Complaint short message intelligent matching method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of finance and medical treatment, and discloses a complaint short message intelligent matching method and device, computer equipment and a storage medium, and the method comprises the steps: receiving a complaint short message; screening out a candidate short message set related to the complaint short message from a short message complaint platform according to the complaint short message content; extracting key features of complaint content from the complaint short message, and extracting key features of each candidate short message from the candidate short message set; performing feature matching on the complaint content key feature and each candidate short message key feature to identify an original short message record corresponding to the complaint short message; and displaying the original short message record. Through automatic screening and intelligent feature matching, automatic identification of the original short message record corresponding to the complaint short message is realized, the bottleneck of information fault caused by telephone number desensitization is especially broken through, and compared with a traditional mode depending on manual screening, the short message complaint processing efficiency and accuracy are greatly improved.

Description

Technical Field

[0001] The present invention relates to the fields of finance and medical technology, and more specifically to a method, device, computer equipment and storage medium for intelligent matching of complaint text messages. Background Art

[0002] In industries such as finance and healthcare, users filing complaints through SMS channels is a common way of providing service feedback. In recent years, with the development of medical informatization, there has been an increasing number of patients filing medical service complaints through SMS channels (such as failed registrations, delayed drug delivery, incorrect diagnosis and treatment information, etc.). Such complaints require quick location of the corresponding original SMS sending records (such as hospital notifications, drug delivery reminders, etc.) so that medical institutions can trace the problems and respond in a timely manner. However, whether in traditional industries or the medical field, existing technical solutions have the following pain points:

[0003] First, because mobile phone numbers are masked, they cannot be directly linked to specific SMS sending records, creating an information silo between complaint information and the actual sending behavior. Second, incomplete information makes it difficult for complaint handlers to quickly locate the source of the complaint, increasing the workload of manual investigation and reducing processing efficiency. Third, the inability to trace specific SMS sending behavior results in a lack of necessary data support for complaint processing, which affects the accurate identification and resolution of the problem. Fourth, the lack of complete mobile phone number information makes it impossible to provide effective basis for complaints, hindering the appeal process and affecting the fairness and accuracy of complaint handling.

[0004] The above problems have seriously affected the efficiency and accuracy of complaint handling, and also caused great trouble to the appeal process and subsequent management. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method, device, computer equipment and storage medium for intelligent matching of complaint text messages, aiming to solve the problems of low efficiency and low accuracy in processing user text message complaints.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for intelligently matching complaint SMS messages, comprising:

[0008] Receive complaint SMS messages;

[0009] Filter out candidate SMS sets related to the complaint SMS from the SMS complaint platform based on the content of the complaint SMS;

[0010] Extracting key features of the complaint content from the complaint SMS, and extracting key features of each candidate SMS from the candidate SMS set;

[0011] Match the key features of the complaint content with the key features of each candidate SMS message to identify the original SMS record corresponding to the complaint SMS message;

[0012] Displays the original SMS records.

[0013] Furthermore, the method of screening out a set of candidate SMS messages related to the complaint SMS message from the SMS complaint platform based on the content of the complaint SMS message includes:

[0014] Extract the masked mobile phone number and the SMS complaint platform sending number from the complaint SMS content;

[0015] The masked mobile phone number, the SMS complaint platform sending number and the set query time range are used as query conditions to filter out candidate SMS that meet the requirements from the SMS complaint platform.

[0016] Furthermore, extracting the key features of the complaint content from the complaint SMS and extracting the key features of each candidate SMS from the candidate SMS set include:

[0017] Use the pre-trained NLP model to pre-process the complaint SMS and candidate SMS respectively;

[0018] The pre-processed text is hierarchically extracted with a pre-trained language model;

[0019] Generate key features of complaint content and key features of each candidate SMS message based on semantic features.

[0020] Furthermore, before generating the complaint content key features and each candidate SMS key features based on the semantic features, the following steps are also included:

[0021] The attention mechanism is used to assign weights to the extracted semantic features.

[0022] Furthermore, before generating the complaint content key features and each candidate SMS key features based on the semantic features, the following steps are also included:

[0023] The extracted semantic features are weighted according to the custom rule base.

[0024] Furthermore, the feature matching of the complaint content key features with the key features of each candidate SMS to identify the original SMS record corresponding to the complaint SMS includes:

[0025] The dual-tower model is used to calculate the multi-dimensional similarity between the key features of the complaint content and the key features of each candidate SMS message;

[0026] The multi-dimensional similarity scores are integrated to obtain the comprehensive scores of the complaint SMS and each candidate SMS;

[0027] The candidate SMS with the highest comprehensive score is used as the final matching result of the complaint SMS.

[0028] Furthermore, before displaying the original SMS record, the process further includes:

[0029] Anonymize the personal information in the original SMS records.

[0030] In a second aspect, the present invention also provides a complaint SMS intelligent matching device, comprising a receiving unit, a screening unit, a key feature extraction unit, a feature matching unit, and a display unit;

[0031] The receiving unit is used to receive complaint text messages;

[0032] The screening unit is used to screen out a set of candidate SMS messages related to the complaint SMS message from the SMS complaint platform based on the content of the complaint SMS message;

[0033] The key feature extraction unit is used to extract the key features of the complaint content from the complaint SMS, and to extract the key features of each candidate SMS from the candidate SMS set;

[0034] The feature matching unit is used to perform feature matching on the key features of the complaint content with the key features of each candidate SMS message to identify the original SMS record corresponding to the complaint SMS message;

[0035] The display unit is used to display the original SMS records.

[0036] In a third aspect, the present invention also 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 above-mentioned intelligent matching method for complaint SMS when executing the computer program.

[0037] In a fourth aspect, the present invention also provides a computer-readable storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the above-mentioned intelligent matching method for complaint SMS.

[0038] The beneficial effects of the present invention compared to the prior art are as follows: the present invention provides an intelligent matching method for complaint SMS messages, comprising: receiving a complaint SMS message; screening a set of candidate SMS messages related to the complaint SMS message from an SMS complaint platform based on the content of the complaint SMS message; extracting key features of the complaint content from the complaint SMS message, and extracting key features of each candidate SMS message from the set of candidate SMS messages; performing feature matching on the key features of the complaint content with the key features of each candidate SMS message to identify the original SMS message record corresponding to the complaint SMS message; and displaying the original SMS message record. The present invention achieves automated identification of the original SMS message record corresponding to the complaint SMS message through automated screening and intelligent feature matching, particularly breaking through the information gap bottleneck caused by phone number desensitization. Compared with the traditional model that relies on manual screening, the efficiency and accuracy of SMS complaint processing are greatly improved.

[0039] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following preferred embodiments are specifically cited and described in detail as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] 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.

[0041] Figure 1 A schematic diagram of the application environment of the intelligent matching method for complaint SMS provided by a specific embodiment of the present invention;

[0042] Figure 2 The process of the intelligent matching method for complaint SMS provided by the specific embodiment of the present invention Figure 1 ;

[0043] Figure 3 The process of the intelligent matching method for complaint SMS provided by the specific embodiment of the present invention Figure 2 ;

[0044] Figure 4 Schematic diagram of the intelligent matching device for complaint SMS provided by a specific embodiment of the present invention Figure 1 ;

[0045] Figure 5 Schematic diagram of the intelligent matching device for complaint SMS provided by a specific embodiment of the present invention Figure 2 ;

[0046] Figure 6 A schematic block diagram of a computer device provided in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0047] 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.

[0048] 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.

[0049] It should also be understood that the terms used in the present specification are only for the purpose of describing particular embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0050] 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.

[0051] Figure 1 This figure illustrates the application environment of the intelligent SMS complaint matching method presented in this invention. This invention can be applied to SMS complaint processing systems in fields such as healthcare and finance. This system, deployed on an enterprise or institution's backend server, forms a complete closed-loop complaint processing loop with network elements such as SMS gateways, user terminals, and service databases. Its core function is to automatically match complaint SMS messages with historical records, improving processing efficiency and accuracy.

[0052] In the application environment, there are users and servers. When the server receives the content of a complaint SMS, it filters out a set of candidate SMS related to the complaint SMS from the SMS complaint platform based on the content; extracts key features of the complaint content from the complaint SMS, and extracts key features of each candidate SMS from the candidate SMS set; performs feature matching between the key features of the complaint content and the key features of each candidate SMS to identify the original SMS record corresponding to the complaint SMS; and displays the original SMS record.

[0053] It should be noted that, in actual operation, the server may include multiple servers, such as a server cluster or a cloud server.

[0054] Figure 2 This is a flowchart of the method for intelligently matching complaint SMS messages provided by a specific embodiment of the present invention, such as Figure 2 As shown, the method includes the following steps: S10-S50.

[0055] S10. Receive complaint SMS.

[0056] Taking the SMS complaint processing system of Ping An Bank Credit Card as an example, the complaint SMS receiving module is deployed on the cloud server and can support multi-channel access. For example, the user complaint SMS received is:

[0057] Dear Mr. / Ms. Zhai, your Ping An Bank credit card is overdue, and your default risk level has increased. To avoid card suspension and the bank requiring a lump-sum payment of the entire outstanding balance, please promptly repay the minimum payment of 915.82 yuan. For more information, please call 02138952580. If payment has already been made, please disregard it.

[0058] S20. Filter out a set of candidate SMS messages related to the complaint SMS message from the SMS complaint platform based on the content of the complaint SMS message.

[0059] In one embodiment, step S20 specifically includes the following steps: S201 - S202 .

[0060] S201. Extract the masked mobile phone number and the SMS complaint platform sending number from the complaint SMS content.

[0061] Using character matching, we can identify masked mobile phone numbers in complaint SMS messages, such as "123******45." We can also use string matching to identify SMS complaint platform sending numbers (SP numbers), such as "10690000" and "95511," from the SMS sender field or SMS text content.

[0062] S202: Using the masked mobile phone number, the SMS complaint platform sending number, and the set query time range as query conditions, filter out candidate SMS messages that meet the requirements from the SMS complaint platform.

[0063] The query time range can be set to the SMS complaint platform sent in the past month, with this month as the query time interval. Of course, it can support custom settings based on actual needs, such as setting the time range to 15 days or 20 days.

[0064] By matching the receiving number, sending number, and sending time fields in the database table, we can filter out SMS records that meet the requirements. For example, we can filter out 1,000 SMS messages that meet the requirements from 100,000 SMS messages in the past month. These 1,000 SMS messages together constitute the candidate SMS set.

[0065] This screening mechanism can reduce the data pressure for subsequent processing and improve the overall processing efficiency.

[0066] S30. Extract the key features of the complaint content from the complaint text messages, and extract the key features of each candidate text message from the candidate text message set.

[0067] In one embodiment, step S30 specifically includes the following steps: S301 - S303.

[0068] S301. Use a pre - trained NLP model to perform text pre - processing on the complaint text messages and each candidate text message respectively.

[0069] In this step, an NLP tool is selected to perform basic processing on the text messages to eliminate the interference of non - key information. Specifically, it includes format standardization, word segmentation, and stop - word filtering.

[0070] Format standardization: Unify full - width / semi - width punctuation marks (such as converting “,” to “,”), and remove invalid characters such as HTML tags and emojis.

[0071] Word segmentation: Use the Jieba word - segmentation tool to split the text into word units. For example, split “

Ping An Bank

Ping An Bank

[0072] Stop - word filtering: By customizing a stop - word list (including words without actual semantic meaning such as “of”, “already”, “in”, etc.), remove the interfering words and retain the core words.

[0073] Through pre - processing, the original text is converted into a structured vocabulary sequence, reducing data noise, enabling subsequent semantic analysis to focus on key content, improving the accuracy of feature extraction, for example, reducing misjudgments caused by emojis or garbled codes, thereby increasing the effective information extraction rate.

[0074] S302. Extract semantic features from the pre - processed text through a pre - trained language model layer by layer.

[0075] In this step, a pre - trained language model (such as BERT) is used for feature extraction. Its multi - layer network based on the Transformer architecture can achieve step - by - step parsing from basic grammar to high - order semantics:

[0076] [[ID=​​Deep network semantic extraction: High-level networks (8-12 layers) use self-attention to capture long-range semantic dependencies and understand the overall meaning of the text. For example, they can extract the core meaning of "abnormal amount" from "spending 1,000 yuan but displaying 2,000 yuan."

[0078] Take the Ping An Bank credit card complaint text message "My credit card was stolen and 5,000 yuan was spent overseas in the early morning" as an example. After processing it with the BERT model:

[0079] Shallow extraction: "credit card" (noun), "stolen card" (verb phrase), "overseas consumption" (noun phrase);

[0080] Deep extraction: The semantic label "credit card fraud" is integrated and a 768-dimensional semantic vector is generated to represent the text feature.

[0081] Through hierarchical feature extraction, the model can not only capture basic grammatical structures but also understand complex semantic relationships. Compared with traditional single-level analysis, it improves the accurate recognition rate of complaint semantics, especially when dealing with implicit expressions (such as "I don't understand this deduction" → abnormal funds).

[0082] S303: Generate key features of the complaint content and key features of each candidate SMS message based on the semantic features.

[0083] Entity Recognition (NER): Utilizes pre-trained financial domain NER models to extract entity information such as names, amounts, and dates. For example, from the sentence "Zhang San's Ping An Bank credit card generated an unidentified deduction of 100 yuan on December 1, 2024," we can extract "Zhang San" (name), "100 yuan" (amount), and "December 1, 2024" (date).

[0084] Keyword extraction: Combining the TF-IDF algorithm and the attention mechanism weight, we select high-frequency and important words as keywords. For example, words such as "fraudulent credit card" and "abnormal deduction" in complaint text messages are included due to their high weight.

[0085] Semantic label generation: Complaint types are annotated using classification models (such as TextCNN fine-tuned with BERT). For example, "credit card fraud" → security risk category, "incorrect bill amount" → fee dispute category.

[0086] The generation of structured key features converts text information into a machine-processable format, facilitating subsequent multi-dimensional matching with complaint SMS messages. For example, with Ping An Bank receiving millions of SMS messages per day, the efficiency of candidate SMS screening can be improved, significantly shortening the complaint handling response time.

[0087] In one embodiment, before step S303, the process further includes step S3025.

[0088] S3025. Use the attention mechanism to assign weights to the extracted semantic features.

[0089] After obtaining the semantic features extracted by a pre-trained language model (such as BERT), this step introduces the self-attention mechanism. By dynamically calculating the importance of each word in the text, it assigns differentiated weights to the semantic features, thereby highlighting key information and suppressing redundant content.

[0090] Specifically, the semantic feature vector (e.g., 768 dimensions) output by BERT is used as input, and the query vector, key vector, and value vector are generated respectively through three learnable weight matrices. The degree of association between each word and other words is calculated through dot product operations to obtain an attention score matrix. In order to avoid the gradient disappearing due to excessive values, a scaling factor is introduced. The attention score is normalized using the Softmax function to generate a weight value in the range of 0-1, which represents the importance of each word in the semantic expression. The higher the weight, the greater the contribution of the word to the overall semantics. The weight matrix is ​​multiplied by the value vector to output the weighted semantic feature vector to complete the reinforcement of key information.

[0091] For example, the SMS complaint handling system for Ping An Bank Credit Cards might contain a user complaint text message: "Ping An Bank Credit Card, I only spent 500 yuan, but the bill shows 1,000 yuan. This is an unreasonable charge!" After weighting, the final output semantic feature vector emphasizes key semantics such as "unreasonable charges" and "abnormal spending amount," while de-weighting general information such as the bank name.

[0092] Through dynamic weight allocation, the model prioritizes the core issues of complaints (such as "random charges" and "fraudulent credit card fraud"), avoiding interference in matching results due to irrelevant words (such as modal particles and bank names), and improving the accuracy of key information recognition; at the same time, it captures long-distance dependencies between words (such as the comparison between "consumption amount" and "bill amount"), making the model's understanding of complex semantics (such as "actual consumption does not match the bill") more accurate and reducing misjudgments.

[0093] In one embodiment, before step S303, the process further includes step S3026.

[0094] S3026. Modify the weights of the extracted semantic features according to the custom rule base.

[0095] This step further optimizes feature weights based on semantic features extracted by the pre-trained language model and combined with a custom rule library. It aims to make up for the model's insufficient understanding of industry-specific terminology and business logic and strengthen domain-related features.

[0096] Specifically, a custom rule base is constructed, and its rule form is: using a key-value pair (Key-Value) structure, where the "key" is an industry term or business keyword, and the "value" contains a weight correction coefficient and an applicable scenario label. At the same time, it supports manual or automatic updating of the rule base to adapt to business changes (such as the launch of new services, the transfer of complaint hotspots). Semantic feature matching and weight correction are specifically as follows: traverse the vocabulary in the semantic features (such as keywords extracted by BERT), search for matching items in the rule base, and if the match is successful, adjust the weight of the corresponding feature according to the weight correction coefficient in the rule base. For example, the weight correction coefficient of "credit card fraud" in the rule base is 0.3. When the keyword is included in the semantic feature, its original weight is multiplied by 1.3; in addition, the scenario label in the rule base (such as "security risk") is attached to the semantic feature to assist in subsequent matching and classification. Of course, the adjusted weights are normalized to ensure that the sum of all weights is 1 to avoid numerical deviation caused by weight accumulation.

[0097] By injecting industry-specific terms (such as "minimum repayment amount" and "overpayment") into the rule base, the problem of insufficient understanding of domain vocabulary by the pre-trained model is solved, and the accuracy of complaint feature recognition is further improved.

[0098] S40: Match the key features of the complaint content with the key features of each candidate SMS message to identify the original SMS record corresponding to the complaint SMS message.

[0099] In one embodiment, step S40 specifically includes the following steps: S401 - S403 .

[0100] S401. Calculate the multi-dimensional similarity between the key features of the complaint content and the key features of each candidate SMS message through the twin-tower model.

[0101] This step uses the Siamese Network model to encode the key features of the complaint SMS and candidate SMS respectively, and calculates the similarity from multiple dimensions such as semantics, time, business attributes, and structure.

[0102] Specifically, the dual-tower model architecture includes the left tower (the complaint SMS tower): which inputs the key feature vectors of complaint SMS messages, including semantic features (such as the 768-dimensional vector extracted by BERT), entity features (name, amount, time), and classification labels (complaint type). The right tower (the candidate SMS tower): which inputs the corresponding features of candidate SMS messages, including SMS template ID, sending time, business line (BU), semantic vectors, and scenario labels annotated by the rule library. A shared parameter layer: The two towers share some parameters at the bottom layer (such as the BERT embedding layer) to ensure feature space alignment.

[0103] Multi-dimensional similarity calculations include semantic similarity, temporal matching, business attribute matching, and structural similarity. Semantic similarity uses cosine similarity to calculate the semantic vector distance between the complaint and the candidate SMS. Temporal matching scores the time interval between the sending time and the complaint time. Business attribute matching compares attributes such as the business line (BU), SMS template ID, and scenario tag. For example, 5 points are awarded for each successful match. Structural similarity: Structural features such as SMS signatures and the number of variable placeholders are examined. For example, 2 points are awarded for each successful match.

[0104] S402: Integrate the multi-dimensional similarity scores to obtain a comprehensive score of the complaint SMS and each candidate SMS.

[0105] This step combines the multi-dimensional similarity scores into a single comprehensive score through a weighted summation method. The formula is: Comprehensive score = α × semantic similarity + β × temporal match + γ × business attribute match + δ × structural similarity. α, β, γ, and δ are adjustable weight parameters set according to business needs (for example, in financial scenarios, semantic similarity has a higher weight). During implementation, the scores of each dimension are first normalized to ensure a uniform numerical range, and then the weighted calculation is performed.

[0106] S403: The candidate SMS with the highest comprehensive score is used as the final matching result of the complaint SMS.

[0107] This step sorts all candidate SMS messages by their comprehensive scores in descending order and selects the message with the highest score as the original record to match the complaint message. If multiple candidate SMS messages have the same score, the one with the most recent time and the closest matching business attributes is prioritized.

[0108] For example, since the comprehensive score of candidate SMS 1 (0.84) is higher than that of candidate SMS 2 (0.45), candidate SMS 1 is determined to be the original SMS record corresponding to the complaint SMS.

[0109] For steps S401-S403, multi-dimensional fusion is used to avoid misjudgment based on a single feature (such as relying solely on time or keywords). A dual-tower model is used to parallelly calculate feature vectors. This significantly improves complaint handling efficiency compared to traditional sentence-by-sentence comparison.

[0110] S50: Display the original SMS record.

[0111] This step displays the successfully matched original SMS records to the user or customer service staff through a visual interface.

[0112] like Figure 3 As shown, in one embodiment, step S50 further includes the following steps: S45.

[0113] S45. Anonymize the personal information in the original SMS record.

[0114] Automatically anonymize sensitive personal information in successfully matched original SMS records to ensure that data display meets compliance standards.

[0115] Specifically, masking processing: retain the first and last parts of sensitive information, and replace the middle part with "*", for example: mobile phone number "13812345678" → "138********78"; replacement processing: replace the name with a common title, such as "Zhang San" → "Mr. So-and-so", or retain the first character + "Mr. / Ms." (such as "Mr. Zhang**").

[0116] In summary: The present invention realizes the automatic identification of the original SMS records corresponding to the complaint SMS through automatic screening and intelligent feature matching, especially breaking through the information gap bottleneck caused by telephone number desensitization. Compared with the traditional mode relying on manual screening, it greatly improves the efficiency and accuracy of SMS complaint processing.

[0117] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0118] The embodiment of the present invention also provides a complaint SMS intelligent matching device, which is used to perform the steps in any embodiment of the above complaint SMS intelligent matching method. Figure 4 , Figure 4 A schematic block diagram of the complaint SMS intelligent matching device 100 provided in an embodiment of the present application is shown. The complaint SMS intelligent matching device 100 specifically includes a receiving unit 110, a screening unit 120, a key feature extraction unit 130, a feature matching unit 140 and a display unit 150.

[0119] The receiving unit 110 is used to receive complaint SMS messages; the screening unit 120 is used to screen out a set of candidate SMS messages related to the complaint SMS messages from the SMS complaint platform based on the content of the complaint SMS messages; the key feature extraction unit 130 is used to extract key features of the complaint content from the complaint SMS messages, and to extract key features of each candidate SMS message from the candidate SMS message set; the feature matching unit 140 is used to perform feature matching between the key features of the complaint content and the key features of each candidate SMS message to identify the original SMS record corresponding to the complaint SMS message; the display unit 150 is used to display the original SMS record.

[0120] like Figure 5 As shown, in one embodiment, the complaint SMS intelligent matching device 100 further includes an anonymization processing unit 145 for anonymizing the personal information in the original SMS record.

[0121] In one embodiment, the screening unit 120 is specifically used to: extract the masked mobile phone number and the SMS complaint platform sending number from the complaint SMS content; use the masked mobile phone number, the SMS complaint platform sending number and the set query time range as query conditions to screen out candidate SMS that meet the requirements from the SMS complaint platform.

[0122] In one embodiment, the key feature extraction unit 130 is specifically used to: use a pre-trained NLP model to perform text preprocessing on the complaint text message and each candidate text message respectively; extract semantic features from the pre-processed text in layers through a pre-trained language model; and generate key features of the complaint content and key features of each candidate text message based on the semantic features.

[0123] In one embodiment, the key feature extraction unit 130 is further specifically configured to: use an attention mechanism to assign weights to the extracted semantic features.

[0124] In one embodiment, the key feature extraction unit 130 is further specifically configured to perform weight correction on the extracted semantic features according to a custom rule base.

[0125] In one embodiment, the feature matching unit 140 is specifically used to: calculate the multi-dimensional similarity between the key features of the complaint content and the key features of each candidate SMS through a dual-tower model; integrate the multi-dimensional similarity scores to obtain the comprehensive score of the complaint SMS and each candidate SMS; and use the candidate SMS with the highest comprehensive score as the final matching result of the complaint SMS.

[0126] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned complaint SMS intelligent matching device 100 and each unit can refer to the corresponding description in the aforementioned method embodiment. For the convenience and brevity of the description, it will not be repeated here.

[0127] The above complaint SMS intelligent matching device can be implemented in the form of a computer program, which can be used in Figure 6 Runs on the computer device shown.

[0128] See also Figure 6 , Figure 6 7 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 700 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0129] like Figure 6 As shown, the computer device includes 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 steps of the above-mentioned method for intelligent matching of complaint SMS are implemented.

[0130] The computer device 700 includes a processor 720 , a memory, and a network interface 750 connected via a system bus 710 , wherein the memory may include a non-volatile storage medium 730 and an internal memory 740 .

[0131] The non-volatile storage medium 730 can store an operating system 731 and a computer program 732. When the computer program 732 is executed, the processor 720 can execute the complaint SMS intelligent matching method.

[0132] The processor 720 is used to provide computing and control capabilities and support the operation of the entire computer device 700.

[0133] The internal memory 740 provides an environment for the operation of the computer program 732 in the non-volatile storage medium 730. When the computer program 732 is executed by the processor 720, the processor 720 can execute the complaint SMS intelligent matching method.

[0134] The network interface 750 is used for network communication, such as sending assigned tasks, etc. It will be understood by those skilled in the art that Figure 6 The structure shown in the figure is merely a block diagram of a portion of the structure related to the present application solution and does not limit the computer device 700 to which the present application solution is applied. The specific computer device 700 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement. The processor 720 is used to execute program code stored in the memory to implement the complaint SMS intelligent matching method.

[0135] Those skilled in the art will understand that Figure 6 The embodiment of the computer device shown in the figure does not constitute a limitation on the specific composition of the computer device. In other embodiments, the computer device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. For example, in some embodiments, the computer device may only include a memory and a processor. In such an embodiment, the structure and function of the memory and processor are the same as those in the figure. Figure 6 The embodiments shown are consistent and will not be described again here.

[0136] It should be understood that in the embodiment of the present application, the processor 720 may be a central processing unit (CPU), and the processor 720 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0137] In another embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium may be a non-volatile computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the complaint SMS intelligent matching method disclosed in an embodiment of the present invention.

[0138] 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.

[0139] 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 device, 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.

[0140] 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.

[0141] 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.

[0142] 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 perform 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.

[0143] 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. The intelligent matching method for complaint SMS is characterized by: include: Receive complaint SMS messages; Filter out candidate SMS sets related to the complaint SMS from the SMS complaint platform based on the content of the complaint SMS; Extracting key features of the complaint content from the complaint SMS, and extracting key features of each candidate SMS from the candidate SMS set; Match the key features of the complaint content with the key features of each candidate SMS message to identify the original SMS record corresponding to the complaint SMS message; Displays the original SMS records.

2. The intelligent matching method for complaint SMS according to claim 1, characterized in that: The candidate SMS set related to the complaint SMS is screened from the SMS complaint platform based on the content of the complaint SMS, including: Extract the masked mobile phone number and the SMS complaint platform sending number from the complaint SMS content; The masked mobile phone number, the SMS complaint platform sending number and the set query time range are used as query conditions to filter out candidate SMS that meet the requirements from the SMS complaint platform.

3. The intelligent matching method for complaint SMS according to claim 1, characterized in that: The step of extracting the key features of the complaint content from the complaint SMS and extracting the key features of each candidate SMS from the candidate SMS set includes: Use the pre-trained NLP model to pre-process the complaint SMS and candidate SMS respectively; The pre-processed text is hierarchically extracted with a pre-trained language model; Generate key features of complaint content and key features of each candidate SMS message based on semantic features.

4. The intelligent matching method for complaint SMS according to claim 3, characterized in that: Before generating the complaint content key features and each candidate SMS key features based on the semantic features, the following steps are also included: The attention mechanism is used to assign weights to the extracted semantic features.

5. The intelligent matching method for complaint SMS according to claim 3, characterized in that: Before generating the complaint content key features and each candidate SMS key features based on the semantic features, the following steps are also included: The extracted semantic features are weighted according to the custom rule base.

6. The intelligent matching method for complaint SMS according to claim 1, characterized in that: The feature matching of the complaint content key features with the key features of each candidate SMS to identify the original SMS record corresponding to the complaint SMS includes: The dual-tower model is used to calculate the multi-dimensional similarity between the key features of the complaint content and the key features of each candidate SMS message; The multi-dimensional similarity scores are integrated to obtain the comprehensive scores of the complaint SMS and each candidate SMS; The candidate SMS with the highest comprehensive score is used as the final matching result of the complaint SMS.

7. The method for intelligently matching complaint SMS messages according to claim 1, characterized in that: Before displaying the original SMS records, the method further includes: Anonymize the personal information in the original SMS records.

8. Complaint SMS intelligent matching device, characterized by: It includes a receiving unit, a screening unit, a key feature extraction unit, a feature matching unit and a display unit; The receiving unit is used to receive complaint text messages; The screening unit is used to screen out a set of candidate SMS messages related to the complaint SMS message from the SMS complaint platform based on the content of the complaint SMS message; The key feature extraction unit is used to extract the key features of the complaint content from the complaint SMS, and to extract the key features of each candidate SMS from the candidate SMS set; The feature matching unit is used to perform feature matching on the key features of the complaint content with the key features of each candidate SMS message to identify the original SMS record corresponding to the complaint SMS message; The display unit is used to display the original SMS records.

9. A computer device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the complaint SMS intelligent matching method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The storage medium stores a computer program, which includes program instructions. When the program instructions are executed by the processor, the processor executes the complaint SMS intelligent matching method according to any one of claims 1 to 7.

Images