Customer complaint problem processing method and device, storage medium and computer equipment

Abstract

The invention discloses a customer complaint problem processing method and device, a storage medium and computer equipment, and relates to the technical field of digital medical treatment and the technical field of financial science and technology. Comprising the steps of obtaining multi-mode customer complaint information fed back by a target customer; inputting the multi-modal customer complaint information into a customer complaint problem classification large model for classification to obtain a target customer complaint problem category to which the customer complaint problem belongs; performing embedded coding on the multi-modal customer complaint information and a plurality of pieces of historical multi-modal customer complaint information under the target customer complaint problem category, correspondingly obtaining a customer complaint feature vector and a first historical customer complaint feature vector, and obtaining a first historical customer complaint feature vector based on the customer complaint feature vector and the first historical customer complaint feature vector; screening out at least one piece of target historical multi-modal customer complaint information matched with the multi-modal customer complaint information; and determining a customer complaint problem processing flow corresponding to the customer complaint problem based on historical customer complaint problem processing flows of the multiple pieces of target historical multi-mode customer complaint information, and calling the customer complaint problem processing flow to process the customer complaint problem of the target customer.

Description

Technical Field

[0001] This invention relates to the fields of digital healthcare technology and financial technology technology, and in particular to a method, apparatus, storage medium and computer equipment for handling customer complaints. Background Technology

[0002] As the company grows, its business logic will become increasingly complex, such as the logic for expense reimbursement and car rental. New configuration items and business processes will also increase accordingly. Customers will also raise numerous complaints while using the services, and these complaints need to be addressed to ensure smooth service for customer support.

[0003] Currently, customer complaints are typically handled manually. However, this manual approach is time-consuming and labor-intensive, and the varying skill levels of staff can lead to errors in handling complaints, thus reducing the customer service experience. Summary of the Invention

[0004] This invention provides a method, apparatus, storage medium, and computer equipment for handling customer complaints, which mainly improves the efficiency and accuracy of handling customer complaints and enhances the customer experience.

[0005] According to a first aspect of the present invention, a method for handling customer complaints is provided, comprising: In response to customer complaint handling signals from target customers using target services, acquire multimodal customer complaint information from the target customers regarding the customer complaints; The multimodal customer complaint information is input into a large customer complaint problem classification model to classify the customer complaint problem and obtain the target customer complaint problem category to which the customer complaint problem belongs. The multimodal customer complaint information and multiple historical multimodal customer complaint information under the target customer complaint problem category are embedded and encoded respectively to obtain a customer complaint feature vector and a first historical customer complaint feature vector. Based on the customer complaint feature vector and the first historical customer complaint feature vector, at least one target historical multimodal customer complaint information that matches the multimodal customer complaint information is selected from the target customer complaint problem category. Based on the historical customer complaint handling process corresponding to multiple target historical multimodal customer complaint information, the customer complaint handling process corresponding to the customer complaint problem is determined, and the customer complaint handling process is invoked to handle the customer complaint problem of the target customer.

[0006] Optionally, before inputting the multimodal customer complaint information into a large-scale customer complaint classification model to classify the customer complaint and obtain the target customer complaint category to which the customer complaint belongs, the method further includes: Obtain historical multimodal customer complaint information and determine the second historical customer complaint feature vector corresponding to the historical multimodal customer complaint information; Initialize the centroid vectors corresponding to multiple clusters; Calculate the distance between the second historical customer complaint feature vector and the centroid vector corresponding to each cluster, and divide the historical multimodal customer complaint information into each cluster based on the distance; Based on the second historical customer complaint feature vector corresponding to the historical multimodal customer complaint information divided into each cluster, the updated centroid vector corresponding to the corresponding cluster is determined. Based on the updated centroid vector, the historical multimodal customer complaint information is reclassified into each cluster until the updated centroid vector remains unchanged. The historical multimodal customer complaint information finally classified into each cluster is then determined as historical multimodal customer complaint information under different customer complaint problem categories. Based on historical multimodal customer complaint information under different customer complaint categories, a customer complaint classification knowledge base is constructed, and the customer complaint classification knowledge base is embedded into a large model to obtain the customer complaint classification large model.

[0007] Optionally, the customer complaint handling process is invoked to handle the customer complaint of the target customer, including: Based on the multimodal customer complaint information, the problem location corresponding to the customer complaint is determined in the business process configuration of the target business; The customer complaint handling process is invoked to repair the business process configuration at the location of the problem.

[0008] Optionally, based on the multimodal customer complaint information, the location of the problem corresponding to the customer complaint is determined in the business process configuration of the target business, including: The source code of the target business is parsed, and an abstract syntax tree of the source code is constructed based on the parsing results; Extract functions and variables from the abstract syntax tree, determine the call relationships between each function and the data flow between each variable, and construct a code heterogeneous graph by using the functions and variables as nodes and the call relationships and data flows as edges. Extract the problem keywords from the multimodal customer complaint information, locate the code node corresponding to the customer complaint problem in the code heterogeneity graph based on the problem keywords, and determine the problem location corresponding to the customer complaint problem based on the location of the code node.

[0009] Optionally, before inputting the multimodal customer complaint information into the large-scale customer complaint problem classification model for customer complaint problem classification, the method further includes: Determine whether there is information to be optimized in the multimodal customer complaint information. If so, determine the information optimization request template corresponding to the information to be optimized, and generate an information optimization request based on the information to be optimized and the information optimization request template. The information to be optimized is at least one of missing information and ambiguous information. The information optimization request is sent to the client terminal where the target customer is located for information optimization interaction, and the information optimization interaction result is received from the client terminal. Based on the information optimization interaction results, the multimodal customer complaint information is optimized to obtain the optimized multimodal customer complaint information; The multimodal customer complaint information is input into a large-scale customer complaint problem classification model for customer complaint problem classification, including: The optimized multimodal customer complaint information is input into the large-scale customer complaint problem classification model for customer complaint problem classification.

[0010] Optionally, after processing the customer complaint issue by invoking the customer complaint handling process, the method further includes: If the customer complaint is not resolved successfully, the type of the unresolved customer complaint will be determined. Based on the problem type, a customer complaint processing terminal is determined, and the customer complaint processing terminal is scheduled to process customer complaints that were not successfully processed. The process of scheduling the customer complaint processing terminal to handle unresolved customer complaints includes: If the customer complaint handling terminal is a product terminal, then a problem handling requirement document is generated based on the multimodal customer complaint information, and the problem handling requirement document is sent to the product terminal for customer complaint handling; If the customer complaint handling terminal is a development terminal, the defect configuration corresponding to the unprocessed customer complaint will be sent to the development terminal for customer complaint handling. If the customer complaint handling terminal is a customer service terminal, then the customer service terminal will schedule multiple terminals to jointly handle the customer complaint.

[0011] Optionally, the multimodal customer complaint information is input into a large-scale customer complaint problem classification model to classify the customer complaint problem, thereby obtaining the target customer complaint problem category to which the customer complaint problem belongs, including: Determine the customer complaint feature vectors corresponding to the multimodal customer complaint information respectively; Feature-level cross processing is performed on the customer complaint feature vectors under multiple modalities to obtain feature cross vectors; element-level cross processing is performed on the customer complaint feature vectors under multiple modalities to obtain element cross vectors; and low-order cross processing is performed on the customer complaint feature vectors under multiple modalities to obtain low-order cross vectors. The feature cross vector, the element cross vector, and the low-order cross vector are combined to obtain the customer complaint cross feature vector. The cross-feature vector of the customer complaint is input into the large-scale customer complaint classification model to classify the customer complaint and obtain the target customer complaint category to which the customer complaint belongs.

[0012] According to a second aspect of the present invention, a customer complaint handling apparatus is provided, comprising: The acquisition unit is used to acquire multimodal customer complaint information from the target customer in response to a customer complaint handling signal when the target customer is using the target service. The classification unit is used to input the multimodal customer complaint information into the large customer complaint problem classification model to classify the customer complaint problem and obtain the target customer complaint problem category to which the customer complaint problem belongs; The matching unit is used to embed and encode the multimodal customer complaint information and multiple historical multimodal customer complaint information under the target customer complaint problem category, respectively, to obtain a customer complaint feature vector and a first historical customer complaint feature vector. Based on the customer complaint feature vector and the first historical customer complaint feature vector, at least one target historical multimodal customer complaint information that matches the multimodal customer complaint information is selected in the target customer complaint problem category. The processing unit is used to determine the customer complaint handling process corresponding to the customer complaint based on the historical customer complaint handling processes corresponding to multiple target historical multimodal customer complaint information, and to call the customer complaint handling process to process the customer complaint of the target customer.

[0013] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described customer complaint handling method.

[0014] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described customer complaint handling method.

[0015] According to the present invention, a customer complaint handling method, apparatus, storage medium, and computer device, compared with the current method of handling customer complaints manually, the present invention obtains multimodal customer complaint information fed back by customers when using the target service, and classifies the customer complaints based on the multimodal customer complaint information using a large model to obtain the target customer complaint category. The multimodal customer complaint information and historical multimodal customer complaint information under the target customer complaint category are encoded. Based on the encoding results, multiple historical multimodal customer complaint information similar to the multimodal customer complaint information are selected from the target customer complaint category. Then, the customer complaint handling process for the current customer complaint is determined based on the historical customer complaint handling processes of multiple historical multimodal customer complaint information. Finally, the customer complaint handling process is used to handle the customer complaint. Thus, by using a large model and embedded encoding, the handling of customer complaints is automatically realized, which can improve the efficiency and accuracy of customer complaint handling. At the same time, by classifying customer complaints in advance, the scope of subsequent customer complaint information matching can be narrowed, thereby improving the efficiency of determining the subsequent customer complaint handling process and reducing computing resources. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart of a customer complaint handling method provided by an embodiment of the present invention is shown; Figure 2 A schematic diagram of a dominance tree structure provided by an embodiment of the present invention is shown; Figure 3 This invention provides a flowchart of another customer complaint handling method according to an embodiment of the invention. Figure 4 This diagram illustrates the structure of a customer complaint handling device according to an embodiment of the present invention. Figure 5 This invention provides a schematic diagram of the structure of another customer complaint handling device according to an embodiment of the invention. Figure 6 A schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0017] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0018] Currently, handling customer complaints manually is time-consuming and labor-intensive. Furthermore, due to the varying skill levels of staff, errors in handling customer complaints can occur, thereby reducing the customer service experience.

[0019] To address the aforementioned problems, embodiments of the present invention provide a method for handling customer complaints, such as... Figure 1 As shown, the method includes: 101. Respond to customer complaint handling signals from target customers when using target services, and obtain multimodal customer complaint information from target customers regarding customer complaint issues.

[0020] The target business can be any business in any scenario, such as system payment, intelligent consultation, car rental, or insurance processing. The business can also be a business within an application. Multimodal customer complaint information includes a description of the problem encountered by customer service representatives when using the target business, screenshots of the problem, and characteristic information such as the customer's ID, age, gender, occupation, and login account for the target business. It should be noted that the above examples are merely illustrative and do not specifically limit the embodiments of this invention.

[0021] Specifically, when a customer complaint handling signal is received, information such as the customer service's description of the complaint, screenshots of the complaint, the user's login account for the target business, and user ID is obtained.

[0022] 102. Input the multimodal customer complaint information into the large customer complaint problem classification model to classify the customer complaint problem and obtain the target customer complaint problem category to which the customer complaint problem belongs.

[0023] In accordance with this embodiment of the invention, to ensure the quality of multimodal customer complaint information, it is necessary to optimize the multimodal customer complaint information. Based on this, the method includes: determining whether there is information to be optimized in the multimodal customer complaint information; if so, determining an information optimization request template corresponding to the information to be optimized, and generating an information optimization request based on the information to be optimized and the information optimization request template, wherein the information to be optimized is at least one of missing information and ambiguous information; sending the information optimization request to the customer terminal where the target customer is located for information optimization interaction, and receiving the information optimization interaction result fed back by the customer terminal; and optimizing the multimodal customer complaint information based on the information optimization interaction result to obtain the optimized multimodal customer complaint information.

[0024] Specifically, determining whether there is information to be optimized in the multimodal customer complaint information, i.e., determining whether there is missing information, ambiguous words, etc., in the multimodal customer complaint information includes: performing word segmentation on the multimodal customer complaint information to obtain individual words; performing part-of-speech tagging on each word; and determining keywords, dependent words, and the dependency relationship between the keywords and dependent words in each word based on each word and its corresponding part of speech; constructing a dominance tree based on the keywords, dependent words, and the dependency relationship between the keywords and dependent words, wherein the keywords are the top-level nodes in the dominance tree, the dependent words are the branch nodes in the dominance tree, and the dependency relationship is the connecting branch between the corresponding top-level nodes and branch nodes; and identifying missing information in the multimodal customer complaint information based on the dominance tree.

[0025] Specifically, firstly, the multimodal customer complaint information is segmented using a word segmentation tool, and then each segmented word is tagged with part-of-speech tags, such as nouns, verbs, and adjectives. Based on each segmented word and its corresponding part-of-speech tagging results, the dominant content words (such as nouns / verbs) in each segmented word are identified as keywords, and the segmented words with modifying components corresponding to the keywords are identified as subordinate words. For example, if the multimodal customer complaint information is "querying car rental information from Xi'an to Shanxi", the word segmentation results are ["query", "from", "Xi'an", "to", "Shanxi", "of", "car rental information"], and the corresponding part-of-speech tagging results are ["verb", "preposition", "place name", "auxiliary word", "place name", "auxiliary word", "noun"]. The identified core word is "query" (verb, governing the entire action), and the dependency words and their dependency relationships with the core word are: "car rental information" (noun, object of the action), "Xi'an" (noun, starting point), "Shanxi" (noun, destination), and "from...to..." (prepositional phrase, indicating spatial relationship). Then, using keywords as top-level nodes, subordinate words as branch nodes, and subordinate relationships as connecting branches, a tree diagram representing the structure of the multimodal customer complaint information is constructed, namely the dominance tree, as shown below. Figure 2The diagram illustrates a visual representation of the dominance tree. Further, a pre-defined dependency structure database matching the business scenario of the multimodal customer complaint information is obtained. This database stores standard dependency structures for that business scenario. The dominance tree is then examined based on these standard dependency structures to identify missing information. For example, if a date parameter exists in the standard dependency structure but is missing from the dominance tree, then date information is determined to be missing from the multimodal customer complaint information. For instance, the multimodal customer complaint "querying car rental information from Xi'an to Shanxi" lacks information specifying which date's car rental information is being queried. This embodiment of the invention identifies missing information in multimodal customer complaint information by constructing a dominance tree. Because the dominance tree structure is clear, simple, and intuitive, it improves the efficiency and accuracy of identifying missing information.

[0026] Furthermore, it is also necessary to identify whether there are ambiguous words in the multimodal customer complaint information. Based on this, the method includes: determining the core words to be calibrated in the multimodal customer complaint information, and constructing sentences for the core words to be calibrated in different scenarios; determining the word vectors of the core words to be calibrated in the sentences of each scenario, and calculating the similarity between each word vector; if the similarity is greater than a preset threshold, the core words to be calibrated are determined to be ambiguous words; otherwise, the core words to be calibrated are determined to be unambiguous words.

[0027] Specifically, the preset threshold is set according to actual needs. First, the multimodal customer complaint information is segmented into individual words. Then, potentially ambiguous core words to be calibrated are initially screened from these segments. For example, a pre-built polysemous word library can be used to identify ambiguous words (core words to be calibrated) in each segment. Then, sentences in different contexts are generated for these core words to be calibrated. For example, if the core word to be calibrated is "Xiaomi," sentences in different contexts are generated as "[Xiaomi] Company releases new products" and "Eating [Xiaomi] porridge is good for the stomach." Then, a context-related vector (word vector) for the core word to be calibrated in each context is generated using a context embedding model. Finally, the cosine similarity between word vectors in different contexts is calculated. If the cosine similarity is greater than the preset threshold, the core word to be calibrated is determined to be an ambiguous word (polysemous word). This embodiment of the invention identifies ambiguous words by constructing sentences of the core word to be calibrated in different contexts, which can quantify the semantic stability of ambiguous words in different contexts and provide a clear basis for judgment. Meanwhile, by identifying ambiguous words in multimodal customer complaint information, this embodiment of the invention avoids misunderstandings of user intent, thereby improving the optimization accuracy of multimodal customer complaint information, ensuring the quality of optimized multimodal customer complaint information, and thus improving the accuracy of handling customer complaints and enhancing user experience.

[0028] Furthermore, if the multimodal customer complaint information contains defects such as missing information or ambiguous words, it is determined that the multimodal customer complaint information needs to be optimized. First, based on the types of defects such as missing information and ambiguous words, the defect types of the multimodal customer complaint information are identified, and the information optimization interaction method with the target customer is determined. Based on the defect type and information optimization interaction method, an information optimization request template is determined, and the defect information is filled into the information optimization request template to obtain the information optimization request. Then, based on the information optimization request, multiple communications are carried out with the target customer to fill in the missing information and modify ambiguous words according to the communication content. This results in the optimized multimodal customer complaint information.

[0029] The defect types include information missing types and ambiguous word types. For example, information missing types include missing key parameters (such as "input parameters?"), and ambiguous types include the presence of polysemous words or vague descriptions (such as unclear time range). Information optimization interaction methods include closed interaction methods, open interaction methods, and suggestive interaction methods. For example, closed interaction methods provide limited options for quick information confirmation, open interaction methods guide users to fill in details, and suggestive interaction methods recommend options based on common scenarios.

[0030] Finally, the optimized multimodal customer complaint information is input into a large-scale customer complaint classification model for customer complaint classification. Based on this, the specific classification method includes: determining the customer complaint feature vectors corresponding to the multimodal customer complaint information; performing feature-level cross processing on the customer complaint feature vectors under multiple modalities to obtain feature cross vectors; performing element-level cross processing on the customer complaint feature vectors under multiple modalities to obtain element cross vectors; performing low-order cross processing on the customer complaint feature vectors under multiple modalities to obtain low-order cross vectors; combining the feature cross vectors, the element cross vectors, and the low-order cross vectors to obtain customer complaint cross feature vectors; and inputting the customer complaint cross feature vectors into the large-scale customer complaint classification model for customer complaint classification to obtain the target customer complaint category to which the customer complaint belongs.

[0031] Specifically, word embedding and other methods are used to determine the customer complaint feature vectors corresponding to the multimodal customer complaint information. Then, to fully utilize the relationships between data, extract more latent features, and simultaneously handle both high-order and low-order data, making data utilization more efficient and the subsequent prediction results more accurate, meeting the needs of practical application scenarios, each customer complaint feature vector needs to be cross-processed. For example, if the first customer complaint feature vector is (a1, a2) and the second customer complaint feature vector is (b1, b2), the specific cross-processing method includes: performing feature-level cross-processing between different customer complaint feature vectors, that is, performing a Hadamard product on all elements of the vectors, followed by a convolution transformation under a certain weight w1, to obtain the feature cross vector f. (w1×(a1×b1,a2×b2)); Simultaneously, element-level cross-vectors are performed on all customer complaint feature vectors. That is, after performing a Hadamard product on each element of the vectors, different weight values ​​w2 and w3 are assigned to the result of each product, and then a linear transformation is performed to obtain the element cross vector f(w2×a1×b1,w3×a2×b2); In addition, all feature vectors are subjected to low-order cross-vector processing, and then the result of the cross-vector processing is assigned a weight coefficient w4, and then a linear transformation is performed to obtain the low-order cross vector f(w4(a1,a2,b1,b2)); Finally, the above feature cross vectors, element cross vectors, and low-order cross vectors are combined, such as by horizontal concatenation, to obtain the customer complaint cross feature vector. It should be noted that the above examples are only illustrative and do not limit the embodiments of this application. Therefore, by cross-processing different multimodal feature vectors, different features can be automatically or explicitly combined to generate new feature combinations. These combined features may contain complex nonlinear relationships between the original features, enabling the model to capture more refined and richer information in the data. In other words, it can make full use of the relationships between various data, extract more latent features, and take into account both high-order and low-order processing, making the data utilization more efficient and the subsequent classification results more accurate, thus meeting the needs of practical application scenarios.

[0032] 103. Embed and encode the multimodal customer complaint information and multiple historical multimodal customer complaint information under the target customer complaint problem category respectively to obtain the customer complaint feature vector and the first historical customer complaint feature vector. Based on the customer complaint feature vector and the first historical customer complaint feature vector, select at least one target historical multimodal customer complaint information that matches the multimodal customer complaint information in the target customer complaint problem category.

[0033] Among them, the target customer complaint category stores a variety of historical multimodal customer complaint information. Each historical multimodal customer complaint information corresponds to a historical customer complaint handling process, and the historical customer complaint handling process is a verified standard handling process.

[0034] In this embodiment of the invention, the input multimodal customer complaint information (including but not limited to text descriptions, user-uploaded images, and voice recordings) is standardized as follows: Text data: Standardized text sequences are generated through operations such as word segmentation, stop word removal, and stemming; Image data: Standardized images are generated through operations such as image cropping, resolution unification, and color space conversion (e.g., RGB to grayscale); Voice data: Voice data is converted into text through speech recognition, and voiceprint features (e.g., Mel-frequency cepstral coefficients, MFCC) are extracted as auxiliary features. Multimodal fusion encoding: The standardized multimodal data is jointly encoded using methods such as word embedding: Text branch: The standardized text is input into language models such as BERT or RoBERTa to extract text semantic feature vectors; Image branch: The standardized images are input into visual models such as ResNet or ViT to extract image visual feature vectors; Voice branch (optional): Voiceprint features are input into 1D-CNN or LSTM networks to extract voice prosody feature vectors; Cross-modal fusion: The text, image, and voice feature vectors are fused into a unified customer complaint feature vector. Similarly, the first historical customer complaint feature vector corresponding to the historical multimodal customer complaint information can be determined in the above manner. Further, based on the customer complaint feature vector and the first historical customer complaint feature vector, the similarity between the multimodal customer complaint information and each historical multimodal customer complaint information is calculated respectively. Then, the historical multimodal customer complaint information with a similarity greater than a preset similarity threshold is selected as the target historical multimodal customer complaint information.

[0035] 104. Based on the historical customer complaint handling process corresponding to multiple target historical multimodal customer complaint information, determine the customer complaint handling process corresponding to the customer complaint issue, and call the customer complaint handling process to handle the customer complaint issue of the target customer.

[0036] The method involves determining the target historical customer complaint handling process corresponding to each target historical multimodal customer complaint information, and integrating these processes to obtain the corresponding customer complaint handling process. Specifically, the integration method involves inputting each target historical customer complaint handling process and multimodal customer complaint information into a preset handling process integration model for integration, thereby obtaining the corresponding customer complaint handling process. To improve the integration accuracy of the preset handling process integration model, it is first necessary to train and construct the preset handling process integration model. Based on this, the method includes: constructing a preset initial handling process integration model; obtaining a sample dataset, wherein the sample dataset includes sample customer complaint information labeled with customer complaint handling processes and multiple sample customer complaint handling processes, where the sample customer complaint handling processes are standard customer complaint handling processes corresponding to customer complaint information similar to the sample customer complaint information; dividing the dataset into a training set and a test set; training the preset initial handling process integration model using the training set; testing the trained preset initial handling process integration model using the test set; and finally, using the trained preset initial handling process integration model that meets the test conditions as the preset handling process integration model.

[0037] Specifically, during model training, a pre-defined initial processing flow integration model is first constructed, followed by the acquisition of a sample dataset. The dataset is ensured to contain all necessary files. The data is converted to a format understandable by the pre-defined initial processing flow integration model, and finally, the model is trained and tested. Specifically, the dataset can be divided first: using random or specific strategies (such as stratified sampling), the sample dataset is divided into a training set and a test set. The training set is then used to train the model, and the test set is used to test the trained model, evaluating its performance on unseen data. Metrics such as mCP, precision, and recall on the test set are calculated and recorded. If the model performance does not meet requirements, it can return to the training phase for further iterations or adjustments. This yields a pre-defined processing flow integration model that meets the requirements. Finally, the customer complaint handling process output by the pre-defined processing flow integration model is called to handle customer complaints from target customers. This embodiment of the invention determines the current customer complaint handling process by analyzing the processing flow of historical customer complaint information. This saves the time and resources wasted on manually rewriting code to build the handling process, and also avoids the risk of human error. Therefore, this embodiment of the invention can improve the efficiency and accuracy of customer complaint handling.

[0038] According to the present invention, a customer complaint handling method, compared with the current method of handling customer complaints manually, obtains multimodal customer complaint information from customer service representatives when using the target service. Based on this multimodal customer complaint information, a large model is used to classify the customer complaints, obtaining a target customer complaint category. The multimodal customer complaint information and historical multimodal customer complaint information under the target customer complaint category are encoded. The encoding results are used to filter out multiple historical multimodal customer complaint information similar to the target customer complaint category. Then, the customer complaint handling process for the current customer complaint is determined based on the historical customer complaint handling processes of these multiple historical multimodal customer complaint information. Finally, this customer complaint handling process is used to handle the customer complaint. Thus, by using a large model and embedded encoding, the handling of customer complaints is automatically achieved, improving the efficiency and accuracy of customer complaint handling. Simultaneously, by pre-classifying customer complaints, the scope of subsequent customer complaint information matching is narrowed, thereby improving the efficiency of determining the subsequent customer complaint handling process and reducing computational resources.

[0039] Furthermore, to better illustrate the process of handling customer complaints described above, and as a refinement and extension of the above embodiments, this invention provides another method for handling customer complaints, such as... Figure 3 As shown, the method includes: 201. Respond to customer complaint handling signals from target customers when using target services, and obtain multimodal customer complaint information from target customers regarding customer complaint issues.

[0040] Specifically, when a customer complaint handling signal is received, multimodal customer complaint information is obtained.

[0041] 202. Input the multimodal customer complaint information into the large customer complaint problem classification model to classify the customer complaint problem and obtain the target customer complaint problem category to which the customer complaint problem belongs.

[0042] In this embodiment of the invention, to improve the classification accuracy of the large-scale customer complaint classification model, it is first necessary to construct the large-scale customer complaint classification model. Based on this, the method includes: acquiring historical multimodal customer complaint information and determining the second historical customer complaint feature vector corresponding to the historical multimodal customer complaint information; initializing the centroid vectors corresponding to multiple clusters; calculating the distance between the second historical customer complaint feature vector and the centroid vector corresponding to each cluster, and based on the distance, dividing the historical multimodal customer complaint information into each cluster; and based on the historical multimodal customer complaint information divided into each cluster... The second historical customer complaint feature vector is used to determine the updated centroid vector corresponding to the corresponding cluster. Based on the updated centroid vector, the historical multimodal customer complaint information is reclassified into each cluster until the updated centroid vector does not change. The historical multimodal customer complaint information finally classified into each cluster is determined as historical multimodal customer complaint information under different customer complaint problem categories. Based on the historical multimodal customer complaint information under different customer complaint problem categories, a customer complaint problem classification knowledge base is constructed, and the customer complaint problem classification knowledge base is embedded into the large model to obtain the customer complaint problem classification large model.

[0043] Historical multimodal customer complaint information refers to the problem description text, screenshots, and voice information corresponding to historical customer complaint issues. Specifically, word embedding and other methods are used to determine the second historical customer complaint feature vector corresponding to each piece of historical multimodal customer complaint information. K clusters are pre-determined, and the centroid vectors corresponding to the centroids of the K clusters are determined. In the first round, based on the distance between the second historical customer complaint feature vector and the centroid vector of each cluster, the historical multimodal customer complaint information corresponding to the second historical customer complaint feature vector is first divided into the nearest cluster, thus obtaining the historical multimodal customer complaint information under each cluster in the first round. In the second round, the centroid vector of each cluster is recalculated based on the historical multimodal customer complaint information under each cluster. This centroid vector is then used as the new centroid vector. The distance between each second historical customer complaint feature vector and the new centroid vector corresponding to each cluster is recalculated, and the historical multimodal customer complaint information corresponding to the second historical customer complaint feature vector is assigned to the nearest cluster. This process is repeated for multiple rounds. After each round, it is determined whether the centroid vector in each cluster has changed. If it has changed, the division continues; otherwise, it stops. The historical multimodal customer complaint information in each cluster after the final round of division is identified as historical multimodal customer complaint information under different customer complaint problem categories. A customer complaint problem classification knowledge base is constructed from the historical multimodal customer complaint information under different customer complaint problem categories, and this knowledge base is embedded into the large model to obtain the large customer complaint problem classification model. Thus, after inputting multimodal customer complaint information into the large model, the model can classify the information based on the distance between the historical multimodal customer complaint information in the customer complaint problem classification knowledge base and the current multimodal customer complaint information.

[0044] 203. Embed and encode the multimodal customer complaint information and multiple historical multimodal customer complaint information under the target customer complaint problem category respectively to obtain the customer complaint feature vector and the first historical customer complaint feature vector. Based on the customer complaint feature vector and the first historical customer complaint feature vector, select at least one target historical multimodal customer complaint information that matches the multimodal customer complaint information in the target customer complaint problem category.

[0045] Specifically, under the target customer complaint category, find multiple historical multimodal customer complaint information that are similar to the multimodal customer complaint information, and finally determine the customer complaint handling process for this customer complaint based on the historical customer complaint handling process corresponding to the multiple historical multimodal customer complaint information.

[0046] 204. Based on the historical customer complaint handling process corresponding to multiple target historical multimodal customer complaint information, determine the customer complaint handling process corresponding to the customer complaint issue.

[0047] Specifically, by integrating the historical customer complaint handling processes corresponding to multiple target historical multimodal customer complaint information, a customer complaint handling process corresponding to the customer complaint issue is obtained.

[0048] 205. Based on multimodal customer complaint information, determine the location of the customer complaint in the business process configuration of the target business.

[0049] In this embodiment of the invention, handling customer complaints requires modifying the business process configuration, i.e., modifying the code in the business process. To improve the efficiency of handling customer complaints, it is first necessary to find the problem location corresponding to the customer complaint, i.e., find the defective code location corresponding to the target business. Based on this, step 205 specifically includes: parsing the source code of the target business and constructing an abstract syntax tree of the source code based on the parsing results; extracting functions and variables from the abstract syntax tree, and determining the call relationship between each function and the data flow between each variable, using the functions and variables as nodes, and the call relationship and the data flow as edges to construct a code heterogeneous graph; extracting problem keywords from the multimodal customer complaint information, locating the code node corresponding to the customer complaint in the code heterogeneous graph based on the problem keywords, and determining the problem location corresponding to the customer complaint based on the location of the code node.

[0050] Specifically, the source code of the target business system (such as an e-commerce order module) is standardized, such as by unifying the file encoding format (e.g., UTF-8) and removing comments and blank lines. The code language type (e.g., Java / Python / C++) is identified, and the corresponding language parser is invoked to generate an abstract syntax tree (AST) for the source code, ensuring the integrity of the syntactic structure. For example, the Java code `public class OrderService` will generate a tree structure containing nodes such as `ClassDeclaration` and `Modifier`. Illegal nodes are filtered through syntax rule validation (e.g., bracket matching, keyword usage). The AST is stored as an intermediate representation in JSON format, and a unique ID is assigned to each node. A node index table is built, recording the node type, file path, and line number. Functions and variables are extracted from the AST. Call relationships, such as calling and called functions, are determined within functions. Edges are defined pointing from variable definition to usage. Functions and variables are treated as nodes, and call relationships between functions and data flows between variables are treated as edges. For example, if `orderService.submitOrder()` appears in `main()`, an edge `main() → submitOrder()` is added. A heterogeneous code graph is then constructed based on the nodes and edges. Furthermore, customer complaint keywords are extracted from the multimodal customer complaint information, such as high-frequency words. Nodes matching these keywords are then searched in the code heterogeneity graph. If a keyword uniquely matches a node, the node's position is directly returned, and the corresponding code location is used as the problem location for the customer complaint. If a keyword matches multiple nodes, a graph traversal algorithm (such as breadth-first search) is used to analyze node relationships and determine the problem location based on these relationships. For example, the keyword "order" matches `createOrder()` and `logOrder()`, but `createOrder()` calls `logOrder()`, and the customer complaint description includes "not recorded." Therefore, `logOrder()` is prioritized, and its corresponding location is used as the problem location for the customer complaint.

[0051] 206. Call the customer complaint handling process to repair the business process configuration at the location of the problem.

[0052] Specifically, after determining the location of the problem, the customer complaint handling process is invoked to repair the business process configuration at the problem location. After repair, it is necessary to check whether the processing was successful. If successful, a processing report is generated and sent to the display terminal for display. If unsuccessful, further processing is required. Based on this, the method includes: if the customer complaint is not successfully processed, determining the problem type of the unsuccessfully processed customer complaint; determining the customer complaint handling terminal based on the problem type, and scheduling the customer complaint handling terminal to process the unsuccessfully processed customer complaint; wherein, scheduling the customer complaint handling terminal to process the unsuccessfully processed customer complaint includes: if the customer complaint handling terminal is a product terminal, generating a problem handling requirement document based on the multimodal customer complaint information, and sending the problem handling requirement document to the product terminal for customer complaint processing; if the customer complaint handling terminal is a development terminal, sending the defect configuration corresponding to the unsuccessfully processed customer complaint to the development terminal for customer complaint processing; if the customer complaint handling terminal is a customer service terminal, scheduling multiple terminals to jointly process the customer complaint through the customer service terminal.

[0053] The problem types include new requirement resolution, development fix, and other problem types. Specifically, if the problem type is a new requirement resolution, the corresponding customer complaint handling terminal is the product terminal; if it's a development fix, the corresponding terminal is the development terminal; and if it's another problem type, the terminal is the customer service terminal. Specifically, if it's the product terminal, a new requirement resolution is needed, automatically connecting to the release system, generating a requirement document for the problem, and submitting the document to the product team for fix implementation. If it's the development terminal, development fixes are needed, automatically submitting defects, creating new fix iterations, and simultaneously having developers modify and release them. If it's the customer service terminal, it requires intervention from customer service and the person responsible for the problem to jointly resolve the issue. This embodiment of the invention personalizes the handling of unresolved problems by problem type, improving problem handling efficiency and accuracy.

[0054] According to another customer complaint handling method provided by this invention, compared with the current method of handling customer complaints manually, this invention obtains multimodal customer complaint information from customer service representatives when using the target service, and classifies the customer complaints based on the multimodal customer complaint information using a large model to obtain the target customer complaint category. It then encodes the multimodal customer complaint information and historical multimodal customer complaint information under the target customer complaint category. Based on the encoding results, it filters out multiple historical multimodal customer complaint information similar to the multimodal customer complaint information in the target customer complaint category. Subsequently, it determines the customer complaint handling process for the current customer complaint based on the historical customer complaint handling processes of these multiple historical multimodal customer complaint information, and finally uses this customer complaint handling process to handle the customer complaint. Therefore, by using a large model and embedded encoding, it automatically handles customer complaints, improving the efficiency and accuracy of customer complaint handling. At the same time, by pre-classifying customer complaints, it can narrow down the scope of subsequent customer complaint information matching, thereby improving the efficiency of determining the subsequent customer complaint handling process and reducing computational resources.

[0055] Furthermore, as Figure 1 In specific implementation, embodiments of the present invention provide a customer complaint handling device, such as... Figure 4 As shown, the device includes: an acquisition unit 31, a classification unit 32, a matching unit 33, and a processing unit 34.

[0056] The acquisition unit 31 can be used to acquire multimodal customer complaint information from the target customer in response to a customer complaint handling signal when the target customer is using the target service.

[0057] The classification unit 32 can be used to input the multimodal customer complaint information into the large customer complaint problem classification model to classify the customer complaint problem and obtain the target customer complaint problem category to which the customer complaint problem belongs.

[0058] The matching unit 33 can be used to embed and encode the multimodal customer complaint information and multiple historical multimodal customer complaint information under the target customer complaint problem category, respectively, to obtain a customer complaint feature vector and a first historical customer complaint feature vector. Based on the customer complaint feature vector and the first historical customer complaint feature vector, at least one target historical multimodal customer complaint information that matches the multimodal customer complaint information is selected in the target customer complaint problem category.

[0059] The processing unit 34 can be used to determine the customer complaint handling process corresponding to the customer complaint based on the historical customer complaint handling process corresponding to multiple target historical multimodal customer complaint information, and call the customer complaint handling process to handle the customer complaint of the target customer.

[0060] In specific application scenarios, in order to build a large-scale model for classifying customer complaints, such as Figure 5 As shown, the device also includes a construction unit 35.

[0061] The construction unit 35 can be used to acquire historical multimodal customer complaint information and determine the second historical customer complaint feature vector corresponding to the historical multimodal customer complaint information; initialize the centroid vectors corresponding to multiple clusters; calculate the distance between the second historical customer complaint feature vector and the centroid vector corresponding to each cluster, and classify the historical multimodal customer complaint information into each cluster based on the distance; determine the updated centroid vector corresponding to the historical multimodal customer complaint information in each cluster based on the second historical customer complaint feature vector corresponding to the historical multimodal customer complaint information in each cluster; classify the historical multimodal customer complaint information into each cluster again based on the updated centroid vector until the updated centroid vector does not change, and determine the historical multimodal customer complaint information finally classified into each cluster as historical multimodal customer complaint information under different customer complaint problem categories; construct a customer complaint problem classification knowledge base based on the historical multimodal customer complaint information under different customer complaint problem categories, and embed the customer complaint problem classification knowledge base into the large model to obtain the customer complaint problem classification large model.

[0062] In specific application scenarios, in order to handle customer complaints from target customers, the processing unit 34 includes a first determining module 341 and a processing module 342.

[0063] The first determining module 341 can be used to determine the problem location corresponding to the customer complaint in the business process configuration of the target business based on the multimodal customer complaint information.

[0064] The processing module 342 can be used to call the customer complaint handling process to repair the business process configuration at the location of the problem.

[0065] In specific application scenarios, in order to determine the location of the problem corresponding to the customer complaint, the first determining module 341 can be used to parse the source code of the target business and construct an abstract syntax tree of the source code based on the parsing results; extract functions and variables from the abstract syntax tree, and determine the call relationship between each function and the data flow between each variable; construct a code heterogeneous graph by using the functions and variables as nodes and the call relationship and the data flow as edges; extract problem keywords from the multimodal customer complaint information, locate the code node corresponding to the customer complaint in the code heterogeneous graph based on the problem keywords, and determine the location of the problem corresponding to the customer complaint based on the location of the code node.

[0066] In specific application scenarios, in order to optimize multimodal customer complaint information, the device also includes an optimization unit 36.

[0067] The optimization unit 36 ​​can be used to determine whether there is information to be optimized in the multimodal customer complaint information. If so, it determines an information optimization request template corresponding to the information to be optimized, and generates an information optimization request based on the information to be optimized and the information optimization request template, wherein the information to be optimized is at least one of missing information and ambiguous information; it sends the information optimization request to the customer terminal where the target customer is located for information optimization interaction, and receives the information optimization interaction result fed back by the customer terminal; it optimizes the multimodal customer complaint information based on the information optimization interaction result to obtain the optimized multimodal customer complaint information.

[0068] In specific application scenarios, in order to classify customer complaints, the classification unit 32 can also be used to input the optimized multimodal customer complaint information into the large customer complaint classification model for customer complaint classification.

[0069] In specific application scenarios, in order to continue processing unresolved customer complaints, the processing unit 34 can also be used to determine the type of the unresolved customer complaint if it has not been successfully processed; determine the customer complaint processing terminal based on the problem type, and schedule the customer complaint processing terminal to process the unresolved customer complaint; wherein, scheduling the customer complaint processing terminal to process the unresolved customer complaint includes: if the customer complaint processing terminal is a product terminal, generating a problem processing requirement document based on the multimodal customer complaint information, and sending the problem processing requirement document to the product terminal for customer complaint processing; if the customer complaint processing terminal is a development terminal, sending the defect configuration corresponding to the unresolved customer complaint to the development terminal for customer complaint processing; if the customer complaint processing terminal is a customer service terminal, scheduling multiple terminals to jointly process the customer complaint through the customer service terminal.

[0070] In specific application scenarios, in order to classify customer complaints, the classification unit 32 includes a second determination module 321, a cross-processing module 322, a combination module 323, and a classification module 324.

[0071] The second determining module 321 can be used to determine the customer complaint feature vectors corresponding to the multimodal customer complaint information respectively.

[0072] The cross-processing module 322 can be used to perform feature-level cross-processing on customer complaint feature vectors under multiple modalities to obtain feature cross vectors, perform element-level cross-processing on customer complaint feature vectors under multiple modalities to obtain element cross vectors, and perform low-order cross-processing on customer complaint feature vectors under multiple modalities to obtain low-order cross vectors.

[0073] The combination module 323 can be used to combine the feature cross vector, the element cross vector, and the low-order cross vector to obtain the customer complaint cross feature vector.

[0074] The classification module 324 can be used to input the customer complaint cross feature vector into the customer complaint problem classification model to classify the customer complaint problem and obtain the target customer complaint problem category to which the customer complaint problem belongs.

[0075] It should be noted that other corresponding descriptions of the functional modules involved in the customer complaint handling device provided in this embodiment of the invention can be found in [reference needed]. Figure 1 The corresponding description of the method shown will not be repeated here.

[0076] Based on the above, Figure 1 Accordingly, this embodiment of the invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps: responding to a customer complaint handling signal from a target customer using a target service, acquiring multimodal customer complaint information from the target customer regarding the complaint; inputting the multimodal customer complaint information into a large-scale customer complaint classification model for complaint classification to obtain the target customer complaint category to which the complaint belongs; embedding and encoding the multimodal customer complaint information and multiple historical multimodal customer complaint information under the target customer complaint category to obtain a customer complaint feature vector and a first historical customer complaint feature vector; based on the customer complaint feature vector and the first historical customer complaint feature vector, selecting at least one target historical multimodal customer complaint information that matches the multimodal customer complaint information from the target customer complaint category; determining the customer complaint handling process corresponding to the complaint based on the historical customer complaint handling processes corresponding to the multiple target historical multimodal customer complaint information, and calling the customer complaint handling process to handle the customer complaint of the target customer.

[0077] Based on the above, Figure 1 The method shown and as Figure 4 The embodiment of the device shown in the invention also provides a physical structure diagram of a computer device, such as... Figure 6As shown, the computer device includes: a processor 41, a memory 42, and a computer program stored in the memory 42 and executable on the processor. Both the memory 42 and the processor 41 are mounted on a bus 43. When the processor 41 executes the program, it performs the following steps: in response to a customer complaint handling signal from a target customer using a target service, it acquires multimodal customer complaint information from the target customer regarding the complaint; it inputs the multimodal customer complaint information into a large customer complaint classification model to classify the customer complaint and obtain the target customer complaint category to which the complaint belongs; and it further classifies the multimodal customer complaint information... Customer complaint information and multiple historical multimodal customer complaint information under the target customer complaint issue category are embedded and encoded to obtain a customer complaint feature vector and a first historical customer complaint feature vector. Based on the customer complaint feature vector and the first historical customer complaint feature vector, at least one target historical multimodal customer complaint information matching the multimodal customer complaint information is selected from the target customer complaint issue category. Based on the historical customer complaint issue handling processes corresponding to multiple target historical multimodal customer complaint information, the customer complaint issue handling process corresponding to the customer complaint issue is determined, and the customer complaint issue handling process is invoked to handle the customer complaint issue of the target customer.

[0078] The present invention obtains multimodal customer complaint information from customer service representatives when using a target service. Based on this information, a large model is used to classify customer complaints, resulting in a target customer complaint category. The multimodal customer complaint information and historical multimodal customer complaint information under the target category are encoded. The encoding results are used to filter out multiple historical multimodal customer complaint information entries similar to the target customer complaint category. Then, the current customer complaint handling process is determined based on the historical customer complaint handling processes of these multiple historical multimodal customer complaint information entries. Finally, this customer complaint handling process is used to process the customer complaint. This automatic processing of customer complaints through large models and embedded encoding improves processing efficiency and accuracy. Furthermore, pre-classifying customer complaints narrows the scope of subsequent customer complaint information matching, thereby improving the efficiency of determining the subsequent customer complaint handling process and reducing computational resources.

[0079] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0080] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for handling customer complaints, characterized in that, include: In response to customer complaint handling signals from target customers using target services, acquire multimodal customer complaint information from the target customers regarding the customer complaints; The multimodal customer complaint information is input into a large customer complaint problem classification model to classify the customer complaint problem and obtain the target customer complaint problem category to which the customer complaint problem belongs. The multimodal customer complaint information and multiple historical multimodal customer complaint information under the target customer complaint problem category are embedded and encoded respectively to obtain a customer complaint feature vector and a first historical customer complaint feature vector. Based on the customer complaint feature vector and the first historical customer complaint feature vector, at least one target historical multimodal customer complaint information that matches the multimodal customer complaint information is selected from the target customer complaint problem category. Based on the historical customer complaint handling process corresponding to multiple target historical multimodal customer complaint information, the customer complaint handling process corresponding to the customer complaint problem is determined, and the customer complaint handling process is invoked to handle the customer complaint problem of the target customer.

2. The method according to claim 1, characterized in that, Before inputting the multimodal customer complaint information into a large-scale customer complaint classification model to classify the customer complaint and obtain the target customer complaint category to which the customer complaint belongs, the method further includes: Obtain historical multimodal customer complaint information and determine the second historical customer complaint feature vector corresponding to the historical multimodal customer complaint information; Initialize the centroid vectors corresponding to multiple clusters; Calculate the distance between the second historical customer complaint feature vector and the centroid vector corresponding to each cluster, and divide the historical multimodal customer complaint information into each cluster based on the distance; Based on the second historical customer complaint feature vector corresponding to the historical multimodal customer complaint information divided into each cluster, the updated centroid vector corresponding to the corresponding cluster is determined. Based on the updated centroid vector, the historical multimodal customer complaint information is reclassified into each cluster until the updated centroid vector remains unchanged. The historical multimodal customer complaint information finally classified into each cluster is then determined as historical multimodal customer complaint information under different customer complaint problem categories. Based on historical multimodal customer complaint information under different customer complaint categories, a customer complaint classification knowledge base is constructed, and the customer complaint classification knowledge base is embedded into a large model to obtain the customer complaint classification large model.

3. The method according to claim 1, characterized in that, The customer complaint handling process is invoked to handle the customer complaint of the target customer, including: Based on the multimodal customer complaint information, the problem location corresponding to the customer complaint is determined in the business process configuration of the target business; The customer complaint handling process is invoked to repair the business process configuration at the location of the problem.

4. The method according to claim 3, characterized in that, Based on the multimodal customer complaint information, the location of the problem corresponding to the customer complaint is determined in the business process configuration of the target business, including: The source code of the target business is parsed, and an abstract syntax tree of the source code is constructed based on the parsing results; Extract functions and variables from the abstract syntax tree, determine the call relationships between each function and the data flow between each variable, and construct a code heterogeneous graph by using the functions and variables as nodes and the call relationships and data flows as edges. Extract the problem keywords from the multimodal customer complaint information, locate the code node corresponding to the customer complaint problem in the code heterogeneity graph based on the problem keywords, and determine the problem location corresponding to the customer complaint problem based on the location of the code node.

5. The method according to claim 1, characterized in that, Before inputting the multimodal customer complaint information into the large-scale customer complaint classification model for customer complaint classification, the method further includes: Determine whether there is information to be optimized in the multimodal customer complaint information. If so, determine the information optimization request template corresponding to the information to be optimized, and generate an information optimization request based on the information to be optimized and the information optimization request template. The information to be optimized is at least one of missing information and ambiguous information. The information optimization request is sent to the client terminal where the target customer is located for information optimization interaction, and the information optimization interaction result is received from the client terminal. Based on the information optimization interaction results, the multimodal customer complaint information is optimized to obtain the optimized multimodal customer complaint information; The multimodal customer complaint information is input into a large-scale customer complaint problem classification model for customer complaint problem classification, including: The optimized multimodal customer complaint information is input into the large-scale customer complaint problem classification model for customer complaint problem classification.

6. The method according to claim 1, characterized in that, After processing the customer complaint from the target customer by invoking the customer complaint handling process, the method further includes: If the customer complaint is not resolved successfully, the type of the unresolved customer complaint will be determined. Based on the problem type, a customer complaint processing terminal is determined, and the customer complaint processing terminal is scheduled to process customer complaints that were not successfully processed. The process of scheduling the customer complaint processing terminal to handle unresolved customer complaints includes: If the customer complaint handling terminal is a product terminal, then a problem handling requirement document is generated based on the multimodal customer complaint information, and the problem handling requirement document is sent to the product terminal for customer complaint handling; If the customer complaint handling terminal is a development terminal, the defect configuration corresponding to the unprocessed customer complaint will be sent to the development terminal for customer complaint handling. If the customer complaint handling terminal is a customer service terminal, then the customer service terminal will schedule multiple terminals to jointly handle the customer complaint.

7. The method according to claim 1, characterized in that, The multimodal customer complaint information is input into a large-scale customer complaint classification model to classify the customer complaint issues, thereby obtaining the target customer complaint category to which the customer complaint issue belongs, including: Determine the customer complaint feature vectors corresponding to the multimodal customer complaint information respectively; Feature-level cross processing is performed on the customer complaint feature vectors under multiple modalities to obtain feature cross vectors; element-level cross processing is performed on the customer complaint feature vectors under multiple modalities to obtain element cross vectors; and low-order cross processing is performed on the customer complaint feature vectors under multiple modalities to obtain low-order cross vectors. The feature cross vector, the element cross vector, and the low-order cross vector are combined to obtain the customer complaint cross feature vector. The cross-feature vector of the customer complaint is input into the large-scale customer complaint classification model to classify the customer complaint and obtain the target customer complaint category to which the customer complaint belongs.

8. A customer complaint handling device, characterized in that, include: The acquisition unit is used to acquire multimodal customer complaint information from the target customer in response to a customer complaint handling signal when the target customer is using the target service. The classification unit is used to input the multimodal customer complaint information into the large customer complaint problem classification model to classify the customer complaint problem and obtain the target customer complaint problem category to which the customer complaint problem belongs; The matching unit is used to embed and encode the multimodal customer complaint information and multiple historical multimodal customer complaint information under the target customer complaint problem category, respectively, to obtain a customer complaint feature vector and a first historical customer complaint feature vector. Based on the customer complaint feature vector and the first historical customer complaint feature vector, at least one target historical multimodal customer complaint information that matches the multimodal customer complaint information is selected in the target customer complaint problem category. The processing unit is used to determine the customer complaint handling process corresponding to the customer complaint based on the historical customer complaint handling processes corresponding to multiple target historical multimodal customer complaint information, and to call the customer complaint handling process to process the customer complaint of the target customer.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images