Sales volume data prediction method and device of e-commerce platform, equipment and medium
By applying a combined model of long short-term memory networks and temporal convolutional networks on e-commerce platforms, the problem of insufficient feature mining in e-commerce sales data prediction is solved, achieving more accurate and flexible sales predictions and supporting the operational decisions of e-commerce platforms.
Patent Information
- Application Number
- CN202510693188.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-27
- Publication Date
- 2025-09-19
AI Technical Summary
Existing e-commerce platform sales data prediction methods are difficult to fully exploit data characteristics, especially nonlinear relationships and long-term dependencies, resulting in low prediction accuracy and traditional models are difficult to adapt to the complex and changing market environment.
Using a preset network structure based on long short-term memory networks and temporal convolutional networks, the system preprocesses e-commerce sales data, extracts features, and trains models to extract multi-dimensional features and optimize the model, including sales trends, seasonality, local features, and external factor features, thereby improving the model's predictive capabilities.
It achieves a comprehensive and detailed description of e-commerce sales data, improves the accuracy and adaptability of forecasts, and can provide sales forecast information in a timely and accurate manner to support the operation and management of e-commerce platforms.
Smart Images
Figure CN120672377A_ABST
Abstract
Description
Technical Field
[0001] This specification relates to the field of machine learning technology, and in particular to a method, device, equipment, and medium for predicting sales data on an e-commerce platform. Background Art
[0002] With the rapid development of internet technology and the growing prosperity of the e-commerce market, e-commerce companies are facing fierce competition and a complex and volatile market environment. In this context, sales forecasting is more than just a data estimation exercise; it is a core component in achieving refined operations and enhancing competitiveness. Therefore, in the e-commerce sector, accurate sales forecasting is crucial for business operations and management.
[0003] Current e-commerce platforms have accumulated vast amounts of sales data, rich in information. However, raw data often suffers from noise and inconsistent formats, making it ineffective for direct forecasting. Furthermore, sales data exhibits distinct time series characteristics, such as trends and seasonality, and is influenced by numerous external factors, such as holidays, promotions, and competitor strategies. Forecasting methods based on traditional statistical models struggle to fully exploit the data characteristics and nonlinear relationships in sales data, making it difficult to extract the underlying information. While forecasting methods based on simple machine learning models such as linear regression and decision trees can handle certain nonlinear relationships, they offer lower accuracy for e-commerce sales data with its long-term dependencies and complex characteristics. Summary of the Invention
[0004] In order to solve the above technical problems, one or more embodiments of this specification provide a sales data prediction method, device, equipment and medium for an e-commerce platform.
[0005] One or more embodiments of this specification adopt the following technical solutions: One or more embodiments of this specification provide a method for predicting sales data on an e-commerce platform, the method comprising: Preprocessing existing e-commerce sales data to obtain time series features corresponding to the existing e-commerce sales data; Extracting features from the time series features based on preset feature requirements to obtain multi-dimensional features corresponding to the existing e-commerce sales data; wherein the multi-dimensional features include: sales trend features, seasonal features, local features, and external factor features; Inputting the multi-dimensional features into a preset network structure for model training to obtain a sales forecast model that meets the requirements; wherein the preset network structure is based on a long short-term memory network and a temporal convolutional network; The sales forecast model is deployed to the corresponding e-commerce platform sales forecast system to regularly forecast the sales data of the current time period collected by the e-commerce platform sales forecast system based on the sales forecast model to obtain sales data for the next time period.
[0006] Optionally, in one or more embodiments of this specification, preprocessing existing e-commerce sales data to obtain time series features corresponding to the existing e-commerce sales data specifically includes: Determine whether the existing e-commerce sales data has missing values, and if so, convert the data format of the existing e-commerce sales data based on a preset nearest neighbor algorithm model, input the converted existing e-commerce sales data into the preset nearest neighbor algorithm model, and obtain the completed existing e-commerce sales data; Calculating the Z-score value corresponding to each of the completed existing e-commerce sales data, comparing the Z-score value with a preset threshold, determining abnormal values in the completed existing e-commerce sales data, filtering the completed existing e-commerce sales data, and obtaining the e-commerce sales data to be extracted; The e-commerce sales data to be extracted are subjected to date analysis to determine the time series features corresponding to the e-commerce sales data to be extracted based on the date features corresponding to each of the e-commerce sales data to be extracted.
[0007] Optionally, in one or more embodiments of this specification, feature extraction is performed on the time series features based on preset feature requirements to obtain multi-dimensional features corresponding to the existing e-commerce sales data, specifically including: Calculating the sales volume average of each window of the time series feature based on a given time window, so as to determine the sales volume trend feature corresponding to the existing e-commerce sales volume data based on the sales volume average of each window; Decomposing the time series features based on a preset seasonal decomposition strategy to determine the seasonal features corresponding to the existing e-commerce sales data; The time series features are moved point by point based on a given time window to determine the local features corresponding to the existing e-commerce sales data based on the data in each window; Collecting external factor data associated with the time series features, and encoding the external factor data to obtain external factor features corresponding to the existing e-commerce sales data; The sales trend characteristics, seasonal characteristics, local characteristics and the external factor characteristics are summarized to obtain multi-dimensional characteristics corresponding to the existing e-commerce sales data.
[0008] Optionally, in one or more embodiments of this specification, before inputting the multi-dimensional features into a preset network structure for model training, the method further includes: Determine the number of neurons and the number of LSTM units in the preset long short-term memory network based on the number of data and the number of feature dimensions corresponding to the multi-dimensional features; Constructing an input layer and an LSTM layer of the preset long short-term memory network based on the neuron data and the number of LSTM units, and adding a fully connected hidden layer after the LSTM layer to obtain a long short-term memory network of the preset network structure; Setting an input layer of a preset temporal convolutional network based on the input layer of the long short-term memory network, and determining TCN layer parameters of the preset temporal convolutional network based on local features within the multi-dimensional features, so as to set the TCN layer according to the TCN layer parameters; wherein the TCN layer parameters include: a convolution kernel and a dilation factor; Adding a fully connected hidden layer after the TCN layer to obtain a temporal convolutional network of the preset network structure; The outputs of the long short-term memory network and the temporal convolutional network are fused in the feature dimension, and a preset network structure is obtained based on a fully connected layer as an output layer.
[0009] Optionally, in one or more embodiments of this specification, the multi-dimensional features are input into a preset network structure for model training to obtain a sales forecast model that meets the requirements, specifically including: Dividing the multi-dimensional features based on a preset ratio to obtain a training set, a validation set, and a test set; Inputting the multi-dimensional features in the training set into the preset network structure in batches for training, and optimizing the hyperparameters of the preset network structure based on a preset tuning strategy; Evaluate the preset network structure in each round based on the validation set to determine whether early stopping is required; If so, the preset network structure is evaluated using the multi-dimensional features corresponding to the test set to obtain a sales forecast model that meets the requirements.
[0010] Optionally, in one or more embodiments of this specification, optimizing the hyperparameters of the preset network structure based on a preset tuning strategy specifically includes: Calculating the loss value after each round of training of the preset network structure according to a preset loss function; wherein the preset loss function is: mean square error or root mean square error; Determine the gradient of the model parameters in the preset network structure using a back-propagation algorithm and the loss function, so as to update the hyperparameters of the preset network structure according to the gradient; The preset network structure is iteratively updated, and the updated multiple preset network structures are integrated to iteratively train to obtain a sales forecast model that meets the requirements.
[0011] Optionally, in one or more embodiments of this specification, the sales forecast model is deployed to a corresponding e-commerce platform sales forecast system, so as to regularly forecast the sales data of the current time period collected by the e-commerce platform sales forecast system based on the sales forecast model to obtain sales data for the next time period, specifically including: Integrating the sales forecast model with the corresponding e-commerce platform sales forecast system based on an interface of the corresponding e-commerce platform sales forecast system to deploy the sales forecast model to the corresponding e-commerce platform sales forecast system; Determining an evaluation period for the sales forecast model based on actual business needs and business feature transformation data of the corresponding e-commerce platform sales forecast system, and performing an indicator test on the sales forecast model according to the evaluation period; Adjusting the sales forecast model based on the test results to obtain an updated sales forecast model; Based on the updated sales forecast model, the sales data of the current time period collected by the sales forecast system of the e-commerce platform is regularly obtained, so as to input the sales data of the current time period into the sales forecast model to obtain the sales data of the next time period.
[0012] One or more embodiments of this specification provide a sales data prediction device for an e-commerce platform, the device comprising: A processing unit, configured to pre-process existing e-commerce sales data to obtain time series features corresponding to the existing e-commerce sales data; An extraction unit is configured to extract features from the time series features based on preset feature requirements to obtain multi-dimensional features corresponding to the existing e-commerce sales data; wherein the multi-dimensional features include: sales trend features, seasonal features, local features, and external factor features; A training unit, configured to input the multi-dimensional features into a preset network structure for model training, thereby optimizing the hyperparameters of the preset network structure based on a preset tuning strategy to obtain a sales forecast model that meets the requirements; wherein the preset network structure is based on a long short-term memory network and a temporal convolutional network; The prediction unit is used to deploy the sales prediction model to the corresponding e-commerce platform sales prediction system, so as to regularly predict the sales data of the current time period collected by the e-commerce platform sales prediction system based on the sales prediction model to obtain the sales data of the next time period.
[0013] One or more embodiments of this specification provide a sales data prediction device for an e-commerce platform, the device comprising: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: perform any of the above methods.
[0014] One or more embodiments of this specification provide a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to execute any of the above-described methods.
[0015] At least one of the above technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects: Based on pre-defined feature requirements, this model extracts multi-dimensional features, describing patterns in e-commerce sales data from various perspectives. It considers both long-term trends and cyclical changes in the data, as well as local details and the influence of external factors. This model comprehensively and meticulously captures the complex nature of sales data, providing rich information for accurate forecasting. Model training utilizes a pre-defined network architecture based on long-short-term memory networks and temporal convolutional networks, leveraging their respective strengths to better uncover potential patterns in the data and effectively capture the long-term dependencies and local characteristics of sales data, improving the model's predictive power and accuracy. Deploying the sales forecast model to the e-commerce platform's sales forecasting system and performing regular sales forecasts can provide the platform with timely and accurate sales forecast information. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some of the embodiments described in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without inventive work. In the drawings: Figure 1 A flowchart of a sales data prediction method for an e-commerce platform provided in an embodiment of this specification; Figure 2 A flowchart of an e-commerce sales data prediction system in an application scenario provided by an embodiment of this specification; Figure 3 A schematic diagram of an e-commerce sales data prediction device in an application scenario provided by an embodiment of this specification; Figure 4 A schematic diagram of the structure of a sales data prediction device for an e-commerce platform provided in an embodiment of this specification; Figure 5 A schematic diagram of the structure of a non-volatile storage medium provided in an embodiment of this specification. DETAILED DESCRIPTION
[0017] The embodiments of this specification provide a method, device, equipment, and medium for predicting sales data of an e-commerce platform.
[0018] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this specification without creative work should fall within the scope of protection of this specification.
[0019] like Figure 1 As shown, the embodiment of this specification provides a method flow chart of a sales data prediction method for an e-commerce platform. Figure 1 It can be seen that in one or more embodiments of this specification, a sales data prediction method for an e-commerce platform includes: S101: Preprocess existing e-commerce sales data to obtain time series features corresponding to the existing e-commerce sales data.
[0020] Current e-commerce platforms have accumulated a large amount of sales data, which contains rich information. However, raw data often suffers from noise and inconsistent formats, making it ineffective for direct prediction. Therefore, in the embodiments of this specification, existing e-commerce sales data will be preprocessed. Moreover, since sales data has obvious time series characteristics, such as trends and seasonality, in order to better achieve sales data prediction, the time series features corresponding to the existing e-commerce sales data will be obtained.
[0021] Specifically, in one or more embodiments of this specification, preprocessing the existing e-commerce sales data to obtain the time series features corresponding to the existing e-commerce sales data specifically includes: Because missing values may appear in fields such as sales quantity, sales amount, and sales date, they can affect data integrity and the accuracy of analysis results. Therefore, it is necessary to determine whether existing e-commerce sales data contains missing values. If so, the existing e-commerce sales data is converted based on the data format of a preset nearest neighbor algorithm model. The converted existing e-commerce sales data is then input into the preset nearest neighbor algorithm model to obtain the completed existing e-commerce sales data. It should be noted that the preset nearest neighbor algorithm can be a K-nearest neighbor algorithm. In this case, the model finds the K data points most similar to the data point containing the missing value and estimates the missing value based on the data values of these K nearest neighbors. Then, the Z-score value corresponding to each completed existing e-commerce sales data is calculated and compared with a preset threshold to identify outliers in the completed existing e-commerce sales data. That is, if the Z-score value of a certain existing e-commerce sales data exceeds the preset threshold, the existing e-commerce sales data is considered outlier and needs to be filtered to obtain the e-commerce sales data to be extracted. This outlier processing process can reduce the interference of outliers on subsequent analysis and model training. Next, we perform date parsing on the e-commerce sales data to be extracted. Based on the date characteristics of each e-commerce sales data point, we determine the time series characteristics corresponding to the data. By performing detailed date parsing and determining time series characteristics based on this, we can fully explore the patterns of e-commerce sales data over time.
[0022] S102: Extracting features from the time series features based on preset feature requirements to obtain multi-dimensional features corresponding to the existing e-commerce sales data; wherein the multi-dimensional features include: sales trend features, seasonal features, local features, and external factor features.
[0023] Since the original time series features may only be a preliminary description of the data and cannot fully cover the various factors that affect e-commerce sales, in order to obtain the potential information in the data, the embodiments of this specification will extract the time series features according to the preset feature requirements to obtain the multi-dimensional features corresponding to the existing e-commerce sales data. Among them, it should be noted that the multi-dimensional features include: sales trend features, seasonal features, local features and external factor features. By extracting multi-dimensional features in a targeted manner, the characteristics of sales data can be portrayed from different angles, including long-term trends, seasonal patterns, short-term fluctuations and the influence of external factors, so that the model can more comprehensively capture the information in the data and improve the accuracy and reliability of sales forecasts.
[0024] Specifically, in one or more embodiments of this specification, feature extraction is performed on time series features based on preset feature requirements to obtain multi-dimensional features corresponding to existing e-commerce sales data, specifically including the following process: First, the sales average of each window of the time series feature is calculated based on a given time window, so as to determine the sales trend characteristics corresponding to the existing e-commerce sales data based on the sales average of each window. Then, the time series feature is decomposed based on the preset seasonal decomposition strategy to determine the seasonal characteristics corresponding to the existing e-commerce sales data. In order to pay attention to the characteristics of the data in different local time periods, such as sudden fluctuations or abnormal changes in sales within a short period of time. In the embodiment of this specification, the time series feature will be moved point by point based on a given time window to determine the local characteristics corresponding to the existing e-commerce sales data based on the data in each window. Then, in order to take the reasons for the changes in the e-commerce sales data into consideration, the embodiment of this specification will collect external factor data associated with the time series feature to encode the external factor data to obtain the external factor characteristics corresponding to the existing e-commerce sales data. Then, by summarizing the above-mentioned sales trend characteristics, seasonal characteristics, local characteristics and the external factor characteristics, the multi-dimensional characteristics corresponding to the existing e-commerce sales data are obtained. This process extracts features from e-commerce sales data across multiple dimensions, including trends, seasonality, local variations, and external factors. This allows for a more comprehensive and detailed description of the data's characteristics and patterns, avoiding the one-sidedness inherent in analyzing data from a single perspective. Furthermore, the diverse processing and feature extraction methods employed for time series data constitute effective data preprocessing, providing more valuable input features for subsequent machine learning or other data analysis algorithms, helping to improve model accuracy and generalization capabilities.
[0025] In addition, in an embodiment under a certain scenario of this specification, feature extraction is performed on time series features based on preset feature requirements to obtain multi-dimensional features corresponding to existing e-commerce sales data. This can also be achieved based on the following processes: calculating the moving average and exponential smoothing of historical sales to capture the long-term trend of sales changes and obtain sales trend features; using seasonal decomposition such as STL decomposition to extract the seasonal pattern of sales and obtain seasonal features; extracting local fluctuation features of sales data through technologies such as sliding windows to obtain local features; integrating external factor data that may affect sales, such as weather, competitor prices, social media popularity, etc., and performing encoding processing to obtain external factor features.
[0026] S103: Input the multi-dimensional features into a preset network structure for model training to obtain a sales forecasting model that meets the requirements; wherein the preset network structure is based on a long short-term memory network and a temporal convolutional network.
[0027] In order to improve the accuracy and efficiency of sales data prediction in the e-commerce field, and solve the problem that it is difficult to capture the long-term dependencies and local features of sales changes in the existing technology, in the embodiment of this specification, multi-dimensional features will be input into the preset network structure for model training, so as to obtain a sales prediction model that meets the requirements. It should be noted that the preset network structure is based on the long short-term memory network and the time convolution network, that is, Figure 2 The preset network structure shown is based on the long short-term memory network LSTM part and the temporal convolutional network TCN part.
[0028] Furthermore, in one or more embodiments of this specification, before inputting the multi-dimensional features into the preset network structure for model training, the method further includes the following process: First, since the amount of data and the number of feature dimensions reflect the scale and complexity of the input data, in order to better adapt to the input data, the embodiments of this specification will determine the number of neurons and the number of LSTM units of the preset long-short-term memory network based on the amount of data and the number of feature dimensions corresponding to the multi-dimensional features. Then, based on the neuron data and the number of LSTM units, the input layer and LSTM layer of the preset long-short-term memory network are constructed, and a fully connected hidden layer is added after the LSTM layer to obtain a long-short-term memory network with a preset network structure. That is, by setting the LSTM layer, since it has a hidden state that runs through the unit, it is determined by the hidden state which moments are memories that can be stored for a long time and which are short-term memories that can be ignored. This enables the subsequent model to enhance its ability to capture complex market dynamics. The input layer of a pre-set temporal convolutional network is set based on the input layer of the long short-term memory network. The parameters of the pre-set temporal convolutional network, such as the convolution kernel and dilation factor, are determined based on local features within the multi-dimensional features. The TCN layer is then set based on these parameters. Specifically, by setting the convolution kernel and dilation factor, the TCN layer uses a dilated convolutional structure, increasing the network's receptive field without increasing the number of parameters, thereby effectively capturing long-term dependencies. Furthermore, TCN can enhance the model's representational power by stacking multiple layers of convolution, enabling it to excel in various time series tasks. A fully connected hidden layer is then added after the TCN layer to obtain a temporal convolutional network with the pre-set network structure. The outputs of the long short-term memory network and the temporal convolutional network are then fused along the feature dimension. This fusion can be performed using methods such as concatenation, addition, or an attention mechanism, with the fully connected layer serving as the output layer to obtain the pre-set network structure.
[0029] In one embodiment of the present specification in a certain application scenario, the preset network structure includes an LSTM part and a TCN part. The LSTM part includes: an input layer that receives preprocessed sales data and its time features; an LSTM layer that uses LSTM units to capture long-term dependencies in sales data, especially trends and periodic changes in time series; and a hidden layer that sets multiple layers of LSTM units to enhance the learning ability of the model. The TCN part includes: an input layer that shares the same input data as the LSTM; a TCN layer that uses a temporal convolutional network to capture local features and short-term fluctuations in sales data through convolution operations; a hidden layer that stacks multiple TCN layers to extract higher-level features; a fusion layer that fuses the outputs of the LSTM and TCN, which can be fused by splicing, addition, or attention mechanism; and an output layer that uses a fully connected layer and a linear activation function to output the predicted sales value.
[0030] Specifically, in one or more embodiments of this specification, the multi-dimensional features are input into a preset network structure for model training to obtain a sales forecast model that meets the requirements, specifically including: Multi-dimensional features are divided based on preset proportions to create training, validation, and test sets. The training set is used to train the model, allowing it to learn patterns in the data; the validation set is used to evaluate model performance during training and adjust hyperparameters; and the test set is used to ultimately evaluate the generalization ability of the trained model. The multi-dimensional features in the training set are then fed into the pre-set network structure in batches for training, reducing memory usage and improving training efficiency. Simultaneously, the hyperparameters of the pre-set network structure are optimized based on a preset tuning strategy, ensuring a better fit to the training data and improving performance. During training, the pre-set network structure is evaluated using the validation set after each round of training to determine whether the model is overfitting or performance is no longer improving, thereby determining whether early stopping is necessary. If early stopping is necessary, it indicates that the model has achieved satisfactory performance on the validation set, and further training may lead to overfitting. After training is stopped early, the pre-set network structure is evaluated using the multi-dimensional features corresponding to the test set, resulting in a final sales forecasting model that meets the requirements. The test set data is not used during training, allowing for an objective assessment of the model's predictive ability on new data. During this process, batching training data effectively utilizes computing resources and reduces memory pressure. This is particularly useful for processing large datasets, improving the speed and efficiency of model training and saving training time and costs. Optimizing hyperparameters based on pre-set tuning strategies allows the model to better adapt to data characteristics and find the optimal combination of model parameters, thereby improving model accuracy and stability and enhancing its predictive performance.
[0031] Furthermore, in one or more embodiments of this specification, optimizing the hyperparameters of the preset network structure based on a preset tuning strategy specifically includes: First, the loss value of the preset network structure after each round of training is calculated based on a preset loss function; the preset loss function can be either mean squared error (MSE) or root mean squared error (RMSE). Then, the backpropagation algorithm, combined with the loss function, is used to determine the gradients of the model parameters in the preset network structure. The backpropagation algorithm propagates the loss value from the output layer to the input layer, calculating the contribution of each parameter to the loss, i.e., the gradient. The hyperparameters of the preset network structure are then updated based on the gradient, optimizing the model towards decreasing the loss value. The model parameters are gradually adjusted to improve prediction accuracy. The preset network structure is continuously updated iteratively, i.e., the above process of calculating the loss value and updating the hyperparameters is repeated multiple times. Simultaneously, the updated multiple preset network structures are integrated, combining the results of multiple models to obtain more accurate and stable predictions. Through this iterative training and model integration approach, a sales forecast model that meets the requirements is ultimately obtained.
[0032] S104: Deploy the sales forecast model to the corresponding e-commerce platform sales forecast system to regularly forecast the sales data of the current time period collected by the e-commerce platform sales forecast system based on the sales forecast model to obtain sales data for the next time period.
[0033] After obtaining the sales forecast model based on the above steps, in order to achieve real-time or periodic sales forecasts, the embodiments of this specification will deploy the sales forecast model to the corresponding e-commerce platform sales forecast system, so that the sales data collected by the e-commerce platform sales forecast system for the current time period will be regularly predicted based on the sales forecast model to obtain sales data for the next time period. Specifically, in one or more embodiments of this specification, the sales forecast model is deployed to the corresponding e-commerce platform sales forecast system to regularly predict the sales data collected by the e-commerce platform sales forecast system for the current time period based on the sales forecast model to obtain sales data for the next time period, specifically including: The sales forecast model is integrated with the corresponding e-commerce platform sales forecast system based on its interface, allowing the model to be integrated into the actual business processes of the e-commerce platform and deployed to the corresponding e-commerce platform sales forecast system. Next, based on the actual business needs and changes in business characteristics of the e-commerce platform sales forecast system, the sales forecast model evaluation cycle is determined, and the sales forecast model is then subjected to indicator testing based on the evaluation cycle. Based on the results of the indicator testing, the sales forecast model is adjusted accordingly, which may include fine-tuning model parameters and optimizing the structure to better adapt it to business needs and data changes, resulting in an updated sales forecast model.
[0034] Based on the updated sales forecast model, sales data for the current time period collected by the e-commerce platform's sales forecasting system is regularly obtained. The current time period sales data is then input into the sales forecast model to obtain sales data for the next time period, providing support for the e-commerce platform's operational decision-making. This process integrates the sales forecast model with the e-commerce platform's sales forecasting system via an interface, enabling seamless integration of the model with the existing system and ensuring smooth integration into the e-commerce platform's business processes. This reduces the cost and risk of system integration and improves the system's overall operational efficiency. Data transformation is performed based on the e-commerce platform's actual business needs and business characteristics to determine the evaluation cycle, enabling the model to adapt promptly to business changes. This helps ensure that the model accurately reflects dynamic business changes and improves the model's adaptability and forecasting accuracy. Regularly acquiring data and performing forecasts using the updated model automates and normalizes sales forecasting. This not only saves manual time and effort but also ensures the timeliness and stability of forecasts, enabling e-commerce platforms to stay abreast of market trends and proactively manage inventory and adjust marketing strategies, thereby improving their operational efficiency and competitiveness.
[0035] like Figure 3 As shown in FIG, the embodiment of this specification provides a schematic diagram of the structure of a sales data prediction device for an e-commerce platform. Figure 3 It can be seen that in one or more embodiments of this specification, a sales data prediction device for an e-commerce platform includes: Processing unit 301 is used to pre-process existing e-commerce sales data to obtain time series features corresponding to the existing e-commerce sales data; Extraction unit 302 is configured to extract features from the time series features based on preset feature requirements to obtain multi-dimensional features corresponding to the existing e-commerce sales data; wherein the multi-dimensional features include sales trend features, seasonal features, local features, and external factor features; A training unit 303 is configured to input the multi-dimensional features into a preset network structure for model training, thereby optimizing the hyperparameters of the preset network structure based on a preset tuning strategy to obtain a sales forecast model that meets the requirements; wherein the preset network structure is based on a long short-term memory network and a temporal convolutional network; The prediction unit 304 is used to deploy the sales prediction model to the corresponding e-commerce platform sales prediction system, so as to regularly predict the sales data of the current time period collected by the e-commerce platform sales prediction system based on the sales prediction model to obtain sales data of the next time period.
[0036] like Figure 4 As shown in FIG, the embodiment of this specification provides a schematic diagram of the sales data prediction device structure of an e-commerce platform. Figure 4It can be seen that in one or more embodiments of this specification, a sales data prediction device for an e-commerce platform includes: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: perform any of the above methods.
[0037] like Figure 5 As shown in FIG, the embodiment of this specification provides a structural diagram of a non-volatile storage medium. Figure 5 It can be seen that in one or more embodiments of this specification, a non-volatile storage medium stores computer-executable instructions 501, and the computer-executable instructions 501 can execute any of the methods described above.
[0038] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simplified. For relevant details, refer to the descriptions of the method embodiments.
[0039] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0040] The foregoing description is merely one or more embodiments of this specification and is not intended to limit this specification. It will be apparent to those skilled in the art that various modifications and variations may be made to one or more embodiments of this specification. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of one or more embodiments of this specification are intended to be within the scope of the claims of this specification.
Claims
1. A sales data prediction method for an e-commerce platform, characterized in that: The method comprises: Preprocessing existing e-commerce sales data to obtain time series features corresponding to the existing e-commerce sales data; Extracting features from the time series features based on preset feature requirements to obtain multi-dimensional features corresponding to the existing e-commerce sales data; wherein the multi-dimensional features include: sales trend features, seasonal features, local features, and external factor features; Inputting the multi-dimensional features into a preset network structure for model training to obtain a sales forecast model that meets the requirements; wherein the preset network structure is based on a long short-term memory network and a temporal convolutional network; The sales forecast model is deployed to the corresponding e-commerce platform sales forecast system to regularly forecast the sales data of the current time period collected by the e-commerce platform sales forecast system based on the sales forecast model to obtain sales data for the next time period.
2. The sales data prediction method of an e-commerce platform according to claim 1, characterized in that: Preprocess existing e-commerce sales data to obtain the time series features corresponding to the existing e-commerce sales data, specifically including: Determine whether the existing e-commerce sales data has missing values, and if so, convert the data format of the existing e-commerce sales data based on a preset nearest neighbor algorithm model, input the converted existing e-commerce sales data into the preset nearest neighbor algorithm model, and obtain the completed existing e-commerce sales data; Calculating the Z-score value corresponding to each of the completed existing e-commerce sales data, comparing the Z-score value with a preset threshold, determining abnormal values in the completed existing e-commerce sales data, filtering the completed existing e-commerce sales data, and obtaining the e-commerce sales data to be extracted; The e-commerce sales data to be extracted are subjected to date analysis to determine the time series features corresponding to the e-commerce sales data to be extracted based on the date features corresponding to each of the e-commerce sales data to be extracted.
3. The sales data prediction method of an e-commerce platform according to claim 1, characterized in that: Feature extraction is performed on the time series features based on preset feature requirements to obtain multi-dimensional features corresponding to the existing e-commerce sales data, specifically including: Calculating the sales volume average of each window of the time series feature based on a given time window, so as to determine the sales volume trend feature corresponding to the existing e-commerce sales volume data based on the sales volume average of each window; Decomposing the time series features based on a preset seasonal decomposition strategy to determine the seasonal features corresponding to the existing e-commerce sales data; The time series features are moved point by point based on a given time window to determine the local features corresponding to the existing e-commerce sales data based on the data in each window; Collecting external factor data associated with the time series features, and encoding the external factor data to obtain external factor features corresponding to the existing e-commerce sales data; The sales trend characteristics, seasonal characteristics, local characteristics and the external factor characteristics are summarized to obtain multi-dimensional characteristics corresponding to the existing e-commerce sales data.
4. The sales data prediction method of an e-commerce platform according to claim 1, characterized in that: Before inputting the multi-dimensional features into a preset network structure for model training, the method further includes: Determine the number of neurons and the number of LSTM units in the preset long short-term memory network based on the number of data and the number of feature dimensions corresponding to the multi-dimensional features; Constructing an input layer and an LSTM layer of the preset long short-term memory network based on the neuron data and the number of LSTM units, and adding a fully connected hidden layer after the LSTM layer to obtain a long short-term memory network of the preset network structure; Setting an input layer of a preset temporal convolutional network based on the input layer of the long short-term memory network, and determining TCN layer parameters of the preset temporal convolutional network based on local features within the multi-dimensional features, so as to set the TCN layer according to the TCN layer parameters; wherein the TCN layer parameters include: a convolution kernel and a dilation factor; Adding a fully connected hidden layer after the TCN layer to obtain a temporal convolutional network of the preset network structure; The outputs of the long short-term memory network and the temporal convolutional network are fused in the feature dimension, and a preset network structure is obtained based on a fully connected layer as an output layer.
5. The sales data prediction method of an e-commerce platform according to claim 1, characterized in that: The multi-dimensional features are input into the preset network structure for model training to obtain a sales forecast model that meets the requirements, specifically including: Dividing the multi-dimensional features based on a preset ratio to obtain a training set, a validation set, and a test set; Inputting the multi-dimensional features in the training set into the preset network structure in batches for training, and optimizing the hyperparameters of the preset network structure based on a preset tuning strategy; Evaluate the preset network structure in each round based on the validation set to determine whether early stopping is required; If so, the preset network structure is evaluated using the multi-dimensional features corresponding to the test set to obtain a sales forecast model that meets the requirements.
6. The sales data prediction method of an e-commerce platform according to claim 1, characterized in that: Optimizing the hyperparameters of the preset network structure based on the preset tuning strategy, specifically including: Calculating the loss value after each round of training of the preset network structure according to a preset loss function; wherein the preset loss function is: mean square error or root mean square error; Determine the gradient of the model parameters in the preset network structure using a back-propagation algorithm and the loss function, so as to update the hyperparameters of the preset network structure according to the gradient; The preset network structure is iteratively updated, and the updated multiple preset network structures are integrated to iteratively train to obtain a sales forecast model that meets the requirements.
7. The sales data prediction method of an e-commerce platform according to claim 1, characterized in that: The sales forecast model is deployed to the corresponding e-commerce platform sales forecast system to regularly forecast the sales data of the current time period collected by the e-commerce platform sales forecast system based on the sales forecast model to obtain sales data for the next time period, specifically including: Integrating the sales forecast model with the corresponding e-commerce platform sales forecast system based on an interface of the corresponding e-commerce platform sales forecast system to deploy the sales forecast model to the corresponding e-commerce platform sales forecast system; Determine an evaluation period for the sales forecast model based on actual business needs and business feature transformation data of the corresponding e-commerce platform sales forecast system, and perform an indicator test on the sales forecast model according to the evaluation period; Adjusting the sales forecast model based on the test results to obtain an updated sales forecast model; Based on the updated sales forecast model, the sales data of the current time period collected by the sales forecast system of the e-commerce platform is regularly obtained, so as to input the sales data of the current time period into the sales forecast model to obtain the sales data of the next time period.
8. A sales data prediction device for an e-commerce platform, characterized in that: The device comprises: A processing unit, configured to pre-process existing e-commerce sales data to obtain time series features corresponding to the existing e-commerce sales data; An extraction unit is configured to extract features from the time series features based on preset feature requirements to obtain multi-dimensional features corresponding to the existing e-commerce sales data; wherein the multi-dimensional features include: sales trend features, seasonal features, local features, and external factor features; A training unit, configured to input the multi-dimensional features into a preset network structure for model training, thereby optimizing the hyperparameters of the preset network structure based on a preset tuning strategy to obtain a sales forecast model that meets the requirements; wherein the preset network structure is based on a long short-term memory network and a temporal convolutional network; The prediction unit is used to deploy the sales prediction model to the corresponding e-commerce platform sales prediction system, so as to regularly predict the sales data of the current time period collected by the e-commerce platform sales prediction system based on the sales prediction model to obtain the sales data of the next time period.
9. A sales data prediction device for an e-commerce platform, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute the method according to any one of claims 1 to 7.
10. A non-volatile storage medium storing computer-executable instructions, characterized in that: The computer executable instructions can execute the method according to any one of claims 1 to 7.