Complaint information processing method and device, computer equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, can be applied to the fields of financial science and technology and medical health, and discloses a complaint information processing method and device, computer equipment and a medium. The recognition content is classified according to the complaint type, the emergency degree and the operator source, the recognition content and the original short message content are matched through the composite matching model, the whole complaint information processing process is automatic, traditional manual operation is replaced, manual intervention is not needed from data cleaning to result output, the processing efficiency is improved, and the user experience is improved. Moreover, the complaint source can be accurately recognized through the composite matching model, the shielding misjudgment rate is reduced, and the workload of invalid manual rechecking is reduced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology and can be applied to fields such as finance and medical care. In particular, it relates to a complaint information processing method, device, computer equipment and medium. Background Art

[0002] With the continuous development of artificial intelligence (AI) technology, SMS has become a crucial channel for communication between businesses and users. Furthermore, SMS services play a vital role in key sectors such as fintech and healthcare, providing services such as bank transaction verification and medical appointment reminders. However, with the growth of SMS traffic, spam, fraudulent text messages, and user complaints have become increasingly prominent. These issues can pose significant risks, particularly in the fintech and healthcare sectors. For example, in the fintech sector, fraudulent text messages sent from fake base stations can trick users into disclosing their account information. In the healthcare sector, spam can disrupt important notifications and delay patient treatment.

[0003] Therefore, the sending of spam and fraudulent text messages is often complained by users. In the existing technology, the processing of complaint information mainly relies on manual review or automated systems based on fixed rules, which are inefficient and difficult to cope with the real-time processing needs of massive data. In addition, traditional complaint classification methods usually use static rules or single-dimensional cluster analysis, which cannot dynamically adapt to different types of complaint scenarios, resulting in insufficient classification accuracy. In terms of SMS content matching, existing technologies mostly use simple text similarity calculations, which are difficult to accurately identify complaint intent and are prone to misjudgment or omission. At the same time, the existing system lacks intelligent analysis of user sending behavior, and is unable to efficiently screen high-risk users and accurately block them, affecting the timeliness and accuracy of complaint processing. Summary of the Invention

[0004] The embodiments of the present invention provide a complaint information processing method, apparatus, computer equipment, and medium, aiming to solve the problem of low processing efficiency and inability to accurately shield complaint information.

[0005] In a first aspect, an embodiment of the present invention provides a method for processing complaint information, including:

[0006] Acquire SMS complaint data in real time and clean the SMS complaint data;

[0007] Use the optimized neural network model to perform intent recognition on the cleaned SMS complaint data and obtain the recognized content;

[0008] Based on the dynamic clustering module, the identified content is classified according to complaint type, urgency and operator source, and classification labels are generated;

[0009] Pre-building a composite matching model, inputting the identified content and the original SMS content into the composite matching model for matching, and generating a blocking instruction based on the classification label if the matching degree between the identified content and the original SMS content exceeds a matching threshold;

[0010] Obtain a user list to which the original SMS content is sent, sort and filter the users in the user list, and block the filtered users according to the blocking instruction.

[0011] In a second aspect, an embodiment of the present invention provides a complaint information processing device, including:

[0012] A cleaning unit, configured to obtain SMS complaint data in real time and clean the SMS complaint data;

[0013] The intent recognition unit is used to use the optimized neural network model to perform intent recognition on the cleaned SMS complaint data to obtain the recognized content;

[0014] a classification unit, configured to classify the identified content according to complaint type, urgency, and operator source based on a dynamic clustering module, and generate classification labels;

[0015] a matching unit configured to pre-build a composite matching model, input the identified content and the original SMS content into the composite matching model for matching, and generate a blocking instruction based on the classification label if the matching degree between the identified content and the original SMS content exceeds a matching threshold;

[0016] The shielding unit is used to obtain the user list to which the original short message content is sent, sort and filter the users in the user list, and shield the filtered users according to the shielding instruction.

[0017] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the complaint information processing method described above when executing the computer program.

[0018] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the complaint information processing method as described above is implemented.

[0019] An embodiment of the present invention provides a complaint information processing method, apparatus, computer equipment and medium. The complaint information processing method uses an optimized neural network model to identify the intent of cleaned SMS complaint data, and based on a dynamic clustering module, classifies the identified content according to complaint type, urgency and operator source, and uses a composite matching model to match the identified content with the original SMS content. The entire complaint information processing is fully automated, replacing traditional manual operations. No human intervention is required from data cleaning to result output, thereby improving processing efficiency. Moreover, the composite matching model can accurately identify the source of the complaint, reduce the false positive rate of shielding, and reduce the workload of invalid manual review. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 A schematic diagram of an application environment of a complaint information processing method according to an embodiment of the present invention;

[0022] Figure 2 A flowchart of a method for processing complaint information according to an embodiment of the present invention;

[0023] Figure 3 for Figure 2 A schematic flow chart of a specific implementation of step S10;

[0024] Figure 4 for Figure 2 A schematic flow chart of a specific implementation of step S20;

[0025] Figure 5 for Figure 2 A schematic flow chart of a specific implementation of step S40;

[0026] Figure 6 for Figure 2 Another specific implementation flow diagram of step S40;

[0027] Figure 7 for Figure 2 A schematic flow chart of a specific implementation of step S50;

[0028] Figure 8 A schematic structural diagram of a complaint information processing device according to an embodiment of the present invention;

[0029] Figure 9A schematic diagram of the structure of a computer device according to an embodiment of the present invention;

[0030] Figure 10 FIG. 2 is another structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0032] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0033] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0034] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0035] See also Figure 1 , Figure 1 A schematic diagram of an application scenario of a complaint information processing method provided by an embodiment of the present invention. The client can communicate with the server via a network; the server can obtain SMS complaint data in real time and clean the SMS complaint data; the cleaned SMS complaint data is subjected to intent recognition using an optimized neural network model to obtain recognized content; based on a dynamic clustering module, the recognized content is classified by complaint type, urgency, and operator source, and classification labels are generated; a composite matching model is pre-constructed, and the recognized content and original SMS content are input into the composite matching model for matching; if the degree of match between the recognized content and the original SMS content exceeds a matching threshold, a blocking instruction is generated according to the classification label; a list of users to whom the original SMS content was sent is obtained, the users in the user list are sorted and filtered, and the filtered users are blocked according to the blocking instruction.

[0036] In the present invention, for the application scenario in the field of financial technology, the example of AA Bank receiving user complaints about "fraudulent text messages" in real time is used for explanation. The bank's SMS customer service system captures the complaint text messages sent by users to the designated number (i.e., SMS complaint data) in real time, such as: "Received a text message from number 1069XXXX, saying 'Your account is about to be frozen, click link xxx to complete verification', suspected to be a fraud!" For the captured SMS complaint data, sensitive information needs to be removed to hide the user's account tail number, link domain name, etc.; therefore, it is necessary to clean the SMS complaint data, and uniformly classify keywords such as "account freeze" and "phishing link" as "fraud-related" terms, and filter invalid content to eliminate garbled characters and avoid repeated complaints.

[0037] The cleaned SMS content is input into the optimized neural network model. Through neural network analysis, the keywords "fraud", "link", and "account freeze" are extracted, and the intention is determined to be "preventing financial fraud" (i.e., identifying content). The identified content is then dynamically clustered and classified. According to the complaint type, the identified content is classified as "fraud complaint"; according to the urgency, the identified content is marked as "urgent (high risk)" because it involves financial security; according to the operator source, the SMS sending number 1069XXXX of the identified content belongs to "sent through mobile channel", so the classification label is generated: [Fraud - Emergency - Mobile Channel].

[0038] The identified content and the original SMS content are input into the composite matching model for verification. The original SMS content is the SMS content sent by the bank to the user. When the matching degree between the identified content and the original SMS content is 92%, it has exceeded the matching threshold of 80%. At this time, a blocking instruction is generated based on the classification label of the identified content, and a list of users to whom the original SMS content was sent is obtained. The list is sorted by priority based on "users who have clicked on the link" or "users who have had financial transaction records". SMS messages are intercepted for user numbers through the mobile operator interface, and risk warnings are pushed to users.

[0039] For application scenarios in the medical and health field, take a tertiary hospital receiving a patient complaint as an example. The hospital complaint platform receives patient text messages (i.e., SMS complaint data) in real time, such as "I received a text message from an unknown number 138XXXX. The content contains my name, medical record number, and diagnosis results. I suspect the hospital has leaked information!" For the received SMS complaint data, sensitive information needs to be removed to hide medical information such as medical record number and diagnosis details. Therefore, the SMS complaint data needs to be cleaned, and "information leakage" and "medical record number" need to be classified as "privacy and security". In addition, invalid content needs to be filtered to eliminate garbled characters and avoid repeated complaints.

[0040] The cleaned SMS content is input into the optimized neural network model. Through neural network analysis, the keywords "medical record number", "diagnosis result", and "leakage" are extracted, and the intention is determined to be "complaint about medical privacy information leakage" (i.e., identified content). The identified content is then dynamically clustered and classified. According to the complaint type, the identified content is classified as "information security complaint"; according to the urgency, the identified content is marked as "urgent (high risk)" because it involves personal health privacy; according to the operator source, the SMS sending number 138XXXX of the identified content is "sent through China Unicom channel", so the classification label is generated: [Information Security - Emergency - China Unicom channel].

[0041] The identified content and the original SMS content are input into the composite matching model for verification. The original SMS content is the SMS content sent by the hospital to the user. When the matching degree between the identified content and the original SMS content is 95%, the matching threshold of 80% has been exceeded. At this time, a blocking instruction is generated according to the classification label of the identified content, and a list of users to whom the original SMS content was sent is obtained. The patient group whose SMS was leaked is confirmed by medical record number, and priority is given to blocking "patients whose information has been publicly leaked" or "high-frequency receiving numbers with a risk of secondary leakage". SMS messages to user numbers are intercepted through the China Unicom operator interface, and privacy protection consulting services are provided to users.

[0042] The present invention fully automates the processing of complaint information, replacing traditional manual operations. From data cleaning to result output, no human intervention is required, which greatly shortens the processing cycle. In addition, the composite matching model can accurately identify the source of the complaint, reduce the shielding misjudgment rate, and reduce the workload of invalid manual review. Among them, the client provides an interactive platform for users, allowing users to clean the cleaned SMS complaint data through the optimized neural network model carried by the server. For example, the user sends SMS complaint data to the optimized neural network model and composite matching model of the server through the client, and the optimized neural network model and composite matching model return the shielding results to the client. Among them, the client can be but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server can be implemented as an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.

[0043] In the existing technology, SMS complaints have been fully issued since 2024. The daily complaint volume from the 12321 platform of the Ministry of Industry and Information Technology and the operator complaint sources has surged from dozens per month to more than a thousand per month. The data volume has increased significantly. Starting from 2024, the Monetary Authority of Singapore will issue thousands of complaints every quarter to review the complaint information of the previous quarter. As the number of complaints increases, the difficulty of analyzing complaint data for operations also increases accordingly. After the complaint is collected, it is currently necessary to manually confirm whether the match is correct one by one. The correct match will be blocked. If there are multiple mobile phone numbers that meet the requirements after the match, no suitable solution has been found. Failed matches need to be marked as failed matches. Therefore, it is hoped that this step of marking the matching status can be automatically identified by intelligent tools, and blocking suggestions will be given for multiple mobile phone numbers after matching. Therefore, please refer to Figure 2 , an embodiment of the present invention provides a complaint information processing method, including S10-S50:

[0044] S10. Acquire SMS complaint data in real time and clean the SMS complaint data;

[0045] In this step, real-time access to SMS complaint data ensures that banks or hospitals can respond promptly to customer feedback, preventing issues from worsening due to delayed processing. For example, in the fintech sector, customers may send urgent SMS reports about account anomalies. Failure to obtain SMS data in a timely manner can increase the risk of financial loss. In healthcare, patients may complain about medical service issues via SMS. Real-time data acquisition helps medical institutions quickly intervene and resolve issues, ensuring patient safety.

[0046] Cleaning SMS complaint data can improve the accuracy of subsequent data processing. This is because non-standard punctuation, garbled characters, and repeated fields can interfere with the semantic understanding of text by subsequent optimized neural network models and composite matching models. For example, during the intent recognition phase, garbled characters in the text can cause the optimized neural network model to misjudge the intent. During the classification phase, repeated fields increase the computational burden and reduce clustering effectiveness. Cleaning SMS complaint data can make it more standardized, improve the recognition accuracy of subsequent optimized neural network models, and reduce misjudgments caused by data noise.

[0047] In one embodiment, if Figure 3 As shown, S10 includes:

[0048] S11. Monitor the SMS channel in real time through the SMS gateway interface, and obtain SMS complaint data through the SMS channel;

[0049] S12. Use regular expressions to replace non-standard punctuation marks in the complaint data with uniform symbols and remove garbled characters and repeated fields in the SMS complaint data to complete data cleaning.

[0050] In S11, the short message gateway interface is a standardized interface provided by the communication operator, which is used to connect the enterprise system and the short message channel to realize the sending and receiving of short messages. Through this interface, the enterprise can directly interface with the operator network to obtain the short message content sent by the user to the specified number. Once a new short message (especially a complaint short message) enters, it is immediately captured and the short message complaint data is extracted to ensure that the short message complaint information is obtained without delay. By real-time monitoring, the complaint information can be avoided to lag behind, and the enterprise can master the user feedback in the first time, for example, when the user complains about network failure or service problems, the system can immediately push the data to the processing process to shorten the response time. In this step, the short message complaint data is obtained through the short message gateway interface of the operator, which is more reliable than the third-party platform, reduces the loss of information caused by interface failure, and ensures the complete collection of short message complaint data. Moreover, real-time monitoring can capture all complaint information through the short message channel, avoiding the omission of urgent complaints caused by manual collection or timed grabbing, and improving the data integrity.

[0051] In S12, the regular expression is a logical formula for matching and processing text, which can quickly locate and replace specific characters according to pre-set rules. For example, non-standard punctuation marks such as "!!!" or "???" can be uniformly replaced by "!" and "?", or non-standard characters can be recognized and deleted (such as ).

[0052] For non-standard punctuation marks, non-standard punctuation marks can be replaced by uniform symbols to achieve uniform text format and avoid data parsing errors caused by punctuation disorder (such as "The service is too bad!!!" standardized to "The service is too bad!"). For non-standard characters, invalid characters (such as binary non-standard characters) that may be generated due to encoding problems in short message transmission can be filtered out to ensure that the text is readable. For repeated fields, such as repeated "The service is not good" in the complaint content, the repeated field "The service is not good" is removed, and only one "The service is not good" is kept to reduce data redundancy. After the uniform punctuation symbols are replaced, the complaint text format input by different channels and different users can be ensured to be uniform, and the complaint type can be quickly classified by keyword search (such as "complaint" and "dissatisfaction") to improve the data analysis efficiency. Removing non-standard characters can avoid system errors (such as database storage abnormalities) caused by invalid characters, and cleaning repeated fields makes the data more concise and reduces storage costs.

[0053] In other embodiments, the cleaned short message complaint data can be directly connected to the natural language processing (NLP) model or the manual review process, but in this embodiment, in order to accurately match the short message complaint data with the original short message content, the short message complaint data also needs to be subjected to intent recognition. Specifically, S20, the cleaned short message complaint data is subjected to intent recognition by using an optimized neural network model to obtain the recognized content.

[0054] In this step, the optimized neural network model is refined based on the original one. This optimized model can distinguish more subtle differences, making subsequent classification more accurate. By using the optimized neural network model to identify intent in the cleaned SMS complaint data, it provides reliable support for subsequent customer service optimization and operational decision-making.

[0055] In one embodiment, if Figure 4 As shown, S20 includes:

[0056] S21. Preload the BERT model as the original neural network model, introduce a GRU layer into the last layer of the BERT model, and add a fully connected layer and a softmax layer after the GRU layer to obtain an optimized neural network model;

[0057] S22, configuring training parameters for the optimized neural network model, and training the optimized neural network model using pre-collected and labeled training data;

[0058] S23. Segment and vectorize the cleaned SMS complaint data and input it into the trained optimized neural network model for intent recognition, and output the recognized content.

[0059] In S21, the BERT model is a Transformer-based pre-trained language model that captures text context information through a bidirectional attention mechanism. For example, in financial complaints, BERT can distinguish the semantic differences between "low returns on financial products" (investment complaints) and "failed credit card repayments" (account complaints). Therefore, the BERT model is preferred as the original neural network model. The GRU layer is introduced to process the sequential characteristics of text, because complaint texts usually contain information in a temporal or logical order, such as "I first bought a financial product, and then found that the returns did not match the promotion." Then, the gating mechanism of the GRU layer can effectively capture long-distance dependencies, avoid the gradient vanishing problem of traditional technologies, and thus more accurately model text sequences.

[0060] The fully connected layer and the Softmax layer are responsible for mapping the features extracted by BERT+GRU to the specific intent category space. The fully connected layer enhances the model's expressiveness through nonlinear transformations, while the Softmax layer outputs the probability distribution of each category, enabling multi-classification tasks. The optimized neural network model combines the semantic understanding capabilities of the BERT model with the sequence modeling capabilities of the GRU layer, significantly improving the accuracy of intent recognition and enabling precise classification in subsequent processes.

[0061] In S22, the training parameters of the optimized neural network model (such as learning rate, batch size, and number of iterations) can be pre-configured to confirm the convergence speed and generalization ability of the optimized neural network model. Use labeled training data for supervised learning to allow the optimized neural network model to learn the mapping relationship between complaint text and intent category. During the training process, the optimized neural network model continuously adjusts the weight parameters through back propagation and minimizes the error between the prediction results and the labeled labels. To prevent overfitting, regularization methods such as early stopping strategy and Dropout can also be used. Through multiple rounds of iterative training, the optimized neural network model gradually optimizes the recognition ability of various types of complaint intentions.

[0062] In S23, the cleaned SMS complaint data is tokenized, breaking the Chinese text into meaningful words or phrases. For example, "I want to complain about this financial product" is tokenized into "I," "want," "complain," and "this financial product." Vectorization converts the tokenized text into numerical vectors, enabling the BERT model to generate word vectors that incorporate contextual information and more accurately represent the text's semantics. The vectorized SMS complaint data is then fed into a trained, optimized neural network model for intent inference, ultimately outputting the recognized content.

[0063] The output of the recognition provides a basis for subsequent classification and processing. In the field of financial technology, accurate intent recognition can guide customer service personnel to respond quickly. For example, for marketing complaints, all marketing text messages from the mobile phone number are immediately blocked, and for collection complaints, all collection and marketing text messages from the mobile phone number are immediately blocked. In the field of healthcare, it can help hospitals distinguish between urgent complaints (such as medical accidents) and general complaints (such as the quality of cafeteria food), giving priority to high-priority issues.

[0064] S30. Based on a dynamic clustering module, classify the identified content according to complaint type, urgency, and operator source, and generate classification labels;

[0065] In this step, a dynamic clustering algorithm is used to automatically classify the identified content in multiple dimensions. The specific dimensions include: complaint type, urgency, and operator source. Among them, the complaint type is extracted through text analysis to extract keywords, and semantically similar complaints are classified into the same category; the urgency needs to be combined with preset rules or historical data training models to dynamically determine the priority; the operator source is directly classified according to the operator identifier (such as China Mobile, China Unicom, etc.) in the SMS complaint data before the identified content is processed.

[0066] Ultimately, each complaint is assigned a classification label (such as "Complaint Type: Marketing - Urgency: Medium - Operator: China Telecom") to facilitate subsequent processing.

[0067] In one embodiment, S30 includes:

[0068] The DBSCAN density clustering algorithm is used as a dynamic clustering module, the identified content is input into the dynamic clustering module, and the identified content is classified according to the preset complaint type keyword library, urgency determination rules and operator identification information, and corresponding classification labels are generated.

[0069] In this embodiment, the DBSCAN density clustering algorithm is a density-based spatial clustering algorithm that divides areas with sufficient density into clusters and finds clusters of arbitrary shapes in noisy data. The DBSCAN density clustering algorithm automatically identifies high-density complaint subject areas to form clusters, while filtering out low-density noise points (such as rare or vague complaints). When clustering, the complaint type keywords, urgency determination rules and operator identification are comprehensively considered to ensure that the classification results cover all dimensions of business needs at the same time. Each cluster corresponds to a complaint pattern, and a classification label is automatically generated. Noise points (abnormal complaints) that are not clustered are marked separately for manual review.

[0070] This embodiment uses the DBSCAN density clustering algorithm for classification. It eliminates the need to pre-set the number of clusters (as with the K-means algorithm, which requires a specific K value) and automatically discovers natural clusters within the input data (i.e., the identified content). This makes it suitable for scenarios with diverse complaint content and dynamically changing topics. Furthermore, the DBSCAN density clustering algorithm can directly filter out low-density noise points (such as unclear or rare complaints), preventing them from interfering with the mainstream classification results, while retaining its anomaly detection capabilities (noise points can be analyzed separately).

[0071] S40: pre-build a composite matching model, input the identified content and the original SMS content into the composite matching model for matching, and if the matching degree between the identified content and the original SMS content exceeds a matching threshold, generate a blocking instruction according to the classification label;

[0072] In this step, a composite matching model is pre-built so that the model can achieve efficient and accurate analysis of SMS content through multi-dimensional feature fusion. The model needs to be trained with a large amount of labeled SMS data (including normal SMS and SMS that need to be blocked) to learn how to distinguish different types of content and optimize the parameters of the matching algorithm. Among them, the original SMS content refers to the SMS content sent by the bank or hospital to the user, that is, the complete SMS text received by the user. The identified content and the original SMS content are input into the composite matching model together. The composite matching model calculates the matching degree between the identified content and the original SMS content based on the preset algorithm and features. This matching degree represents the degree of similarity between the two.

[0073] The matching degree between the obtained recognition content and the original SMS content is judged against the pre-set matching threshold. When the matching degree exceeds this matching threshold, it is considered that the recognition content is highly correlated with the original SMS content. Then, a corresponding blocking instruction is generated according to the classification label of the recognition content. This blocking instruction includes operations such as moving the SMS to the trash, directly deleting the SMS, and sending a warning message to the user.

[0074] In practice, the matching threshold setting needs to be adjusted based on the actual application scenario and requirements to generate the optimal blocking instructions. This is because a too high matching threshold may result in missed messages (i.e., failure to block messages that should be blocked), while a too low matching threshold may result in false positives (i.e., incorrectly blocking normal messages).

[0075] In one embodiment, if Figure 5 As shown, S40 includes:

[0076] S41, constructing a TF-IDF text similarity calculation module and a BiLSTM semantic matching module into a composite matching model;

[0077] S42, inputting the identified content and the original SMS content into a TF-IDF text similarity calculation module to calculate word frequency similarity and obtain a similarity score;

[0078] S43: Input the similarity score into the BiLSTM semantic matching module for deep semantic matching.

[0079] In the S41, the TF-IDF text similarity calculation module measures the importance of a word in a document. It evaluates word importance by calculating the product of term frequency (TF) and inverse document frequency (IDF). The BiLSTM semantic matching module captures contextual information and long-range dependencies in the text. Using bidirectional LSTM, this model simultaneously considers both forward and backward information in the text, thereby better understanding the text's semantics. In the composite matching model, the TF-IDF module primarily captures shallow text features, such as the frequency and distribution of keywords, to calculate the word frequency similarity between the identified content and the original text message. The BiLSTM module primarily captures deeper semantic features of the text, learning contextual representations of the text to achieve more accurate semantic matching. The composite matching model is constructed by integrating the TF-IDF and BiLSTM modules. The TF-IDF module first calculates the word frequency similarity of the text, and then uses this result as input to the BiLSTM module for deeper semantic matching. This fusion approach takes into account both shallow features of the text (such as keyword matching) and deep semantic features (such as context understanding), thereby improving the accuracy and robustness of matching.

[0080] In S42, the cosine similarity (i.e., term frequency similarity) between the identified content and the original short message content is calculated using the TF-IDF algorithm to obtain a similarity score. This similarity score is used as input for the next step of the BiLSTM semantic matching module.

[0081] In S43, the similarity score, the identified content, and the original short message content are input into the BiLSTM model. The BiLSTM model generates context-aware text representations for both the identified content and the original short message content. Based on the text representations generated by the BiLSTM, the semantic similarity between the two pieces of text is calculated. The matching degree of the identified content and the original short message content is determined based on the semantic similarity and the similarity score.

[0082] In an embodiment, as shown in FIG. 4B, S40 further includes: Figure 6

[0083] S44, according to the matching situation of historical short message complaint data, the matching threshold is confirmed by cross-validation method;

[0084] S45, when the matching degree of the identified content and the original short message content exceeds the matching threshold, the classification label of the current identified content is obtained;

[0085] S46, according to the complaint type in the classification label, a corresponding shielding instruction is generated.

[0086] In S44, the historical short message complaint data includes short message content that has been complained by users in the past and its classification labels (such as spam short messages, fraudulent short messages, etc.). It is used to train and verify the composite matching model, helping the model learn how to distinguish between normal short messages and short messages that need to be shielded. Through cross-validation method, the setting of the matching threshold has robustness and generalization ability, avoiding overfitting or underfitting. The specific implementation process is as follows: the historical short message complaint data is divided into training set, validation set and test set. The training set is used to train the composite matching model, different matching thresholds are tried on the validation set, the model performance indicators (such as accuracy, recall rate, F1 score, etc.) under each matching threshold are calculated, and the matching threshold with the best performance on the validation set is selected, usually the matching threshold that balances the accuracy and recall rate. The performance of the final model is evaluated on the test set to ensure that its performance on new data meets the expectations.

[0087] ​In S45, the matching degree output by the composite matching model is obtained. If the matching degree exceeds the matching threshold, the identified content is considered to be highly correlated with the original SMS content. At this time, the classification label of the current identified content is obtained, which provides the basis for the subsequent generation of the blocking instruction. If the matching degree is lower than the matching threshold, the source of the complaint is confirmed. If the source of the complaint is the MIIT 12321 platform or the operator, the matching degree between the identified content and the original SMS content is determined. If the matching degree is greater than or equal to the predetermined matching score but less than the matching threshold, the message is set to be temporarily not blocked and awaits manual review. If the matching degree is less than the predetermined matching score, the message is set to match failure and no manual review is required. If the source of the complaint is the Monetary Authority of Singapore, no manual review is required, and the matching degree between the identified content and the original SMS content is determined. If the matching degree is greater than or equal to the predetermined matching score but less than the matching threshold, the message is set to be temporarily not blocked. If the matching degree is less than the predetermined matching score, the message is set to match failure. The predetermined matching score is less than the matching threshold.

[0088] In S46, the classification tag may contain multiple complaint types, each requiring different handling strategies. Therefore, based on the pre-set rules between the classification tag and the blocking instruction, a corresponding blocking instruction is generated. For example, "Marketing SMS" → blocks the user's number; "Fraudulent SMS" → deletes the message and warns the user; "Collection SMS" → marks the message as a collection message and restricts the sender. The generated blocking instruction is sent to the SMS system's execution module, which completes the specific blocking operation.

[0089] In specific implementations, when the degree of match between the identified content and the original text message exceeds the matching threshold, the platform may also obtain the text message sending time recorded. If this time is before the text message complaint time and is closest to the text message complaint time, the number will be blocked for text messages. However, this embodiment also performs the operation of S50. Specifically, S50 obtains a list of users to whom the original text message content was sent, sorts and filters the users in the user list, and blocks the filtered users according to the blocking instruction.

[0090] In this step, you can obtain all the user information that sent the original SMS content from the database or log system of the SMS service provider to form a user list. The user list usually contains the user's unique identifier (such as mobile phone number, user ID, etc.), sending time, sending status and other related information. Sort the user list according to business needs. For example: sort according to the time sequence of when the user receives the SMS, giving priority to users who have received SMS recently, or sort according to the user's activity (such as login frequency, number of SMS sent, etc.), giving priority to users with higher activity (in some scenarios, you want to block active users first to reduce the scope of impact), or sort according to the user's past complaint history, giving priority to users with complaint records.

[0091] The sorted users can be further filtered, and filtering conditions can be set to narrow the scope of users that need to be blocked. For example: only block users from a specific region, or only block users from a certain operator (such as China Unicom users, etc.). Filtering helps reduce unnecessary blocking operations and improve the efficiency and accuracy of the execution of blocking instructions. Match the filtered user list with the previously generated blocking instructions to determine the blocking operation that should be performed on each user. Record the execution status of the blocking operation, including execution time, executing user, blocking instruction type and other information, for subsequent auditing and tracking, and regularly evaluate the execution effect of the blocking instruction, including indicators such as the number of blocked users, user feedback, and complaint rate. According to the evaluation results, adjust the sorting, filtering and blocking strategies to improve the accuracy and effectiveness of the blocking instructions.

[0092] In one embodiment, if Figure 7 As shown, S50 includes:

[0093] S51. If there is only one user in the obtained user list, directly block the user according to the blocking instruction;

[0094] S52: If there is more than one user in the obtained user list, obtain the historical number of complaints and credit scores of all users, and sort each user according to the historical number of complaints and credit scores;

[0095] S53. Screening users whose number of historical complaints exceeds a preset number and whose credit scores are lower than a score threshold after sorting;

[0096] S54. Through the SMS gateway interface, the SMS sending to the screened users is blocked according to the blocking instruction.

[0097] In S51 , processing is performed for a single user, that is, when the user list contains only one user, there is no need to perform complex sorting and screening, and the blocking instruction is directly executed on the user.

[0098] In S52, processing is performed for multiple users. When the user list contains multiple users, the number of historical complaints and credit scores of each user are obtained from the user database or related system, and the users are sorted according to the number of historical complaints and credit scores.

[0099] In S53, a threshold for the number of complaints is set as a preset number, and users exceeding the preset number are considered high-risk users for complaints; a threshold for the credit score is set as a scoring threshold, and users below the scoring threshold are considered low-credit users. From the sorted user list, users who meet both the "number of historical complaints exceeding the preset number" and "credit score below the scoring threshold" conditions are screened out, and then those users who meet the conditions are blocked according to subsequent blocking instructions. In specific implementation, if the number of historical complaints of all users in the user list does not exceed the preset number, but the credit scores of some users are below the scoring threshold, users with credit scores below the scoring threshold are screened out. Similarly, if the credit scores of all users in the user list are not below the scoring threshold, but the number of historical complaints of some users exceeds the preset number, users with historical complaints exceeding the preset number are screened out. In specific implementation, a comprehensive score (e.g., complaint number weight + credit score weight) can be calculated based on the number of complaints and credit score, and users with the highest sum of the two weights are ranked first, giving priority to these users.

[0100] In S54, by calling the SMS gateway interface, the blocking instruction is used to block SMS sending to the screened user, and SMS will not be sent to the user in the future. The user will not receive subsequent related similar SMS, which improves user satisfaction and reduces subsequent complaints.

[0101] In this embodiment of the present invention, fully automated processing replaces traditional manual operations, eliminating the need for human intervention from data cleaning to result output, significantly shortening the processing cycle. Furthermore, a composite matching model accurately identifies the source of complaints, reducing the rate of false negatives and the workload of ineffective manual review.

[0102] During specific implementation, the embodiment of the present invention also makes use of intelligent tools (such as Doubao and DeepSeek) to automatically generate daily / weekly / monthly reports through intelligent tools, and supports cross-analysis by dimensions such as business units (BUs), suppliers, and communication gateways, and conducts in-depth tracking of high-frequency complaint BUs by refining specific SMS templates and associating responsible persons and affiliated institutions. In addition, intelligent tools can view complaint data according to classification labels, among which, for collection scenarios, separate marking is required. In addition, intelligent tools can also display core modules, which include complaint volume statistics, complaint type distribution, complaint trend analysis, etc. As for SMS templates, for BUs with high complaint volumes and complaint rates, it is necessary to count which SMS templates have had high complaints recently, as well as the applicants and application institutions corresponding to the SMS templates. If the complaints are high, the SMS templates will be improved. Specifically, complaint volume statistics can be displayed in a statistical table, complaint type distribution can be displayed in a pie chart, and complaint trend analysis includes changes in complaint volume and complaint rate. These two changes can be displayed in a change curve chart.

[0103] Intelligent tools can also assist in decision-making. They can identify and uncover root causes, predict short-term complaint fluctuations and future trends, and automatically generate corrective action suggestions (such as locating the root cause of the problem, channel closure prompts, and resource allocation plans). For example, if the root cause is that the tone of a certain SMS template is causing user dissatisfaction, the tone of that SMS template will be improved. If a communication channel has received 8 complaints that month, and has been shut down after reaching 10 complaints per month, a warning prompt will be issued for the channel closure. Furthermore, through rule configuration, if the complaint rate of a communication channel reaches a threshold, channel resources will be automatically switched to other channels. For example, if the number of complaints on the China Mobile channel reaches the threshold, it will automatically switch to China Unicom or China Telecom. For complaints in collection scenarios, sensitive word libraries are extracted to generate guidance on optimized text and SMS content, and the sending strategy is revised in conjunction with new regulations from regulatory authorities. Channel complaint rates are also evaluated monthly, and if a channel's complaint rate reaches a red line threshold, an email alert will be sent to the operations team.

[0104] The embodiment of the present invention further provides a complaint information processing device, which is used to execute any embodiment of the above-mentioned complaint information processing method. Figure 8 , Figure 8 6 is a schematic block diagram of a complaint information processing device provided by an embodiment of the present invention. The complaint information processing device 600 includes:

[0105] A cleaning unit 610 is used to obtain SMS complaint data in real time and clean the SMS complaint data;

[0106] Intent recognition unit 620, configured to perform intent recognition on the cleaned SMS complaint data using an optimized neural network model to obtain recognized content;

[0107] a classification unit 630 for classifying the identified content by complaint type, urgency, and operator source based on a dynamic clustering module, and generating classification labels;

[0108] a matching unit 640 configured to pre-build a composite matching model, input the identified content and the original SMS content into the composite matching model for matching, and generate a blocking instruction based on the classification label if the matching degree between the identified content and the original SMS content exceeds a matching threshold;

[0109] The shielding unit 650 is configured to obtain a list of users to whom the original SMS content is sent, sort and filter the users in the list, and shield the filtered users according to the shielding instruction.

[0110] In one embodiment, the cleaning unit 610 is used to:

[0111] Monitor the SMS channel in real time through the SMS gateway interface and obtain SMS complaint data through the SMS channel;

[0112] Regular expressions are used to replace non-standard punctuation marks in the complaint data with uniform symbols and to remove garbled characters and repeated fields in the SMS complaint data to complete data cleaning.

[0113] In one embodiment, the intention recognition unit 620 is configured to:

[0114] Pre-loading a BERT model as the original neural network model, introducing a GRU layer into the last layer of the BERT model, and adding a fully connected layer and a softmax layer after the GRU layer to obtain an optimized neural network model;

[0115] Configuring training parameters for the optimized neural network model and training the optimized neural network model using pre-collected and labeled training data;

[0116] The cleaned SMS complaint data is segmented and vectorized and then input into the trained optimized neural network model for intent recognition, and the recognized content is output.

[0117] In one embodiment, the classification unit 630 is configured to:

[0118] The DBSCAN density clustering algorithm is used as a dynamic clustering module, the identified content is input into the dynamic clustering module, and the identified content is classified according to the preset complaint type keyword library, urgency determination rules and operator identification information, and corresponding classification labels are generated.

[0119] In one embodiment, the matching unit 640 is configured to:

[0120] The TF-IDF text similarity calculation module and the BiLSTM semantic matching module are constructed into a composite matching model;

[0121] Input the identified content and the original text message content into the TF-IDF text similarity calculation module to calculate the word frequency similarity and obtain a similarity score;

[0122] The similarity score is input into the BiLSTM semantic matching module for deep semantic matching.

[0123] In one embodiment, the matching unit 640 is further configured to:

[0124] Based on the matching results of historical SMS complaint data, the matching threshold is confirmed through cross-validation method;

[0125] When the matching degree between the identified content and the original SMS content exceeds the matching threshold, obtaining a classification label of the current identified content;

[0126] Generate corresponding blocking instructions based on the complaint type in the classification label.

[0127] In one embodiment, the shielding unit 650 is used to:

[0128] If there is only one user in the obtained user list, directly block the user according to the blocking instruction;

[0129] If there is more than one user in the obtained user list, obtain the historical number of complaints and credit scores of all users, and sort each user according to the historical number of complaints and credit scores;

[0130] Screening out users whose number of historical complaints exceeds a preset number and whose credit scores are below the score threshold after sorting;

[0131] Through the SMS gateway interface, SMS sending is blocked for the screened users according to the blocking instructions.

[0132] The specific definition of a complaint information processing device can be found in the definition of a complaint information processing method above and will not be repeated here. The various modules in the above-mentioned complaint information processing device can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0133] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 9 As shown. The computer device includes a processor, memory, network interface and database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a classification method server.

[0134] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 10As shown. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of a classification method.

[0135] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are performed:

[0136] Acquire SMS complaint data in real time and clean the SMS complaint data;

[0137] Use the optimized neural network model to perform intent recognition on the cleaned SMS complaint data and obtain the recognized content;

[0138] Based on the dynamic clustering module, the identified content is classified according to complaint type, urgency and operator source, and classification labels are generated;

[0139] Pre-building a composite matching model, inputting the identified content and the original SMS content into the composite matching model for matching, and generating a blocking instruction based on the classification label if the matching degree between the identified content and the original SMS content exceeds a matching threshold;

[0140] Obtain a user list to which the original SMS content is sent, sort and filter the users in the user list, and block the filtered users according to the blocking instruction.

[0141] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0142] Acquire SMS complaint data in real time and clean the SMS complaint data;

[0143] Use the optimized neural network model to perform intent recognition on the cleaned SMS complaint data and obtain the recognized content;

[0144] Based on the dynamic clustering module, the identified content is classified according to complaint type, urgency and operator source, and classification labels are generated;

[0145] Pre-building a composite matching model, inputting the identified content and the original SMS content into the composite matching model for matching, and generating a blocking instruction based on the classification label if the matching degree between the identified content and the original SMS content exceeds a matching threshold;

[0146] Obtain a user list to which the original SMS content is sent, sort and filter the users in the user list, and block the filtered users according to the blocking instruction.

[0147] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant description in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0148] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0149] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0150] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, or units with the same function may be combined into one unit. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices or units, or may be an electrical, mechanical or other form of connection.

[0151] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected according to actual needs to achieve the objectives of the embodiments of the present invention.

[0152] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0153] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0154] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for processing complaint information, characterized in that: include: Acquire SMS complaint data in real time and clean the SMS complaint data; Use the optimized neural network model to perform intent recognition on the cleaned SMS complaint data and obtain the recognized content; Based on the dynamic clustering module, the identified content is classified according to complaint type, urgency and operator source, and classification labels are generated; Pre-building a composite matching model, inputting the identified content and the original SMS content into the composite matching model for matching, and generating a blocking instruction based on the classification label if the matching degree between the identified content and the original SMS content exceeds a matching threshold; Obtain a user list to which the original SMS content is sent, sort and filter the users in the user list, and block the filtered users according to the blocking instruction.

2. The method for processing complaint information according to claim 1, characterized in that: The real-time acquisition of SMS complaint data and cleaning of the SMS complaint data include: Monitor the SMS channel in real time through the SMS gateway interface and obtain SMS complaint data through the SMS channel; Regular expressions are used to replace non-standard punctuation marks in the complaint data with uniform symbols and to remove garbled characters and repeated fields in the SMS complaint data to complete data cleaning.

3. The method for processing complaint information according to claim 1, wherein: The optimized neural network model is used to perform intent recognition on the cleaned SMS complaint data to obtain the recognized content, including: Pre-loading a BERT model as the original neural network model, introducing a GRU layer into the last layer of the BERT model, and adding a fully connected layer and a softmax layer after the GRU layer to obtain an optimized neural network model; Configuring training parameters for the optimized neural network model and training the optimized neural network model using pre-collected and labeled training data; The cleaned SMS complaint data is segmented and vectorized and then input into the trained optimized neural network model for intent recognition, and the recognized content is output.

4. The method for processing complaint information according to claim 1, wherein: The dynamic clustering module classifies the identified content by complaint type, urgency, and operator source, and generates classification labels, including: The DBSCAN density clustering algorithm is used as a dynamic clustering module, the identified content is input into the dynamic clustering module, and the identified content is classified according to the preset complaint type keyword library, urgency determination rules and operator identification information, and corresponding classification labels are generated.

5. The method for processing complaint information according to claim 1, wherein: The pre-built composite matching model, inputting the identified content and the original SMS content into the composite matching model for matching, includes: The TF-IDF text similarity calculation module and the BiLSTM semantic matching module are constructed into a composite matching model; Input the identified content and the original text message content into the TF-IDF text similarity calculation module to calculate the word frequency similarity and obtain a similarity score; The similarity score is input into the BiLSTM semantic matching module for deep semantic matching.

6. The method for processing complaint information according to claim 1, wherein: If the matching degree between the identified content and the original text message content exceeds a matching threshold, generating a blocking instruction according to the classification label includes: Based on the matching results of historical SMS complaint data, the matching threshold is confirmed through cross-validation method; When the matching degree between the identified content and the original SMS content exceeds the matching threshold, obtaining a classification label of the current identified content; Generate corresponding blocking instructions based on the complaint type in the classification label.

7. The method for processing complaint information according to claim 1, characterized in that: The obtaining of the user list to which the original SMS content is sent, sorting and screening the users in the user list, and screening the screened users according to the screening instruction, includes: If there is only one user in the obtained user list, directly block the user according to the blocking instruction; If there is more than one user in the obtained user list, obtain the historical number of complaints and credit scores of all users, and sort each user according to the historical number of complaints and credit scores; Screening out users whose number of historical complaints exceeds a preset number and whose credit scores are below the score threshold after sorting; Through the SMS gateway interface, SMS sending is blocked for the screened users according to the blocking instructions.

8. A complaint information processing device, characterized in that: include: A cleaning unit, configured to obtain SMS complaint data in real time and clean the SMS complaint data; The intent recognition unit is used to use the optimized neural network model to perform intent recognition on the cleaned SMS complaint data to obtain the recognized content; a classification unit, configured to classify the identified content according to complaint type, urgency, and operator source based on a dynamic clustering module, and generate classification labels; a matching unit configured to pre-build a composite matching model, input the identified content and the original SMS content into the composite matching model for matching, and generate a blocking instruction based on the classification label if the matching degree between the identified content and the original SMS content exceeds a matching threshold; The shielding unit is used to obtain the user list to which the original short message content is sent, sort and filter the users in the user list, and shield the filtered users according to the shielding instruction.

9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for processing complaint information as claimed in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the complaint information processing method according to any one of claims 1 to 7.

Images