Complaint data processing method and device and server

Abstract

The invention provides a complaint data processing method and device and a server. The method comprises the following steps: after complaint record data of a current time period is obtained, determining attribute data of a plurality of complaint lists according to the complaint record data of the current time period; dividing the plurality of complaint sheets into a plurality of data groups according to complaint handling channels; wherein each data group at least corresponds to one complaint handling channel; based on a preset processing rule, according to the plurality of data groups, constructing a plurality of time sequence feature sequences respectively corresponding to the plurality of complaint handling channels; and then a preset complaint amount prediction model which integrates various model structures of Prophet, LSTM and ARIMA at the same time is utilized to process a plurality of time sequence feature sequences, and the complaint amount of a specified time window of the complaint handling channel is determined in advance. Therefore, the future complaint amount of each complaint handling channel can be efficiently and accurately predicted, so that preparation is made in advance, related complaints are handled in time, and the service experience of customers is improved.

Description

Technical Field

[0001] This manual belongs to the field of artificial intelligence technology, and in particular relates to complaint data processing methods, devices and servers. Background Technology

[0002] Large financial institutions such as banks receive a large number of customer complaints every day, which need to be handled promptly.

[0003] Because large financial institutions often simultaneously launch and implement a large number of products or services, and their organizational structures are typically large and complex, these products or services often involve multiple specific business departments or even subsidiaries, resulting in a relatively complex overall complaint handling process. When a particular business department experiences a sudden surge in complaints about a specific product or service at a particular time, it is often difficult to allocate relevant resources in a timely manner to effectively handle customer complaints, thereby impacting the customer's service experience.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This manual provides a complaint data processing method, apparatus, and server that can efficiently and accurately predict the future complaint volume of various complaint handling channels, so as to prepare in advance, handle relevant complaints in a timely manner, and improve the customer service experience.

[0006] This manual provides a method for processing complaint data, including:

[0007] Obtain complaint record data for the current time period; wherein the complaint record data for the current time period contains multiple complaint forms received within the current time period;

[0008] Based on the complaint record data for the current time period, the attribute data of multiple complaint forms is determined; wherein, the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text;

[0009] Based on the complaint handling channels, the multiple complaint forms are divided into multiple data groups; each data group corresponds to at least one complaint handling channel.

[0010] Based on preset processing rules, multiple time-series feature sequences are constructed according to multiple data groups and the attribute data of the corresponding complaint forms; wherein, each time-series feature sequence corresponds to at least one complaint handling channel.

[0011] Multiple time-series feature sequences are processed using a pre-defined complaint volume prediction model to obtain corresponding target prediction results; wherein, the pre-defined complaint volume prediction model is a hybrid model that integrates Prophet, LSTM, and ARIMA.

[0012] Based on the target prediction results, the number of complaints during the designated time window for the complaint handling channel is determined.

[0013] In one embodiment, the construction of multiple time-series feature sequences based on preset processing rules, multiple data groups, and the attribute data of corresponding complaint forms includes: constructing a corresponding current time-series feature sequence based on the current data group according to preset processing rules in the following manner:

[0014] Based on preset processing rules, the current data group is divided into multiple sub-data groups according to the complaint time; wherein, the sub-data group corresponds to a point in time within the current time period;

[0015] Based on the complaint forms in each sub-data group and the corresponding attribute data of the complaint forms, at least the number of complaint forms, the proportion distribution of complaint source channels, and the proportion distribution of target complaint objects in each sub-data group should be determined.

[0016] Based on the number of complaint forms, the proportion of complaint source channels, and the proportion of target complaint subjects in each sub-data group, the complaint feature labels for each sub-data group are determined.

[0017] Based on the complaint feature labels of each sub-data group, multiple corresponding feature elements are determined;

[0018] Arrange multiple feature elements in chronological order to obtain the corresponding current time-series feature sequence.

[0019] In one embodiment, after obtaining the corresponding current time-series feature sequence, the method further includes:

[0020] Retrieve complaint record data for the previous time period regarding the current complaint handling channel; where the current complaint handling channel refers to the complaint handling channel corresponding to the current data group.

[0021] Based on the complaint record data of the previous time period, generate the current lagged feature sequence for the current time series feature sequence;

[0022] According to the preset processing rules, the current time-series feature sequence and the current lagging feature sequence are concatenated to obtain the current joint feature sequence for the current complaint handling channel;

[0023] By using a pre-defined complaint volume prediction model and processing the current joint feature sequence, the complaint volume of the current complaint handling channel within a specified time window is determined.

[0024] In one embodiment, the complaint handling channels include multiple complaint handling channels.

[0025] In one embodiment, after determining the number of complaints within a designated time window for the complaint handling channel based on the target prediction results, the method further includes:

[0026] The system detects whether the number of complaints received during the designated time window of the complaint handling channel exceeds a preset first complaint threshold, and obtains the corresponding first detection result.

[0027] Based on the first test results, complaint handling channels where the number of complaints in a specified time window exceeds the preset first complaint volume threshold are identified as target complaint risk channels.

[0028] Based on the current time-series characteristic sequence of the target complaint risk channels, statistical analysis is performed on the proportion distribution of complaint source channels and / or the proportion distribution of target complaint targets within the current time period to obtain the statistical analysis results for the current time period.

[0029] Based on the statistical analysis results of the current time period, a matching target complaint handling strategy is determined from the preset complaint handling strategy set;

[0030] Generate a first complaint risk alert regarding the target complaint risk channel; and send the first complaint risk alert and the target complaint handling strategy to the target business department associated with the target complaint risk channel.

[0031] In one embodiment, after determining the number of complaints within a designated time window for the complaint handling channel based on the target prediction results, the method further includes:

[0032] The system detects whether the number of complaints received during the designated time window of the complaint handling channel exceeds a preset second complaint threshold, and obtains the corresponding second detection result.

[0033] Based on the second test results, complaint handling channels with a complaint volume exceeding the preset second complaint volume threshold within a specified time window are identified as candidate complaint risk channels.

[0034] Based on the current time-series characteristics of candidate complaint risk channels, count the number of target complaint subjects;

[0035] Based on the number of target complaints, target risk targets are identified;

[0036] Based on multiple time-series characteristic sequences, relevant complaint risk channels involving the target risk object are identified from multiple complaint handling channels;

[0037] Generate a second complaint risk alert regarding the target risk object; and send the second complaint risk alert to the relevant business departments associated with the relevant complaint risk channels.

[0038] In one embodiment, the method further includes:

[0039] Obtain and extract corresponding sample data based on historical complaint records;

[0040] Construct an initial complaint volume prediction model; wherein the initial complaint volume prediction model includes at least: an initial master model based on Prophet, an initial first auxiliary model based on LSTM, and an initial second auxiliary model based on ARIMA, wherein the output of the initial master model is connected to the input of the initial first auxiliary model and the input of the initial second auxiliary model;

[0041] Construct a joint objective loss function for the main model, the first auxiliary model, and the second auxiliary model;

[0042] Based on the objective joint loss function, an initial complaint volume prediction model is trained using the sample data to obtain a preset complaint volume prediction model that meets the requirements.

[0043] This specification also provides a complaint data processing device, including:

[0044] The acquisition module is used to acquire complaint record data for the current time period; wherein, the complaint record data for the current time period contains multiple complaint forms received within the current time period;

[0045] The first determining module is used to determine the attribute data of multiple complaint forms based on the complaint record data of the current time period; wherein, the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text;

[0046] The grouping module is used to divide the multiple complaint forms into multiple data groups according to the complaint handling channels; wherein each data group corresponds to at least one complaint handling channel;

[0047] The construction module is used to construct multiple time-series feature sequences based on preset processing rules, multiple data groups, and the attribute data of corresponding complaint forms; wherein each time-series feature sequence corresponds to at least one complaint handling channel.

[0048] The processing module is used to process multiple time-series feature sequences using a preset complaint volume prediction model to obtain the corresponding target prediction results; wherein, the preset complaint volume prediction model is a hybrid model that integrates Prophet, LSTM and ARIMA.

[0049] The second determining module is used to determine the number of complaints within a specified time window of the complaint handling channel based on the target prediction results.

[0050] This specification also provides a server, including a processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to implement the steps of the complaint data processing method.

[0051] This specification also provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of the complaint data processing method.

[0052] Based on the complaint data processing method, apparatus, and server provided in this specification, before specific implementation, a preset complaint volume prediction model suitable for analyzing and predicting the complaint volume of different complaint handling channels can be constructed and trained, which simultaneously integrates Prophet, LSTM, and ARIMA. In specific implementation, complaint record data for the current time period can be acquired first; wherein the complaint record data packet for the current time period contains multiple complaint forms received within the current time period; and based on the complaint record data for the current time period, the attribute data of the multiple complaint forms is determined; wherein the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text; then, based on the complaint handling channel, the multiple complaint forms are divided into multiple data groups; wherein each data group corresponds to at least one complaint handling channel; based on preset processing rules, multiple time-series feature sequences are constructed according to the multiple data groups and the corresponding complaint form attribute data; wherein each time-series feature sequence corresponds to at least one complaint handling channel; the preset complaint volume prediction model is used to process the multiple time-series feature sequences to obtain the corresponding target prediction results; and based on the target prediction results, the complaint volume for a specified time window of the complaint handling channel is determined. This allows it to be well-suited for complex complaint handling scenarios involving a massive number of complaints. It can efficiently and accurately predict the number of complaints in each complaint handling channel during a specified time window in the future, so as to determine the corresponding handling strategies in advance, allocate resources, make relevant preparations, and handle the received customer complaints in a timely and effective manner, thereby improving the customer service experience. Attached Figure Description

[0053] To more clearly illustrate the embodiments of this specification, the accompanying drawings used in the embodiments will be briefly introduced below. The drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart illustrating a complaint data processing method provided in one embodiment of this specification;

[0055] Figure 2 This is a schematic diagram illustrating one embodiment of the complaint data processing method provided in the embodiments of this specification, applied in a scenario example.

[0056] Figure 3This is a schematic diagram illustrating one embodiment of the complaint data processing method provided in the embodiments of this specification, applied in a scenario example.

[0057] Figure 4 This is a schematic diagram illustrating one embodiment of the complaint data processing method provided in the embodiments of this specification, applied in a scenario example.

[0058] Figure 5 This is a schematic diagram illustrating one embodiment of the complaint data processing method provided in the embodiments of this specification, applied in a scenario example.

[0059] Figure 6 This is a schematic diagram illustrating one embodiment of the complaint data processing method provided in the embodiments of this specification, applied in a scenario example.

[0060] Figure 7 This is a schematic diagram of the structural composition of a server provided in one embodiment of this specification;

[0061] Figure 8 This is a schematic diagram of the structural composition of a complaint data processing device provided in one embodiment of this specification. Detailed Implementation

[0062] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0063] It should be noted that the information and data related to users involved in the embodiments of this specification are all information and data authorized by the user or fully authorized by the relevant parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with relevant laws, regulations, and standards, and necessary confidentiality measures have been taken. They do not violate public order and good morals, and corresponding operation entry points are provided for users or relevant parties to choose to authorize or refuse.

[0064] It should also be noted that in the embodiments of this specification, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0065] See Figure 1As shown in the embodiments of this specification, a method for processing complaint data is provided. Specifically, this method is applied to the server side. In specific implementation, the method may include the following:

[0066] S101: Obtain complaint record data for the current time period; wherein the complaint record data for the current time period contains multiple complaint forms received within the current time period;

[0067] S102: Based on the complaint record data for the current time period, determine the attribute data of multiple complaint forms; wherein, the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text;

[0068] S103: Divide the multiple complaint forms into multiple data groups according to the complaint handling channels; wherein each data group corresponds to at least one complaint handling channel;

[0069] S104: Based on preset processing rules, multiple time-series feature sequences are constructed according to multiple data groups and the attribute data of the corresponding complaint forms; wherein, the time-series feature sequences correspond to at least one complaint handling channel;

[0070] S105: Process multiple time-series feature sequences using a preset complaint volume prediction model to obtain the corresponding target prediction results; wherein, the preset complaint volume prediction model is a hybrid model that integrates Prophet, LSTM, and ARIMA.

[0071] S106: Based on the target prediction results, determine the number of complaints during the designated time window of the complaint handling channel.

[0072] The aforementioned "current time period" can specifically refer to the most recent week or the most recent month. In practice, an appropriate current time period can be set based on the current customer complaint situation.

[0073] The complaint record data for the current time period mentioned above can specifically include complaint forms received by the system platform through multiple complaint source channels during the current time period, as well as the timestamps corresponding to the complaint forms. The timestamps can be specifically determined based on the time the complaint form was received.

[0074] The aforementioned complaint form can be understood as a customer complaint form submitted to the system platform through the complaint channel, containing specific demands and the customer's identification. At a minimum, the complaint form should include the customer's identification and a detailed text of the complaint.

[0075] In practice, based on the complaint text of the aforementioned complaint form, relevant data such as the complaint handling channel, the source of the complaint, and the text of the complaint content can be identified and extracted. The complaint time can also be identified and extracted based on the timestamp of the aforementioned complaint form.

[0076] The aforementioned complaint channels can be understood as the channels through which the system platform directly receives complaint forms initiated by customers. Specifically, these complaint channels can include at least one of the following: in-person complaints at branch offices, mobile banking complaints, complaints transferred from regulatory agencies, complaints through media channels, etc. The different complaint channels used by customers can, to some extent, reflect the degree of customer dissatisfaction and the severity of the complaint. For example, complaints received in person at branch offices are often more serious than those received through mobile banking; similarly, complaints transferred from regulatory agencies tend to reflect greater customer dissatisfaction than those received in person at branch offices.

[0077] The aforementioned complaint handling channels can be understood as the channels responsible for receiving and processing complaint forms. Specifically, these channels may include at least one of the following: third-party channels, self-service machines, intelligent customer service robots, tellers, specific business departments, etc.

[0078] Of course, it should be noted that the complaint sources and handling channels listed above are only illustrative. In practice, depending on the specific circumstances and processing needs, the aforementioned complaint sources and handling channels may include other suitable channels and paths. This instruction manual does not limit this.

[0079] Each of the above data groups corresponds to at least one complaint handling channel. Specifically, each data group may contain multiple complaint forms related to its corresponding complaint handling channel.

[0080] The aforementioned time-series feature sequences correspond to the aforementioned data sets and at least one complaint handling channel. Specifically, each time-series feature sequence may contain multiple feature elements arranged in chronological order. Each feature element corresponds to a point in time within the current time period.

[0081] The aforementioned feature elements may specifically include one or more feature tags related to the following characteristics: the number of complaint forms at the corresponding time point, the percentage distribution of complaint source channels at the corresponding time point, and the percentage distribution of target complaint recipients at the corresponding time point. The aforementioned feature elements may also be empty, indicating that no complaint forms were received for that complaint handling channel at the corresponding time point. The aforementioned target complaint recipients may specifically include the specific service and / or product being complained about in the complaint form.

[0082] The aforementioned pre-set complaint volume prediction model can be understood as a pre-trained algorithm model that can predict the future specified time window of the corresponding complaint handling channel based on the input time-series feature sequence.

[0083] The specified time window mentioned above can specifically include a time window corresponding to a single future period, such as within a future day. The specified time window can also specifically include time windows corresponding to multiple future periods, such as within a future day, within a future week, and within a future month.

[0084] The aforementioned pre-defined complaint volume prediction model can be a hybrid model that integrates at least several model structures, including Prophet, LSTM, and ARIMA.

[0085] The aforementioned complaint volume prediction model includes at least: a Prophet-based main model, a first auxiliary model based on LSTM, and a second auxiliary model based on ARIMA. The output of the main model is connected to the inputs of the first and second auxiliary models. Specifically, the input of the first auxiliary model can be connected to the total revenue module of the preset complaint volume prediction model and the residual module of the main model, respectively; the input of the second auxiliary model can be connected to the total revenue module of the preset complaint volume prediction model and the residual module of the main model, respectively.

[0086] Specifically, Prophet can be understood as a time series model based on time and variable values, combining time series decomposition and machine learning fitting. Compared to conventional time series models, Prophet emphasizes the influence of trend lines, seasonality, periodicity, and exogenous variables during the modeling process. It is relatively more suitable for business scenarios such as financial institution system platforms that process complaint forms on a daily, weekly, or monthly basis, and periodically organize and monitor the processing of complaint forms, achieving relatively better predictive results; and it can exhibit higher timeliness when processing massive amounts of data.

[0087] The aforementioned LSTM (Long Short-Term Memory) network can be understood as a type of time-recurrent neural network, which is better at handling long-term temporal dependencies compared to other models.

[0088] The aforementioned ARIMA (Autoregressive Integrated Moving Average) can be understood as an autoregressive integrated moving average model that combines autoregressive (AR) and moving average (MA) methods, and incorporates the concept of differencing (I) to make non-stationary time series data stationary, thereby enabling effective prediction. Compared to Prophet, ARIMA is more suitable for handling short-term dynamic changes, and this model does not consider factors such as periodicity, holidays, and exogenous regressors.

[0089] In the aforementioned pre-defined complaint volume prediction model, firstly, by using Prophet as the main model, the overall model becomes more adaptable to complex complaint handling scenarios involving a massive number of complaint requests. This ensures the model's prediction accuracy and meets the requirements of the relevant scenarios while improving the overall efficiency of the model's operation. Furthermore, by using LSTM, which excels at handling long-term temporal dependencies, as the first auxiliary model, and ARIMA, which excels at handling short-term dynamic changes, as the second auxiliary model, the model can work in conjunction with the main model to perform more comprehensive and refined feature extraction and analysis. The prediction results output by the main model are supplemented and corrected, resulting in higher accuracy in the final prediction results and improved robustness of the model. In this way, the advantages of different model structures such as Prophet, LSTM, and ARIMA can be fully utilized, making the model well-suited to complex complaint handling scenarios involving a massive number of complaint requests and efficiently and accurately predicting future complaint volumes.

[0090] The aforementioned complaint data processing method can be specifically applied to the server side. The server can specifically include a backend server applied to a system platform of a financial institution, capable of data transmission, data processing, and other functions. Specifically, the server can be, for example, an electronic device with data processing, storage, and network interaction capabilities. Alternatively, the server can be a software program running on the electronic device, providing support for data processing, storage, and network interaction. In this embodiment, the number of servers is not specifically limited. The server can be a single server, several servers, or a server cluster formed by several servers.

[0091] In practice, the system platform's server (e.g., a customer service server) can acquire complaint record data for the current time period at preset screening intervals (e.g., every week or every month) according to preset screening rules. Based on this data, multiple time-series feature sequences corresponding to various complaint handling channels are constructed. A preset complaint volume prediction model is then used to process these time-series feature sequences to obtain corresponding target prediction results. Based on these prediction results, the complaint volume for each of the multiple complaint handling channels within a specified future time window is determined. Furthermore, based on the future complaint volume for each channel within that specified time window, appropriate prompts are sent to each channel to remind them to allocate resources in advance and prepare for timely and effective handling of future complaints. This prevents a sudden surge in the number of complaints that, due to limited resources, could result in a large number of complaints being unable to be processed promptly and effectively, thus impacting customer experience.

[0092] Based on the above embodiments, by distinguishing between various complaint source channels and complaint handling channels, and constructing multiple corresponding time-series feature sequences for each complaint handling channel, and by training and utilizing a preset complaint volume prediction model that integrates multiple model structures such as Prophet, LSTM, and ARIMA to process the above multiple time-series feature sequences, it is possible to efficiently and accurately predict the future complaint volume of each complaint handling channel, so as to prepare in advance, handle relevant complaints in a timely manner, and improve the customer service experience.

[0093] In some embodiments, the complaint record data for the current time period may include at least: multiple complaint forms received during the current time period, and timestamps corresponding to the complaint forms; wherein the timestamps are used to indicate the time when the complaint forms were received.

[0094] Accordingly, the above-mentioned determination of attribute data for multiple complaint forms based on complaint record data within the current time period can, in practice, include the following: processing the complaint record data within the current time period according to preliminary processing rules to obtain multiple complaint forms within the current time period and their corresponding timestamps; then, for each complaint form, determining the type of complaint form by extracting and based on key indicators; determining a matching rule template from a preset rule template set based on the type of complaint form; and processing each complaint form according to the matching rule template to extract data such as complaint handling channel, complaint source channel, complaint time, and complaint content text, which serve as the attribute data of the complaint form.

[0095] Of course, it should be noted that the attribute data listed above is only illustrative. In actual implementation, depending on the specific application scenario and processing requirements, the attribute data may also include other information data related to the complaint form, such as the customer type of the customer who initiated the complaint form. This instruction manual does not limit this.

[0096] Before implementation, a large number of sample complaint forms of different types can be obtained; cluster learning and information statistics can be performed on the large number of sample complaint forms of different types; based on the cluster learning results and information statistics results, preset rule templates for sample complaint forms of different types can be constructed; based on multiple preset rule templates, a preset rule template set can be constructed.

[0097] In some embodiments, reference may be made to Figure 2 As shown, based on preset processing rules, multiple time-series feature sequences are constructed according to multiple data groups and the attribute data of the corresponding complaint forms. These include: constructing the corresponding current time-series feature sequence based on the current data group according to the preset processing rules in the following manner:

[0098] S1: Based on preset processing rules, the current data group is divided into multiple sub-data groups according to the complaint time; wherein, the sub-data group corresponds to a point in time within the current time period;

[0099] S2: Based on the complaint forms in each sub-data group and the corresponding attribute data of the complaint forms, at least determine the number of complaint forms in each sub-data group, the proportion distribution of complaint source channels, and the proportion distribution of target complaint objects;

[0100] S3: Determine the complaint feature labels for each sub-data group based on the number of complaint forms, the proportion of complaint source channels, and the proportion of target complaint subjects.

[0101] S4: Based on the complaint feature labels of each sub-data group, determine the corresponding multiple feature elements;

[0102] S5: Arrange multiple feature elements in chronological order to obtain the corresponding current time-series feature sequence.

[0103] Following the above processing method, multiple other data groups can be processed to obtain multiple other time-series feature sequences corresponding to the other multiple data groups respectively.

[0104] Each sub-data set corresponds to a point in time within the current time period. Each feature element corresponds to a sub-data set. Correspondingly, each feature element corresponds to a point in time within the current time period.

[0105] The aforementioned feature elements may specifically include multiple complaint feature tags. These multiple complaint feature tags include at least: a tag indicating the number of complaint forms determined based on the corresponding sub-data group; a tag indicating the proportion of complaint source channels determined based on the attribute data of complaint forms in the corresponding sub-data group; and a tag indicating the proportion of target complaint subjects determined based on the attribute data of complaint forms in the corresponding sub-data group.

[0106] Of course, it should be noted that the complaint characteristic tags listed above are only illustrative. In actual implementation, depending on the specific circumstances and processing needs, the above complaint characteristic tags may also include: additional explanatory tags for the target complaint object, customer sentiment statistics tags, and other types of complaint characteristic tags. This instruction manual does not limit this.

[0107] Specifically, the aforementioned percentage distribution labels of target complaint subjects and customer sentiment statistics labels can be determined based on the complaint content text of the complaint forms in the attribute data of the complaint forms in the corresponding sub-data groups.

[0108] The additional explanatory labels for the aforementioned target complaint subjects can be determined based on the associated data of the corresponding sub-data groups at the corresponding time points.

[0109] In practice, a pre-defined semantic recognition model can be used to perform semantic recognition on the text content of each complaint form in the sub-data group to obtain the semantic recognition results of the complaint forms. Based on the semantic recognition results of the complaint forms, the object identifier of the target complaint object in the complaint form, as well as the corresponding emotional state parameters, are determined. The emotional state parameters include: emotional type and emotional level. Then, based on the object identifier of the target complaint object in the complaint forms of the sub-data group, data statistical analysis is performed to determine the proportion distribution of the target complaint object, and based on the proportion distribution of the target complaint object, corresponding proportion distribution labels of the target complaint object are generated. At the same time, based on the emotional state parameters of the complaint forms in the sub-data group, data statistical analysis is performed to determine the customer emotional statistical results, and based on the customer emotional statistical results, corresponding customer emotional statistical labels are generated.

[0110] The preset semantic recognition model is trained by deep learning on a large number of sample complaint texts using NLP and large language models.

[0111] The aforementioned NLP (Natural Language Processing) model can be understood as an algorithmic model based on NLP technology that can understand, generate, and process human language. Specifically, the NLP model used here can be a Hidden Markov Model (HMM), etc.

[0112] The aforementioned Large Language Model (LLM) can be understood as a natural language processing model based on deep learning. By learning language rules through pre-training on massive amounts of text data, the model has the ability to understand, generate, and reason about text.

[0113] In practice, after identifying the object identifier of the target complaint in the complaint form, the time point corresponding to the sub-data group can be determined and recorded as the sub-time point. Then, based on the object identifier of the target complaint and the sub-time point corresponding to the sub-data group, internal and external databases are searched to obtain descriptive information about the adjacent time intervals of the sub-time point, as well as descriptive information about the target complaint within the adjacent time intervals, as associated data. Based on the above associated data, special circumstances information about the adjacent time intervals of the sub-time point (e.g., during holidays, during cybersecurity exercises, during system platform maintenance, etc.) and / or special circumstances information about the target complaint within the adjacent time intervals (e.g., the target complaint has just been launched, the target complaint has just been taken offline, the complaint is in an iterative update state, etc.) are determined. Based on the above special circumstances information, corresponding additional description tags for the target complaint are generated.

[0114] By introducing and using the aforementioned multiple complaint feature labels to construct feature elements corresponding to each time point, the causal relationship between the occurrence of complaints at each time point within the current time period can be characterized more comprehensively and precisely. This allows the model to mine and utilize the aforementioned causal relationship when processing time-series feature sequences, and to more accurately and effectively predict future complaint situations.

[0115] Based on the above embodiments, it is possible to efficiently process the massive amount of complaint record data for the current time period, and accurately construct multiple time-series feature sequences that correspond to each complaint handling channel and are suitable for subsequent processing by a preset complaint volume prediction model.

[0116] In some embodiments, after obtaining the corresponding current time-series feature sequence, refer to Figure 3 As shown, in specific implementations, the method may also include the following:

[0117] S1: Retrieve complaint record data for the previous time period regarding the current complaint handling channel; where the current complaint handling channel is the complaint handling channel corresponding to the current data group;

[0118] S2: Based on the complaint record data of the previous time period, generate the current lagged feature sequence for the current time series feature sequence;

[0119] S3: According to the preset processing rules, the current time-series feature sequence and the current lagging feature sequence are concatenated to obtain the current joint feature sequence for the current complaint handling channel;

[0120] S4: By using a preset complaint volume prediction model to process the current joint feature sequence, determine the complaint volume of the current complaint handling channel within a specified time window.

[0121] Specifically, the complaint record data from the previous time period can be understood as the historical complaint record data from a preset time period adjacent to the current time period.

[0122] The inclusion of complaint records from the previous time period is due to two considerations: First, complaints may have a certain lag in their occurrence. For example, if a customer was dissatisfied with a behavior by the target party in the previous time period, they might only file a complaint in the current time period. Second, the processing of complaints may have a certain continuity. For example, if a customer filed a complaint about the target party in the previous time period but was not satisfied with the outcome, they might change the source of their complaint and file another complaint against the same target party in the current time period.

[0123] Based on the above embodiments, by introducing complaint record data from the previous time period and jointly using the time-series feature sequence obtained from complaint record data in the current time period and the lagged feature sequence obtained from complaint record data in the previous time period, a joint feature sequence for the complaint handling channel is constructed. This enables the model to more comprehensively analyze and utilize the long-term trend of complaint changes during operation, thereby more accurately predicting the number of complaints in a specified window for the complaint handling channel in the future, and further improving the model's prediction accuracy.

[0124] In some embodiments, the complaint handling channels may specifically include multiple complaint handling channels. Thus, by utilizing the solution provided in this description and embodiment, the future complaint volume of multiple complaint handling areas on the system platform can be predicted more comprehensively and accurately, avoiding omissions.

[0125] In some embodiments, after determining the number of complaints within a designated time window for the complaint handling channel based on the target prediction results, refer to Figure 4 As shown, in specific implementations, the method may also include the following:

[0126] S1: Detect whether the number of complaints in the designated time window of the complaint handling channel exceeds the preset first complaint threshold, and obtain the corresponding first detection result;

[0127] S2: Based on the first detection result, the complaint handling channels where the number of complaints in a specified time window exceeds the preset first complaint volume threshold are identified as target complaint risk channels;

[0128] S3: Based on the current time-series characteristic sequence of the target complaint risk channels, statistically analyze the proportion distribution of complaint source channels and / or the proportion distribution of target complaint objects within the current time period to obtain the statistical analysis results for the current time period;

[0129] S4: Based on the statistical analysis results of the current time period, determine the matching target complaint handling strategy from the preset complaint handling strategy set;

[0130] S5: Generate a first complaint risk alert regarding the target complaint risk channel; and send the first complaint risk alert and the target complaint handling strategy to the target business department associated with the target complaint risk channel.

[0131] Specifically, the aforementioned preset first complaint threshold can be determined in advance through statistical analysis of a large number of test experiments, taking into account the processing resources of the complaint handling channels and the complexity of the services and / or products involved. In particular, the first complaint threshold can be different for different complaint handling channels.

[0132] In practice, if the number of complaints received by a complaint handling channel during a designated time window exceeds the preset first complaint threshold, it can be determined that the complaint handling channel will face significant processing pressure in handling complaints in the future. There is a high probability that the received complaints will not be processed in a timely and effective manner, which may lead to customer dissatisfaction and even further complaints from customers.

[0133] The aforementioned set of preset complaint strategies may specifically include multiple preset complaint handling strategies; wherein each preset complaint handling strategy corresponds to at least one complaint business scenario.

[0134] In practice, the corresponding complaint business scenario can be determined based on the corresponding mapping rules and the statistical analysis results of the current time period. Then, based on the complaint business scenario, the preset complaint handling strategy set can be retrieved, and the matching preset complaint handling strategy can be determined as the target complaint handling strategy.

[0135] Before implementation, a large number of historical complaint handling records can be collected. Simultaneously, based on these historical records, corresponding extended complaint handling records are generated through simulation. These historical and extended records are then combined to obtain a mixed complaint handling record. Multiple complaint handling samples are extracted from the mixed record. Clustering learning is then performed on these samples to construct multiple preset complaint handling strategies for different complaint business scenarios. These preset strategies are then combined to construct a preset complaint handling set.

[0136] Based on the above embodiments, a first complaint risk alert carrying a matching target complaint handling strategy can be intelligently determined for the target complaint channel, and the first complaint risk alert can be sent to the target business department associated with the target complaint risk channel, so that the relevant department can make targeted preparations for complaint handling in advance, so as to ensure that the received complaint forms can be processed efficiently in the future and improve the customer service experience.

[0137] In some embodiments, after determining the number of complaints within a designated time window for the complaint handling channel based on the target prediction results, refer to Figure 5 As shown, in specific implementations, the method may also include the following:

[0138] S1: Detect whether the number of complaints in the designated time window of the complaint handling channel is greater than the preset second complaint threshold, and obtain the corresponding second detection result;

[0139] S2: Based on the second detection results, the complaint handling channels with a complaint volume exceeding the preset second complaint volume threshold within a specified time window are identified as candidate complaint risk channels;

[0140] S3: Based on the current time-series characteristic sequence of candidate complaint risk channels, count the number of target complaint objects;

[0141] S4: Identify the target risk targets based on the number of target complaints;

[0142] S5: Based on multiple time-series characteristic sequences, identify the relevant complaint risk channels involving the target risk object from multiple complaint handling channels;

[0143] S6: Generate a second complaint risk alert regarding the target risk object; and send the second complaint risk alert to the relevant business departments associated with the relevant complaint risk channels.

[0144] The aforementioned preset second complaint threshold is less than the preset first complaint threshold.

[0145] The aforementioned target risk objects can be specifically understood as risk services and / or risk products that are likely to receive the most and most frequent customer complaints in the near future.

[0146] The aforementioned second complaint risk warning must at least include the object identifier of the target risk object.

[0147] Based on the above embodiments, the target risk object that will cause a surge in complaints in the future can be identified from the source. A second complaint risk warning carrying the object's identifier can be sent in advance to the relevant departments and channels involving the target risk object. This allows the relevant departments and channels to conduct self-inspection and improvement on the services and / or products involved in the target risk object in advance, so as to avoid customers initiating a large number of complaints due to the target risk object in the future. This can effectively reduce the number of complaints in the future from the source and further improve the customer service experience.

[0148] In some embodiments, the specified time window may specifically include multiple time windows corresponding to different time periods. Correspondingly, the method may further include the following in its specific implementation:

[0149] S1: Receive time window adjustment information;

[0150] S2: According to the preset splicing rules, multiple time series feature sequences are spliced ​​with time window adjustment information to obtain multiple spliced ​​time series feature sequences;

[0151] S3: Using a pre-defined complaint volume prediction model, the complaint volume for a specified time window after the adjustment of the complaint handling channel is determined by processing multiple concatenated time-series feature sequences; wherein, the adjusted specified time window is determined based on the time window adjustment information.

[0152] Based on the above embodiments, it is possible to customize and adjust the specific time period corresponding to the specified future time window to meet diverse business needs.

[0153] In some embodiments, the complaint feature labels may further include multi-level labels. For example, customer sentiment statistics labels may specifically include a primary label (e.g., an anger label) to characterize the type of customer sentiment, and a secondary label (e.g., an anger secondary label) to characterize the sentiment level based on the primary label. Using the complaint feature labels to construct feature elements allows the information contained in the feature elements to be more detailed and hierarchical, thereby enabling the model to more accurately predict the number of complaints based on the feature elements.

[0154] In some embodiments, see Figure 6 As shown, in specific implementations, the method may also include the following:

[0155] S1: Obtain and extract the corresponding sample data based on historical complaint records;

[0156] S2: Construct an initial complaint volume prediction model; wherein the initial complaint volume prediction model includes at least: an initial master model based on Prophet, an initial first auxiliary model based on LSTM, and an initial second auxiliary model based on ARIMA, wherein the output of the initial master model is connected to the input of the initial first auxiliary model and the input of the initial second auxiliary model.

[0157] S3: Construct a joint objective loss function for the main model, the first auxiliary model, and the second auxiliary model;

[0158] S4: Based on the objective joint loss function, train an initial complaint volume prediction model using the sample data to obtain a preset complaint volume prediction model that meets the requirements.

[0159] Specifically, the initial master model may include a Prophet-based residual module, which can be connected to the inputs of the initial first auxiliary model and the initial second auxiliary model, respectively. The residual module is also connected to the outputs of the initial first and second auxiliary models, respectively. Both the initial first and second auxiliary models can directly receive the temporal feature sequences input to the model.

[0160] The aforementioned joint objective function is a hybrid loss function that integrates the model errors of the initial main model, the initial first auxiliary model, and the initial second auxiliary model.

[0161] Based on the above model structure, during specific model training, sample data can be output to the initial complaint volume prediction model based on the target joint loss function for ensemble learning. During ensemble learning, multiple loss values ​​can be calculated based on the target joint loss function for the initial main model, the initial first auxiliary model, the initial second service main model, and the initial complaint volume prediction model as a whole. Then, based on these multiple loss values ​​and the model's ensemble weights, the network parameters of the initial main model, the initial first auxiliary model, the initial second auxiliary model, and the residual module of the main model connecting the main model, the first auxiliary model, and the second auxiliary model are adjusted accordingly, until a pre-defined complaint volume prediction model that meets the requirements is obtained based on the target joint loss function.

[0162] Based on the above model structure, when processing time-series feature sequences using the preset complaint volume prediction model, the first primary prediction result can be output by the primary model processing the time-series feature sequences. The residual module then feeds the first primary prediction result into the first and second auxiliary models respectively. The first auxiliary model first obtains the corresponding long-term time-series dependencies based on the input time-series feature sequences. Based on these long-term time-series dependencies, the first primary prediction result is corrected to obtain a first corrected result. This first corrected result is then fed back to the residual module of the primary model. Simultaneously, the second auxiliary model first obtains the corresponding short-term dynamic changes based on the input time-series feature sequences. Based on these short-term dynamic changes, the first primary prediction result is corrected to obtain a second corrected result. This second corrected result is then fed back to the residual module of the primary model. After receiving the first correction result and the second correction result, the residual module of the main model can use the knowledge learned through ensemble learning to determine the corresponding fusion error based on the first main prediction result, the first correction result, and the second correction result. Then, based on the fusion error, the first main prediction result, the first correction result, and the second correction result are fused to obtain the corresponding second main prediction result, which is the final output target prediction result.

[0163] Based on the above embodiments, the advantages and characteristics of different types of model structures can be fully considered. By constructing corresponding joint loss functions and performing ensemble learning, a pre-set complaint volume prediction model that is adapted to complex complaint handling scenarios and has good performance can be efficiently trained.

[0164] In some embodiments, after training an initial complaint volume prediction model using the sample data based on the target joint loss function to obtain a preset complaint volume prediction model that meets the requirements, when a new business scenario arises, the method may further include the following in its specific implementation:

[0165] S1: Obtain sample data for new business scenarios; among them, the amount of sample data for new business scenarios is relatively small;

[0166] S2: Based on the sample data of the new business scenario, perform scenario matching detection on the sample data of the preset complaint volume prediction model previously used to train the old business scenario, and find the sample data of the old business scenario with a scenario matching degree greater than the preset matching degree threshold, as the sample data of the supplementary scenario.

[0167] S3: Mix the sample data of the new business scenario with the sample data of the supplementary scenario to obtain the mixed sample data;

[0168] S4: Using the mixed sample data, transfer learning is performed on the previously preset complaint volume prediction model, and the preset complaint volume prediction model is updated to be suitable for the new business scenario.

[0169] Based on the above embodiments, by using a good amount of sample data from new business scenarios, a new, pre-set complaint volume prediction model that is suitable for new business scenarios and meets the requirements can be efficiently trained through transfer learning on the basis of the previously preset complaint volume prediction model.

[0170] As can be seen from the above, the complaint data processing method provided in the embodiments of this specification can, before specific implementation, construct and train a preset complaint volume prediction model suitable for analyzing and predicting the complaint volume of different complaint handling channels, and simultaneously integrate Prophet, LSTM, and ARIMA. In specific implementation, complaint record data for the current time period is first acquired; wherein, the complaint record data packet for the current time period contains multiple complaint forms received within the current time period; and based on the complaint record data for the current time period, the attribute data of the multiple complaint forms is determined; wherein, the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text; then, based on the complaint handling channel, the multiple complaint forms are divided into multiple data groups; wherein, each data group corresponds to at least one complaint handling channel; based on preset processing rules, multiple time-series feature sequences are constructed according to the multiple data groups and the corresponding complaint form attribute data; wherein, each time-series feature sequence corresponds to at least one complaint handling channel; the multiple time-series feature sequences are processed using the preset complaint volume prediction model to obtain the corresponding target prediction result; based on the target prediction result, the complaint volume of the specified time window for the complaint handling channel is determined. This allows it to be well-suited for complex complaint handling scenarios, efficiently and accurately predicting the number of complaints in each complaint handling channel during designated time windows in the future. This enables the determination of appropriate handling strategies in advance, the allocation of resources, and the preparation of relevant measures to promptly handle received customer complaints and improve customer service experience.

[0171] This specification provides an embodiment of a server, see below. Figure 7 As shown. The server includes a network communication port 701, a processor 702, and a memory 703. These structures are connected by internal cables so that they can perform specific data interaction.

[0172] Specifically, the network communication port 701 can be used to acquire complaint record data for the current time period; wherein the complaint record data packet for the current time period contains multiple complaint forms received within the current time period.

[0173] The processor 702 is specifically configured to determine attribute data of multiple complaint forms based on complaint record data for the current time period; wherein the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text; divide the multiple complaint forms into multiple data groups according to the complaint handling channel; wherein each data group corresponds to at least one complaint handling channel; construct multiple time-series feature sequences based on preset processing rules, according to the multiple data groups and the corresponding complaint form attribute data; wherein each time-series feature sequence corresponds to at least one complaint handling channel; process the multiple time-series feature sequences using a preset complaint volume prediction model to obtain corresponding target prediction results; wherein the preset complaint volume prediction model is a hybrid model integrating Prophet, LSTM, and ARIMA; and determine the complaint volume of a specified time window for the complaint handling channel based on the target prediction results.

[0174] The memory 703 can be used to store the corresponding instruction program and related intermediate data.

[0175] Based on the above methods, the relevant structural performance of the server can be effectively utilized to improve the data processing speed of electronic devices and efficiently realize complaint data processing.

[0176] In this embodiment, the network communication port 701 can be a virtual port bound to different communication protocols, thereby enabling the sending or receiving of different data. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for email data communication. Furthermore, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM or CDMA; it can also be a Wi-Fi chip; or it can be a Bluetooth chip.

[0177] In this embodiment, the processor 702 can be implemented in any suitable manner. For example, the processor can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers, etc. This specification is not limiting.

[0178] In this embodiment, the memory 703 may include multiple layers. In a digital system, anything that can store binary data can be a memory. In an integrated circuit, a circuit with storage function but no physical form is also called a memory, such as RAM, FIFO, etc. In a system, a storage device with a physical form is also called a memory, such as a memory stick, TF card, etc.

[0179] This specification also provides a computer-readable storage medium based on the above-described complaint data processing method. The computer-readable storage medium stores computer program instructions that, when executed, perform the following: acquiring complaint record data for the current time period; wherein the complaint record data for the current time period contains multiple complaint forms received within the current time period; determining attribute data for the multiple complaint forms based on the complaint record data for the current time period; wherein the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text; dividing the multiple complaint forms into multiple data groups based on the complaint handling channel; wherein each data group corresponds to at least one complaint handling channel; constructing multiple time-series feature sequences based on preset processing rules, according to the multiple data groups and the corresponding complaint form attribute data; wherein each time-series feature sequence corresponds to at least one complaint handling channel; processing the multiple time-series feature sequences using a preset complaint volume prediction model to obtain a corresponding target prediction result; wherein the preset complaint volume prediction model is a hybrid model integrating Prophet, LSTM, and ARIMA; and determining the complaint volume for a specified time window of the complaint handling channel based on the target prediction result.

[0180] In this embodiment, the storage medium includes, but is not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Cache, Hard Disk Drive (HDD), or Memory Card. The memory can be used to store computer program instructions. The network communication unit can be an interface configured according to standards specified in the communication protocol for network connection communication.

[0181] In this embodiment, the specific functions and effects implemented by the program instructions stored in the computer-readable storage medium can be explained in comparison with other embodiments, and will not be repeated here.

[0182] This specification also provides a computer program product, comprising at least a computer program, which, when executed by a processor, implements the following method steps: acquiring complaint record data for the current time period; wherein the complaint record data for the current time period contains multiple complaint forms received within the current time period; determining attribute data of the multiple complaint forms based on the complaint record data for the current time period; wherein the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text; dividing the multiple complaint forms into multiple data groups according to the complaint handling channel; wherein each data group corresponds to at least one complaint handling channel; constructing multiple time-series feature sequences based on preset processing rules, according to the multiple data groups and the corresponding complaint form attribute data; wherein each time-series feature sequence corresponds to at least one complaint handling channel; processing the multiple time-series feature sequences using a preset complaint volume prediction model to obtain a corresponding target prediction result; wherein the preset complaint volume prediction model is a hybrid model integrating Prophet, LSTM, and ARIMA; and determining the complaint volume of a specified time window for the complaint handling channel based on the target prediction result.

[0183] See Figure 8 As shown in the embodiments of this specification, a complaint data processing device is also provided, which may specifically include the following structural modules:

[0184] The acquisition module 801 can be specifically used to acquire complaint record data for the current time period; wherein, the complaint record data for the current time period contains multiple complaint forms received within the current time period;

[0185] The first determining module 802 can be used to determine the attribute data of multiple complaint forms based on the complaint record data of the current time period; wherein, the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text;

[0186] The grouping module 803 can be used to divide the multiple complaint forms into multiple data groups according to the complaint handling channels; wherein each data group corresponds to at least one complaint handling channel;

[0187] The construction module 804 can be specifically used to construct multiple time-series feature sequences based on preset processing rules, multiple data groups, and the attribute data of the corresponding complaint forms; wherein, the time-series feature sequences correspond to at least one complaint handling channel;

[0188] The processing module 805 can be used to process multiple time-series feature sequences using a preset complaint volume prediction model to obtain the corresponding target prediction results; wherein, the preset complaint volume prediction model is a hybrid model that integrates Prophet, LSTM and ARIMA.

[0189] The second determining module 806 can be used to determine the number of complaints within a specified time window of the complaint handling channel based on the target prediction results.

[0190] In some embodiments, when the above-mentioned construction module 804 is specifically implemented, it can construct a corresponding current time-series feature sequence based on preset processing rules and the current data group in the following manner: Based on preset processing rules, the current data group is divided into multiple sub-data groups according to the complaint time; wherein, the sub-data group corresponds to a point in time within the current time period; based on the complaint forms in each sub-data group and the corresponding attribute data of the complaint forms, at least the number of complaint forms, the proportion distribution of complaint source channels, and the proportion distribution of target complaint objects in each sub-data group are determined; based on the number of complaint forms, the proportion distribution of complaint source channels, and the proportion distribution of target complaint objects in each sub-data group, complaint feature labels of each sub-data group are determined; based on the complaint feature labels of each sub-data group, multiple corresponding feature elements are determined; multiple feature elements are arranged in chronological order to obtain the corresponding current time-series feature sequence.

[0191] In some embodiments, after obtaining the corresponding current time-series feature sequence, the device can be used to: acquire complaint record data for the previous time period regarding the current complaint handling channel; wherein the current complaint handling channel is the complaint handling channel corresponding to the current data group; generate a current lag feature sequence for the current time-series feature sequence based on the complaint record data for the previous time period; concatenate the current time-series feature sequence and the current lag feature sequence according to preset processing rules to obtain a current joint feature sequence for the current complaint handling channel; and determine the complaint volume for a specified time window of the current complaint handling channel by processing the current joint feature sequence using a preset complaint volume prediction model.

[0192] In some embodiments, the complaint handling channels may specifically include multiple complaint handling channels.

[0193] In some embodiments, after determining the number of complaints within a specified time window of the complaint handling channel based on the target prediction result, the device may further be used to: detect whether the number of complaints within the specified time window of the complaint handling channel is greater than a preset first complaint volume threshold, and obtain a corresponding first detection result; based on the first detection result, identify the complaint handling channel whose number of complaints within the specified time window is greater than the preset first complaint volume threshold as a target complaint risk channel; based on the current time-series characteristic sequence of the target complaint risk channel, statistically analyze the proportion distribution of complaint source channels and / or the proportion distribution of target complaint objects within the current time period of the target complaint risk channel, and obtain the statistical analysis result for the current time period; based on the statistical analysis result for the current time period, determine a matching target complaint handling strategy from a preset set of complaint handling strategies; generate a first complaint risk warning for the target complaint risk channel; and send the first complaint risk warning and the target complaint handling strategy to the target business department associated with the target complaint risk channel.

[0194] In some embodiments, after determining the number of complaints within a specified time window of the complaint handling channel based on the target prediction result, the device may be specifically implemented as follows: detecting whether the number of complaints within the specified time window of the complaint handling channel is greater than a preset second complaint volume threshold, and obtaining a corresponding second detection result; determining, based on the second detection result, complaint handling channels whose complaint volume within the specified time window is greater than the preset second complaint volume threshold as candidate complaint risk channels; counting the number of target complaint objects based on the current temporal feature sequence of the candidate complaint risk channels; determining the target risk object based on the number of target complaint objects; determining the relevant complaint risk channels involving the target risk object from multiple complaint handling channels based on multiple temporal feature sequences; generating a second complaint risk alert regarding the target risk object; and sending the second complaint risk alert to the relevant business departments associated with the relevant complaint risk channels.

[0195] In some embodiments, the device can also be used to: acquire and extract corresponding sample data based on historical complaint record data; construct an initial complaint volume prediction model; wherein the initial complaint volume prediction model includes at least: an initial master model based on Prophet, an initial first auxiliary model based on LSTM, and an initial second auxiliary model based on ARIMA, the output of the initial master model being connected to the input of the initial first auxiliary model and the input of the initial second auxiliary model; construct a target joint loss function for the master model, the first auxiliary model, and the second auxiliary model; and train the initial complaint volume prediction model using the sample data based on the target joint loss function to obtain a preset complaint volume prediction model that meets the requirements.

[0196] It should be noted that the units, devices, or modules described in the above embodiments can be implemented by computer chips or physical entities, or by products with certain functions. For ease of description, the above devices are described by dividing them into various modules according to their functions. Of course, in implementing this specification, the functions of each module can be implemented in one or more software and / or hardware, or the module that implements the same function can be implemented by a combination of multiple sub-modules or sub-units, etc. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection between the devices or units shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0197] As can be seen from the above, the complaint data processing device provided in the embodiments of this specification can be well adapted to complex complaint handling scenarios. It can efficiently and accurately predict the number of complaints in a specified time window for each complaint handling channel in the future, so as to determine the matching handling strategy in advance, allocate resources, make relevant preparations, and handle the received customer complaints in a timely manner, thereby improving the customer service experience.

[0198] In a specific scenario example, the complaint data processing method provided in this manual is applied to predict the volume of bank customer complaints across all channels, optimized for a small sample scenario. The specific implementation process can be found below.

[0199] This scenario example employs a method for predicting the volume of customer complaints across all channels in a bank, optimized for small-sample scenarios. It presents the design of a customer complaint volume prediction model that integrates time-series multi-dimensional features. Real-time updates of product information from various bank business lines are incorporated to optimize the model for new business scenarios. Finally, the complaint volume prediction information is displayed by channel, and warnings are issued for channels exceeding set thresholds, effectively improving the precision and efficiency of customer complaint management by bank management departments. Specific implementation may include the following steps.

[0200] Step 1: Establish a tiered list of channels for handling bank customer complaints and channels for categorizing complaints.

[0201] This paper presents a statistical classification label design method for classifying historical complaint data and predicting across all channels. Label categories include complaint processing channels (e.g., complaint handling channels) and complaint source channels. Each label (complaint feature label) category can have single or multiple layers of labels, specifically subdividing the channels for subsequent complaint processing and the sources of complaint data. This allows for refined statistical analysis and prediction of complaints across all channels.

[0202] Step 2: Generation of time-series-based multidimensional fusion feature dataset.

[0203] Collect historical complaint data text information from all banking channels, and obtain automatically generated tag information such as customer tags, complaint business tags, complaint business processing channels, complaint reasons, complaint products, and complaint handling departments based on NLP technology.

[0204] Based on the complaint handling channel category, historical complaint data is aggregated and calculated in chronological order to construct a dataset (e.g., a chronological feature sequence) containing multi-dimensional chronological feature labels such as date information, complaint volume information, complaint source channel information, whether there is product launch, delisting or change information, business channel information to which the product change belongs, and holiday and workday identification information.

[0205] Step 3: Use a time series model to train and fit the historical dataset.

[0206] The multidimensional fused time-series feature dataset is trained and fitted using time-series models. These time-series models include, but are not limited to, the Prophet model, the ARIMA differential ensemble moving average autoregressive model; tree-based regression models such as random forest, XGboost, lightgbm, and Catboost; Seq2Seq models such as LSTM long short-term memory network and Transformer neural network; and ensemble learning models based on the above models. Finally, the time-series model with the best prediction performance (e.g., the preset complaint volume prediction model) is selected.

[0207] In practical implementation, taking the prediction of categorized complaints based on the Prophet model as an example, the core input data includes timestamps (by day) and historical complaint volumes by category. External variables such as whether it's a holiday, whether a new product is launched, weekday / holiday status, and daily channel sales volume are added. Simultaneously, lagged feature fusion (such as complaint volume from the previous N days, complaint growth rate from the previous N days, etc.) is performed to optimize the model at the data feature level. The residuals output by the Prophet model are then used to train the model, and the output of the residual prediction model is used as feature input to the Prophet model to predict the original complaint volume, correcting system errors. Ensemble learning, on the other hand, integrates LSTM, ARIMA, and other models based on the Prophet main model, using error weights to reduce the prediction error of a single model.

[0208] Step 4: Predict and optimize the number of complaints for each business processing channel.

[0209] Using a pre-trained prediction model, the number of complaints for different business processing channels at a set time granularity is predicted, and the predicted complaint volume for each channel at different time windows is output. The prediction model is also input with real-time data on the launch, delisting, and changes of products from various business lines of the bank, and real-time statistical information is collected by channel. The prediction model is optimized for small sample launch of new business scenarios, adding time-series characteristics of the business and improving the accuracy of the prediction.

[0210] Step 5: Display and alert on the volume of complaints across all channels.

[0211] The predicted complaint volume, along with corresponding channels and time-granularity tags, is added to a database. Visualization is then achieved through querying, filtering, statistics, and graphing. This enhances the management department's evaluation and management of complaint handling quality. Furthermore, separate warning thresholds can be set for complaint volumes from different channels. When these thresholds are exceeded, the monitoring and management personnel for the corresponding channels receive an alert, prompting them to prepare appropriate contingency plans. This improves the efficiency and accuracy of complaint volume management and provides a basis for measures to improve product and service practices.

[0212] The above scenario examples validate the complaint data processing method provided in this manual. By utilizing artificial intelligence technology to predict the volume of complaints from bank customers across all channels, the method can anticipate the inflow of complaints through each channel, improving the efficiency of complaint triage and processing by bank management. Furthermore, in practical application scenarios, the method addresses the impact of changes in business scenarios and the small sample size of newly added business scenarios on prediction performance. By introducing time-series business feature information, the method eliminates the need to retrain the prediction model, improving its accuracy and reducing iteration costs after the model goes live.

[0213] While this specification provides the steps of operation for the methods described in the embodiments or flowcharts, more or fewer steps may be included based on conventional or non-inventive means. The order of steps listed in the embodiments is merely one possible order of execution among many steps and does not represent the only possible order. In actual device or client product execution, the methods shown in the embodiments or drawings may be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment, or even a distributed data processing environment). The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitations, the presence of other identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded. The terms "first," "second," etc., are used to denote names and do not indicate any particular order.

[0214] Those skilled in the art will also know that, besides implementing the controller using purely computer-readable program code, the same functions can be achieved by logically programming the method steps, making the controller function as logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers (PLCs), and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices within it used to implement various functions can also be considered structures within that hardware component. Alternatively, the devices used to implement various functions can be considered as both software modules implementing the method and structures within a hardware component.

[0215] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside on local and remote computer-readable storage media, including storage devices.

[0216] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this specification can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions of this specification can essentially be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, mobile terminal, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments of this specification.

[0217] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. This specification can be used in numerous general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0218] Although this specification has been described by way of examples, those skilled in the art will recognize that many variations and modifications are possible without departing from the spirit of this specification, and it is intended that the appended claims cover such variations and modifications without departing from the spirit of this specification.

Claims

1. A method for processing complaint data, characterized in that, include: Obtain complaint record data for the current time period; wherein the complaint record data for the current time period contains multiple complaint forms received within the current time period; Based on the complaint record data for the current time period, the attribute data of multiple complaint forms is determined; wherein, the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text; Based on the complaint handling channels, the multiple complaint forms are divided into multiple data groups; each data group corresponds to at least one complaint handling channel. Based on preset processing rules, multiple time-series feature sequences are constructed according to multiple data groups and the attribute data of the corresponding complaint forms; wherein each time-series feature sequence corresponds to at least one complaint handling channel. Multiple time-series feature sequences are processed using a pre-defined complaint volume prediction model to obtain corresponding target prediction results; wherein, the pre-defined complaint volume prediction model is a hybrid model that integrates Prophet, LSTM, and ARIMA. Based on the target prediction results, the number of complaints during the designated time window for the complaint handling channel is determined.

2. The method according to claim 1, characterized in that, Based on preset processing rules, multiple time-series feature sequences are constructed according to multiple data groups and the attribute data of corresponding complaint forms. This includes constructing a corresponding current time-series feature sequence based on the current data group according to the preset processing rules in the following manner: Based on preset processing rules, the current data group is divided into multiple sub-data groups according to the complaint time; wherein, the sub-data group corresponds to a point in time within the current time period; Based on the complaint forms in each sub-data group and the corresponding attribute data of the complaint forms, at least the number of complaint forms, the proportion distribution of complaint source channels, and the proportion distribution of target complaint objects in each sub-data group should be determined. Based on the number of complaint forms, the proportion of complaint source channels, and the proportion of target complaint subjects in each sub-data group, the complaint feature labels for each sub-data group are determined. Based on the complaint feature labels of each sub-data group, multiple corresponding feature elements are determined; Arrange multiple feature elements in chronological order to obtain the corresponding current time-series feature sequence.

3. The method according to claim 2, characterized in that, After obtaining the corresponding current time-series feature sequence, the method further includes: Retrieve complaint record data for the previous time period regarding the current complaint handling channel; where the current complaint handling channel refers to the complaint handling channel corresponding to the current data group. Based on the complaint record data of the previous time period, generate the current lagged feature sequence for the current time series feature sequence; According to the preset processing rules, the current time-series feature sequence and the current lagging feature sequence are spliced ​​together to obtain the current joint feature sequence for the current complaint handling channel; By using a pre-defined complaint volume prediction model and processing the current joint feature sequence, the complaint volume of the current complaint handling channel within a specified time window is determined.

4. The method according to claim 1, characterized in that, The complaint handling channels include multiple channels.

5. The method according to claim 4, characterized in that, After determining the number of complaints within a designated time window for the complaint handling channel based on the target prediction results, the method further includes: The system detects whether the number of complaints received during the designated time window of the complaint handling channel exceeds a preset first complaint threshold, and obtains the corresponding first detection result. Based on the first test results, complaint handling channels where the number of complaints in a specified time window exceeds the preset first complaint volume threshold are identified as target complaint risk channels. Based on the current time-series characteristic sequence of the target complaint risk channels, statistical analysis is performed on the proportion distribution of complaint source channels and / or the proportion distribution of target complaint targets within the current time period to obtain the statistical analysis results for the current time period. Based on the statistical analysis results of the current time period, a matching target complaint handling strategy is determined from the preset complaint handling strategy set; Generate a first complaint risk alert regarding the target complaint risk channel; and send the first complaint risk alert and the target complaint handling strategy to the target business department associated with the target complaint risk channel.

6. The method according to claim 4, characterized in that, After determining the number of complaints within a designated time window for the complaint handling channel based on the target prediction results, the method further includes: The system detects whether the number of complaints received during the designated time window of the complaint handling channel exceeds a preset second complaint threshold, and obtains the corresponding second detection result. Based on the second test results, complaint handling channels with a complaint volume exceeding the preset second complaint volume threshold within a specified time window are identified as candidate complaint risk channels. Based on the current time-series characteristics of candidate complaint risk channels, count the number of target complaint subjects; Based on the number of target complaints, target risk targets are identified; Based on multiple time-series characteristic sequences, relevant complaint risk channels involving the target risk object are identified from multiple complaint handling channels; Generate a second complaint risk alert regarding the target risk object; and send the second complaint risk alert to the relevant business departments associated with the relevant complaint risk channels.

7. The method according to claim 1, characterized in that, The method further includes: Obtain and extract corresponding sample data based on historical complaint records; Construct an initial complaint volume prediction model; wherein the initial complaint volume prediction model includes at least: an initial master model based on Prophet, an initial first auxiliary model based on LSTM, and an initial second auxiliary model based on ARIMA, wherein the output of the initial master model is connected to the input of the initial first auxiliary model and the input of the initial second auxiliary model; Construct a joint objective loss function for the main model, the first auxiliary model, and the second auxiliary model; Based on the objective joint loss function, an initial complaint volume prediction model is trained using the sample data to obtain a preset complaint volume prediction model that meets the requirements.

8. A complaint data processing device, characterized in that, include: The acquisition module is used to acquire complaint record data for the current time period; wherein, the complaint record data for the current time period contains multiple complaint forms received within the current time period; The first determining module is used to determine the attribute data of multiple complaint forms based on the complaint record data of the current time period; wherein, the attribute data includes at least: complaint handling channel, complaint source channel, complaint time, and complaint content text; The grouping module is used to divide the multiple complaint forms into multiple data groups according to the complaint handling channels; wherein each data group corresponds to at least one complaint handling channel; The construction module is used to construct multiple time-series feature sequences based on preset processing rules, multiple data groups, and the attribute data of corresponding complaint forms; wherein each time-series feature sequence corresponds to at least one complaint handling channel. The processing module is used to process multiple time-series feature sequences using a preset complaint volume prediction model to obtain the corresponding target prediction results; wherein, the preset complaint volume prediction model is a hybrid model that integrates Prophet, LSTM and ARIMA. The second determining module is used to determine the number of complaints within a specified time window of the complaint handling channel based on the target prediction results.

9. A server, characterized in that, It includes a processor and a memory for storing processor-executable instructions, wherein the processor, when executing the instructions, implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images