Customer feedback information classification method, device and system and medium

Abstract

The invention discloses a client feedback information classification method, apparatus and system, and a medium. The method comprises the steps of obtaining to-be-classified client feedback information; extracting structured information and unstructured information in the client feedback information according to a preset data type; performing data fusion processing on the structured information and the unstructured information; and inputting the fused data into the trained text classification model, and outputting a corresponding scene appeal classification result. Structured information and non-structured information in client feedback information are extracted and fused, fused data are classified by using a text classification model, and non-structured semantics and structured semantics can be fused, so that the model pays more attention to context information, and the user experience is improved. Structured and unstructured factors in the client feedback information can be efficiently utilized, rapid and accurate client feedback classification is achieved, client feedback classification efficiency is improved, and improvement of appeal response speed is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of financial technology, and in particular to a method, device, system and medium for classifying customer feedback information. Background Art

[0002] At present, with the development of banking business, the continuous expansion of customer base and the diversification of customer service touchpoints, banks have more and more channels to receive customer feedback, such as social media, email, online chat, telephone hotlines and other interactive channels, which generate a large amount of customer feedback and problem reports, including suggestions, consultations, complaints, etc. Timely processing of feedback information can improve customer satisfaction, reduce customer complaint rates, timely discover product defects, and improve customer experience.

[0003] The traditional manual method of classifying these feedbacks is not only time-consuming and laborious, but also difficult to respond promptly and accurately to complex multi-dimensional demands. Staff are prone to omissions or misclassifications when processing large amounts of information, which reduces the efficiency and accuracy of classifying customer feedback information, thereby affecting the speed of response to demands. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the object of the present invention is to provide a method, device, system and medium for classifying customer feedback information that can be applied to financial technology or other related fields. Its main purpose is to improve the classification efficiency and accuracy of customer feedback information to increase the response speed of demands.

[0005] The technical solution of the present invention is as follows:

[0006] A first aspect of the present invention provides a method for classifying customer feedback information, comprising:

[0007] Obtain customer feedback information to be classified;

[0008] Extracting structured information and unstructured information from the customer feedback information according to preset data types;

[0009] Performing data fusion processing on the structured information and the unstructured information;

[0010] Input the fused data into the trained text classification model and output the corresponding scene demand classification results.

[0011] In one embodiment, extracting structured information and unstructured information from the customer feedback information according to a preset data type includes:

[0012] Extracting the feedback source, page entry, feedback case link and / or feedback case number from the customer feedback information as the structured information;

[0013] The conversation text in the customer feedback information is extracted as the unstructured information.

[0014] In one embodiment, the data fusion processing of the structured information and the unstructured information specifically includes:

[0015] The data values ​​of the feedback source, page entry, feedback case link and / or feedback case number are connected with the data value of the dialogue text through a separator to generate a corresponding fusion description text.

[0016] In one embodiment, the fused data is input into a trained text classification model to output corresponding scene demand classification results, including:

[0017] Inputting the fused description text into a trained text classification model;

[0018] Performing vector encoding on the fused description text by using an encoder in the text classification model to generate a fused vector;

[0019] Performing classification prediction on the fusion vector through the fully connected layer in the text classification model, and outputting the probability that the fusion vector belongs to each preset appeal category;

[0020] The category with the highest probability is taken as the scene appeal classification result of the fused description text.

[0021] In one embodiment, after inputting the fused data into a trained text classification model and outputting the corresponding scene demand classification result, the method further includes:

[0022] Collect statistics on the classification results of scene demands within a specified time period;

[0023] Confirm the processing priority of different scenario demand classifications based on statistical results.

[0024] In one embodiment, after inputting the fused data into a trained text classification model and outputting the corresponding scene demand classification result, the method further includes:

[0025] Get the processing priority of all unprocessed scene requests;

[0026] Based on the processing priorities of all unprocessed scenario requests, a dynamically sorted request processing plan is generated and pushed to the client.

[0027] In one embodiment, the text classification model is a RoBERTa model.

[0028] A second aspect of the present invention provides a classification device for customer feedback information, comprising:

[0029] An acquisition module, used to acquire customer feedback information to be classified;

[0030] An extraction module, used to extract structured information and unstructured information from the customer feedback information according to preset data types;

[0031] A fusion module, used for performing data fusion processing on the structured information and the unstructured information;

[0032] The classification module is used to input the fused data into the trained text classification model and output the corresponding scene demand classification results.

[0033] A third aspect of the present invention provides a customer feedback information classification system, the system comprising at least one processor; and

[0034] a memory communicatively connected to the at least one processor; wherein,

[0035] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned customer feedback information classification method.

[0036] A fourth aspect of the present invention provides a non-volatile computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the one or more processors can execute the above-mentioned customer feedback information classification method.

[0037] Beneficial effects: The present invention discloses a classification method, device, system and medium for customer feedback information. Compared with the prior art, the embodiments of the present invention obtain customer feedback information to be classified; extract structured information and unstructured information in the customer feedback information according to preset data types; perform data fusion processing on the structured information and unstructured information; input the fused data into a trained text classification model, and output the corresponding scene demand classification results. By extracting structured information and unstructured information from customer feedback information and fusing them, and using a text classification model to classify the fused data, unstructured semantics can be integrated with structured semantics, allowing the model to pay more attention to contextual information, and can efficiently utilize structured and unstructured factors in customer feedback information to achieve fast and accurate customer feedback classification, which improves the efficiency of customer feedback classification while also helping to improve the speed of demand response. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0039] Figure 1A flow chart of a method for classifying customer feedback information provided by an embodiment of the present invention;

[0040] Figure 2 A flow chart of step S102 in the method for classifying customer feedback information provided by an embodiment of the present invention;

[0041] Figure 3 A flow chart of step S104 in the method for classifying customer feedback information provided by an embodiment of the present invention;

[0042] Figure 4 Another flow chart of the method for classifying customer feedback information provided by an embodiment of the present invention;

[0043] Figure 5 Another flow chart of the method for classifying customer feedback information provided by an embodiment of the present invention;

[0044] Figure 6 A schematic diagram of the functional modules of a device for classifying customer feedback information provided by an embodiment of the present invention;

[0045] Figure 7 A schematic diagram of the hardware structure of a customer feedback information classification system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. The embodiments of the present invention are described below in conjunction with the accompanying drawings.

[0047] At present, with the development of banking business, the continuous expansion of customer base and the diversification of customer service touchpoints, banks have more and more channels to receive customer feedback, such as social media, email, online chat, telephone hotlines and other interactive channels, which generate a large amount of customer feedback and problem reports, including suggestions, consultations, complaints, etc. Timely processing of feedback information can improve customer satisfaction, reduce customer complaint rates, timely discover product defects, and improve customer experience.

[0048] The traditional manual method of classifying these feedbacks is not only time-consuming and laborious, but also difficult to respond promptly and accurately to complex multi-dimensional demands. Staff are prone to omissions or misclassifications when processing large amounts of information, which reduces the efficiency and accuracy of classifying customer feedback information, thereby affecting the speed of response to demands.

[0049] In order to solve the above problems, the present invention proposes a classification method for customer feedback information. Figure 1 , Figure 1The flowchart of an embodiment of the method for classifying customer feedback information provided by the present invention. The system architecture of the method for classifying customer feedback information provided by this embodiment may include a first terminal device, a second terminal device, a third terminal device, a network and a server. The network is used to provide a medium for a communication link between the first terminal device, the second terminal device, the third terminal device and the server. The network may include various connection types, such as wired and / or wireless communication links, etc.

[0050] The user may use the first terminal device, the second terminal device, and the third terminal device to interact with the server through the network to receive or send messages, etc. Various communication client applications may be installed on the first terminal device, the second terminal device, and the third terminal device, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, etc. (only as examples).

[0051] The first terminal device, the second terminal device, and the third terminal device may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0052] The server may be a server that provides various services, such as a background management server that provides support for the content browsed by the user using the first terminal device, the second terminal device, and the third terminal device (for example only). The background management server may analyze and process the received data such as the user request, and feed back the processing results (such as web pages, information, or data obtained or generated according to the user request) to the terminal device. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server for a distributed system, or a server combined with a blockchain.

[0053] It should be noted that the classification of customer feedback information provided by the embodiment of the present invention can generally be performed by the first terminal device, the second terminal device or the third terminal device. Correspondingly, the classification of customer feedback information provided by the embodiment of the present invention can also be set in the first terminal device, the second terminal device or the third terminal device.

[0054] Alternatively, the method for classifying customer feedback information provided by the embodiment of the present invention may also be generally executed by a server. Accordingly, the device for classifying customer feedback information provided by the embodiment of the present invention may generally be disposed in a server. The method for classifying customer feedback information provided by the embodiment of the present invention may also be executed by a server or server cluster that is different from the server and that can communicate with the first terminal device, the second terminal device, the third terminal device and / or the server. Accordingly, the device for classifying customer feedback information provided by the embodiment of the present invention may also be disposed in a server or server cluster that is different from the server and that can communicate with the first terminal device, the second terminal device, the third terminal device and / or the server.

[0055] It should be understood that the numbers of the above terminal devices, networks and servers are only illustrative. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0056] like Figure 1 As shown, the classification method of customer feedback information provided in this embodiment specifically includes the following steps:

[0057] S101. Obtain customer feedback information to be classified.

[0058] In this embodiment, customers can raise their demands through multiple channels, and the customer service system can collect the corresponding customer feedback information to be classified so as to automatically classify the demands and process them. Specifically, it can be through customer service hotline channels, where customers directly contact customer service by phone to raise inquiries or feedback; social media channels, where customers leave messages or send private messages to the company's social media accounts to feedback problems; email channels, where customers send detailed feedback or complaints via email; online form channels, where customers fill out online feedback forms on the company's website; mobile application channels, where customers submit feedback through mobile applications developed by the company; offline store channels, where customers go directly to offline stores to communicate and give feedback to staff, etc.

[0059] Taking insurance companies as an example, customers’ demands usually include service demands, operational demands, functional demands, etc. The specific customer feedback information collected can include, for example, claims support, customers’ questions, dissatisfaction or suggestions about the claims process; or billing issues, customers’ questions about premium calculation, bills or payment issues; or product consultation, customers’ detailed consultation on auto insurance products, such as insurance coverage, clause explanation, etc.; or case inquiry, customers’ inquiries on the progress of the claim case or results; or value-added services, customers’ feedback on additional services such as roadside assistance, safety inspections, etc. By collecting all pending customer feedback information for automatic, efficient and accurate classification, it is beneficial to improve the response speed and guidance accuracy of customer service.

[0060] S102: extracting structured information and unstructured information from the customer feedback information according to preset data types.

[0061] In this embodiment, structured information and unstructured information in customer feedback information are extracted according to preset data types, wherein the extraction of structured information helps the machine to better understand the specific content of the data, such as specific values ​​such as date, time, and amount, and structured information facilitates statistical analysis and data mining; and unstructured information (such as the sentiment and context of the text) provides deeper customer insights, helps to understand the real needs and feelings of customers, identifies topics and keywords from unstructured text, and also helps to quickly classify and understand the main content of feedback. In customer feedback information, since the content of customer feedback often exists in the form of unstructured text, which contains rich semantic information, key information can be extracted from the unstructured text and potential demands can be mined, thereby providing data analysis and decision support for classification. In addition to text content, customer feedback also includes a series of structured factors, such as feedback sources (service hotlines, WeChat public accounts, web portals, etc.), case links, specific case numbers, etc. These structured data have clear relevance and directionality. For example, different sources of voice may correspond to different types of customer problems or satisfaction levels; there is an intrinsic connection between specific service labels (such as "service", "operation", "function", etc.) and specific source channels. For example, data shows that 35% of "service"-related feedback comes from the official customer service hotline, 30% comes from operational interactions on the WeChat platform, and 17% is raised through online customer service telephone channels, etc. Therefore, by extracting structured and unstructured information from customer feedback information, a more comprehensive data perspective can be provided to enhance the value of data utilization. The combination of the two can enrich the use of data from different angles and improve the classification accuracy of customer feedback information.

[0062] In one embodiment, Figure 2 As shown, the extracting structured information and unstructured information from the customer feedback information according to a preset data type includes:

[0063] S201, extracting the feedback source, page entry, feedback case link and / or feedback case number from the customer feedback information as the structured information;

[0064] S202: Extracting the dialogue text in the customer feedback information as the unstructured information.

[0065] In this embodiment, the corresponding data is extracted from the collected customer feedback information according to the preset data type, so as to form structured information and unstructured information for subsequent demand classification, wherein the feedback source, page entry, feedback case link and / or feedback case number of the customer feedback information are extracted as structured information, which is easy to be processed and analyzed by computer, and can quickly sort, search and count data, and the structured data has clear relevance and directionality. By extracting these structured information, the model can better mine the direct relevance and directionality of different types of data. For example, different voice sources may correspond to different types of customer problems or satisfaction levels; there is an intrinsic connection between specific service labels (such as "service", "operation", "function", etc.) and specific source channels. For example, data shows that 35% of the feedback related to "service" comes from the official customer service hotline, 30% comes from the operation interaction of the WeChat platform, and 17% is raised through the online customer service phone channel, etc. By extracting these structured information with certain regularity, its guiding role in customer service strategy, response speed and accuracy can be fully exerted.

[0066] At the same time, the conversation text in the customer feedback information is extracted as unstructured information. This structured information includes the specific demands of the customer and has rich semantic information, such as product consultation dialogues, case inquiry dialogues, complaint and suggestion dialogues between customers and customer service, etc. By extracting the conversation text in the customer feedback information, the subsequent demand classification model is provided with comprehensive and rich semantic information, thereby accurately mining the customer's feedback demands and achieving efficient and accurate feedback classification and response.

[0067] S103: Perform data fusion processing on the structured information and the unstructured information.

[0068] In this embodiment, after the structured information and unstructured information are extracted according to the preset data type, the two types of information are subjected to data fusion processing. In specific implementation, the extracted structured information including the data value of the feedback source, page entry, feedback case link and / or feedback case number is connected with the data value of the unstructured information dialogue text through a separator to generate a corresponding fusion description text. For example, the fusion processing is performed according to the following rules:

[0069] row['last_concat_sentence']="Feedback channel is"+row['data_src']+"|"+"Page entry is"+row['task_src_page_etrn']+"|"+"Case link is"+row['lead_senario']+"|"+row['is_rpt_no']+"Case number"+"|"+row['cust_dlg_text']

[0070] Among them, a new column last_concat_sentence is created as the fused description text. This column is formed by connecting multiple columns of data using a specific separator ("|" in this example), extracting the value of the column named data_src from the current row as the feedback channel, extracting the value of the column named task_src_page_etrn from the current row as the page entry, extracting the value of the column named lead_senario from the current row as the case link, extracting the value of the column named is_rpt_no from the current row as the case number, and extracting the value of the column named cust_dlg_text from the current row as the conversation text, thereby merging multiple data points into a single readable string.

[0071] This embodiment fuses structured information with unstructured information before sending them into the model for encoding, which has the following advantages: on the one hand, considering the semantic issue, the structured information in this embodiment also has certain semantics, so that the unstructured semantics can be integrated with the structured semantics, allowing the model to pay more attention to contextual information; on the other hand, it can optimize data processing efficiency: although direct processing of unstructured data is often slow and complicated, through preprocessing and fusion, unstructured data can be converted into structured or semi-structured form, which is convenient for subsequent use of unified algorithms and models for efficient analysis. This means that in the encoding stage, the algorithm can focus more on mining the value of data rather than processing the complexity of data, thereby improving processing efficiency.

[0072] S104: Input the fused data into the trained text classification model and output the corresponding scene demand classification results.

[0073] In this embodiment, the fused data is input into the trained text classification model for automatic scene demand classification prediction, and then the corresponding scene demand classification results are output, for example, the output demand label is function, and the specific sub-classification is APP basic account_real-name authentication||process_cannot be authenticated / authentication failed service, at this time, the customer service can quickly respond to the customer's demand for real-name authentication to help the customer complete the real-name authentication; or the output demand label is service, and the specific sub-classification is car value-added service_ETC||process_cancel application / cancel ETC, at this time, the customer service can quickly respond to the customer's demand for canceling application or canceling ETC, to help the customer cancel or cancel the ETC function. By extracting the structured information and unstructured information in the customer feedback information and fusing them, and using the text classification model to classify the fused data, the unstructured semantics can be integrated with the structured semantics, so that the model pays more attention to the context information, and the structured and unstructured factors in the customer feedback information can be efficiently used to achieve fast and accurate customer feedback classification, which improves the efficiency of customer feedback classification and is also conducive to improving the speed of demand response.

[0074] Specifically, the text classification model adopts the RoBERTa model, which is an improved version of the BERT model, and it has been more finely tuned in the pre-training stage. This optimization includes strategies such as dynamic masking, removing the NSP (next sentence prediction) target and increasing the amount of training data. These improvements make the model more powerful in understanding and generating language, and can improve the efficiency of model training. Compared with the original BERT model, RoBERTa improves the training efficiency of the model by optimizing the training process; in addition, it can enhance the generalization ability of the model. Since RoBERTa uses a larger data set and longer training time during pre-training, the model has better generalization ability. In the scene demand classification task, this means that the model can better handle unseen or complex scene demands, thereby improving the accuracy of classification. Compared with the traditional CNN+RNN structure, the use of RoBERTa for text classification has the advantage of pre-training. RoBERTa is a pre-trained language model based on the Transformer architecture. It has been deeply pre-trained on a large-scale corpus and can capture rich language structure and semantic information. This means that the model has a strong language understanding ability before fine-tuning for specific tasks. In contrast, the CNN+RNN combination needs to learn these features from scratch, and the training cost and time are usually higher. At the same time, the Transformer architecture in RoBERTa can effectively capture long-distance dependencies in the text through the self-attention mechanism, while the traditional RNN may encounter the problem of gradient disappearance or explosion when processing long sequences. Although CNN can capture local features through sliding windows, its understanding of the global context is not as strong as the Transformer model. The Transformer architecture used by RoBERTa naturally supports parallel computing, especially in the self-attention mechanism, where calculations at all positions can be performed simultaneously, which greatly speeds up the training speed. In contrast, RNN is difficult to achieve the same degree of parallelization due to its sequence-dependent characteristics. Therefore, in this embodiment, the RoBERTa model is used to perform text classification processing on the fused data, so as to efficiently and accurately realize customer feedback classification, which improves the efficiency of customer feedback classification and is also conducive to improving the speed of appeal response.

[0075] In one embodiment, Figure 3 As shown, the fused data is input into the trained text classification model, and the corresponding scene demand classification results are output, including:

[0076] S301, inputting the fusion description text into a trained text classification model;

[0077] S302, performing vector encoding on the fused description text by an encoder in the text classification model to generate a fused vector;

[0078] S303, performing classification prediction on the fusion vector through the fully connected layer in the text classification model, and outputting the probability that the fusion vector belongs to each preset appeal category;

[0079] S304: Taking the category with the highest probability as the scene appeal classification result of the fused description text.

[0080] In this embodiment, the fused description text with structured information and unstructured information is input into the trained text classification model. Preferably, the fused description text can also be preprocessed before being input into the text classification model, such as text cleaning, that is, removing irrelevant characters in the text, such as special characters, HTML tags, URLs, etc.; word segmentation, that is, dividing the text into token sequences that can be understood by the RoBERTa model; stop word removal, that is, removing common stop words as needed; case conversion, that is, converting all texts to lowercase or uppercase to reduce vocabulary variants; padding, ensuring that all input sequences have the same length, usually by adding special tokens (such as [PAD]) at the end of shorter sequences, etc. Text preprocessing can help the model understand the meaning of the text more accurately by cleaning and normalizing text data. Removing irrelevant characters and special symbols can reduce noise in the data and improve the operating efficiency of the model; removing stop words and performing feature extraction can reduce the dimension of the feature space, which helps to avoid model overfitting, etc., thereby improving the overall model effect.

[0081] After the fused description text is input into the text classification model, the fused description text is vector-encoded by the encoder in the model. Specifically, the encoder in the RoBERTa model is the core component of the Transformer architecture, which is responsible for encoding the input text sequence and generating a high-level representation of the text, namely the fused vector. The RoBERTa model contains multiple Transformer layers, each of which contains the above-mentioned self-attention and feedforward network structure. These layers are stacked together, allowing the model to abstract and extract text information layer by layer and generate deep feature representations. The encoder converts the input text into a series of high-level feature vectors, which can be used for various downstream natural language processing tasks such as text classification, question answering, etc. In this embodiment, based on the fused vector after vector encoding, the fused vector is classified and predicted through the fully connected layer, and the probability that the fused vector belongs to each preset demand category is output. The classification output of the specific fully connected layer is used to generate the probability of each category through a softmax function. By using the Softmax function, when dealing with multiple categories that need to be considered at the same time, the probability of belonging to each category can be given, and the category with the highest probability is used as the scene demand classification result of the fused description text, and the customer feedback demand is automatically confirmed as a sub-category demand under service, operation, function or other classification labels, so that while efficiently and accurately classifying customer feedback information, the customer service's response speed to customer demands is improved, thereby enhancing customer experience.

[0082] In one embodiment, Figure 4 As shown, after the fused data is input into the trained text classification model and the corresponding scene demand classification result is output, the method further includes:

[0083] S401, collecting statistics on the scene demand classification results within a specified time period;

[0084] S402: Determine the processing priority of different scenario demand classifications based on the statistical results.

[0085] In this embodiment, based on the scene demand classification results obtained by collecting and automatically classifying, the scene demand classification results received historically are statistically analyzed. For example, customer demands within the waiting time period within the past three months, six months, and one year can be statistically analyzed, including the type of customer demands, the scope of impact of the demands, the length of time it takes to resolve the demands, etc. Based on the statistical results of the scene demand classification results, the number, frequency, scope of impact, resolution time, and other factors of various customer demands can be analyzed, and then the processing priority of different scene demand classifications can be confirmed. For example, the occurrence frequency of different types of demands can be analyzed, and frequently occurring demands have a higher processing priority; different levels of severity (such as minor, medium, and severe) are defined according to the impact of the demands on the business, and the higher the severity, the higher the processing priority; the average time required for demand resolution is analyzed, and demands with a long resolution time may need to be processed first, so the processing priority is higher, etc., and each statistical result can be further assigned a weight, and the processing priority of different scene demand classifications can be obtained after calculating the comprehensive score of each scene demand classification based on the statistical results, so that the priority result comprehensively considers different factors and is more in line with actual processing needs.

[0086] In one embodiment, Figure 5 As shown, after the fused data is input into the trained text classification model and the corresponding scene demand classification result is output, the method further includes:

[0087] S501, obtaining the processing priority of all unprocessed scene demands;

[0088] S502: Based on the processing priorities of all unprocessed scenario demands, a dynamically sorted demand processing plan is generated and pushed to the client end.

[0089] In this embodiment, based on the processing priority obtained by statistical analysis, when manual intervention is required to process customer demands, since there is usually more than one unresolved scenario demand, in order to make a more reasonable and efficient work arrangement, the processing priority of all unresolved scenario demands can be obtained first, and based on the processing priority of all unresolved scenario demands, a dynamically sorted demand processing plan is generated and pushed to the customer service end. The higher the priority of the unresolved scenario demand in the demand processing plan, the earlier the planned time is, and vice versa, the lower the priority of the unresolved scenario demand, the later the actual plan is. The plan is dynamically updated based on the latest processing priority of the unresolved scenario demand and synchronized to the customer service end, for example, it is updated once for each classification of one or a fixed number of unresolved scenario demands, or it can also be updated once at regular intervals, so that the customer service end can receive the latest scenario demand resolution plan. When manual intervention is performed to process customer demands, the demand processing plan generated based on the processing priority is used as a reference for its work arrangement, and more important or urgent scenario demands are processed first, and the processing order of scenario demands is arranged more reasonably, so as to avoid the impact of important or urgent customer demands on business operations, customer satisfaction, system stability, etc. caused by unresolved important or urgent customer demands, thereby improving the processing efficiency and rationality of customer demand processing.

[0090] It should be noted that there is not necessarily a certain order between the above-mentioned steps. A person skilled in the art can understand, based on the description of the embodiments of the present invention, that in different embodiments, the above-mentioned steps may have different execution orders, that is, they may be executed in parallel, may be executed interchangeably, and so on.

[0091] Another embodiment of the present invention provides a classification device for customer feedback information, such as Figure 6 As shown, the device 1 comprises:

[0092] An acquisition module 11 is used to acquire customer feedback information to be classified;

[0093] An extraction module 12, used to extract structured information and unstructured information from the customer feedback information according to a preset data type;

[0094] A fusion module 13, used for performing data fusion processing on the structured information and the unstructured information;

[0095] The classification module 14 is used to input the fused data into the trained text classification model and output the corresponding scene demand classification results.

[0096] The module referred to in the present invention refers to a series of computer program instruction segments that can complete specific functions, which is more suitable for describing the classification and execution process of customer feedback information than a program. For the specific implementation of each module, please refer to the corresponding method embodiment above, which will not be repeated here.

[0097] In one embodiment, the extraction module 12 includes:

[0098] A first extraction unit is used to extract the feedback source, page entry, feedback case link and / or feedback case number from the customer feedback information as the structured information;

[0099] The second extraction unit is used to extract the dialogue text in the customer feedback information as the unstructured information.

[0100] In one embodiment, the fusion module 13 is specifically used to:

[0101] The data values ​​of the feedback source, page entry, feedback case link and / or feedback case number are connected with the data value of the dialogue text through a separator to generate a corresponding fusion description text.

[0102] In one embodiment, the classification module 14 includes:

[0103] An input unit, used for inputting the fused description text into a trained text classification model;

[0104] An encoding unit, used for performing vector encoding on the fused description text through an encoder in the text classification model to generate a fused vector;

[0105] A prediction unit, configured to perform classification prediction on the fused vector through a fully connected layer in the text classification model, and output a probability that the fused vector belongs to each preset appeal category;

[0106] The classification unit is used to take the category with the highest probability as the scene appeal classification result of the fused description text.

[0107] In one embodiment, the device 1 further comprises:

[0108] The statistics module is used to collect statistics on the scene demand classification results within a specified time period;

[0109] The priority module is used to confirm the processing priority of different scenario demand classifications based on statistical results.

[0110] In one embodiment, the device 1 further comprises:

[0111] The priority acquisition module is used to obtain the processing priority of all unprocessed scene demands;

[0112] The processing plan generation module is used to generate a dynamically sorted request processing plan based on the processing priorities of all unprocessed scenario requests and push it to the client.

[0113] In one embodiment, the text classification model is a RoBERTa model.

[0114] Another embodiment of the present invention provides a classification system for customer feedback information, such as Figure 7 As shown, the system 10 includes:

[0115] One or more processors 110 and memory 120, Figure 7 A processor 110 is used as an example for description. The processor 110 and the memory 120 may be connected via a bus or other means. Figure 7 The example of connecting through bus is taken in the following.

[0116] The processor 110 is used to complete various control logics of the system 10, and it can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a single-chip microcomputer, an ARM (Acorn RISC Machine) or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of these components. In addition, the processor 110 can also be any traditional processor, microprocessor or state machine. The processor 110 can also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a DSP and / or any other such configuration.

[0117] The memory 120 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions corresponding to the customer feedback information classification method in the embodiment of the present invention. The processor 110 executes various functional applications and data processing of the system 10 by running the non-volatile software programs, instructions and units stored in the memory 120, that is, implements the customer feedback information classification method in the above method embodiment.

[0118] The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the system 10, etc. In addition, the memory 120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 120 may optionally include a memory remotely arranged relative to the processor 110, and these remote memories may be connected to the system 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0119] One or more units are stored in the memory 120, and when executed by one or more processors 110, the customer feedback information classification method in any of the above method embodiments is executed, for example, the following steps are executed:

[0120] Obtain customer feedback information to be classified;

[0121] Extracting structured information and unstructured information from the customer feedback information according to preset data types;

[0122] Performing data fusion processing on the structured information and the unstructured information;

[0123] Input the fused data into the trained text classification model and output the corresponding scene demand classification results.

[0124] In one embodiment, extracting structured information and unstructured information from the customer feedback information according to a preset data type includes:

[0125] Extracting the feedback source, page entry, feedback case link and / or feedback case number from the customer feedback information as the structured information;

[0126] The conversation text in the customer feedback information is extracted as the unstructured information.

[0127] In one embodiment, the data fusion processing of the structured information and the unstructured information specifically includes:

[0128] The data values ​​of the feedback source, page entry, feedback case link and / or feedback case number are connected with the data value of the dialogue text through a separator to generate a corresponding fusion description text.

[0129] In one embodiment, the fused data is input into a trained text classification model to output corresponding scene demand classification results, including:

[0130] Inputting the fused description text into a trained text classification model;

[0131] Performing vector encoding on the fused description text by using an encoder in the text classification model to generate a fused vector;

[0132] Performing classification prediction on the fusion vector through the fully connected layer in the text classification model, and outputting the probability that the fusion vector belongs to each preset appeal category;

[0133] The category with the highest probability is taken as the scene appeal classification result of the fused description text.

[0134] In one embodiment, after inputting the fused data into a trained text classification model and outputting the corresponding scene demand classification result, the method further includes:

[0135] Collect statistics on the classification results of scene demands within a specified time period;

[0136] Confirm the processing priority of different scenario demand classifications based on statistical results.

[0137] In one embodiment, after inputting the fused data into a trained text classification model and outputting the corresponding scene demand classification result, the method further includes:

[0138] Get the processing priority of all unprocessed scene requests;

[0139] Based on the processing priorities of all unprocessed scenario requests, a dynamically sorted request processing plan is generated and pushed to the client.

[0140] In one embodiment, the text classification model is a RoBERTa model.

[0141] An embodiment of the present invention provides a non-volatile computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the customer feedback information classification method in any of the above method embodiments is executed, for example, the following steps are executed:

[0142] Obtain customer feedback information to be classified;

[0143] Extracting structured information and unstructured information from the customer feedback information according to preset data types;

[0144] Performing data fusion processing on the structured information and the unstructured information;

[0145] Input the fused data into the trained text classification model and output the corresponding scene demand classification results.

[0146] In one embodiment, extracting structured information and unstructured information from the customer feedback information according to a preset data type includes:

[0147] Extracting the feedback source, page entry, feedback case link and / or feedback case number from the customer feedback information as the structured information;

[0148] The conversation text in the customer feedback information is extracted as the unstructured information.

[0149] In one embodiment, the data fusion processing of the structured information and the unstructured information specifically includes:

[0150] The data values ​​of the feedback source, page entry, feedback case link and / or feedback case number are connected with the data value of the dialogue text through a separator to generate a corresponding fusion description text.

[0151] In one embodiment, the fused data is input into a trained text classification model to output corresponding scene demand classification results, including:

[0152] Inputting the fused description text into a trained text classification model;

[0153] Performing vector encoding on the fused description text by using an encoder in the text classification model to generate a fused vector;

[0154] Performing classification prediction on the fusion vector through the fully connected layer in the text classification model, and outputting the probability that the fusion vector belongs to each preset appeal category;

[0155] The category with the highest probability is taken as the scene appeal classification result of the fused description text.

[0156] In one embodiment, after inputting the fused data into a trained text classification model and outputting the corresponding scene demand classification result, the method further includes:

[0157] Collect statistics on the classification results of scene demands within a specified time period;

[0158] Confirm the processing priority of different scenario demand classifications based on statistical results.

[0159] In one embodiment, after inputting the fused data into a trained text classification model and outputting the corresponding scene demand classification result, the method further includes:

[0160] Get the processing priority of all unprocessed scene requests;

[0161] Based on the processing priorities of all unprocessed scenario requests, a dynamically sorted request processing plan is generated and pushed to the client.

[0162] In one embodiment, the text classification model is a RoBERTa model.

[0163] As an example, non-volatile storage media can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) as external cache memory. By way of illustration and not limitation, RAM can be obtained in many forms such as synchronous RAM (SRAM), dynamic RAM, (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The disclosed memory components or memories of the operating environment described herein are intended to include one or more of these and / or any other suitable types of memory.

[0164] In summary, the present invention discloses a classification method, device, system and medium for customer feedback information, the method comprising: obtaining customer feedback information to be classified; extracting structured information and unstructured information in the customer feedback information according to preset data types; performing data fusion processing on the structured information and unstructured information; inputting the fused data into a trained text classification model, and outputting the corresponding scene demand classification results. By extracting structured information and unstructured information from customer feedback information and fusing them, and using a text classification model to classify the fused data, unstructured semantics can be fused with structured semantics, allowing the model to pay more attention to contextual information, and efficiently utilizing structured and unstructured factors in customer feedback information to achieve fast and accurate customer feedback classification, which improves the efficiency of customer feedback classification while also helping to improve the speed of demand response.

[0165] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium, and the computer program can include the processes of the above-mentioned method embodiments when executed. The storage medium can be a memory, a disk, a floppy disk, a flash memory, an optical storage device, etc.

[0166] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A classification method for customer feedback information, characterized in that: include: Obtain customer feedback information to be classified; Extracting structured information and unstructured information from the customer feedback information according to preset data types; Performing data fusion processing on the structured information and the unstructured information; Input the fused data into the trained text classification model and output the corresponding scene demand classification results.

2. The method for classifying customer feedback information according to claim 1, characterized in that: The extracting structured information and unstructured information from the customer feedback information according to a preset data type includes: Extracting the feedback source, page entry, feedback case link and / or feedback case number from the customer feedback information as the structured information; The conversation text in the customer feedback information is extracted as the unstructured information.

3. The method for classifying customer feedback information according to claim 2, characterized in that: The data fusion processing of the structured information and the unstructured information specifically includes: The data values ​​of the feedback source, page entry, feedback case link and / or feedback case number are connected with the data value of the dialogue text through a separator to generate a corresponding fusion description text.

4. The method for classifying customer feedback information according to claim 3, characterized in that: The fused data is input into the trained text classification model to output the corresponding scene demand classification results, including: Inputting the fused description text into a trained text classification model; Performing vector encoding on the fused description text by using an encoder in the text classification model to generate a fused vector; Performing classification prediction on the fusion vector through the fully connected layer in the text classification model, and outputting the probability that the fusion vector belongs to each preset appeal category; The category with the highest probability is taken as the scene appeal classification result of the fused description text.

5. The method for classifying customer feedback information according to claim 1, characterized in that: After the fused data is input into the trained text classification model and the corresponding scene demand classification result is output, the method further includes: Collect statistics on the classification results of scene demands within a specified time period; Confirm the processing priority of different scenario demand classifications based on statistical results.

6. The method for classifying customer feedback information according to claim 5, characterized in that: After the fused data is input into the trained text classification model and the corresponding scene demand classification result is output, the method further includes: Get the processing priority of all unprocessed scene requests; Based on the processing priorities of all unprocessed scenario requests, a dynamically sorted request processing plan is generated and pushed to the client.

7. The method for classifying customer feedback information according to any one of claims 1 to 6, characterized in that: The text classification model is the RoBERTa model.

8. A classification device for customer feedback information, characterized in that: include: An acquisition module, used to acquire customer feedback information to be classified; An extraction module, used to extract structured information and unstructured information from the customer feedback information according to preset data types; A fusion module, used for performing data fusion processing on the structured information and the unstructured information; The classification module is used to input the fused data into the trained text classification model and output the corresponding scene demand classification results.

9. A classification system for customer feedback information, characterized in that: The system includes at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the customer feedback information classification method according to any one of claims 1 to 7.

10. A non-volatile computer-readable storage medium, characterized in that: The non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by one or more processors, enable the one or more processors to execute the method for classifying customer feedback information according to any one of claims 1 to 7.

Images