Complaint quantity prediction method and device, equipment, storage medium and product

Abstract

The invention discloses a complaint amount prediction method and device, equipment, a storage medium and a product, belongs to the field of data processing, and is used for improving the accuracy of predicted complaint amount. The method comprises the steps that historical complaint amount time sequence data are acquired, the historical complaint amount time sequence data comprise historical complaint amount data of a plurality of continuous time points, and the historical complaint amount data comprise complaint time and complaint amount; performing feature extraction on the historical complaint amount time sequence data to obtain a time sequence feature vector which is a feature vector of the historical complaint amount time sequence data; frequency domain conversion operation is carried out on the time sequence feature vector, a time sequence period and a period feature vector are determined based on a frequency domain conversion result, the time sequence period is a period included by the time sequence feature vector, and the period feature vector is a feature vector of the time sequence feature vector in the time sequence period; and inputting the periodic feature vector into a pre-trained complaint amount prediction model to obtain a complaint amount prediction result.

Description

Technical Field

[0001] This application belongs to the field of data processing, specifically relating to a method, apparatus, equipment, storage medium, and product for predicting the number of complaints. Background Technology

[0002] In recent years, with the development of artificial intelligence technology, complaint volume prediction has been widely used in areas such as customer service optimization and resource allocation. Complaint volume prediction mainly involves analyzing historical data to forecast the number of complaints in the future, providing a basis for corporate decision-making. Currently, most complaint volume prediction methods employ traditional statistical analysis methods, such as moving averages, ARIMA models (time series algorithms), and regression analysis. While these methods can capture the periodicity and trends of complaint data and quantify the impact of single factors, their accuracy is low when dealing with non-linear and complex complaint data.

[0003] Therefore, a method is needed to improve the accuracy of predicting complaint volume. Summary of the Invention

[0004] This application provides a method for predicting the number of complaints, which can improve the accuracy of the predicted number of complaints. In a first aspect, embodiments of this application provide a method for predicting the number of complaints. The method includes: acquiring historical complaint volume time-series data, wherein the historical complaint volume time-series data includes historical complaint volume data at multiple consecutive time points, and the historical complaint volume data includes complaint time and complaint volume; extracting features from the historical complaint volume time-series data to obtain a time-series feature vector, wherein the time-series feature vector is a feature vector of the historical complaint volume time-series data; performing a frequency domain transformation operation on the time-series feature vector, and determining a time-series period and a periodic feature vector based on the frequency domain transformation result, wherein the time-series period is the period included in the time-series feature vector, and the periodic feature vector is a feature vector of the time-series feature vector within the time-series period; and inputting the periodic feature vector into a pre-trained complaint volume prediction model to obtain a complaint volume prediction result. Secondly, embodiments of this application provide a complaint volume prediction device, which includes: a first acquisition module, configured to acquire historical complaint volume time-series data, the historical complaint volume time-series data including historical complaint volume data at multiple consecutive time points, the historical complaint volume data including complaint time and complaint volume; a first extraction module, configured to perform feature extraction on the historical complaint volume time-series data to obtain a time-series feature vector, the time-series feature vector being the feature vector of the historical complaint volume time-series data; a first determination module, configured to perform a frequency domain transformation operation on the time-series feature vector, and determine a time-series period and a periodic feature vector based on the frequency domain transformation result, the time-series period being the period included in the time-series feature vector, the periodic feature vector being the feature vector of the time-series feature vector within the time-series period; and a first prediction module, configured to input the periodic feature vector into a pre-trained complaint volume prediction model to obtain a complaint volume prediction result.

[0005] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0006] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0007] Fifthly, embodiments of this application provide a computer program product that, when executed by a processor, implements the steps of the method described in the first aspect.

[0008] In a sixth aspect, embodiments of this application provide a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the method as described in the first aspect.

[0009] In this embodiment, historical complaint volume time-series data is acquired, which includes complaint volume data at multiple consecutive time points, including complaint time and complaint volume. Feature extraction is performed on the historical complaint volume time-series data to obtain a time-series feature vector, which is the feature vector of the historical complaint volume time-series data. A frequency domain transformation operation is performed on the time-series feature vector, and the time-series period and periodic feature vector are determined based on the frequency domain transformation result. The time-series period is the period included in the time-series feature vector, and the periodic feature vector is the feature vector of the time-series feature vector within the time-series period. The periodic feature vector is input into a pre-trained complaint volume prediction model to obtain the complaint volume prediction result, which can improve the accuracy of the predicted complaint volume. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a method for predicting the number of complaints provided in an embodiment of this application; Figure 2 This is a flowchart illustrating the second method for predicting complaint volume provided in this application embodiment; Figure 3 This is a flowchart illustrating the third method for predicting complaint volume provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of a complaint volume prediction system provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a complaint volume prediction device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a complaint volume prediction device provided in an embodiment of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0013] The complaint volume prediction method provided in this application will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0014] Figure 1 This illustration shows an embodiment of a complaint volume prediction method provided by the present invention. The method can be executed by an electronic device, which may include a server and / or a terminal device, wherein the terminal device may be, for example, a vehicle-mounted terminal or a mobile phone terminal. In other words, the method can be executed by software or hardware installed in the complaint volume prediction device, and the method includes the following steps: Step 102: Obtain historical complaint volume time series data.

[0015] The historical complaint volume time series data includes historical complaint volume data at multiple consecutive time points, and the historical complaint volume data includes complaint time and complaint volume.

[0016] The entity implementing the complaint volume prediction method described in this application can be a complaint volume prediction system, complaint volume prediction software, or other entities. Here, we will take a complaint volume prediction system as an example.

[0017] The complaint volume prediction system acquires historical complaint volume time-series data, which consists of complaint volume data from multiple consecutive time points. The historical complaint volume data is the complaint volume data at a specific point in time (or time period), including both the complaint time and the number of complaints. Therefore, historical complaint volume time-series data is determined by arranging historical complaint volume data from multiple consecutive time points (or time periods) in chronological order.

[0018] Furthermore, the complaint volume prediction data can be data from the same province, city, and business type, mainly including the date (accurate to the hour, i.e., hourly statistics) and the number of complaints. Additionally, the complaint volume prediction data can also include a reference value (calculated as the average number of complaints in the same dimension at the current time over the past 30 days, excluding holiday data) and a date type (whether it's a working day, where working days take into account compensatory leave).

[0019] After obtaining historical complaint volume time series data, the complaint volume prediction model can perform data cleaning on the historical complaint volume time series data, for example, by using linear interpolation to supplement the missing values ​​in the historical complaint volume time series data.

[0020] Specifically, the complaint volume prediction system first determines the complaint volume data for multiple consecutive time points (or time periods), such as the complaint volume data for each hour within 30 days. Then, it integrates the complaint volume data for multiple consecutive time points (or time periods) and determines the overall data obtained from the integration as the historical complaint volume time series data.

[0021] Step 104: Extract features from the historical complaint volume time series data to obtain a time series feature vector.

[0022] The time-series feature vector is the feature vector of the historical complaint volume time-series data.

[0023] After acquiring historical complaint volume time series data, the complaint volume prediction system performs feature extraction on the historical complaint volume time series data to obtain a time series feature vector, where the time series feature vector is the feature vector corresponding to the entire historical complaint volume time series data.

[0024] In other words, the complaint volume prediction system treats historical complaint volume time-series data as a whole, determines the feature vector corresponding to the whole, and uses the determined feature vector as the time-series feature vector. When determining the time-series feature vector, the complaint volume prediction system can directly determine the feature vector corresponding to the whole by treating historical complaint volume time-series data as a whole. It can also determine multiple historical complaint volume data points included in the historical complaint volume time-series data, determine the sub-feature vectors corresponding to each historical complaint volume data point, and then integrate these sub-feature vectors to determine the feature vector corresponding to the historical complaint volume time-series data (i.e., the time-series feature vector).

[0025] Specifically, when performing feature extraction based on historical complaint volume time-series data and directly determining the time-series feature vector based on the extraction results, the time-series feature vector determined by the complaint volume prediction system can be x_enc, where the dimension of x_enc is [B, T, N] (B=32, T=seq_len, N=3), and seq_len is the data length of the historical complaint volume time-series data.

[0026] Step 106: Perform a frequency domain transformation operation on the time-series feature vector, and determine the time-series period and periodic feature vector based on the frequency domain transformation result.

[0027] Wherein, the time series period is the period included in the time series feature vector, and the periodic feature vector is the feature vector of the time series feature vector within the time series period.

[0028] After determining the time-series feature vector corresponding to the historical complaint volume time-series data, the complaint volume prediction system performs frequency domain transformation on the time-series feature vector and determines the time-series period and periodic feature vector based on the frequency domain transformation result. The time-series period is determined by dividing the time-series feature vector into periods; that is, the time-series period is the period included in the time-series feature vector. The periodic feature vector is the feature vector corresponding to the time-series period within the overall time-series feature vector; that is, the periodic feature vector is a subset of the feature vectors within the time-series period. Furthermore, the number of time-series periods included in the time-series feature vector can be one or more, therefore the determined periodic feature vector data can also be one or more, but each time-series period has a corresponding periodic feature vector.

[0029] When determining the time series period and the periodic feature vector, the complaint volume prediction system first performs a frequency domain transformation operation on the time series feature vector to obtain frequency domain features. Then, it performs period identification on the frequency domain features to determine one or more periods. Then, based on the start and end times of the period, it determines the feature vectors included between the start and end times from the time series feature vectors, and determines the determined part of the feature vectors as the periodic feature vectors of the time series period.

[0030] Specifically, when determining multiple time series periods based on the frequency domain conversion results, the multiple time series periods determined by the complaint volume prediction system may or may not have overlapping time series periods. Here, time overlap is used to characterize that the time periods included in two time series periods overlap. For example, if the first time series period is from 0:00 to 23:00 on December 10 and the second time series period is from 12:00 on December 10 to 12:00 on December 11, then it can be determined that the two time series periods have overlapping time series (i.e., from 12:00 on December 10 to 23:00 on December 10).

[0031] Furthermore, when identifying periods based on frequency domain transformation results (frequency domain features), the complaint volume prediction model identifies periods according to frequency domain features, rather than according to a pre-defined, single, fixed period identification method. This results in a greater diversity of time series period types, enabling complaint volume prediction based on various period feature vectors. For example, the identified time series periods can be daily periods, weekly periods, data change periods, etc. Daily periods can be determined based on different numbers of days, weekly periods can be determined based on different weeks, and data change periods can be determined based on data change patterns.

[0032] Step 108: Input the periodic feature vector into the pre-trained complaint volume prediction model to obtain the complaint volume prediction result.

[0033] After determining the periodic feature vector corresponding to the time series period, the complaint volume prediction system inputs the periodic feature vector into the pre-trained complaint volume prediction model so that the complaint volume prediction model can make predictions based on the input periodic feature vector and output the prediction results, where the prediction results are the number of complaints corresponding to a future time point (or time period).

[0034] Specifically, when the complaint volume prediction system determines multiple periodic feature vectors, it can input all of these feature vectors into a pre-trained complaint volume prediction model, enabling the model to predict complaint volumes. Because the multiple periodic feature vectors input to the model correspond to various types of time-series periods, the pre-trained model can make predictions based on these diverse periodic feature vectors, thereby improving the accuracy of complaint volume prediction.

[0035] The complaint volume prediction model is trained based on sample periodic feature vectors. These sample periodic feature vectors can also contain feature vectors corresponding to periods with overlapping time intervals. During model training, the complaint volume prediction system predicts the number of complaints based on the sample periodic feature vectors, obtaining the predicted sample complaint volume. Then, based on the actual sample complaint volume and the predicted sample complaint volume, the model is iteratively updated to obtain the pre-trained complaint volume prediction model.

[0036] The complaint volume prediction method provided in this invention obtains historical complaint volume time-series data, which includes complaint volume data at multiple consecutive time points, including complaint time and complaint volume. Features are extracted from the historical complaint volume time-series data to obtain a time-series feature vector, which is the feature vector of the historical complaint volume time-series data. A frequency domain transformation operation is performed on the time-series feature vector, and a time-series period and a periodic feature vector are determined based on the frequency domain transformation result. The time-series period is the period included in the time-series feature vector, and the periodic feature vector is the feature vector of the time-series feature vector within the time-series period. The periodic feature vector is input into a pre-trained complaint volume prediction model to obtain the complaint volume prediction result, which can improve the accuracy of the predicted complaint volume.

[0037] In one implementation, the step of extracting features from the historical complaint volume time-series data to obtain a time-series feature vector (step 104) can be performed by steps A1-A3: Step A1: Normalize the number of complaints and the time of complaints in the historical complaint data to obtain the first time series feature and the second time series feature.

[0038] When extracting features from historical complaint volume time series data to obtain time series feature vectors, the complaint volume prediction system first normalizes the complaint volume and complaint time in each historical complaint volume data included in the historical complaint volume data to obtain the first time series feature and the second time series feature.

[0039] In other words, the complaint volume prediction system first normalizes the various types of data included in the historical complaint volume time series data to obtain the corresponding time series characteristics. When the data types included in the historical complaint volume time series data are complaint time and complaint volume, the system normalizes the complaint volume and complaint time separately to obtain the corresponding first and second time series characteristics. When the historical complaint volume data includes complaint time, complaint volume, reference value, and date type, the system normalizes the complaint volume and reference value to obtain the first time series characteristic, and normalizes the complaint time and date type to obtain the second time series characteristic.

[0040] Specifically, during normalization, the complaint volume prediction system determines the standardized parameter value μ and variance of the complaint volume and reference value, and then performs Z-score standardization on the data. When processing complaint time, the system processes the date information, parsing hourly time information into structured features. The table below shows the corresponding structured features for the date.

[0041]

[0042] After structuring the complaint time, the complaint volume prediction system standardizes the extracted time features, encoding the corresponding time features as values ​​in the range of [-0.5, 0.5]. This transforms the original time information into numerical features suitable for model training, helping to capture periodic patterns in the time dimension and providing basic support for time series data analysis and prediction.

[0043] Specifically, the complaint volume prediction system can determine x_enc as the first time series feature and x_mark_enc as the second time series feature. The dimensions of x_mark_enc are [B, T, N] (B=32, T=seq_len, N=4), where seq_len is the data length of the historical complaint volume time series data.

[0044] Step A2: Perform location encoding processing on the historical complaint data to obtain the third time-series feature.

[0045] When extracting features from historical complaint volume time series data to obtain time series feature vectors, the complaint volume prediction system can also perform location encoding processing on each historical complaint volume data in the historical complaint volume time series data to obtain location encoding (i.e., the third time series feature).

[0046] Specifically, the complaint volume prediction system encodes the first time-series feature x_enc and the second time-series feature x_mark_enc, converting the original data values ​​into a feature representation of a unified dimension. That is, it extracts local features of the original sequence through the convolutional layer (3*3 convolutional kernel) in value_embedding, maps them to the target dimension d_model (e.g., set to 16 dimensions), and then uses position_embedding to generate position encoding, giving the sequence temporal position information. At the same time, it maps the realized features to the same dimension (16 dimensions), where d_model is the target embedding dimension.

[0047] Step A3: Determine the time series feature vector based on the first time series feature, the second time series feature, and the third time series feature.

[0048] After determining the first, second, and third time-series features, the complaint volume prediction system can add the first, second, and third time-series features together and determine the result as the time-series feature vector.

[0049] Specifically, the complaint volume prediction system adds up the three feature components (sequence value embedding, time feature embedding, and location embedding) and regularizes them using Dropout (dropout rate 0.1) to output a unified feature vector that integrates temporal patterns, temporal context, and location dependencies.

[0050] When determining the time-series feature vector based on the first, second, and third time-series features, the time-series feature vector determined by the complaint volume prediction system can be enc_emb, where the dimension of enc_emb is [B, T, N] (B=32, T=seq_len, N=3), and seq_len is the data length of the historical complaint volume time-series data [B, T, N] (B=32, T=seq_len, N=3), where seq_len is the data length of the historical complaint volume time-series data.

[0051] In one implementation, determining the time-series feature vector based on the first time-series feature, the second time-series feature, and the third time-series feature (step A3) can be performed via steps B1-B3: Step B1: Add the first time series feature, the second time series feature, and the third time series feature together to obtain the fourth time series feature.

[0052] The fourth time-series feature is used to characterize the original data features of the historical complaint volume data.

[0053] After determining the first, second, and third time-series features, the complaint volume prediction system can add the first, second, and third time-series features together and determine the result as the fourth time-series feature. The fourth time-series feature is used to characterize the original data features of the historical complaint volume time-series data (or multiple historical complaint volume data included in the historical complaint volume time-series data).

[0054] In other words, the fourth time series feature is a data feature directly determined by normalization or feature extraction operations based on the historical complaint volume time series data (or multiple historical complaint volume data included in the historical complaint volume time series data).

[0055] Step B2: Input the fourth temporal feature into a preset self-attention network to obtain the fifth temporal feature.

[0056] The fifth time-series feature is used to characterize the long-distance data features of the historical complaint volume data.

[0057] After determining the fourth time-series feature, the complaint volume prediction system can directly identify the fourth time-series feature as the time-series feature vector. It can also continue processing based on the fourth time-series feature to determine the time-series feature vector based on the results of the continued processing.

[0058] When further processing the fourth time-series feature, the complaint volume prediction system can input the fourth time-series feature into a preset self-attention network to process the fourth time-series feature through the self-attention network and obtain the output result, namely the fifth time-series feature. The fifth time-series feature is used to characterize the long-distance data features of historical complaint volume time-series data (or multiple historical complaint volume data included in the historical complaint volume time-series data).

[0059] In other words, the complaint volume prediction system inputs the fourth time series feature into a preset self-attention network so that the self-attention network can determine the long-distance data features of the historical complaint volume time series data (or multiple historical complaint volume data included in the historical complaint volume time series data) based on the fourth time series feature.

[0060] Specifically, when determining the fifth temporal feature based on a self-attention network, the complaint volume prediction system first performs linear projection on the input fourth temporal feature using ProbAttention to generate a Q / K,V matrix. Then, it employs a sparse sampling strategy to calculate the sparsity score for each query vector, retaining only the k = log(T) query vectors with the highest scores. The remaining query vectors are approximated through interpolation or shared weights, reducing the computational complexity from O(T^2) to O(K^2). 2 The value is reduced to O(TlogT). Then, the complaint volume prediction system is based on the attention mechanism Attention(Q, K, V) = softmax(QK). T / √d)*V,where QK T The algorithm is executed only on k queries, and the remaining locations are approximated and accelerated through feature mapping, resulting in the fifth time-series feature.

[0061] Specifically, the fifth time-series feature determined by the complaint volume prediction system can be represented as x_attn, where the dimension of x_attn is [B, T, N] (B=32, T=seq_len, N=3), and seq_len is the data length of the historical complaint volume time-series data.

[0062] Step B3: Determine the time series feature vector based on the fourth time series feature and the fifth time series feature.

[0063] After determining the fourth and fifth time-series features, the complaint volume prediction system determines the time-series feature vector based on the fourth and fifth time-series features. For example, the fourth and fifth time-series features are added together, and the sum is determined as the time-series feature vector.

[0064] Specifically, the complaint volume prediction system adds the fourth and fifth time-series features and then normalizes them using LayerNorm to obtain the output data time-series feature vector. This allows for lossless fusion of the long-distance features of historical complaint volume time-series data (or multiple historical complaint volume data included in the historical complaint volume time-series data) with the original time-series information.

[0065] Specifically, when determining the time series feature vector based on the fourth and fifth time series features, the complaint volume prediction system can determine X by adding the fourth and fifth time series features, i.e., X = enc_emb + x_attn.

[0066] In one implementation, the step of performing a frequency domain transformation on the time-series feature vector and determining the time-series period and periodic feature vector based on the frequency domain transformation result (step 106) can be performed by steps C1-C4: Step C1: Perform a frequency domain transformation operation on the time-series feature vector to obtain the time-series frequency features.

[0067] Wherein, the time-series frequency feature is the frequency feature of the time-series feature vector.

[0068] The complaint volume prediction system performs a frequency domain transformation operation on the time series feature vector to obtain the frequency domain transformation result, which is the time series frequency feature, where the time series frequency feature is the frequency feature of the time series feature vector.

[0069] Specifically, the complaint volume prediction system performs a Fast Fourier Transform on the time-series feature vector based on Informer self-attention in the time dimension (i.e., dim=1), converting the time-domain signal into a frequency-domain representation. It then extracts the amplitude and phase from the output frequency domain, both of which have dimensions of [B, T / 2, N] (B=32, T=seq_len, N=3), where seq_len is the data length of the historical complaint volume time-series data.

[0070] Furthermore, in order to improve computational efficiency and reduce the interference of high-frequency noise on periodic identification, the complaint volume prediction system adopts the method of averaging the amplitude and phase along the feature dimension and compressing it into a one-dimensional frequency domain feature, resulting in a frequency feature of dimension [B, (T / 2)+1, N] (B=32, T=seq_len, N=3), which retains the overall energy and phase trend of each frequency component.

[0071] Step C2: Perform a periodicity identification operation on the time-series frequency characteristics to determine multiple time-series periods.

[0072] After determining the time-series frequency domain characteristics, the complaint volume prediction system performs periodic identification on these characteristics to obtain multiple time-series periods. Specifically, the system inputs the compressed amplitude and phase features into a multilayer perceptron (MLP) for nonlinear transformation. The MLP automatically learns the mapping relationship between frequency domain features and period importance through two layers of linear mapping and the ReLU activation function. Through backpropagation optimization, it adaptively identifies key periods (such as the 24-hour daily period and the 7-day weekly period) and noise components.

[0073] Step C3: Extract features from the historical complaint volume data within each of the aforementioned time periods to obtain complaint volume data features.

[0074] After determining multiple time series periods, the complaint volume prediction system performs feature extraction on the historical complaint volume data within each time series period, thereby determining the complaint volume data characteristics corresponding to each time series period.

[0075] Specifically, during feature extraction, the complaint volume prediction system iterates through the time series data corresponding to each historical complaint volume data period in the historical complaint volume time series data. When the length of the complaint volume data feature is not an integer multiple of period, the complaint volume data feature is filled to the nearest multiple, and then it is reshaped into a four-dimensional structure of [B, N, num_periods, period] (num_periods=T / period, T=seq_len). Then, the complaint volume prediction system extracts features within the period through a two-dimensional convolution module. This module adopts a dual-branch structure of the Inception architecture, with each branch containing 6 2D convolution kernels of different scales (1*1 to 11*11). It captures patterns of different granularities within the period through parallel multi-scale convolution (such as 1*1 convolution to extract point features; 3*3 convolution to extract local features; and 7*7 and above convolution to capture long-distance dependencies). The first Inception layer expands the number of channels from N to 2N (such as 16-32) to enhance the feature representation capability. After GELU activation, the second Inception layer compresses the number of channels back to N (such as 32-16) to balance the number of parameters. After convolution, it is inversely reshaped to the original dimension and truncated back to length T, thereby determining the complaint volume data features.

[0076] Step C4: Determine the periodic feature vector based on the complaint volume data characteristics and the time-series feature vector.

[0077] After determining the complaint volume data characteristics and time series feature vectors, the complaint volume prediction system can add the complaint volume data characteristics and the feature vector corresponding to the time series period in each time series period, and determine the sum as the periodic feature vector. Alternatively, it can directly determine the complaint volume data characteristics of each time series period as the periodic feature vector of that time series period.

[0078] In one implementation, determining the periodic feature vector based on the complaint volume data characteristics and the time-series feature vector (step C4) can be performed by executing steps D1-D3: Step D1: Score each time series period based on the preset periodic scoring strategy to obtain the periodic score.

[0079] After determining the periodic feature vector, the complaint volume prediction system scores each time series period based on a preset periodic scoring strategy, and obtains the periodic score corresponding to each time series period. The complaint volume prediction system can score the time series period based on the periodic feature vector of the time series period.

[0080] Specifically, the complaint volume prediction system independently calculates scores for each periodic feature vector, adapting to a personalized periodic pattern. This step outputs period scores for the time series, with dimensions of [B, T / 2] (B=32, T=seq_len), where seq_len is the data length of the historical complaint volume time series data.

[0081] Step D2: Based on the periodic score, perform a weighted operation on the complaint volume data features within each time series period to obtain the complaint volume weighted features.

[0082] After determining the periodic score for each time series period, the complaint volume prediction system performs a weighted operation on the complaint volume data features corresponding to each time series period based on the periodic score of each time series period, thereby obtaining the complaint volume weighted features corresponding to the complaint volume data features. The complaint volume weighted features are the data features obtained after weighting the complaint volume data features.

[0083] When weighting complaint volume data features based on periodic scores, the complaint volume prediction system first determines the normalized weights based on the periodic scores, and then weights the complaint volume data features according to the normalized weights.

[0084] Specifically, the complaint volume prediction system determines the periodic score as an unnormalized score, and the larger the value, the more important the time series period. The system uses the Softmax function to process the periodic score to obtain normalized weights (soft weights) with dimensions [B, T / 2] (B=32, T=seq_len), where seq_len is the length of the historical complaint volume time series data. After determining the normalized weights, the system weights the data features corresponding to each frequency point within the time series period based on the normalized weights (soft weights), thus obtaining the weighted complaint volume features.

[0085] Step D3: Determine the periodic feature vector based on the weighted feature of the complaint volume and the time-series feature vector.

[0086] After determining the weighted features of complaint volume, the complaint volume prediction system can either add the weighted features of complaint volume and the time-series feature vector and use the result as the periodic feature vector, or directly use the determined weighted features of complaint volume as the periodic feature vector.

[0087] Specifically, the periodic feature vector determined by the complaint volume prediction system based on the weighted features of complaint volume can be represented as x_time, with dimensions [B, T, N] (B=32, T=seq_len, N=3), where seq_len is the data length of the historical complaint volume time series data. By determining the weighted features of complaint volume and determining the periodic feature vector based on these features, adaptive decomposition and feature extraction of different time series periodic components can be achieved, while the expression of key periods is strengthened through a weighting mechanism.

[0088] Furthermore, after determining the periodic feature vector, the complaint volume prediction system can normalize the output periodic feature vector. For example, the periodic feature vector x_time can be input into a preset feedforward network for normalization to obtain its output result enc_out. The dimension of the output result enc_out is [1, perd_len, N], where perd_len is the number of future time steps to be predicted.

[0089] Figure 2 This is a flowchart illustrating the second complaint volume prediction method provided in one embodiment of this specification, as shown below. Figure 2 As shown, the schematic diagram includes: Step 202: Obtain historical complaint volume time series data.

[0090] The historical complaint volume time series data includes historical complaint volume data at multiple consecutive time points, and the historical complaint volume data includes complaint time and complaint volume.

[0091] Step 204: Normalize the number of complaints and the time of complaints in the historical complaint data to obtain the first time series feature and the second time series feature.

[0092] Step 206: Perform location encoding processing on the historical complaint data to obtain the third time-series feature.

[0093] Step 208: Add the first time series feature, the second time series feature, and the third time series feature together to obtain the fourth time series feature.

[0094] The fourth time-series feature is used to characterize the original data features of the historical complaint volume data.

[0095] Step 210: Input the fourth temporal feature into a preset self-attention network to obtain the fifth temporal feature.

[0096] The fifth time-series feature is used to characterize the long-distance data features of the historical complaint volume data.

[0097] Step 212: Determine the time series feature vector based on the fourth time series feature and the fifth time series feature.

[0098] The time-series feature vector is the feature vector of the historical complaint volume time-series data.

[0099] Step 214: Perform a frequency domain transformation operation on the time-series feature vector to obtain the time-series frequency features.

[0100] Wherein, the time-series frequency feature is the frequency feature of the time-series feature vector.

[0101] Step 216: Perform a periodicity identification operation on the time-series frequency features to determine multiple time-series periods.

[0102] Step 218: Extract features from the historical complaint volume data within each of the time periods to obtain complaint volume data features.

[0103] Step 220: Score each of the time series periods based on the preset periodic scoring strategy to obtain the periodic score.

[0104] Step 222: Based on the periodic score, perform a weighted operation on the complaint volume data features within each time series period to obtain the complaint volume weighted features.

[0105] Step 224: Determine the periodic feature vector based on the weighted feature of the complaint volume and the time-series feature vector.

[0106] Wherein, the time series period is the period included in the time series feature vector, and the periodic feature vector is the feature vector of the time series feature vector within the time series period.

[0107] Step 226: Input the periodic feature vector into the pre-trained complaint volume prediction model to obtain the complaint volume prediction result.

[0108] In the embodiments described in the specification, by scoring the time series period and determining the complaint volume weighted features of the time series period based on the scoring results, the period feature vector can be determined, which can automatically filter key periods, avoid redundant feature learning, and improve the efficiency of complaint volume prediction.

[0109] In one implementation, before inputting the periodic feature vector into the pre-trained complaint volume prediction model to obtain the complaint volume prediction result (step 108), steps E1-E5 may also be performed: Step E1: Determine multiple historical complaint data points at preset positions in the historical complaint volume time series data as sample complaint volume time series data.

[0110] Before making predictions based on the pre-trained complaint volume prediction model, the complaint volume prediction system needs to train the model to obtain the pre-trained complaint volume prediction model. During model training, the complaint volume prediction system first determines multiple historical complaint volume data at preset positions in the historical complaint volume time series data, and uses the determined data as sample complaint volume time series data. The preset position is the historical complaint volume data that appears earlier in the multiple historical complaint volume data included in the historical complaint volume time series data.

[0111] In other words, the complaint volume prediction system trains its model based on the historical complaint volume data at the beginning of the historical complaint volume time series data. This allows the model to be iteratively updated based on the model prediction results and the historical complaint volume data at the end of the historical complaint volume time series data, thereby obtaining the pre-trained sample complaint volume time series data.

[0112] Specifically, during model training, the complaint volume prediction system can divide the time-series data of sample complaints into a training set, a validation set, and a test set. The model is then trained using the training set, and the training results are validated using the validation set. Finally, after the model training is completed, the pre-trained complaint volume prediction model is tested using the test set.

[0113] Step E2: Extract features from the time-series data of the sample complaint volume to obtain the sample feature vector.

[0114] The sample feature vector is the feature vector of the time series data of the sample complaint volume.

[0115] After determining the time-series data of sample complaints, the complaint volume prediction system extracts features from the sample complaint volume time-series data to obtain a sample feature vector. This sample feature vector represents the feature vector at a future time, where the sample feature vector is the feature vector of the sample complaint volume time-series data. The method by which the complaint volume prediction system extracts features from the sample complaint volume time-series data to obtain the sample feature vector can be the same as the method described above for extracting features from historical complaint volume time-series data to obtain time-series feature vectors.

[0116] Specifically, the complaint volume prediction system first performs Z-score standardization on the reference values ​​in the historical complaint volume time series data. The resulting data can be represented as x_dec, where x_dec has dimensions [B, T, N] (B=32, T=seq_len+label_len, N=16), where seq_len is the length of the historical complaint volume time series data and label_len is the length of the historical labels used. Then, the system extracts features from the complaint time in the historical complaint volume time series data, resulting in x_mark_dec, which also has dimensions [B, T, N] (B=32, T=per_len+label_len, N=16). Finally, the system adds x_dec and x_mark_dec together to obtain the result dec_emb, which also has dimensions [B, T, N] (B=32, T=per_len+label_len, N=16).

[0117] After determining dec_emb, the complaint volume prediction system applies ProbAttention with a triangular mask to dec_emb and obtains the output dec_attn. The residuals are then normalized after connection. The processing logic is the same as the Informer self-attention processing mentioned above. The output data sample feature vector is x_dec = dec_attn + dec_emb, where the dimension of the output x_dec is [B, T, N] (B = 32, T = per_len + label_len, N = 16).

[0118] Step E3: Input the periodic feature vector and the sample feature vector into the complaint volume prediction model to be trained to obtain the first prediction result.

[0119] The complaint volume prediction model to be trained is composed of a self-attention network.

[0120] After determining the sample feature vector, the complaint volume prediction system inputs the periodic feature vector and the sample feature vector into the complaint volume prediction model to be trained, so that the complaint volume prediction model to be trained can predict the complaint volume based on the periodic feature vector and the sample feature vector, and output the first prediction result. The complaint volume prediction model to be trained is composed of a self-attention network.

[0121] Specifically, the complaint volume prediction system uses x_dec as the Q-coefficient of the self-attention network and enc_out as the K / V-coefficient within the self-attention network. It calculates attention to fuse encoded information and outputs x_cross. After residual connections, it normalizes the result to obtain x_dec_cross. Then, the system processes x_dec_cross using the FusionTimesBlock module, obtaining the output data dec_sorces and dec_time. Finally, the system applies an FNN to dec_time, normalizes the residual connections, and outputs dec_out.

[0122] After obtaining the output results, the complaint volume prediction system maps the output results to the target dimension and generates the prediction results, namely the first prediction results. Specifically, the complaint volume prediction system maps the d_model dimension features to the c_out dimension output through the linear layer Linear(d_model, c_out), thereby completing the prediction of future time series data.

[0123] Step E4: Determine the loss result based on the first prediction result and the first true result using a preset loss function.

[0124] Wherein, the first true result is the actual number of complaints corresponding to the first prediction result.

[0125] After obtaining the first prediction result, the complaint volume prediction system determines the predicted loss result based on the first prediction result and the first true result using a preset loss function. The first true result is the actual complaint volume data corresponding to the first prediction result, which is determined based on historical complaint volume time series data.

[0126] Specifically, the preset loss function can be total_loss = base_loss + sparsity_penalty - diversity_penalty, where the base loss is calculated as base_loss = MSE = (1 / n)∑(y 真实i -y 预测i ) 2 , n represents the batch size, y 真实i Let y represent the true label value of the i-th sample. 预测 This represents the predicted value of the i-th sample.

[0127] Step E5: Iteratively update the complaint volume prediction model to be trained based on the loss result to obtain the pre-trained complaint volume prediction model.

[0128] After determining the loss result, the complaint volume prediction system iteratively updates the complaint volume prediction model to be trained based on the determined loss result until the loss result is less than the preset threshold. At this point, the pre-trained complaint volume prediction model is obtained.

[0129] Figure 3 This is a flowchart illustrating the third method for predicting complaint volume provided in one embodiment of this specification, as shown below. Figure 3 As shown, the schematic diagram includes: Step 302: Obtain historical complaint volume time series data.

[0130] The historical complaint volume time series data includes historical complaint volume data at multiple consecutive time points, and the historical complaint volume data includes complaint time and complaint volume.

[0131] Step 304: Extract features from the historical complaint volume time series data to obtain a time series feature vector.

[0132] The time-series feature vector is the feature vector of the historical complaint volume time-series data.

[0133] Step 306: Perform a frequency domain transformation operation on the time-series feature vector, and determine the time-series period and periodic feature vector based on the frequency domain transformation result.

[0134] Wherein, the time series period is the period included in the time series feature vector, and the periodic feature vector is the feature vector of the time series feature vector within the time series period.

[0135] Step 308: Determine multiple historical complaint data points at preset positions in the historical complaint volume time series data as sample complaint volume time series data.

[0136] Step 310: Extract features from the time-series data of the sample complaint volume to obtain the sample feature vector.

[0137] The sample feature vector is the feature vector of the time series data of the sample complaint volume.

[0138] Step 312: Input the periodic feature vector and the sample feature vector into the complaint volume prediction model to be trained to obtain the first prediction result.

[0139] The complaint volume prediction model to be trained is composed of a self-attention network.

[0140] Step 314: Determine the loss result based on the first prediction result and the first true result using a preset loss function.

[0141] Wherein, the first true result is the actual number of complaints corresponding to the first prediction result.

[0142] Step 316: Iteratively update the complaint volume prediction model to be trained based on the loss result to obtain the pre-trained complaint volume prediction model.

[0143] Step 318: Input the periodic feature vector into the pre-trained complaint volume prediction model to obtain the complaint volume prediction result.

[0144] In the embodiments described in the specification, a self-attention mechanism is used to determine long-distance data features in the time series data of sample complaints for model training. Compared with traditional single models, this solves the problems of insufficient modeling ability for nonlinear complex data and insufficient extraction of periodic features, thereby improving the accuracy of complaint volume prediction.

[0145] Figure 4 This is a schematic diagram of the structure of a complaint volume prediction system provided in one embodiment of this specification, as shown below. Figure 4 As shown, the system includes an input layer, an embedding layer, a periodic feature extraction module, an encoder, a decoder, and a prediction layer. The input layer takes historical complaint volume time-series data as input. The embedding layer fuses the original data features with time-stamped features through convolution, linear transformation, and positional encoding, outputting an embedding vector of dimension [B, T, N] (B=32, T=seq_len, N=3); its inputs are the original data x_enc, x_dec and the time stamps x_mark_enc, x_mark_dec, and its output is sent to the encoder / decoder. The periodic feature extraction module extracts periodic features from the time-series data through FFT transformation, multi-scale convolution (Inception module), and learnable periodic scoring (MLP), preserving the original information through residual connections; it receives intermediate features from the encoder / decoder and outputs a sequence with fused periodic features and a periodic score. The encoder captures long-range dependencies using a self-attention network, extracts periodic features using a time block, and outputs the encoded sequence `enc_out` and `period_scores`. It consists of multiple `FusionEncoderLayer` layers, each containing a self-attention layer, a time block, and a feedforward network (FFN). The decoder generates feature representations for future time steps based on the encoder output and the sequence to be predicted. It also consists of multiple `FusionEncoderLayer` layers, each containing a self-attention layer (with masking), a time block, and a feedforward network (FFN). The prediction layer maps the decoder output to the target dimension using a linear layer, generating the final prediction result.

[0146] It should be noted that the complaint volume prediction method provided in this application embodiment can be executed by a complaint volume prediction device or a control module within that device for executing the complaint volume prediction method. This application embodiment uses the execution of the complaint volume prediction method by a complaint volume prediction device as an example to illustrate the complaint volume prediction device provided in this application embodiment.

[0147] Figure 5 This is a schematic diagram of the structure of a complaint volume prediction device according to an embodiment of the present invention. Figure 5 As shown, the complaint volume prediction device includes: a first acquisition module 502, a first extraction module 504, a first determination module 506, and a first prediction module 508.

[0148] The first acquisition module 502 is used to acquire historical complaint volume time series data, which includes historical complaint volume data at multiple consecutive time points, and includes complaint time and complaint volume. The first extraction module 504 is used to extract features from the historical complaint volume time series data to obtain a time series feature vector, wherein the time series feature vector is the feature vector of the historical complaint volume time series data. The first determining module 506 is used to perform a frequency domain transformation operation on the time-series feature vector, and determine the time-series period and the periodic feature vector based on the frequency domain transformation result. The time-series period is the period included in the time-series feature vector, and the periodic feature vector is the feature vector of the time-series feature vector within the time-series period. The first prediction module 508 is used to input the periodic feature vector into a pre-trained complaint volume prediction model to obtain the complaint volume prediction result.

[0149] The complaint volume prediction device in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This application embodiment does not impose specific limitations.

[0150] The complaint volume prediction device in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0151] The complaint volume prediction device provided in this application embodiment can achieve... Figures 1 to 4 The various processes implemented in the method embodiments are not described in detail here to avoid repetition.

[0152] Based on the same technical concept, embodiments of this application also provide an electronic device for performing the above-described complaint volume prediction method. Figure 6 This is a schematic diagram of the structure of an electronic device to implement various embodiments of this application. The electronic device can vary significantly due to differences in configuration or performance, and may include a processor 602, a communications interface 604, a memory 606, and a communication bus 608. The processor 602, communications interface 604, and memory 606 communicate with each other via the communication bus 608. The processor 602 can call a computer program stored in the memory 606 and executable on the processor 602 to perform the following steps: Obtain historical complaint volume time series data, which includes historical complaint volume data at multiple consecutive time points, including complaint time and complaint volume; Feature extraction is performed on the historical complaint volume time series data to obtain a time series feature vector, which is the feature vector of the historical complaint volume time series data; A frequency domain transformation operation is performed on the time-series feature vector, and a time-series period and a periodic feature vector are determined based on the frequency domain transformation result. The time-series period is the period included in the time-series feature vector, and the periodic feature vector is the feature vector of the time-series feature vector within the time-series period. The periodic feature vector is input into a pre-trained complaint volume prediction model to obtain the complaint volume prediction result.

[0153] In one implementation, the step of extracting features from the historical complaint volume time-series data to obtain a time-series feature vector includes: The number of complaints and the time of complaints in the historical complaint data are normalized to obtain a first time-series feature and a second time-series feature. The historical complaint data is subjected to location encoding to obtain the third time-series feature; The time series feature vector is determined based on the first time series feature, the second time series feature, and the third time series feature.

[0154] In one implementation, determining the time-series feature vector based on the first time-series feature, the second time-series feature, and the third time-series feature includes: The first time-series feature, the second time-series feature, and the third time-series feature are added together to obtain a fourth time-series feature, which is used to characterize the original data features of the historical complaint volume data. The fourth temporal feature is input into a preset self-attention network to obtain a fifth temporal feature, which is used to characterize the long-distance data features of the historical complaint volume data. The time series feature vector is determined based on the fourth time series feature and the fifth time series feature.

[0155] In one implementation, the step of performing a frequency domain transformation on the time-series feature vector and determining the time-series period and periodic feature vector based on the frequency domain transformation result includes: The time-series feature vector is subjected to a frequency domain transformation operation to obtain a time-series frequency feature, which is the frequency feature of the time-series feature vector. Perform a periodicity identification operation on the time-series frequency characteristics to determine multiple time-series periods; Feature extraction is performed on the historical complaint volume data within each of the aforementioned time periods to obtain complaint volume data features; The periodic feature vector is determined based on the complaint volume data characteristics and the time-series feature vector.

[0156] In one implementation, determining the periodic feature vector based on the complaint volume data features and the time-series feature vector includes: Each time series period is scored based on a preset periodic scoring strategy to obtain a periodic score. Based on the periodic score, the complaint volume data features within each of the time series periods are weighted to obtain the complaint volume weighted features. The periodic feature vector is determined based on the weighted feature of the complaint volume and the time-series feature vector.

[0157] In one implementation, before inputting the periodic feature vector into a pre-trained complaint volume prediction model to obtain the complaint volume prediction result, the method further includes: Multiple historical complaint data points at preset positions in the historical complaint volume time series data are determined as sample complaint volume time series data; Feature extraction is performed on the time-series data of the sample complaint volume to obtain a sample feature vector, which is the feature vector of the time-series data of the sample complaint volume. The periodic feature vector and the sample feature vector are input into the complaint volume prediction model to be trained to obtain a first prediction result. The complaint volume prediction model to be trained is composed of a self-attention network. By using a preset loss function, a loss result is determined based on the first prediction result and the first actual result, where the first actual result is the actual number of complaints corresponding to the first prediction result. The complaint volume prediction model to be trained is iteratively updated based on the loss result to obtain the pre-trained complaint volume prediction model.

[0158] The specific implementation steps can be found in the various steps of the above-described complaint volume prediction method embodiment, and can achieve the same technical effect. To avoid repetition, they will not be repeated here.

[0159] It should be noted that the electronic devices in the embodiments of this application include: servers, terminals, or other devices besides terminals.

[0160] The above electronic device structure does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or arrange them differently. For example, an input unit may include a Graphics Processing Unit (GPU) and a microphone, and a display unit may use a liquid crystal display (LCD), organic light-emitting diode (OLED), or other similar display panels. User input units include at least one of a touch panel and other input devices. A touch panel is also called a touchscreen. Other input devices may include, but are not limited to, physical keyboards, function keys (such as volume control buttons, power buttons, etc.), trackballs, mice, and joysticks, which will not be elaborated further here.

[0161] Memory can be used to store software programs and various data. Memory can primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area can store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, memory can include volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0162] The processor may include one or more processing units; optionally, the processor integrates an application processor and a modem processor, wherein the application processor mainly handles operations related to the operating system, user interface, and applications, while the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into the processor.

[0163] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described complaint volume prediction method embodiments and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0164] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0165] This application also provides a computer program product. When the computer program product is executed by a processor, it implements the various processes of the above-described complaint volume prediction method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0166] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described complaint volume prediction method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0167] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0168] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, 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, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0169] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0170] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for predicting the number of complaints, characterized in that, The method includes: Obtain historical complaint volume time series data, which includes historical complaint volume data at multiple consecutive time points, including complaint time and complaint volume; Feature extraction is performed on the historical complaint volume time series data to obtain a time series feature vector, which is the feature vector of the historical complaint volume time series data; A frequency domain transformation operation is performed on the time-series feature vector, and a time-series period and a periodic feature vector are determined based on the frequency domain transformation result. The time-series period is the period included in the time-series feature vector, and the periodic feature vector is the feature vector of the time-series feature vector within the time-series period. The periodic feature vector is input into a pre-trained complaint volume prediction model to obtain the complaint volume prediction result.

2. The method according to claim 1, characterized in that, The step of extracting features from the historical complaint volume time-series data to obtain a time-series feature vector includes: The number of complaints and the time of complaints in the historical complaint data are normalized to obtain a first time-series feature and a second time-series feature. The historical complaint data is subjected to location encoding to obtain the third time-series feature; The time series feature vector is determined based on the first time series feature, the second time series feature, and the third time series feature.

3. The method according to claim 2, characterized in that, Determining the time series feature vector based on the first time series feature, the second time series feature, and the third time series feature includes: The first time-series feature, the second time-series feature, and the third time-series feature are added together to obtain a fourth time-series feature, which is used to characterize the original data features of the historical complaint volume data. The fourth temporal feature is input into a preset self-attention network to obtain a fifth temporal feature, which is used to characterize the long-distance data features of the historical complaint volume data. The time series feature vector is determined based on the fourth time series feature and the fifth time series feature.

4. The method according to claim 1, characterized in that, The step of performing a frequency domain transformation on the time-series feature vector and determining the time-series period and periodic feature vector based on the frequency domain transformation result includes: The time-series feature vector is subjected to a frequency domain transformation operation to obtain a time-series frequency feature, which is the frequency feature of the time-series feature vector. Perform a periodicity identification operation on the time-series frequency characteristics to determine multiple time-series periods; Feature extraction is performed on the historical complaint volume data within each of the aforementioned time periods to obtain complaint volume data features; The periodic feature vector is determined based on the complaint volume data characteristics and the time-series feature vector.

5. The method according to claim 4, characterized in that, The step of determining the periodic feature vector based on the complaint volume data features and the time-series feature vector includes: Each time series period is scored based on a preset periodic scoring strategy to obtain a periodic score. Based on the periodic score, the complaint volume data features within each of the time series periods are weighted to obtain the complaint volume weighted features. The periodic feature vector is determined based on the weighted feature of the complaint volume and the time-series feature vector.

6. The method according to claim 1, characterized in that, Before inputting the periodic feature vector into the pre-trained complaint volume prediction model to obtain the complaint volume prediction result, the method further includes: Multiple historical complaint data points at preset positions in the historical complaint volume time series data are determined as sample complaint volume time series data; Feature extraction is performed on the time-series data of the sample complaint volume to obtain a sample feature vector, which is the feature vector of the time-series data of the sample complaint volume. The periodic feature vector and the sample feature vector are input into the complaint volume prediction model to be trained to obtain a first prediction result. The complaint volume prediction model to be trained is composed of a self-attention network. By using a preset loss function, a loss result is determined based on the first prediction result and the first actual result, where the first actual result is the actual number of complaints corresponding to the first prediction result. The complaint volume prediction model to be trained is iteratively updated based on the loss result to obtain the pre-trained complaint volume prediction model.

7. A complaint volume prediction device, characterized in that, The device includes: The first acquisition module is used to acquire historical complaint volume time series data, which includes historical complaint volume data at multiple consecutive time points, and includes complaint time and complaint volume. The first extraction module is used to extract features from the historical complaint volume time series data to obtain a time series feature vector, wherein the time series feature vector is the feature vector of the historical complaint volume time series data. The first determining module is used to perform a frequency domain transformation operation on the time-series feature vector, and determine the time-series period and the periodic feature vector based on the frequency domain transformation result. The time-series period is the period included in the time-series feature vector, and the periodic feature vector is the feature vector of the time-series feature vector within the time-series period. The first prediction module is used to input the periodic feature vector into a pre-trained complaint volume prediction model to obtain the complaint volume prediction result.

8. An electronic device, characterized in that, The device includes: Processor; and A memory configured to store computer-executable instructions configured to be executed by the processor, the executable instructions including steps for performing the complaint volume prediction method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is used to store computer-executable instructions that cause the computer to perform the complaint volume prediction method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the complaint volume prediction method according to any one of claims 1 to 6.

Images