An artificial intelligence-based short message filtering system and method

Abstract

The application relates to the technical field of information filtering, in particular to a short message filtering system and method based on artificial intelligence, which comprises an information detection module, an information processing module, an intelligent analysis module, a dynamic adjustment module and an information recovery module; the information detection module is used for acquiring characteristic information and risk parameters of short messages; the information processing module is used for calculating risk coefficients of each short message and filtering out short messages with risk coefficients not meeting requirements; the intelligent analysis module is used for acquiring behavior habits of users and classifying short messages; the dynamic adjustment module is used for adjusting risk parameters corresponding to characteristics in a database according to characteristic information and classification conditions of the short messages; and the information recovery module is used for providing a channel for recovering information and further adjusting the system; the application can intelligently judge personal preferences of users, dynamically change filtering standards of short messages according to the personal preferences of the users, and realize a personalized short message filtering function.

Description

Technical Field

[0001] This invention relates to the field of information filtering technology, specifically to an artificial intelligence-based SMS filtering system and method. Background Technology

[0002] The widespread adoption of smart mobile devices has facilitated communication, but it has also brought many problems, one of which is the proliferation of spam. Numerous advertisements, sales pitches, and even scams bother mobile phone users. This spam not only wastes the phone's memory space but may also cause users to miss out on truly useful information.

[0003] Existing SMS filtering systems filter spam messages through keyword or contextual semantic analysis, but these systems rely on overly simplistic criteria. Each user's interest in different SMS messages varies due to differences in their identity, occupation, or interests. Even messages that are advertisements may be helpful to users; a single criterion cannot meet the personalized needs of users.

[0004] In addition, the false alarm rate of existing SMS filtering systems is also very high. In many cases, some information that is very important to users is judged as spam and blocked. Even if it can be recovered, the overly simplistic standard will cause the system to continue to block similar messages in the future, causing unnecessary losses to users. Summary of the Invention

[0005] The purpose of this invention is to provide an artificial intelligence-based SMS filtering system and method to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an artificial intelligence-based SMS filtering system, comprising: an information detection module, an information processing module, an intelligent analysis module, a dynamic adjustment module, and an information recovery module;

[0007] The information detection module is used to segment the SMS message after receiving it, obtain the feature information of the SMS message, search the database, and give the risk parameters for each feature.

[0008] The information processing module is used to calculate the risk coefficient of each SMS message based on the feature information of the SMS message and the risk parameter of each feature. It compares the risk coefficient with the user's preference coefficient, filters out all SMS messages whose risk coefficients do not meet the requirements, sends a reply message to the source, and sends the remaining SMS messages that meet the requirements to the user.

[0009] The intelligent analysis module is used to obtain the user's behavioral habits after the user receives the text message, determine the user's acceptance level of the text message based on the user's behavioral habits, classify the text messages according to the acceptance level, and update the user's preference coefficient.

[0010] The dynamic adjustment module is used to extract features from the classified SMS messages, obtain feature information of SMS messages in different categories, and adjust the risk parameters of the corresponding features in the database according to the feature information of the SMS messages and their classification. If a feature extracted by the system is not recorded in the database, the feature is added to the database and the risk parameters of the feature are given.

[0011] The information recovery module provides users with a channel to recover information and stores filtered text messages. When a user recovers a text message stored in the module, the system will extract the characteristics of the text message and change the risk parameters of the corresponding characteristics in the database and the user's preference coefficient based on the extracted characteristics.

[0012] Furthermore, the information detection module includes: a source detection unit, a text message segmentation unit, a feature extraction unit, and a risk detection unit;

[0013] The source detection unit is used to detect the source of a text message after it is received;

[0014] The SMS segmentation unit is used to obtain SMS content, segment the SMS into several paragraphs, delete emoticons and punctuation marks, store the segmented content, and mark it as a segmented SMS.

[0015] The feature extraction unit is used to extract the number of characters, keywords and paragraphs of each text message based on the stored segmented text messages. The keywords are calculated based on the number of times the words appear in the text messages and the daily usage frequency of the words.

[0016] The risk detection unit is used to obtain the characteristics of the SMS message, including: the source of the message, the word count range, keywords and the number of paragraphs. Based on the obtained characteristic information, the unit searches the database and provides the risk parameters for each characteristic. If the information of a characteristic is not stored in the database, a record of that characteristic is added to the database, and the default risk parameters for that characteristic are set according to the user's preference coefficient.

[0017] Furthermore, the information processing module includes: a risk calculation unit and an SMS filtering unit;

[0018] The risk calculation unit is used to further calculate the risk coefficient of the SMS based on the obtained SMS characteristics and risk parameters;

[0019] The SMS filtering unit compares the risk coefficient of an SMS with the user's preference coefficient. If the risk coefficient of an SMS is greater than the user's preference coefficient, the SMS is marked as a filtered message and will not be sent to the user. If the risk coefficient of an SMS is less than or equal to the user's preference coefficient, the SMS is sent to the user.

[0020] Furthermore, the intelligent analysis module includes: an access analysis unit, an operation analysis unit, and an SMS classification unit;

[0021] The access analysis unit is used to obtain the user's access data after the user receives a text message. The access data includes: the time from when the user receives the text message to when the user views the text message, the number of times the user views the text message, and the speed at which the user reads the text message.

[0022] The operation analysis unit is used to obtain the user's operation data after the user finishes reading the text message. The operation data includes: the time when the user deleted the text message and the user's marking of the text message.

[0023] The SMS classification unit is used to calculate the importance coefficient of SMS messages based on the obtained access data and operation data, and classify SMS messages according to the importance coefficient.

[0024] Furthermore, the dynamic adjustment module includes: a preference analysis unit, a preference statistics unit, and a data adjustment unit;

[0025] The preference analysis unit is used to extract the feature information of each SMS message and, based on the importance coefficient of the SMS message, give the feature adjustment value of each feature.

[0026] The preference statistics unit is used to count the number of SMS messages in each category, and to calculate the user's preference coefficient adjustment value based on the number of SMS messages of different categories.

[0027] The data adjustment unit is used to adjust the risk parameter of the feature in the database based on the feature extracted by the preference analysis unit and the category of the SMS containing the feature, and to adjust the user's preference coefficient set by the system according to the number of SMS in each category.

[0028] Furthermore, the information recovery module includes: a filter residue storage unit and a misjudgment adjustment unit;

[0029] The filter residue storage unit is used to store information marked as filter residue SMS messages. If the stored information exceeds the preset storage capacity, the earliest filter residue SMS message will be cleared from the storage space.

[0030] The misjudgment adjustment unit is used to extract the feature information of a text message after a user recovers it from the filter residue storage unit, and adjust the risk parameters of the corresponding feature in the database.

[0031] An AI-based SMS filtering method includes the following steps:

[0032] S100. After receiving the SMS, the system obtains the content of the SMS and segments the content to obtain the feature information of the SMS. Based on the extracted feature information, the system retrieves the risk parameters of each feature from the database. When a feature extracted by the system is not recorded in the database, the system adds the feature to the database and provides the risk parameters of the feature.

[0033] S200. Based on the risk parameters of each feature retrieved in step S100, calculate the overall risk coefficient of each SMS message, compare the risk coefficient of the SMS message with the user's preference coefficient, filter out SMS messages whose risk coefficients do not meet the conditions, and automatically send a reply message to the source. For SMS messages that meet the conditions, send them to the user.

[0034] S300. The user receives the information sent in step S200. The system collects the user's behavioral habits after receiving the information, obtains the user's viewing data and operation data, analyzes the user's preference for the text message, classifies the text message according to the calculated preference, and further calculates the user's preference coefficient based on the classification.

[0035] S400. Obtain the classified SMS messages from step S300, obtain the feature information of different types of SMS messages respectively, adjust the risk parameters of each feature in the database in step S100 based on the feature information of different types of SMS messages, and further adjust the user's preference coefficient.

[0036] S500. Store the SMS messages filtered out in step S200 and mark them as "filtered SMS messages". Users can restore the SMS messages at any time. If a user restores one of the SMS messages, the system will extract and analyze the features of the SMS message and adjust the risk parameters in the database according to the analysis results. When the filtered SMS messages are stored to the preset capacity, the earliest filtered SMS message will be cleared.

[0037] Furthermore, step S100 includes:

[0038] Step S101. After receiving the SMS, the system obtains the source of the SMS and reads the content of the SMS. It identifies the punctuation marks in the SMS, breaks the SMS at the punctuation marks, searches for preset connecting words in the broken content, breaks the SMS again at the connecting words, deletes the emoticons and punctuation marks, cuts the SMS into several paragraphs, stores the cut SMS, and marks it as a cut SMS.

[0039] Step S102. Count the word count range and the number of paragraphs of each short message according to the stored cut short messages, compare a plurality of cut short messages, find out the word with the highest occurrence frequency, and mark it as a keyword. If multiple words have the same occurrence frequency at the same time, determine the daily use frequency of the words according to a preset daily vocabulary list, and mark the word with the highest daily use frequency as a keyword, wherein the daily use frequency of each word in the daily vocabulary list is different;

[0040] Step S103. Acquire feature information of the short message, where the feature information includes: message sending source, word count range, keyword and number of paragraphs, retrieve risk parameters corresponding to each piece of feature information of the short message from a database, wherein the retrieved risk parameter of the message sending source is C1, the risk parameter of the word count range is C2, the risk parameter of the keyword is C3, and the risk parameter of the number of paragraphs is C4. If a feature is not stored in the database, add the feature to the database, and set a default risk parameter C5 for the feature according to a user preference coefficient S, where C5=S / 4, and the initial value of the preference coefficient S is set by the user;

[0041] Further, in step S200, a risk coefficient C of the short message is calculated, where C=C1+C2+C3+C4. If C>P, the short message is marked as a filtered spam short message, not sent to the user, and an automatic reply is sent to the sending source. If C<P or C=P, the short message is marked as a normal short message, and all normal short messages are sent to the user;

[0042] Further, step S300 includes:

[0043] Step S301. After the user receives the information sent in step S200, acquire access data of the user for each piece of information, where the access data includes: the time T from when the user receives the short message to when the user views the short message, the number of times P the user browses the short message, and the speed V at which the user reads the short message;

[0044] Wherein, when T>T0, it represents that the user has not viewed the short message, let T=T0, where T0 represents the maximum time for the user to view the short message;

[0045] The speed V of reading the short message is V=H / t, where G represents the total number of words of the short message, t represents the time taken by the user from opening the short message to closing the short message, and G>0 and T>0;

[0046] Step S302. After the user finishes reading the short message, the user can choose to delete or mark the short message, and the system acquires operation data of the user, where the operation data includes: the time H when the user deletes the short message and the marking status of the short message by the user. When H>H0, it represents that the user has not deleted the short message, let H=H0, where H0 represents the maximum time for the user to delete the short message;

[0047] Step S303. Calculate the importance coefficient K of each short message according to the obtained access data and operation data using the following formula:

[0048]

[0049] Wherein, T represents the time from when the user receives the short message to when the user views the information, P represents the number of times the user browses the short message, V represents the user's reading speed for the short message, H represents the time when the user deletes the short message, H1 represents the average time for the user to delete short messages in historical data, Q represents the user's marking status of the short message; when the user marks the short message as an important short message, Q=Q1, when the user does not mark the short message, Q=Q2, when the user marks the short message as uninteresting, Q=Q3, and all of Q1, Q2 and Q3 are constants, with 0<Q3<Q2<Q1;

[0050] Step S304. Classify the short messages according to the user's marking and the value of K, classify a short message marked as important by the user or a short message with an importance coefficient K greater than K0 as a first-type short message, where K0 is the average value of the importance coefficients of all short messages historically received by the user, classify a short message with K equal to K0 or K less than K0 and not marked by the user as a second-type short message, and classify a short message marked as uninteresting by the user as a third-type short message;

[0051] Further, step S400 includes:

[0052] Step S401. Extract the feature information of each short message; if a short message containing a certain feature is classified as a first-type short message, add C0 to the risk parameter corresponding to the feature in the database, where C0 represents a risk parameter adjustment value; if a short message containing a certain feature is classified as a second-type short message, do not adjust the data in the database; if a short message containing a certain feature is classified as a third-type short message, subtract C0 from the risk parameter corresponding to the feature in the database;

[0053] Step S402. Adjust the user's preference coefficient according to the classification of the short message, and calculate the adjusted preference coefficient S0 using the following formula:

[0054]

[0055] Where S represents the preference coefficient before adjustment, n1 represents the number of Category 1 SMS messages, n2 represents the number of Category 2 SMS messages, n3 represents the number of Category 3 SMS messages, r1 represents the influence parameter of Category 1 SMS messages, r2 represents the influence parameter of Category 2 SMS messages, r3 represents the influence parameter of Category 3 SMS messages, i represents the number of Category 1 SMS messages, Ki represents the importance coefficient of SMS message number i in Category 1 SMS messages, K0 represents the average importance coefficient of all SMS messages received by the user in history, u represents the number of Category 2 SMS messages, Ku represents the importance coefficient of SMS message number u in Category 2 SMS messages, and n1, n2 and n3 are positive integers, and r1, r2 and r3 are constants greater than 0 set by the system;

[0056] Adjust the user's preference coefficient to S0;

[0057] Furthermore, step S500 includes:

[0058] Step S501. Store the information marked as spam messages. If the number of stored messages exceeds the preset storage capacity, clear the earliest spam message from the storage space.

[0059] Step S502. When a user recovers a text message from the filter residue storage unit, extract the feature information of the text message and subtract C0 from the risk parameter of the corresponding feature in the database;

[0060] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0061] 1. This invention can analyze the source and content of information, determine whether the information is what the user needs, filter out irrelevant junk information, reduce the amount of information the user needs to process, help the user identify the validity of information, and also save the user's mobile phone memory.

[0062] 2. This invention comprehensively analyzes multiple dimensions, including the number of times a user views text messages, the speed at which they view messages, and the time it takes for them to delete messages. It intelligently determines the user's personal preferences and dynamically adjusts the filtering criteria based on these preferences, ensuring that the filtered messages meet the user's actual needs. For messages that the user is not interested in, it can automatically reply to the sender and extract semantics and sensitive words from the message content for learning, thus achieving a more humanized text message filtering process.

[0063] 3. This invention provides an information recovery channel, enabling users to recover important information in the event of a false alarm. Furthermore, when a user recovers information, the system can self-adjust based on the characteristics of the recovered information, reducing the probability of false alarms when receiving similar information subsequently. Attached Figure Description

[0064] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0065] Figure 1 This is a schematic diagram of the structure of an artificial intelligence-based SMS filtering system according to the present invention;

[0066] Figure 2 This is a schematic diagram illustrating the steps of an AI-based SMS filtering method according to the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Please see Figure 1 The present invention provides a technical solution: an artificial intelligence-based SMS filtering system, comprising: an information detection module, an information processing module, an intelligent analysis module, a dynamic adjustment module, and an information recovery module;

[0069] The information detection module is used to segment the SMS message after receiving it, obtain the feature information of the SMS message, search the database, and give the risk parameters for each feature.

[0070] The information detection module includes: a source detection unit, a text message segmentation unit, a feature extraction unit, and a risk detection unit;

[0071] The source detection unit is used to detect the source of a text message after it is received;

[0072] The SMS segmentation unit is used to obtain SMS content, segment the SMS into several paragraphs, delete emoticons and punctuation marks, store the segmented content, and mark it as a segmented SMS.

[0073] The feature extraction unit is used to extract the number of characters, keywords and paragraphs of each text message based on the stored segmented text messages. The keywords are calculated based on the number of times the words appear in the text messages and the daily usage frequency of the words.

[0074] The risk detection unit is used to obtain the characteristics of the SMS message, including: the source of the message, the word count range, keywords and the number of paragraphs. Based on the obtained characteristic information, the unit searches the database and provides the risk parameters for each characteristic. If the information of a characteristic is not stored in the database, a record of that characteristic is added to the database, and the default risk parameters for that characteristic are set according to the user's preference coefficient.

[0075] The information processing module is used to calculate the risk coefficient of each SMS message based on the feature information of the SMS message and the risk parameter of each feature. It compares the risk coefficient with the user's preference coefficient, filters out all SMS messages whose risk coefficients do not meet the requirements, sends a reply message to the source, and sends the remaining SMS messages that meet the requirements to the user.

[0076] The information processing module includes: a risk calculation unit and an SMS filtering unit;

[0077] The risk calculation unit is used to further calculate the risk coefficient of the SMS based on the obtained SMS characteristics and risk parameters;

[0078] The SMS filtering unit compares the risk coefficient of an SMS with the user's preference coefficient. If the risk coefficient of an SMS is greater than the user's preference coefficient, the SMS is marked as a filtered message and will not be sent to the user. If the risk coefficient of an SMS is less than or equal to the user's preference coefficient, the SMS is sent to the user.

[0079] The intelligent analysis module is used to obtain the user's behavioral habits after the user receives the text message, determine the user's acceptance level of the text message based on the user's behavioral habits, classify the text messages according to the acceptance level, and update the user's preference coefficient.

[0080] The intelligent analysis module includes: an access analysis unit, an operation analysis unit, and an SMS classification unit;

[0081] The access analysis unit is used to obtain the user's access data after the user receives a text message. The access data includes: the time from when the user receives the text message to when the user views the text message, the number of times the user views the text message, and the speed at which the user reads the text message.

[0082] The operation analysis unit is used to obtain the user's operation data after the user finishes reading the text message. The operation data includes: the time when the user deleted the text message and the user's marking of the text message.

[0083] The SMS classification unit is used to calculate the importance coefficient of SMS messages based on the obtained access data and operation data, and classify SMS messages according to the importance coefficient.

[0084] The dynamic adjustment module is used to extract features from the classified SMS messages, obtain feature information of SMS messages in different categories, and adjust the risk parameters of the corresponding features in the database according to the feature information of the SMS messages and their classification. If a feature extracted by the system is not recorded in the database, the feature is added to the database and the risk parameters of the feature are given.

[0085] The dynamic adjustment module includes: a preference analysis unit, a preference statistics unit, and a data adjustment unit;

[0086] The preference analysis unit is used to extract the feature information of each SMS message and, based on the importance coefficient of the SMS message, give the feature adjustment value of each feature.

[0087] The preference statistics unit is used to count the number of SMS messages in each category, and to calculate the user's preference coefficient adjustment value based on the number of SMS messages of different categories.

[0088] The data adjustment unit is used to adjust the risk parameter of the feature in the database based on the feature extracted by the preference analysis unit and the category of the SMS containing the feature, and to adjust the user's preference coefficient set by the system according to the number of SMS in each category.

[0089] The information recovery module provides users with a channel to recover information and stores filtered text messages. When a user recovers a text message stored in the module, the system will extract the characteristics of the text message and change the risk parameters of the corresponding characteristics in the database and the user's preference coefficient based on the extracted characteristics.

[0090] The information recovery module includes: a filter residue storage unit and a misjudgment adjustment unit;

[0091] The filter residue storage unit is used to store information marked as filter residue SMS messages. If the stored information exceeds the preset storage capacity, the earliest filter residue SMS message will be cleared from the storage space.

[0092] The misjudgment adjustment unit is used to extract the feature information of a text message after a user recovers it from the filter residue storage unit, and adjust the risk parameters of the corresponding feature in the database.

[0093] like Figure 2 As shown, an AI-based SMS filtering method includes the following steps:

[0094] S100. After receiving the SMS, the system obtains the content of the SMS and segments the content to obtain the feature information of the SMS. Based on the extracted feature information, the system retrieves the risk parameters of each feature from the database. When a feature extracted by the system is not recorded in the database, the system adds the feature to the database and provides the risk parameters of the feature.

[0095] Step S100 includes:

[0096] Step S101. After the system receives a short message, it acquires the sending source of the short message, reads the content of the short message, identifies punctuation marks in the short message, splits the short message at the punctuation marks, retrieves preset conjunctions in the split content, splits the short message again at the conjunctions, deletes emoticons and punctuation marks therein, cuts the short message into a plurality of segments, and stores the cut short message, marking it as a cut short message;

[0097] Step S102. Based on the stored cut short messages, count the word count range and the number of segments of each short message, compare the multiple cut short messages to find the word with the highest occurrence frequency and mark it as a keyword. If multiple words have the same occurrence frequency at the same time, determine the daily usage frequency of the words according to a preset daily vocabulary list, and mark the word with the highest daily usage frequency as a keyword, wherein in the daily vocabulary list, the daily usage frequency of each word is different;

[0098] Step S103. Acquire feature information of the short message, where the feature information includes: a sending source, a word count range, a keyword and the number of segments, retrieve a risk parameter corresponding to each piece of the feature information of the short message from a database, where the retrieved risk parameter of the sending source is C1, the risk parameter of the word count range is C2, the risk parameter of the keyword is C3, and the risk parameter of the number of segments is C4. If a piece of feature is not stored in the database, add the feature to the database, and set a default risk parameter C5 for the feature according to a user's preference coefficient S, where C5=S / 4, and the initial value of the preference coefficient S is set by the user;

[0099] S200. Calculate the overall risk coefficient of each short message according to the risk parameter of each feature retrieved in step S100, compare the risk coefficient of the short message with the user's preference coefficient, filter out short messages whose risk coefficients do not meet the condition, and automatically send a reply message to the source, and send the short messages meeting the condition to the user;

[0100] In step S200, the risk coefficient C of the short message is calculated, where C=C1+C2+C3+C4. If C>P, the short message is marked as a filtered waste short message, not sent to the user, and an automatic reply is performed to the sending source. If C<P or C=P, the short message is marked as a normal short message, and all normal short messages are sent to the user;

[0101] S300. The user receives the information sent in step S200, the system collects the user's behavior habits after receiving the information, acquires the user's viewing data and operation data, analyzes the user's preference degree for the short message, classifies the short message according to the calculated preference degree, and further calculates the user's preference coefficient according to the classification situation;

[0102] Step S300 includes:

[0103] Step S301. After the user receives the information sent in step S200, the access data of the user for each piece of information is acquired, where the access data includes: the time T from when the user receives the short message to when the user views the short message, the number of times P the user browses the short message, and the speed V at which the user reads the short message;

[0104] Wherein, when T>T0, it means that the user has not viewed the short message, let T=T0, where T0 represents the maximum time for the user to view the short message;

[0105] The speed V of reading the short message is V=H / t, wherein G represents the total number of characters of the short message, t represents the time taken by the user from opening the short message to closing the short message, G>0 and T>0;

[0106] Step S302. After the user finishes reading the short message, the user can choose to delete or mark the short message, and the system acquires the operation data of the user, where the operation data includes: the time H when the user deletes the short message and the marking status of the short message by the user. When H>H0, it means that the user has not deleted the short message, let H=H0, where H0 represents the maximum time for the user to delete the short message;

[0107] Step S303. According to the acquired access data and operation data, the importance coefficient K of each short message is calculated according to the following formula:

[0108]

[0109] Wherein, T represents the time from when the user receives the short message to when the user views the information, P represents the number of times the user browses the short message, V represents the speed at which the user reads the short message, H represents the time when the user deletes the short message, H1 represents the average time for the user to delete short messages in historical data, Q represents the marking status of the short message by the user. When the user marks the short message as an important short message, Q=Q1; when the user does not mark the short message, Q=Q2; when the user marks the short message as uninteresting, Q=Q3. All of Q1, Q2 and Q3 are constants, and 0<Q3<Q2<Q1;

[0110] Step S304. Classify the short messages according to the user's marking and the magnitude of K, classify the short messages marked as important short messages by the user or the short messages with the importance coefficient K>K0 as first-class short messages, where K0 is the average value of the importance coefficients of all short messages historically received by the user, classify the short messages with K=K0 or K<K0 and not marked by the user as second-class short messages, and classify the short messages marked as uninteresting by the user as third-class short messages;

[0111] S400. Obtain the classified SMS messages from step S300, obtain the feature information of different types of SMS messages respectively, adjust the risk parameters of each feature in the database in step S100 based on the feature information of different types of SMS messages, and further adjust the user's preference coefficient.

[0112] Step S400 includes:

[0113] Step S401. Extract the feature information of each SMS message. If an SMS message containing a feature is classified into Category 1 SMS messages, add C0 to the risk parameter corresponding to that feature in the database. C0 represents the risk parameter adjustment value. If an SMS message containing a feature is classified into Category 2 SMS messages, do not adjust the data in the database. If an SMS message containing a feature is classified into Category 3 SMS messages, subtract C0 from the risk parameter corresponding to that feature in the database.

[0114] Step S402. Adjust the user's preference coefficient according to the SMS category, and calculate the adjusted preference coefficient S0 according to the following formula:

[0115]

[0116] Where S represents the preference coefficient before adjustment, n1 represents the number of Category 1 SMS messages, n2 represents the number of Category 2 SMS messages, n3 represents the number of Category 3 SMS messages, r1 represents the influence parameter of Category 1 SMS messages, r2 represents the influence parameter of Category 2 SMS messages, r3 represents the influence parameter of Category 3 SMS messages, i represents the number of Category 1 SMS messages, Ki represents the importance coefficient of SMS message number i in Category 1 SMS messages, K0 represents the average importance coefficient of all SMS messages received by the user in history, u represents the number of Category 2 SMS messages, Ku represents the importance coefficient of SMS message number u in Category 2 SMS messages, and n1, n2 and n3 are positive integers, and r1, r2 and r3 are constants greater than 0 set by the system;

[0117] Adjust the user's preference coefficient to S0;

[0118] S500. Store the SMS messages filtered out in step S200 and mark them as "filtered SMS messages". Users can restore the SMS messages at any time. If a user restores one of the SMS messages, the system will extract and analyze the features of the SMS message and adjust the risk parameters in the database according to the analysis results. When the filtered SMS messages are stored to the preset capacity, the earliest filtered SMS message will be cleared.

[0119] Step S500 includes:

[0120] Step S501. Store the information marked as spam messages. If the number of stored messages exceeds the preset storage capacity, clear the earliest spam message from the storage space.

[0121] Step S502. When a user recovers a text message from the filter residue storage unit, extract the feature information of the text message and subtract C0 from the risk parameter of the corresponding feature in the database.

[0122] Example:

[0123] The system received three messages: Message 1: "A meeting will be held tomorrow evening at 6 PM in the company conference room." (Source: A); Message 2: "Welcome to purchase our company's products. For details, please inquire..." (Source: B); Message 3: "Please choose green travel to support environmental protection. For details on green travel points, please inquire..." (Source: C). "in," "held," "contact," "inquire," and "our company" are preset conjunctions in the database. The initial user preference coefficients are S = 8, H0 = 10 hours, H1 = 5 hours, K0 = 20, Q2 = 2, C0 = 0.5, and r1 = 0.1.

[0124] After receiving the text messages, the system processes them. Text message 1 is segmented into "Tomorrow night at 6 PM" and "Meeting to be held in the company conference room". The segmented paragraphs are marked as the segmented paragraphs of text message 1. Text messages 2 and 3 are segmented together.

[0125] Extract keywords from the segmented paragraphs: SMS 1 is "meeting", SMS 2 is "purchase", and SMS 3 is "green travel". The word count of SMS 1 is 15-20 characters, SMS 2 is 10-15 characters, and SMS 3 is 20-30 characters. The number of paragraphs is 2, 2, and 3 respectively, and the sources of the messages are A, B, and C respectively.

[0126] The system performs a risk parameter retrieval and finds the risk parameters C1 for SMS 1, SMS 2, and SMS 3 to be 1, 3, and 3 respectively; C2 to be 2, 3, and 1; C3 to be 2, 2, and 3; and C4 to be 1, 3, and 2. The system then calculates the risk coefficients for the three SMS messages to be 6, 11, and 9 respectively. Since the user preference coefficient S = 8, SMS 2 and SMS 3 are filtered out, and SMS 1 is sent to the user.

[0127] When a user receives SMS 1, the system detects that the time from receiving the SMS to viewing it is T = 10 seconds, the number of times the user views the SMS is P = 5, the user's reading speed is V = 5 characters / second, the user did not delete the SMS, and it was not marked. Calculate the importance coefficient of SMS 1. Since K > K0, the system classifies SMS 1 into a single category and lowers the risk parameters of SMS 1's characteristics (sender source A, character count range of 15-20 characters, number of paragraphs 2, and keyword "meeting") by 0.5, and calculates a new user preference coefficient. Then adjust the user's preference coefficient to 11;

[0128] SMS 2 and SMS 3 are marked as filtered messages and stored. When the user recovers SMS 3 from the filtered messages, C0 is subtracted from the risk parameter of SMS 3.

[0129] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0130] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-based SMS filtering method, characterized in that, The method includes the following steps: S100. After receiving the SMS, the system obtains the content of the SMS and segments the content to obtain the feature information of the SMS. Based on the extracted feature information, the system retrieves the risk parameters of each feature from the database. When a feature extracted by the system is not recorded in the database, the system adds the feature to the database and provides the default risk parameters of the feature. S200. Based on the risk parameters of each feature retrieved in step S100, calculate the risk coefficient of the SMS message, compare the risk coefficient of the SMS message with the user's preference coefficient, filter out SMS messages whose risk coefficients do not meet the conditions, and automatically send a reply SMS message to the source, and send SMS messages that meet the conditions to the user. S300. The user receives the text message sent in step S200. The system collects the user's behavioral habits after receiving the text message, obtains the user's viewing data and operation data, analyzes the user's preference for the text message, classifies the text message according to the calculated preference, and further calculates the user's preference coefficient based on the classification. S400. Obtain the classified SMS messages from step S300, obtain the feature information of different types of SMS messages respectively, adjust the risk parameters of each feature in the database in step S100 based on the feature information of different types of SMS messages, and further adjust the user's preference coefficient. S500. Store the SMS messages filtered out in step S200 and mark them as "filtered SMS messages". Users can restore the SMS messages at any time. If a user restores one of the SMS messages, the system will extract and analyze the features of the SMS message and adjust the risk parameters in the database according to the analysis results. When the filtered SMS messages are stored to the preset capacity, the earliest filtered SMS message will be cleared.

2. The SMS filtering method based on artificial intelligence according to claim 1, characterized in that: Step S100 includes: Step S101. After receiving the SMS, the system obtains the source of the SMS and reads the content of the SMS. It identifies the punctuation marks in the SMS, breaks the SMS at the punctuation marks, searches for preset connecting words in the broken content, breaks the SMS again at the connecting words, deletes the emoticons and punctuation marks, cuts the SMS into several paragraphs, stores the cut SMS, and marks it as a cut SMS. Step S102. Based on the stored segmented text messages, count the number of characters and paragraphs in each text message, compare multiple segmented text messages, find the word that appears most frequently, and mark it as a keyword. If multiple words appear the same number of times, determine the daily usage frequency of the word based on the preset daily vocabulary list, and mark the word with the highest daily usage frequency as a keyword. In the daily vocabulary list, the daily usage frequency of each word is different. Step S103. Obtain feature information of the short message, where the feature information includes: sending source, word count, keyword, and paragraph count, retrieve risk parameters corresponding to each piece of feature information of the short message from a database, wherein the retrieved risk parameter for the sending source is C1, the risk parameter for the word count range is C2, the risk parameter for the keyword is C3, and the risk parameter for the paragraph count is C4. If a piece of feature is not stored in the database, add the feature to the database, and set a default risk parameter C5 for the feature according to a user's preference coefficient S, where C5=S / 4, and the initial value of the preference coefficient S is set by the user; In step S200, a risk coefficient C of the short message is calculated, where C=C1+C2+C3+C4. If C>S, the short message is marked as a filtered short message, not sent to the user, and an automatic reply is sent to the sending source; if C<S or C=S, the short message is marked as a normal short message, and all normal short messages are sent to the user.

3. The SMS filtering method based on artificial intelligence according to claim 1, characterized in that: Step S300 includes: Step S301. After the user receives the short message sent in step S200, obtain access data of the user for each short message, where the access data includes: time T from when the user receives the short message to when the user views the short message, number of times P the user browses the short message, and speed V at which the user reads the short message; Wherein, when T>T0, it indicates that the user has not viewed the short message, T is set to T0, and T0 represents the maximum time for the user to view the short message; The speed V of reading the short message is V=G / t, where G represents the word count of the short message, t represents the time the user takes from opening the short message to closing the short message, and G>0 and t>0; Step S302. After the user finishes reading the short message, the user may choose to delete or mark the short message, and the system obtains operation data of the user, where the operation data includes: time H when the user deletes the short message and the user's marking status of the short message. When H>H0, it indicates that the user has not deleted the short message, H is set to H0, and H0 represents the maximum time for the user to delete the short message; Step S303. According to the obtained access data and operation data, calculate an importance coefficient K of each short message according to the following formula: *(H-H1)*Q Wherein, T represents the time from when the user receives the short message to when the user views the short message, P represents the number of times the user browses the short message, V represents the speed at which the user reads the short message, H represents the time when the user deletes the short message, H1 represents the average time for the user to delete short messages in historical data, Q represents the user's marking status of the short message, when the user marks the short message as an important short message, Q=Q1, when the user does not mark the short message, Q=Q2, when the user marks the short message as uninteresting, Q=Q3, and Q1, Q2 and Q3 are all constants, and 0<Q3<Q2<Q1; Step S304. Classify the short messages according to the user's marking and the magnitude of K, classify a short message marked as important by the user or a short message with an importance coefficient K>K0 as a first-type short message, where K0 is the average value of the importance coefficients of all short messages historically received by the user, classify a short message with K=K0 or K<K0 that has not been marked by the user as a second-type short message, and classify a short message marked as uninteresting by the user as a third-type short message.

4. The SMS filtering method based on artificial intelligence according to claim 1, characterized in that: Step S400 includes: Step S401. Extract the feature information of each SMS message. If an SMS message containing a feature is classified into Category 1 SMS messages, subtract C0 from the risk parameter corresponding to that feature in the database. C0 represents the risk parameter adjustment value. If an SMS message containing a feature is classified into Category 2 SMS messages, no adjustment is made to the data in the database. If an SMS message containing a feature is classified into Category 3 SMS messages, add C0 to the risk parameter corresponding to that feature in the database. Step S402. Adjust the user's preference coefficient according to the SMS category, and calculate the adjusted preference coefficient S0 according to the following formula: ； Where S represents the preference coefficient before adjustment, n1 represents the number of Category 1 SMS messages, n2 represents the number of Category 2 SMS messages, n3 represents the number of Category 3 SMS messages, r1 represents the influence parameter of Category 1 SMS messages, r2 represents the influence parameter of Category 2 SMS messages, r3 represents the influence parameter of Category 3 SMS messages, i represents the number of Category 1 SMS messages, Ki represents the importance coefficient of SMS message number i in Category 1 SMS messages, K0 represents the average importance coefficient of all SMS messages received by the user in history, u represents the number of Category 2 SMS messages, Ku represents the importance coefficient of SMS message number u in Category 2 SMS messages, and n1, n2 and n3 are positive integers, and r1, r2 and r3 are constants greater than 0 set by the system; Adjust the user's preference coefficient to S0.

5. The SMS filtering method based on artificial intelligence according to claim 1, characterized in that: Step S500 includes: Step S501. Store the SMS messages marked as "filtered SMS messages". If the number of stored SMS messages exceeds the preset storage capacity, then clear the earliest "filtered SMS message" from the storage space. Step S502. When a user recovers a text message from the filter residue storage unit, extract the feature information of the text message and subtract C0 from the risk parameter of the corresponding feature in the database.

6. An artificial intelligence-based SMS filtering system, characterized in that, The system includes the following modules: information detection module, information processing module, intelligent analysis module, dynamic adjustment module, and information recovery module; The information detection module is used to segment the SMS message after receiving it, obtain the feature information of the SMS message, search the database, and give the risk parameters for each feature. The information processing module is used to calculate the risk coefficient of each SMS message based on the feature information of the SMS message and the risk parameter of each feature. It compares the risk coefficient with the user's preference coefficient, filters out all SMS messages whose risk coefficients do not meet the requirements, sends a reply SMS message to the source, and sends the remaining SMS messages that meet the requirements to the user. The intelligent analysis module is used to obtain the user's behavioral habits after the user receives the text message, determine the user's acceptance level of the text message based on the user's behavioral habits, classify the text messages according to the acceptance level, and update the user's preference coefficient. The dynamic adjustment module is used to extract features from the classified SMS messages, obtain feature information of SMS messages in different categories, and adjust the risk parameters of the corresponding features in the database according to the feature information of the SMS messages and their classification. If a feature extracted by the system is not recorded in the database, the feature is added to the database and the risk parameters of the feature are given. The message recovery module provides users with a channel to recover SMS messages. It stores filtered SMS messages. When a user attempts to recover a stored SMS message, the system extracts the message's features and adjusts the risk parameters and user preference coefficients of the corresponding features in the database based on these features.

7. The SMS filtering system based on artificial intelligence according to claim 6, characterized in that: The information detection module includes: a source detection unit, a text message segmentation unit, a feature extraction unit, and a risk detection unit; The source detection unit is used to detect the source of a text message after it is received; The SMS segmentation unit is used to obtain SMS content, segment the SMS into several paragraphs, delete emoticons and punctuation marks, store the segmented content, and mark it as a segmented SMS. The feature extraction unit is used to extract the number of characters, keywords and paragraphs of each text message based on the stored segmented text messages. The keywords are calculated based on the number of times the words appear in the text messages and the daily usage frequency of the words. The risk detection unit is used to obtain the characteristics of the SMS message, including: the source of the message, the number of characters, keywords, and the number of paragraphs. Based on the obtained characteristic information, the unit searches the database and provides a risk parameter for each characteristic. If the information of a characteristic is not stored in the database, a record for that characteristic is added to the database, and a default risk parameter for that characteristic is set according to the user's preference coefficient.

8. The SMS filtering system based on artificial intelligence according to claim 6, characterized in that: The information processing module includes: a risk calculation unit and an SMS filtering unit; The risk calculation unit is used to further calculate the risk coefficient of the SMS based on the obtained SMS characteristics and risk parameters; The SMS filtering unit compares the risk coefficient of an SMS with the user's preference coefficient. If the risk coefficient of an SMS is greater than the user's preference coefficient, the SMS is marked as a spam SMS and will not be sent to the user. If the risk coefficient of an SMS is less than or equal to the user's preference coefficient, the SMS is sent to the user. The intelligent analysis module includes: an access analysis unit, an operation analysis unit, and an SMS classification unit; The access analysis unit is used to obtain the user's access data after the user receives a text message. The access data includes: the time from when the user receives the text message to when the user views the text message, the number of times the user views the text message, and the speed at which the user reads the text message. The operation analysis unit is used to obtain the user's operation data after the user finishes reading the text message. The operation data includes: the time when the user deleted the text message and the user's marking of the text message. The SMS classification unit is used to calculate the importance coefficient of SMS messages based on the obtained access data and operation data, and then classify the SMS messages according to the importance coefficient.

9. The SMS filtering system based on artificial intelligence according to claim 6, characterized in that: The dynamic adjustment module includes: a preference analysis unit, a preference statistics unit, and a data adjustment unit; The preference analysis unit is used to extract the feature information of each SMS message and, based on the importance coefficient of the SMS message, provide the risk parameters for each feature. The preference statistics unit is used to count the number of SMS messages in each category, and to calculate the user's preference coefficient adjustment value based on the number of SMS messages of different categories. The data adjustment unit is used to adjust the risk parameters of the feature extracted by the preference analysis unit and the category of the SMS containing the feature in the database, and adjust the user's preference coefficient set by the system according to the number of SMS in each category. The information recovery module includes: a filter residue storage unit and a misjudgment adjustment unit; The filter residue storage unit is used to store SMS messages marked as filter residue SMS messages. If the stored SMS messages exceed the preset storage capacity, the earliest filter residue SMS message will be cleared from the storage space. The misjudgment adjustment unit is used to extract the feature information of a text message after a user recovers it from the filter residue storage unit, and adjust the risk parameters of the corresponding features in the database.

Images