Abnormal SMS Identification Methods, Devices, Equipment, Storage Media and Products
By performing text segmentation and rule verification on multilingual mixed text messages, the problem of existing technologies being unable to identify text messages that mix foreign languages and native languages has been solved, and efficient identification of abnormal text messages has been achieved.
Patent Information
- Application Number
- CN202611413323.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-11
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to effectively identify unusual text messages that combine multiple languages, especially those relying on monolingual text processing methods, which are unable to handle text messages that mix foreign and native languages.
By splitting the text of the SMS message to be identified, first and second texts in different languages are obtained. Based on a preset strategy, the two texts are checked according to rules, including semantic correlation and behavioral compliance checks, to determine whether the SMS message is abnormal.
It improves the recognition accuracy of multilingual mixed text messages, enhances the ability to identify abnormal information, and improves the accuracy of identifying abnormal text messages.
Smart Images

Figure CN123123906A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of network communication technology, and in particular to methods, apparatus, devices, storage media, and products for identifying abnormal text messages. Background Technology
[0002] With the increasing prevalence of cross-border communication, international logistics, and overseas financial services, suspicious text messages often use a mix of languages. For example, senders commonly compose messages in a combination of foreign languages and their native language to lure users into clicking on links.
[0003] Existing technologies for identifying abnormal text messages mostly rely on recognizing the sender's identifier and comparing it with a feature database to determine whether it is a dangerous message; or, they extract keywords and perform semantic analysis on the text message to determine whether it is abnormal. However, the above methods are all designed for processing single-language text messages and do not address the processing of multilingual text messages. Summary of the Invention
[0004] This invention provides a method, apparatus, device, storage medium, and product for identifying abnormal text messages, so as to accurately identify abnormal multilingual mixed text messages.
[0005] According to one aspect of the present invention, an abnormal SMS identification method is provided, comprising: The SMS message to be identified is split into a first text and a second text; wherein the SMS message to be identified includes text in multiple language types; the first text and the second text belong to different languages; The first text and the second text are subjected to rule verification based on a preset strategy to obtain rule verification results, and the identification result of the SMS to be identified is determined based on the rule verification results; wherein, the identification result is used to indicate whether the SMS to be identified is an abnormal SMS.
[0006] According to another aspect of the present invention, an abnormal text message identification device is provided, comprising: The text segmentation module is used to segment the SMS message to be identified into a first text and a second text; wherein the SMS message to be identified includes text in multiple language types; the first text and the second text belong to different languages; The identification result determination module is used to perform rule verification on the first text and the second text based on a preset strategy, obtain the rule verification result, and determine the identification result of the SMS to be identified based on the rule verification result; wherein, the identification result is used to indicate whether the SMS to be identified is an abnormal SMS.
[0007] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the abnormal SMS identification method according to any embodiment of the present invention.
[0008] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the abnormal SMS identification method according to any embodiment of the present invention.
[0009] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the abnormal SMS identification method according to any embodiment of the present invention.
[0010] The technical solution of this invention involves splitting the text of a text message to be identified into a first text and a second text. The text message to be identified includes text in multiple language types. The first text and the second text belong to different languages. A rule-based verification is performed on the first text and the second text based on a preset strategy to obtain a rule verification result. The identification result of the text message to be identified is then determined based on the rule verification result. The identification result indicates whether the text message to be identified is an abnormal text message. This technical solution splits the text message to be identified into two texts in different languages and performs rule verification on the two texts based on a preset strategy, thereby completing the identification of abnormal text messages. Compared to existing technologies that rely solely on single language features to detect text messages, this invention, based on a preset strategy, can cross-verify the two texts to enhance the ability to identify abnormal information carried in the text message to be identified, thereby effectively improving the accuracy of identifying abnormal text messages.
[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1This is a flowchart of an abnormal SMS identification method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of an abnormal SMS identification method provided in Embodiment 2 of the present invention; Figure 3 This is a flowchart of an abnormal SMS identification method provided in Embodiment 3 of the present invention; Figure 4 This is a flowchart of an abnormal SMS identification method provided according to another embodiment of the present invention; Figure 5 This is a schematic diagram of an abnormal SMS identification device according to Embodiment 4 of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device that implements the abnormal SMS identification method of this invention. Detailed Implementation
[0014] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.
[0015] It should be noted that the terms "first," "second," "initial," and "target," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0016] Example 1 Figure 1 This is a flowchart illustrating an abnormal SMS identification method provided in Embodiment 1 of the present invention. This embodiment is applicable to the identification of anomalies in multilingual mixed text SMS messages. The method can be executed by an abnormal SMS identification device, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes: S101. The SMS message to be identified is split into a first text and a second text; wherein the SMS message to be identified includes text in multiple language types; the first text and the second text belong to different languages.
[0017] In this embodiment, the SMS message to be identified can be a multilingual mixed text message obtained through devices such as SMS gateways or core network elements, which may contain text with mixed language types. The SMS message to be identified can be split based on language type to obtain a first text and a second text in two different language types; for example, the language of the first text can be the native language of the user receiving the SMS message, and the language of the second text can be a non-native language of the user's region.
[0018] For example, a pre-defined language detection model can be used to identify the language of the SMS message to be identified, resulting in multiple text fragments. Each text fragment can carry corresponding language type and location information. The location information can include the start and end positions of the text fragment within the SMS message. The pre-defined language detection model can be a deep learning model, whose input can be the SMS message to be identified, and whose output can be a text fragment carrying language type and location information. Based on the language type of the text fragments, the resulting multiple text fragments are divided into native language text fragments and non-native language text fragments. Then, based on the location information of the text fragments, the native language text fragments are concatenated into a first text, and the non-native language text fragments are concatenated into a second text. For example, for each native language text fragment, the text fragments are arranged in order according to their location information (e.g., start or end position), and the concatenated text fragments are concatenated into the first text; correspondingly, for each non-native language text fragment, the text fragments are arranged in order according to their location information (e.g., start or end position), and the concatenated text fragments are concatenated into the second text.
[0019] S102. Perform rule verification on the first text and the second text based on a preset strategy to obtain rule verification results, and determine the identification result of the SMS to be identified based on the rule verification results; wherein, the identification result is used to indicate whether the SMS to be identified is an abnormal SMS.
[0020] In this embodiment, the preset strategy may include performing semantic correlation verification on the first text and the second text; for example, using a preset semantic correlation model to calculate the semantic correlation degree between the first text and the second text, and judging whether there is a semantic correlation between the first text and the second text based on the semantic correlation degree; wherein, the preset semantic correlation model may be a deep learning model, the input of which may be the feature vectors corresponding to the first text and the second text, and the output may be the semantic correlation degree between the first text and the second text.
[0021] The preset strategy may also include behavioral compliance verification of the first text and the second text; for example, using preset business rules to determine the degree of matching of business behaviors between the first text and the second text, and judging whether the business behaviors between the first text and the second text are consistent based on the degree of matching of business behaviors; wherein, the preset business rules may include relevant rules for judging whether the business processing flow contained in the first text and the second text is consistent; for example, judging whether the business processing flow contained in the first text and the second text is the same; or, judging whether there is a logical relationship between the business processing flow contained in the first text and the second text, that is, whether the business processing flow contained in the first text and the second text is a sub-process in the same processing flow.
[0022] The rule validation results can include validation results determined according to a preset strategy, such as semantic relevance and / or business behavior matching degree. The identification results can be used to indicate whether the SMS message to be identified is an abnormal SMS message, and the identification results can be determined based on the rule validation results. For example, if the rule validation results include semantic relevance, it can be determined whether the semantic relevance exceeds the corresponding relevance threshold (e.g., 0.7); if it exceeds, it indicates that the first text and the second text have a semantic relationship, and the identification result can be: the SMS message to be identified is a normal SMS message; if it does not exceed, it indicates that the first text and the second text do not have a semantic relationship, and the identification result can be: the SMS message to be identified is an abnormal SMS message. If the rule validation results include business behavior matching degree, it can be determined whether the business behavior matching degree exceeds the corresponding behavior matching degree threshold (e.g., 0.7); if it exceeds, it indicates that the business behaviors contained in the first text and the second text are consistent, and the identification result can be: the SMS message to be identified is a normal SMS message; if it does not exceed, it indicates that the business behaviors contained in the first text and the second text are inconsistent, and the identification result can be: the SMS message to be identified is an abnormal SMS message. If the rule verification result includes semantic relevance and business behavior matching degree, then it can be determined whether the semantic relevance exceeds the corresponding relevance threshold and whether the business behavior matching degree exceeds the corresponding behavior matching degree threshold respectively; if both exceed, the identification result is: the SMS to be identified is a non-abnormal SMS; otherwise, the identification result is: the SMS to be identified is a non-abnormal SMS.
[0023] This invention provides a method for identifying abnormal text messages. The method involves splitting the text message to be identified into a first text and a second text. The text message to be identified includes text in multiple languages. The first text and the second text belong to different languages. A rule-based verification is performed on the first text and the second text based on a preset strategy to obtain a rule verification result. The identification result is then used to indicate whether the text message to be identified is abnormal. This technical solution splits the text message to be identified into two texts in different languages and performs rule verification on the two texts based on a preset strategy, thereby completing the identification of abnormal text messages. Compared to existing technologies that rely solely on single-language features to detect text messages, this invention uses a preset strategy to cross-verify the two texts, enhancing the ability to identify abnormal information carried in the text message to be identified, thus effectively improving the accuracy of abnormal text message identification. Furthermore, the preset strategy includes at least one of semantic correlation verification and behavioral compliance verification on the first text and the second text. This allows for cross-verification of the semantic relevance and / or behavioral compliance represented by the verification of the first and second texts based on a preset strategy, thereby further enhancing the ability to identify abnormal information carried in the SMS messages to be identified.
[0024] Example 2 Figure 2 This is a flowchart of an abnormal SMS identification method provided in Embodiment 2 of the present invention. This embodiment further refines the above embodiment. Figure 2 As shown, the method includes: S201. Preprocess the SMS message to be identified to obtain structured text.
[0025] In this embodiment, preprocessing may include text cleaning of the SMS message to be identified; for example, removing spaces, repeated words, and special symbols from the SMS message to be identified.
[0026] For example, the SMS message to be identified might be: "
XXX
[0027] Optionally, to improve the accuracy of identifying abnormal information in the SMS message to be identified, preprocessing may also include extracting key information from the SMS message, such as links, phone numbers, amounts, and sensitive words. For example, the key information in the SMS message to be identified may be "http: / / xxx-fake.com", "ID card", "return", and "please click", etc.
[0028] S202. Using a preset language detection model, the structured text is identified character by character to obtain the language label corresponding to each text character in the structured text; wherein, the language label includes native language label and non-native language label.
[0029] In this embodiment, the preset language detection model can be a multilingual detection model, such as the third-generation Compact Language Detector v3 (CLD3). For example, structured text can be input into the preset language detection model to obtain language labels corresponding to each text character in the structured text.
[0030] S203. Based on the language tags corresponding to each text character in the structured text, divide the structured text into a first text and a second text.
[0031] For example, text characters with the same language tag can be merged based on the language tag corresponding to each text character in the structured text to obtain a first text and a second text; wherein, the first text can be the text merged from the text characters corresponding to the native language tag; and the second text can be the text merged from the text characters corresponding to the non-native language tag.
[0032] Specifically, this step may include the following steps A1 to A2: A1. Based on the language tags corresponding to each text character in the structured text and the position of each text character in the structured text, merge the text characters to obtain several text triples.
[0033] The text triple includes a text segment, a segment position corresponding to the text segment, and a language tag corresponding to the text segment; the text segment is a segment formed by merging text characters with the same language tag; the segment position includes a position interval, the starting position of the position interval is the position of the text character at the starting position in the text segment in the structured text, and the ending position of the position interval is the position of the text character at the ending position in the text segment in the structured text.
[0034] For example, text characters can be associated with character indices, which represent the position of the text character in the structured text. Character indices can start from 0 and increment sequentially. The current text segment is initialized as follows: the text character with index 0 is added to the current text segment, the language label corresponding to the text character with index 0 is determined as the language label of the current text segment, and the character index 0 is determined as the start and end positions of the position interval of the current text segment. After initialization, starting from the text character with index 1 in the structured text, the system traverses sequentially, checking if the language label of the current text character matches the language label of the current text segment. If they match, the current text character is appended to the current text segment, and the end position of the position interval corresponding to the current text segment is updated to the character index corresponding to the current text character, continuing the traversal. If they do not match, the current text segment is output, obtaining the corresponding text triplet. The current text segment is then re-initialized based on the character index and language label of the current text character, continuing the traversal until all text characters in the structured text have been traversed, resulting in several text triplets.
[0035] For example, if the structured text is "XXXDearCustomer Your package is abnormal, please click http: / / xxx-fake.com to replace your ID card or it will be returned", the above steps will yield the following four text triples.
[0036] Text triple 1: XXXDearCustomer, [0-15], English.
[0037] Text triplet 2: Your package is abnormal, please click [16-24], Chinese.
[0038] Text triple 3: http: / / xxx-fake.com, [25-43], English.
[0039] Text triple 4: Replace ID card or return, [44-51], Chinese.
[0040] A2. Based on the segment positions corresponding to the text segments in the text triples, the text segments with the same language tags in the plurality of text triples are concatenated to obtain the first text and the second text; wherein, the first text is the text concatenated after the text segments corresponding to the native language tags are concatenated; the second text is the text concatenated after the text segments corresponding to the non-native language tags are concatenated.
[0041] For example, according to the start and end positions of the text segments in the text triplet, the text segments with the same language tag in several text triplets are sequentially concatenated to obtain the first text and the second text.
[0042] Taking the example of native language tags corresponding to Chinese and non-native language tags corresponding to languages other than Chinese, the above four text triples are concatenated sequentially to obtain the following first and second texts.
[0043] The first message reads: "Your package is abnormal. Please click to request a replacement ID card; otherwise, it will be returned."
[0044] The second text reads: "XXXDearCustomerhttp: / / xxx-fake.com".
[0045] S204. Perform rule verification on the first text and the second text based on a preset strategy to obtain the rule verification result, and determine the recognition result of the SMS to be identified based on the rule verification result.
[0046] In this embodiment, the preset strategy may include semantic correlation verification and behavioral compliance verification of the first text and the second text.
[0047] Optionally, the preset strategy may also include intentional analysis of the first and second texts. Figure 1 Consistency verification; for example, determining the semantic similarity between the first text and the second text, and judging whether there is a semantic similarity between the first text and the second text based on the semantic similarity. Figure 1 Consistency. Accordingly, the rule verification result can include semantic similarity. The semantic relevance (determined through the semantic relevance verification in the aforementioned embodiments), behavioral matching degree (determined through the behavioral compliance verification in the aforementioned embodiments), and semantic similarity can be weighted and summed to obtain the rule verification result. Then, the identification result of the SMS message to be identified can be determined based on the rule verification result. The advantage of this setup is that, based on the semantic similarity between the first text and the second text, it can be determined whether the core purpose contained in the first text and the second text is consistent, thereby further improving the accuracy of identifying abnormal SMS messages.
[0048] For example, the semantic relevance, behavioral matching, and semantic similarity are normalized, and the normalized results are weighted and summed to obtain a weighted sum result. The weighted sum result is then compared with a preset risk threshold (which can be set based on an empirical value, for example, 0.5). If the weighted sum result is greater than or equal to the preset risk threshold, the identification result of the SMS message to be identified is determined to be a non-abnormal SMS message; if the weighted sum result is less than the preset risk threshold, the identification result of the SMS message to be identified is determined to be an abnormal SMS message.
[0049] This invention preprocesses the SMS message to be identified to obtain structured text, providing a clean data foundation for subsequent character-level language recognition using a preset language detection model. By recording the text content, position, and language tag of each text segment using text triples, it prevents text segment disorder during the subsequent splicing of the first and second texts, thus ensuring that the first and second texts retain their original semantics, improving the accuracy of subsequent rule verification of the first and second texts, and guaranteeing the reliability of the SMS message recognition results.
[0050] Example 3 Figure 3 This is a flowchart of an abnormal SMS identification method provided in Embodiment 3 of the present invention. This embodiment further refines the above embodiments. Figure 3 As shown, the method includes: S301. Perform text splitting on the SMS message to be identified to obtain the first text and the second text.
[0051] Wherein, the first text is the native language text and the second text is the non-native language text; S302 and S304 are executed respectively.
[0052] S302. Perform feature extraction on the first text and the second text respectively to obtain the first feature vector corresponding to the first text and the second feature vector corresponding to the second text.
[0053] For example, a pre-trained first feature extraction model can be used to extract features from the first text to obtain a first feature vector corresponding to the first text; a pre-trained second feature extraction model can be used to extract features from the second text to obtain a second feature vector corresponding to the second text.
[0054] The training process of the first feature extraction model can be as follows: A general semantic model pre-trained in the mother tongue can be used as the first base model. For example, if the mother tongue is Chinese, a Bidirectional Encoder Representations from Transformers (BERT) model can be selected as the first base model; if the mother tongue is not Chinese, such as Vietnamese, a Cross-Lingual Language Model-Robustly Optimized BERT Pretraining Approach (XLM-RoBERTa) model can be selected as the first base model. The first base model is then fine-tuned using a labeled first training set. This first training set can include mother tongue text messages and their corresponding label information, which can include normal and abnormal labels. Further refinement of the abnormal labels is also possible, such as labeling the abnormal type. During fine-tuning training, the input to the first base model is native-language text messages; the output is the anomaly probability; the optimization objective is to minimize the loss value of the classification result, allowing the first base model to focus on anomalous features in native-language text messages (e.g., money transfers, verification codes, clicked links, and information filling). After fine-tuning training, a first feature extraction model is obtained. This first feature extraction model can accurately extract features such as native-language expressions, sentence structures, and keywords from native-language text messages, thus ensuring accurate extraction of anomalous features from the first text message input. This allows the numerical dimension of the first feature vector to correspond to anomalous features such as package anomalies, clicked links, and ID card replacement.
[0055] Correspondingly, the second base model corresponding to the second feature extraction model can be XLM-RoBERTa. The training process of the second feature extraction model can be divided into a pre-training process and a fine-tuning training process. The pre-training process can be as follows: unsupervised pre-training of XLM-RoBERTa is performed using a large-scale public text dataset containing more than 100 languages (with hundreds of millions or billions of texts). Specifically, a pre-training task can be constructed based on Masked Language Modeling (MLM) to enable the second base model to map features of different languages to the same vector space, ensuring the compatibility of cross-language feature computation. After the unsupervised pre-training is completed, the second base model can be further fine-tuned using a labeled second training set. The second training set can include non-native language text messages and corresponding label information, which can include normal labels and abnormal labels. During the fine-tuning training process, the input of the second base model is non-native language text messages; the output is the anomaly probability; the optimization objective is to minimize the loss value of the classification result, so that the second base model can focus on the abnormal features in non-native language text messages (e.g., transfers, verification codes, clicked links, and information completion). After fine-tuning and training, a second feature extraction model can be obtained. This second feature extraction model can accurately extract non-native language expressions, sentence structures, and keywords from non-native language text messages, thereby ensuring that when inputting second text, abnormal features in the second text can be accurately extracted, so that the numerical dimension in the second feature vector can correspond to abnormal features such as package anomalies, clicked links, and replacement of ID cards.
[0056] The data dimensions of the first and second feature vectors can be kept consistent, for example, both being [missing data]. .
[0057] S303. Input the first feature vector and the second feature vector into the preset semantic association model to obtain the semantic association score; execute S305.
[0058] For example, the preset semantic association model can be built based on XLM-RoBERTa, with the input being a feature vector concatenated from the first and second feature vectors, and the output being a semantic association score from 0 to 1.
[0059] A third training set can be used to train a pre-defined semantic association model. This third training set can include paired texts and corresponding semantic association labels. The paired texts include corresponding native language texts and non-native language texts. The semantic association labels are values from 0 to 1, representing the semantic association degree between the native language texts and non-native language texts in the paired texts. For example, if both the native language texts and non-native language texts correspond to logistics, the semantic association label value can be closer to 1, such as 0.9. If the native language text corresponds to money transfer, and the non-native language text corresponds to daily communication, the semantic association label value can be closer to 0, such as 0.1. During training, features can be extracted from the paired texts in the third training set to obtain a third feature vector corresponding to the native language text and a fourth feature vector corresponding to the non-native language text. These third and fourth feature vectors are then input into the pre-defined semantic association model, and a regression training method is used to train the model with the optimization objective of minimizing the mean squared error between the model's predicted semantic association score and the semantic association label.
[0060] S304. Based on preset business rules, perform business behavior verification on the first text and the second text to obtain the business behavior matching score between the first text and the second text.
[0061] The preset business rules include at least one of the following: rules for verifying the compliance of business logic, rules for verifying the compliance of operation channels, and rules for verifying the reasonableness of prompts.
[0062] For example, rules for verifying business logic compliance may include determining whether the business processing logic contained in the first text and the second text is consistent, and whether the business entities contained in the first text and the second text are consistent. Rules for verifying operational channel compliance may include determining whether the operational channels contained in the first text and the second text are consistent. Rules for verifying the reasonableness of prompts may include determining whether the prompt information represented by the first text and the second text is consistent.
[0063] The business behaviors contained in the first and second texts can be scored based on preset business rules to obtain a business behavior matching score between the two texts; the scoring range can be 0 to 1. For example, if the business behavior contained in the first text is inputting identity information, while the business behavior contained in the second text is logistics and express delivery, there is no business logic connection between the two, meaning their business processing logic is inconsistent. Therefore, the business logic compliance corresponding to the business behavior can be 0.1. If the business behavior contained in the first text is clicking a non-known link, while the business behavior contained in the second text does not involve an operation channel, the operation channels contained in the two texts are inconsistent. Therefore, the operation channel compliance corresponding to the business behavior can be 0.05. If the prompt information represented by the first and second texts does not involve threatening, coercive, or urging language, then the prompt information represented by the two texts is consistent. Therefore, the prompt reasonableness corresponding to the business behavior can be 0.2. The business behavior matching score between the first and second texts can be determined based on the score of the business behavior. For example, the average of the three scores corresponding to the business behavior can be used as the business behavior matching score.
[0064] S305. Determine the rule verification result based on the semantic relevance score and the business behavior matching score, and determine the recognition result of the SMS to be identified based on the rule verification result.
[0065] For example, the rule verification result may include a semantic relevance score and a business behavior matching score. The identification result of the SMS message to be identified is determined based on the rule verification result. Specifically, this may involve determining whether the semantic relevance score exceeds the corresponding relevance threshold and whether the business behavior matching score exceeds the corresponding behavior matching threshold; if so, the identification result of the SMS message to be identified is determined to be a non-abnormal SMS message; otherwise, the identification result of the SMS message to be identified is determined to be an abnormal SMS message.
[0066] In some embodiments, before determining the rule verification result based on the semantic relevance score and the business behavior matching score, the method further includes: determining whether either the semantic relevance score or the business behavior matching score is within a corresponding preset score range; if so, inputting the first text and the first feature vector into a first preset intent classification model to obtain a first intent category vector, inputting the second text and the second feature vector into a second preset intent classification model to obtain a second intent category vector, and determining the intent category based on the similarity between the first intent category vector and the second intent category vector. Figure 1 Consistency score; wherein, determining the rule verification result based on the semantic relevance score and the business behavior matching score includes: determining the rule verification result based on the semantic relevance score, the business behavior matching score, and the consistency score; Figure 1The consistency score determines the rule verification result. The advantage of this setup is that by setting a preset score range, it can identify situations where semantic relevance or business behavior matching is abnormal, and further calculate the semantic consistency between the first and second texts. Figure 1 Consistency scores are used to determine whether the core intent of the first and second texts is consistent, thereby improving the reliability of rule validation results.
[0067] In this embodiment, the semantic relevance score and the business behavior matching score each correspond to independent preset score ranges, which may or may not be equal. For example, the preset score range for the semantic relevance score can be [0.4, 0.6], which can be used to filter out semantic relevance scores with unclear semantic relevance. Similarly, the preset score range for the business behavior matching score can be [0.4, 0.6], which can be used to filter out business behavior matching scores with unclear business behavior matching.
[0068] For example, such as Figure 4 As shown, after obtaining the semantic relevance score and the business behavior matching score, S405 can be executed to determine whether either the semantic relevance score or the business behavior matching score is within the corresponding preset score range; if so, S407 and S408 are executed to obtain the recognition result of the SMS to be identified; otherwise, S406 is executed to obtain the recognition result of the SMS to be identified.
[0069] For example, if the native language is Chinese, the first pre-defined intent classification model can be built based on the BERT model; if the native language is not Chinese, the first pre-defined intent classification model can be built based on XLM-RoBERTa. The input to the first pre-defined intent classification model is a first text and a first feature vector, and the output is a first intent category vector. The data contained in the first intent category vector is used to represent the probability of each predicted anomalous intent label. For example, the first intent category vector can be... The vector contains data that represent the probabilities of four pre-defined anomalous intent labels.
[0070] The first pre-defined intent classification model can be trained using a fourth training set. This fourth training set can include native language text and corresponding anomalous intent labels; for example, anomalous intent labels could be inducement labels, secret inquiry labels, money transfer labels, and threat labels. During training, features can be extracted from the native language text in the fourth training set to obtain a fifth feature vector. The native language text and the fifth feature vector are then input into the first pre-defined intent classification model, and supervised training is employed to minimize the model's classification loss as the optimization objective.
[0071] Accordingly, the second pre-defined intent classification model can be built based on XLM-RoBERTa. The input to the second pre-defined intent classification model is the second text and the second feature vector, and the output is the second intent category vector. The data contained in the second intent category vector represents the probability of each predicted anomalous intent label. The second intent category vector has the same dimension as the first intent category vector, thus ensuring ease of subsequent calculation of their similarity. For example, the second intent category vector could be... The vector contains data that represent the probabilities of four pre-defined anomalous intent labels.
[0072] The second pre-defined intent classification model can be trained using a fifth training set. This fifth training set can include non-native language text and corresponding anomalous intent labels; for example, anomalous intent labels could be inducement labels, secret inquiry labels, money transfer labels, and threat labels. During training, features can be extracted from the non-native language text in the fifth training set to obtain a sixth feature vector corresponding to the non-native language text. The non-native language text and the sixth feature vector are then input into the second pre-defined intent classification model, and supervised training is employed to minimize the model's classification loss as the optimization objective.
[0073] After obtaining the first intent category vector and the second intent category vector, their cosine similarity can be calculated to determine the intent. Figure 1 Consistency score. Then, based on the semantic relevance score, business behavior matching score, and intention... Figure 1 The consistency score determines the rule validation result; the rule validation result may include semantic relevance score, business behavior matching score, and intention score. Figure 1 A vector of consistency scores. The recognition result of the SMS message to be identified is determined based on the rule verification results. Specifically, this can be based on the semantic relevance score, business behavior matching score, and intention... Figure 1 The consistency scores are weighted and summed (the sum of the weights of the three is 1), and it is determined whether the weighted sum exceeds the corresponding preset risk threshold (e.g., 0.5). If it exceeds the threshold, the identification result can be that the SMS to be identified is a non-abnormal SMS; if it does not exceed the threshold, the identification result can be that the SMS to be identified is an abnormal SMS.
[0074] In some embodiments, before determining the recognition result of the SMS message to be identified based on the rule verification result, the method further includes: inputting the first text and the first feature vector into a preset anomaly recognition model to obtain the anomaly probability corresponding to the first text; wherein, determining the recognition result of the SMS message to be identified based on the rule verification result includes: fusing the rule verification result and the anomaly probability to obtain the recognition result of the SMS message to be identified. The advantage of this setup is that by using the preset anomaly recognition model to perform anomaly recognition on the first text and the first feature vector corresponding to the first text after removing interfering text (e.g., non-native language text, i.e., the second text) in the SMS message to be identified, interference from irrelevant information in the non-native language text can be reduced, thereby accurately identifying the anomaly features present in the native language and obtaining the anomaly probability corresponding to the first text; and fusing the anomaly probability with the rule verification result can further enrich the basis for identifying the SMS message to be identified and improve the accuracy of the recognition result.
[0075] For example, the preset anomaly detection model can be a classification model, such as a bidirectional long short-term memory neural network (CNN-BiLSTM) with superimposed convolutional neural networks. The first text and the first feature vector can be input into the preset anomaly detection model to obtain the anomaly probability corresponding to the first text; where the anomaly probability ranges from [0,1]. The rule verification results and the anomaly probability are fused, specifically by performing a weighted summation of the rule verification results and the anomaly probability; for example, by weighting the semantic relevance score, business behavior matching score, and other relevant metrics. Figure 1 The consistency score and the probability of anomalies are weighted and summed; the sum of the weights of the four factors is 1.
[0076] In some embodiments, before fusing the rule verification result and the anomaly probability to obtain the identification result of the SMS to be identified, the method further includes: extracting risk features from the second text; wherein the risk features include at least one of link text, institutional text, text with pressure-related semantics, and semantically incomplete text; determining a risk score for the second text based on the risk features; wherein fusing the rule verification result and the anomaly probability to obtain the identification result of the SMS to be identified includes: normalizing the rule verification result, the anomaly probability, and the risk score, and performing a weighted summation of the normalization result based on a preset weight; comparing the weighted summation result with a preset risk threshold to obtain the identification result of the SMS to be identified. The advantage of this setting is that if the semantic relevance score, business behavior matching score, and intention... Figure 1If the consistency scores are all high and no abnormal features are identified, then the risk features of the second text can be extracted and the risk score of the second text can be determined based on the risk features to achieve secondary verification of the rule verification results and further improve the accuracy of the identification results.
[0077] In this embodiment, link-type text may include unknown links, such as unregistered links, unofficial links, and links to domains marked as high-risk by third parties. For example, organization-type text may include fake organization text or misspelled organization text. Text with pressure-related semantics may include text with semantics of urgency, threat, and coercion. Semantically incomplete text may include fragmented words without complete semantics, and misspelled words.
[0078] The second text can be scored based on the four types of text in the above risk characteristics. For example, if a certain type of text exists, the risk score is increased by 1, and if a certain type of text does not exist, the risk score is increased by 0. The initial value of the risk score is 0, and its maximum value can be 4, which means that the second text includes the above four types of text. The minimum risk score can be 0, which means that the second text does not include the above four types of text.
[0079] It can evaluate the semantic relevance score, business behavior matching score, and intention in the rule validation results. Figure 1 The consistency score, anomaly probability, and risk score are normalized so that their values all fall within the range of [0,1]. Then, the normalized results are weighted and summed based on preset weights; each of the five data points can correspond to a different preset weight, and the sum of the five preset weights is 1. For example, the semantic relevance score can represent the core features of abnormal SMS messages, so it can be assigned the largest preset weight, for example, 0.3; the business behavior matching score can represent the spoofing features of abnormal SMS messages, so it can be assigned the second largest preset weight, for example, 0.25; Figure 1 Consistency score can characterize the correlation features between the native language and non-native language, and can be assigned the third largest preset weight, for example, 0.2; anomaly probability can characterize the anomalous features corresponding to the native language text, and can be assigned the fourth largest preset weight, for example, 0.15; risk score can characterize the explicit risk features in the non-native language text, and its corresponding preset weight can be 0.1. The above preset weights are not unique and can be set based on empirical values in practical applications.
[0080] The preset risk threshold can be set based on empirical values, such as 0.5. If the weighted sum is greater than or equal to the preset risk threshold, the SMS message to be identified is determined to be a non-abnormal SMS message; if the weighted sum is less than the preset risk threshold, the SMS message to be identified is determined to be an abnormal SMS message.
[0081] Example 4 Figure 5 This is a schematic diagram of an abnormal SMS identification device provided in Embodiment 4 of the present invention. Figure 5 As shown, the device includes a text segmentation module 501 and a recognition result determination module 502.
[0082] The text segmentation module is used to segment the SMS message to be identified into a first text and a second text; wherein the SMS message to be identified includes text in multiple language types; the first text and the second text belong to different languages; The identification result determination module is used to perform rule verification on the first text and the second text based on a preset strategy, obtain the rule verification result, and determine the identification result of the SMS to be identified based on the rule verification result; wherein, the identification result is used to indicate whether the SMS to be identified is an abnormal SMS.
[0083] This invention provides an abnormal SMS identification device. It splits the SMS to be identified into a first text and a second text by text segmentation. The SMS to be identified includes text in multiple language types. The first text and the second text belong to different languages. A rule-based verification is performed on the first text and the second text based on a preset strategy to obtain a rule verification result. The identification result is then used to indicate whether the SMS to be identified is abnormal. This device splits the SMS to be identified into two texts with different languages and performs rule verification on the two texts based on a preset strategy, thereby completing the identification of abnormal SMS. Compared to existing technologies that rely solely on single language features to detect SMS, this invention allows for cross-verification of the two texts based on a preset strategy, enhancing the ability to identify abnormal information carried in the SMS to be identified, thus effectively improving the accuracy of abnormal SMS identification. Furthermore, the preset strategy includes at least one of semantic correlation verification and behavioral compliance verification on the first text and the second text. This allows for cross-verification of the semantic relevance and / or behavioral compliance represented by the verification of the first and second texts based on a preset strategy, thereby further enhancing the ability to identify abnormal information carried in the SMS messages to be identified.
[0084] Optionally, the preset strategy includes at least one of semantic relevance verification and behavioral compliance verification of the first text and the second text.
[0085] Optionally, the text splitting module includes: The preprocessing unit is used to preprocess the SMS message to be identified to obtain structured text; The language identification unit is used to perform character-by-character language identification on the structured text using a preset language detection model to obtain the language label corresponding to each text character in the structured text; wherein, the language label includes native language label and non-native language label; The text segmentation unit is used to divide the structured text into a first text and a second text based on the language tags corresponding to each text character in the structured text.
[0086] Optionally, the text segmentation unit includes: The triplet determination subunit is used to merge text characters according to the language tags corresponding to each text character in the structured text and the position of each text character in the structured text to obtain several text triplets; wherein, the text triplet includes a text segment, the segment position corresponding to the text segment, and the language tag corresponding to the text segment; the text segment is a segment formed by merging text characters with the same language tag; The text determination subunit is used to concatenate text fragments with the same language tag in several text triples based on the fragment position corresponding to the text fragment in the text triples to obtain a first text and a second text; wherein, the first text is the text concatenated after concatenating text fragments corresponding to the native language tag; and the second text is the text concatenated after concatenating text fragments corresponding to the non-native language tag.
[0087] Optionally, the identification result determination module includes: The semantic relevance determination unit is used to extract features from the first text and the second text respectively, to obtain a first feature vector corresponding to the first text and a second feature vector corresponding to the second text; and to input the first feature vector and the second feature vector into a preset semantic relevance model to obtain a semantic relevance score. A business behavior matching degree determination unit is used to perform business behavior verification on the first text and the second text based on preset business rules to obtain a business behavior matching score between the first text and the second text; wherein, the preset business rules include at least one of the following: rules for verifying the compliance of business logic, rules for verifying the compliance of operation channels, and rules for verifying the reasonableness of prompts; The rule verification result determination unit is used to determine the rule verification result based on the semantic relevance score and the business behavior matching score; The identification result determination unit is used to determine the identification result of the SMS message to be identified based on the verification result of the rules.
[0088] Optionally, the identification result determination module further includes: The first verification unit is used to determine whether either the semantic relevance score or the business behavior matching score is within the corresponding preset score range before determining the rule verification result based on the semantic relevance score and the business behavior matching score. meaning Figure 1 The consistency determination unit is configured to, if so, input the first text and the first feature vector into a first preset intent classification model to obtain a first intent category vector, input the second text and the second feature vector into a second preset intent classification model to obtain a second intent category vector, and determine the intent category based on the similarity between the first intent category vector and the second intent category vector. Figure 1 Consistency score; Specifically, the rule verification result determination unit is used for: Based on the semantic relevance score, the business behavior matching score, and the meaning Figure 1 Consistency score determination rule verification results.
[0089] Optionally, the identification result determination module further includes: An anomaly probability determination unit is used to input the first text and the first feature vector into a preset anomaly recognition model before determining the recognition result of the SMS to be identified based on the rule verification result, so as to obtain the anomaly probability corresponding to the first text. Specifically, the identification result determination unit is used for: The rule verification result and the anomaly probability are fused to obtain the identification result of the SMS message to be identified.
[0090] Optionally, the identification result determination module further includes: The risk feature determination unit is used to extract risk features from the second text before fusing the rule verification result and the anomaly probability to obtain the identification result of the SMS to be identified; wherein, the risk features include at least one of link text, organization text, text with pressure semantics, and semantically incomplete text; A risk score determination unit is used to determine the risk score of the second text based on the risk characteristics; Specifically, the identification result determination unit is used for: The rule verification result, the anomaly probability, and the risk score are normalized, and the normalization result is weighted and summed based on a preset weight. The weighted summation result is compared with a preset risk threshold to obtain the identification result of the SMS to be identified.
[0091] The abnormal SMS identification device provided in this embodiment of the invention can execute the abnormal SMS identification method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.
[0092] Example 5 Figure 6 A schematic diagram of an electronic device 600 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.
[0093] like Figure 6 As shown, the electronic device 600 includes at least one processor 601 and a memory, such as a read-only memory (ROM) 602 or a random access memory (RAM) 603, communicatively connected to the at least one processor 601. The memory stores computer programs executable by the at least one processor. The processor 601 can perform various appropriate actions and processes based on the computer program stored in the ROM 602 or loaded into the RAM 603 from storage unit 608. The RAM 603 may also store various programs and data required for the operation of the electronic device 600. The processor 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.
[0094] Multiple components in electronic device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of displays, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows electronic device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0095] Processor 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 601 performs the various methods and processes described above, such as the abnormal SMS identification method.
[0096] In some embodiments, the abnormal SMS identification method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by processor 601, one or more steps of the abnormal SMS identification method described above may be performed. Alternatively, in other embodiments, processor 601 may be configured to perform the abnormal SMS identification method by any other suitable means (e.g., by means of firmware).
[0097] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0098] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0099] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.
[0100] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).
[0101] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.
[0102] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.
[0103] This disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the abnormal SMS identification method provided in the above embodiments.
[0104] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.
[0105] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.
Claims
1. A method for identifying abnormal text messages, characterized in that, include: The SMS message to be identified is split into a first text and a second text; wherein the SMS message to be identified includes text in multiple language types; the first text and the second text belong to different languages; The first text and the second text are subjected to rule verification based on a preset strategy to obtain rule verification results, and the identification result of the SMS to be identified is determined based on the rule verification results; wherein, the identification result is used to indicate whether the SMS to be identified is an abnormal SMS.
2. The abnormal SMS identification method according to claim 1, characterized in that, The preset strategy includes at least one of the following: semantic relevance verification of the first text and the second text, and behavioral compliance verification.
3. The abnormal SMS identification method according to claim 1, characterized in that, The text segmentation of the SMS message to be identified, resulting in a first text and a second text, includes: The SMS messages to be identified are preprocessed to obtain structured text; The structured text is analyzed character by character using a pre-defined language detection model to obtain language tags corresponding to each text character in the structured text; wherein, the language tags include native language tags and non-native language tags; Based on the language tags corresponding to each text character in the structured text, the structured text is divided into a first text and a second text.
4. The abnormal SMS identification method according to claim 3, characterized in that, The method of dividing the structured text into a first text and a second text based on the language tags corresponding to each text character in the structured text includes: Based on the language tags corresponding to each text character in the structured text and the position of each text character in the structured text, the text characters are merged to obtain several text triples; wherein, the text triple includes a text fragment, the fragment position corresponding to the text fragment, and the language tag corresponding to the text fragment; the text fragment is a fragment formed by merging text characters with the same language tag; Based on the segment positions corresponding to the text segments in the text triples, the text segments with the same language tags in the plurality of text triples are concatenated to obtain the first text and the second text; wherein, the first text is the text concatenated after the text segments corresponding to the native language tags are concatenated; and the second text is the text concatenated after the text segments corresponding to the non-native language tags are concatenated.
5. The abnormal SMS identification method according to claim 1, characterized in that, The step of performing rule validation on the first text and the second text based on a preset strategy to obtain rule validation results includes: Feature extraction is performed on the first text and the second text respectively to obtain a first feature vector corresponding to the first text and a second feature vector corresponding to the second text; the first feature vector and the second feature vector are input into a preset semantic association model to obtain a semantic association score; Based on preset business rules, the first text and the second text are subjected to business behavior verification to obtain a business behavior matching score between the first text and the second text; wherein, the preset business rules include at least one of the following: rules for verifying the compliance of business logic, rules for verifying the compliance of operation channels, and rules for verifying the reasonableness of prompts; The rule verification result is determined based on the semantic relevance score and the business behavior matching score.
6. The abnormal SMS identification method according to claim 5, characterized in that, Before determining the rule verification result based on the semantic relevance score and the business behavior matching score, the method further includes: Determine whether either the semantic relevance score or the business behavior matching score falls within the corresponding preset score range; If so, the first text and the first feature vector are input into the first preset intent classification model to obtain the first intent category vector, the second text and the second feature vector are input into the second preset intent classification model to obtain the second intent category vector, and the intent consistency score is determined based on the similarity between the first intent category vector and the second intent category vector. The step of determining the rule verification result based on the semantic relevance score and the business behavior matching score includes: The rule verification result is determined based on the semantic relevance score, the business behavior matching score, and the intent consistency score.
7. The abnormal SMS identification method according to claim 5, characterized in that, Before determining the identification result of the SMS message to be identified based on the rule verification result, the method further includes: Input the first text and the first feature vector into a preset anomaly recognition model to obtain the anomaly probability corresponding to the first text; The step of determining the identification result of the SMS message to be identified based on the rule verification result includes: The rule verification result and the anomaly probability are fused to obtain the identification result of the SMS message to be identified.
8. The abnormal SMS identification method according to claim 7, characterized in that, Before fusing the rule verification result and the anomaly probability to obtain the identification result of the SMS message to be identified, the method further includes: Extract risk features from the second text; wherein, the risk features include at least one of the following: link text, institutional text, text with pressure-related semantics, and text with semantic incompleteness; Determine the risk score of the second text based on the aforementioned risk characteristics; The step of fusing the rule verification result and the anomaly probability to obtain the identification result of the SMS message to be identified includes: The rule verification results, the anomaly probability, and the risk score are normalized, and the normalization results are weighted and summed based on preset weights. The weighted summation result is compared with a preset risk threshold to obtain the identification result of the SMS message to be identified.
9. An abnormal SMS identification device, characterized in that, include: The text segmentation module is used to segment the SMS message to be identified into a first text and a second text; wherein the SMS message to be identified includes text in multiple language types; the first text and the second text belong to different languages; The identification result determination module is used to perform rule verification on the first text and the second text based on a preset strategy, obtain the rule verification result, and determine the identification result of the SMS to be identified based on the rule verification result; wherein, the identification result is used to indicate whether the SMS to be identified is an abnormal SMS.
10. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the abnormal SMS identification method according to any one of claims 1-8.
11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the abnormal SMS identification method according to any one of claims 1-8.
12. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the abnormal SMS identification method according to any one of claims 1-8.
