when converting the marketing promotion information contained in the text information into an image or video for display, adding interference noise to the image or video frame to interfere with the SMS interception program's text recognition of the SMS frame without affecting human eye reading of the information.

5. The intelligent SMS anti-interception method based on AI deep learning according to claim 1, characterized in that, Step S4-1 includes the following sub-steps: Step S4-1-1: Collect SMS samples containing various types and accurately label them. The labels are divided into content theme labels, behavioral feature labels, sending source labels, and language style labels according to type. Step S4-1-2: Select a deep learning algorithm and build an initial SMS classification model; Step S4-1-3: Input the labeled SMS samples into the SMS classification model for training.

6. The intelligent SMS anti-interception method based on AI deep learning according to claim 1, characterized in that, Step S4-2 includes the following sub-steps: Step S4-2-1: Preprocess the input SMS message, including removing special characters from the text, word segmentation, subtitle and audio recognition for video, and content and text recognition for image; Step S4-2-2: Input the preprocessed SMS data into the trained SMS classification model to obtain the feature vector of the SMS. Step S4-2-3: Add tags to the input SMS based on the feature vector, and determine the type of SMS based on the tags to determine whether it is a spam SMS, a fraudulent spam SMS, or a promotional spam SMS.

7. The intelligent SMS anti-interception method based on AI deep learning according to claim 1, characterized in that, Step S4-3 includes the following sub-steps: Step S4-3-1: The interception probability assessment module first reads key information about the SMS message, including the sending number, sending time, and text content; Step S4-3-2: Based on the SMS type, make a preliminary judgment and call the corresponding blocking rules. If the number is a known spam SMS sender or contains fraudulent keywords, the corresponding blocking rules will be called first for judgment. Step S4-3-3: For new text messages, further analyze the text content and use a deep learning model to determine whether they meet the characteristics of spam, fraud, or promotional text messages. Step S4-3-4: Based on the classification results, evaluate the probability of SMS being blocked. If the probability of being blocked is higher than a set threshold, mark the reason for the SMS being blocked and feed it back to the SMS reconstruction model to optimize the SMS reconstruction method. Step S4-3-5: When the probability of being intercepted is lower than the preset value, it is judged as a high-quality SMS. The features of the high-quality SMS are extracted and fed back to the SMS reconstruction model to optimize the SMS reconstruction method, so that the SMS generated later is closer to the standard of high-quality SMS, further reducing the possibility of being intercepted and improving the success rate and effectiveness of SMS delivery.

8. A smart SMS anti-interception device based on AI deep learning, characterized in that, It includes a memory and a processor; the memory is used to store SMS reconstruction models and SMS classification models; the processor executes an AI deep learning-based intelligent SMS anti-interception method as described in any one of claims 1-7.

9. A storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the steps of the AI ​​deep learning-based intelligent SMS anti-interception method as described in any one of claims 1-7.

Images