Data prediction method, model generation method, and device

Abstract

Embodiments of the present application provide a data prediction method, a model generation method and equipment. In the method, product features are extracted from a plurality of information associated with a target behavior corresponding to a target product; a target product level to which the target product belongs is determined in combination with the product features and product side attribute information; a target prediction model corresponding to the target product level is determined from prediction models respectively constructed for different product levels; and a first behavior quantity of performing the target behavior for the target product is predicted based on the product features by using the target prediction model. The technical solution provided by the embodiments of the present application improves the data prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a data prediction method, a model generation method, and an apparatus. Background Technology

[0002] With the development of computer and network technologies, online product processing systems have rapidly emerged. With the help of these systems, product sellers can publish products, and users can browse, favorite, and purchase them. To better evaluate products, the number of corresponding actions performed on them is often predicted. For example, predicting the sales volume of products generated by purchasing behavior in the future helps with product warehousing management.

[0003] Taking product sales volume forecasting as an example, the current method is usually time series analysis, which uses historical sales volume analysis to obtain the sales volume for a future period. However, this method has low accuracy when historical sales volume is lacking or when there is little information. Summary of the Invention

[0004] This application provides a data prediction method, a model generation method, and an apparatus to solve the technical problem of low data prediction accuracy in the prior art.

[0005] Firstly, this application provides a data prediction method, including:

[0006] Extract product features from multiple pieces of information related to the target product and the target behavior;

[0007] Based on the product characteristics and product attribute information, the target product level to which the target product belongs is determined;

[0008] From the prediction models constructed for different product levels, determine the target prediction model corresponding to the target product level;

[0009] Using the target prediction model, the number of first actions to be performed on the target product is predicted based on the product characteristics.

[0010] Secondly, this application provides a model generation method, including:

[0011] Construct prediction models corresponding to different product levels;

[0012] For any given product level, identify the sample products belonging to that product level;

[0013] Extract product sample features from multiple pieces of information related to the target behavior corresponding to the sample products;

[0014] Based on the product sample features and the corresponding number of behavior predictions, a prediction model corresponding to the product level is trained.

[0015] Thirdly, embodiments of this application provide a computing device, including a storage component and a processing component, wherein the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement the data prediction method as described in the first aspect above or the model generation method as described in the second aspect above.

[0016] In this embodiment, product features are extracted from multiple pieces of information associated with the target product and the target behavior. Combining these product features with product attribute information, the target product level to which the target product belongs is first determined. Then, a target prediction model corresponding to the target product level is selected from pre-built prediction models corresponding to different product levels. This target prediction model is then used to predict the quantity of the first behavior corresponding to the target behavior based on the product features. By extracting product features from multiple pieces of information associated with the target behavior, the data information involved in the prediction is enriched, which helps improve the accuracy of data prediction. Furthermore, by combining product features with product attribute information, the target product can be stratified, and separate prediction models can be built for different product levels. This allows for targeted prediction of the number of behaviors based on the product features using the target prediction model corresponding to the target product, further improving the accuracy of data prediction.

[0017] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 An exemplary system architecture diagram is shown, in which the technical solutions of the embodiments of this application are applied.

[0020] Figure 2 A flowchart of one embodiment of the data prediction method provided in this application is shown;

[0021] Figure 3 This illustration shows a schematic diagram of the data prediction process in a practical application, based on an embodiment of this application.

[0022] Figure 4A flowchart of an embodiment of a model generation method provided in this application is shown;

[0023] Figure 5 This invention provides a schematic diagram of the structure of one embodiment of a data prediction device.

[0024] Figure 6 This application provides a schematic diagram illustrating the structure of one embodiment of a computing device.

[0025] Figure 7 A schematic diagram of one embodiment of a model generation apparatus provided in this application is shown. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0027] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0028] The technical solution of this application embodiment can be applied to online product processing scenarios to predict the number of target behaviors performed on products provided online, such as the number of product sales, number of views, number of favorites, or number of items added to cart generated by purchasing, browsing, collecting, or adding to cart.

[0029] It is understood that in online product processing scenarios, the products, shopping carts, etc. described in the embodiments of this application are all virtual objects, and are usually represented in different data forms in online processing systems.

[0030] Taking the prediction of product sales volume (hereinafter often referred to as sales volume) as an example, current sales volume forecasting methods are generally based on historical sales volume to predict future sales volume. A common method is to use time series analysis, such as the EMA (Exponential Moving Average). However, this method cannot accurately predict products with a lack of historical sales volume or limited information. For example, for newly released products or products released in a short period of time, such as less than the scheduled number of days (these products are usually called new products), there may be no historical sales volume, or the historical sales volume may not show obvious patterns or may be discontinuous, all of which will affect the accuracy of the final prediction.

[0031] To improve the accuracy of data prediction, the inventors, through a series of studies, innovatively proposed the technical solution of this application. In the embodiments of this application, product features are extracted from multiple pieces of information related to the target behavior corresponding to the target product, not only limited to the historical behavior quantity of the target behavior, thereby enriching the data information involved in the prediction and helping to improve the accuracy of data prediction. Furthermore, by combining product features and product attribute information, the target product can be stratified, and prediction models can be built separately for different product levels. Thus, the target prediction model corresponding to the target product can be used to make targeted predictions of the number of behaviors based on product features, further improving the accuracy of data prediction.

[0032] 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, and 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.

[0033] Figure 1 This is an exemplary system architecture diagram that can be applied to embodiments of this application. The system architecture may include a client 101 and a server 102. The server 102 can construct the relevant models involved in the embodiments of this application, and the trained model can be deployed on the server 102 or on the client 101 to implement the data prediction method provided in the embodiments of this application. Generally, the data prediction method provided in the embodiments of this application can be executed by the server 102.

[0034] Client 101 can generally be understood as an application deployed on an electronic device, which can be one or more of a smartphone, tablet, and laptop, or of course, a desktop computer, etc. For ease of understanding, Figure 1The client is primarily represented by a device image. Various other types of applications can also be configured on electronic devices, such as search engines and instant messaging applications. Of course, the client 101 can also be a browser, a web application such as an H5 (HyperText Markup Language 5) application, a lightweight application (also known as a mini-program), or a cloud application, etc.

[0035] Server 102 can be either hardware or software. When the server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services), or as a single software program or software module. No specific limitations are made here. The server can also be a server in a distributed system, or a server integrated with blockchain technology. The server can also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology, etc. Figure 1 The server is primarily represented by a cloud server. The model building process and the prediction process using the model can be implemented by different servers or different software modules within server 102.

[0036] The implementation details of the technical solutions in the embodiments of this application are described in detail below.

[0037] Figure 2 A flowchart of one embodiment of a data prediction method provided in this application is shown. The method may include the following steps:

[0038] 201: Extract product features from multiple pieces of information related to the target product and the target behavior.

[0039] The target product can refer to any type of product.

[0040] Furthermore, in a practical application, the target product can refer to a product categorized according to its release time. For example, a target product could be a product released less than a predetermined time, with the target type being a "new product." In practical applications, this means a new product, such as a product released less than 28 days ago. Since historical data for new products is scarce or limited, the technical solution presented in this application can achieve effective and accurate prediction.

[0041] Target behaviors can refer to purchase behavior, browsing behavior, favorite behavior, or add-to-cart behavior, etc. Add-to-cart behavior can lead to product sales, browsing behavior can lead to user traffic, favorite behavior can lead to the number of product favorites, and add-to-cart behavior can lead to the number of add-to-carts, etc. Therefore, the technical solution of this application embodiment can be applied to predict product sales, user traffic, number of favorites, and number of add-to-carts, etc. The prediction of product sales has guiding significance for product warehouse management. In one or more embodiments below, product sales are mainly used as an example to describe the technical solution of this application. Of course, those skilled in the art will understand that the technical solution of this application is not limited to the prediction of product sales. Any time-sequential behavior quantity resulting from any behavior performed on the product can be predicted according to the technical solution of this application.

[0042] The multiple pieces of information associated with the target behavior can refer to information that can influence the user to perform the target behavior, such as information that helps incentivize the performance of the target behavior. In one optional implementation, these multiple pieces of information may include one or more of the following: product text information, product attribute information, product conversion information generated in the first historical time period, related product information, and product-related information. The first time period can refer to the period formed by a historical point in time and the current prediction time, such as the period consisting of the 28 days prior to the current prediction time. For new products, the first time period can also refer to the period from the product launch time to the current prediction time.

[0043] Product text information may include, for example, product title text and / or multiple product review texts generated during the first time period. Product title text refers to the text provided by the product provider when publishing the product in the product processing system. Users can obtain relevant information about the product based on the product title text. Product title text is usually displayed on product aggregation pages corresponding to multiple products or product description pages corresponding to a single product. The specific content of the product title text will affect the probability of performing the target action on the product.

[0044] In the product processing system, after purchasing a product online, the product provider entrusts a logistics provider to mail the actual product to the buyer. The product review text can refer to the buyer's feelings about the received product quality, logistics and delivery service quality, and the product provider's service quality. This information is recorded in the system, and any user browsing the product can see the product review text. Users can refer to the product review text to decide whether to perform the target action. Therefore, the product review text also affects the probability of performing the target action on the product.

[0045] Product attribute information may include, for example, the product category, product brand, and the dates of any promotional activities participated in.

[0046] Product conversion information can include time-series data on the number of the target behavior and at least one related behavior generated within the first time period. The target behavior can be, for example, a purchase, and the corresponding related behavior can include browsing, favorites, adding to cart, etc., which contribute to the execution of the target behavior. Therefore, the product conversion information can specifically include, for example, time-series data on product sales, traffic conversion rates, and the number of products added to cart. Furthermore, product conversion information can also include product transaction volume, etc. Specifically, the time-series data on product sales can be composed of product sales at each point in the first time period arranged in chronological order; the time-series data on user traffic conversion rates can be composed of user traffic conversion rates at each point in the first time period arranged in chronological order; and the time-series data on the number of products added to cart can be composed of the number of products added to cart at each point in the first time period arranged in chronological order. Each point in time can be, for example, a day, a month, a quarter, a year, or another specific time period.

[0047] Related product information can refer to relevant information about products related to the target product, such as the time-series data of the number of target behaviors corresponding to related products in the first time period. It can also include attribute information of related products. Related products can be products in the same category as the target product, where the category can be divided according to product nature, such as tops, pants, hats, etc.

[0048] The "product provider" refers to the supplier of the target product. Product provider information may include, for example, time-series data on the number of target behaviors generated by multiple products of the same type as the target product within the first time period, as well as time-series data on the number of behaviors of at least one related product. For instance, if the target product is new, then all of these products are also new. Furthermore, product provider information may also include product provider attribute information, such as the average sales volume per unit time or the total sales volume within a certain time period. The product provider's sales volume is obtained by statistically analyzing the sales volumes of multiple products provided by the product provider.

[0049] Product characteristics can be obtained from multiple pieces of information associated with target behavior through statistical analysis, data extraction, cleaning, and normalization.

[0050] The product features include sub-features extracted from multiple pieces of information, and the specific extraction methods will be described in detail in the examples below.

[0051] 202: Based on the product characteristics and the corresponding product owner attribute information, determine the target product level to which the target product belongs.

[0052] 203: Determine the target prediction model corresponding to the target product level from the prediction models constructed for different product levels.

[0053] Since the probability of target behavior may vary for different products from different product manufacturers, in order to make more accurate data predictions in this embodiment, products can be segmented according to product manufacturer attribute information and product characteristics, and prediction models for data prediction can be built for different product segments.

[0054] Product attribute information may include, for example, the product name and product sales volume. Product sales volume may refer to the average sales volume of all products within a unit of time or the total sales volume of all products within a certain period of time.

[0055] For example, product popularity can be identified based on product characteristics, such as whether it is likely to be a "hot-selling" product. Generally, hot-selling products are more likely to be popular with users, thus triggering a series of user behaviors. The product seller's level can be determined based on their attributes, such as whether they are a top seller. Generally, products offered by top sellers are more likely to be popular with users, triggering a series of user behaviors. Based on this, for example, four product tiers can be defined: hot-selling product + top seller; non-hot-selling product + top seller; hot-selling product + non-top seller; and non-hot-selling product + non-top seller. Of course, this is just an example of how to perform product tiering; in practical applications, there are other possible implementation methods, and this application is not limited to this.

[0056] By building prediction models for different product levels, more targeted predictions can be made for different products, thereby further improving the accuracy of data prediction.

[0057] Therefore, by using the product characteristics identified above and the product attribute information of the supplier of the target product, the target product level to which the target product belongs can be determined, and the target prediction model corresponding to the target product level can be determined.

[0058] The prediction model corresponding to each product level can be pre-trained. For example, it can be trained in a supervised manner using the product characteristics of the sample products corresponding to each product level and the number of corresponding prediction behaviors. This will be described in more detail in the corresponding embodiments below.

[0059] 204: Using a target prediction model, predict the number of first actions to be performed on the target product based on product characteristics.

[0060] By using the established target prediction model and inputting product characteristics into it, the number of first actions to be performed on the target product can be predicted. Specifically, the number of first actions can be predicted for a future second time period starting from the current prediction time, such as the next day, week, or month. This second time period can be a unit of time corresponding to a point in the first time period, such as a day, or a length of time consisting of multiple units of time, such as a week. The number of first actions can include the number of target actions that may occur at each point in time within this second time period.

[0061] In this embodiment, by extracting product features from multiple sources, not only limiting the historical behavior quantity of the target behavior, the data information involved in the prediction is enriched, which helps to improve the accuracy of data prediction. Furthermore, by combining product features with product attribute information, the target product can be stratified, and a separate prediction model can be built for different product strata. Thus, the target prediction model corresponding to the target product can be used to make targeted predictions of behavior quantity based on product features, which further improves the accuracy of data prediction.

[0062] As described above, the target product's tier can be determined by combining the product's popularity type and the seller's level. As an optional method, the seller's level can be determined based on seller attribute information, such as sales volume and seller name. Assuming seller levels can be categorized as top sellers and non-top sellers, a seller's status as a top seller can be determined by whether their name corresponds to a specific level category or whether their sales volume exceeds a certain threshold. For example, if the seller's name matches a pre-defined top seller name and / or their sales volume exceeds a certain threshold, they are considered a top seller; otherwise, they are considered a non-top seller.

[0063] In addition, as an alternative approach, the product's level type can be determined by combining the product's attribute information and the corresponding behavioral trend information.

[0064] This behavior quantity trend information indicates the changes in the number of behaviors of the product provider in response to the target behavior. This behavior quantity trend information may include an upward trend, a downward trend, or a stable trend.

[0065] This allows us to combine product attribute information with the trend information of the behavior to determine the corresponding level type. For example, if at least one of the following conditions is met: the product name is a top seller name, the product sales volume is greater than a certain threshold, and the behavior trend information is on an upward trend, then the product seller can be identified as a top seller; otherwise, it is a non-top seller.

[0066] One way to implement behavioral quantity trend information is as follows:

[0067] Statistical analysis of the historical number of target behaviors corresponding to the target product is conducted to obtain trend information on the number of behaviors.

[0068] The historical target behavior quantity can refer to the time-series data of the target product's target behavior generated in a specific time period, such as the time-series data of the target behavior quantity for each day of the previous 28 days. By statistically analyzing the historical target behavior quantity corresponding to the target product, it is possible to determine whether the target product's target behavior quantity is increasing, decreasing, or remaining stable. This trend information of the target product's target behavior quantity can then be used as the product's behavior quantity trend information. The trend judgment of the historical target behavior quantity can be obtained using cluster analysis or other statistical analysis methods; this application does not specifically limit this method.

[0069] To further improve the accuracy of information, another approach could be:

[0070] Statistical analysis is performed on the historical target behavior count of at least one product of the same type as the target product provided by the product provider to obtain behavior count trend information;

[0071] The target product may or may not be included, at least one product of the same type as the target product.

[0072] The historical target behavior count of at least one product can be time-series data of the target behavior count of at least one product in a first time period, such as the time-series data of the target behavior count of each day in the previous 28 days.

[0073] By statistically analyzing the historical target behavior counts of at least one product, such as obtaining trend information on the behavior counts of each product, and performing cluster analysis on the trend information of the behavior counts of each product, the trend information of the behavior counts with the highest proportion can be used as the trend information of the behavior counts of the product. Alternatively, if the historical target behavior counts are time series data, the average number of target behaviors or the total number of target behaviors at each time point (e.g., each day) can be statistically analyzed based on the time series data of the behavior counts of at least one product. Thus, the trend information of the behavior counts of the product can be obtained based on the average number of target behaviors or the total number of target behaviors at each time point.

[0074] As described above, multiple pieces of information may include one or more of the following: product text information, product attribute information, product conversion information generated in the first time period, related product information, and product-related information.

[0075] In some embodiments, product features may be composed of product sub-features obtained by extraction according to one or more of the following implementation methods:

[0076] A: Extract target entity words belonging to the target entity type from the product title text; according to the evaluation data of multiple heat evaluation indicators corresponding respectively before and after the appearance of the target entity word in multiple product title texts, statistically obtain the gain data of the target entity word corresponding to multiple heat evaluation indicators; generate product features based on the gain data.

[0077] First, a named entity recognition model can be used to identify entity words and the entity type corresponding to each entity word from the product title file. [[ID=,8]]

[0078] After that, entity calibration can be performed on the entity extraction results, and calibration can be carried out according to the calibration rules. For example, for an entity word recognized for a certain entity type, the calibration word matching it in the calibration rules can be found as the entity word corresponding to this entity type. For instance, if the entity word recognized for the entity type "obvious" is "Li Yi", and the calibration rule defines "star" as "Li Yisan", then the recognized "Li Yi" can be calibrated to "Li Yisan".

[0079] In addition, the obtained entity types and entity words can be screened according to the entity screening rules, screening out entity types and entity words that do not conform to the screening rules, etc., and finally obtaining the target entity words corresponding to the target entity type, etc. ,

[0080] After that, for each heat evaluation indicator, according to the evaluation data of the corresponding heat evaluation indicator before and after the appearance of each target entity word in multiple product title texts, statistically obtain the gain data of the target entity word corresponding to the heat evaluation indicator before and after the appearance in each product title text, and then the gain data corresponding to each product title text can be weighted and fused, etc., to obtain the final gain data corresponding to the heat evaluation indicator. Among them, each product title text can correspond to a product, multiple product title texts can correspond to different products, etc., and the multiple product title texts can be multiple products historically recorded in the product processing system, etc.

[0081] Among them, the gain data can be expressed as the difference or ratio, etc., of two evaluation data of the same heat evaluation indicator.

[0082] Of course, the gain data can also be statistically obtained in advance, that is, the gain data of the target entity word corresponding to multiple heat evaluation indicators can be directly obtained, etc.

[0083] These multiple popularity evaluation indicators may include, for example, product sales and user traffic conversion rate. User traffic conversion rate refers to the percentage of purchases generated by that user traffic, where user traffic refers to the number of visits or views of the product. For ease of calculation, product sales can refer to the average sales volume of the product per unit of time, i.e., at each point in time.

[0084] In addition, the aforementioned target entity words can also refer to entity words in the product title file that correspond to user traffic conversion rates or product sales exceeding a certain value.

[0085] Among them, gain data can be used as product sub-features, or it can be normalized, transformed into feature vectors, and then used as product sub-features.

[0086] The above example illustrates one way to obtain product sub-features based on product title text. Of course, other methods can also be used, such as extracting keywords from the product title text (e.g., target entity words or those extracted using a pre-trained extraction model) and directly using the vector representations corresponding to the keywords as product sub-features. This application does not impose too many restrictions on this.

[0087] B: Analyze the sentiment polarity of multiple product review texts for at least one target sentiment aspect; count the number of occurrences of at least one target sentiment aspect in multiple product review texts, the number of corresponding sentiment polarities, and the percentage values, and generate statistical data; based on the statistical data, generate product features.

[0088] Among them, the sentiment polarity of multiple product review texts in at least one target sentiment aspect can be mined using ABSA (aspect-based sentiment analysis).

[0089] Optionally, multiple emotional aspects corresponding to multiple product review texts and the emotional polarity of each emotional aspect can be obtained first. Emotional aspects may include, for example, taste, quality, packaging, and expiration date. Emotional polarity can mainly include three polarities: positive, negative, and neutral.

[0090] Then, the multiple sentiment aspects obtained from the analysis can be combined with the target product's category and industry, and filtered according to sentiment filtering rules to obtain at least one target sentiment aspect. These sentiment filtering rules may specify sentiment aspects corresponding to specific categories or industries.

[0091] Next, statistical data is generated by counting the number of occurrences of at least one target sentiment aspect in multiple product review texts, the number of occurrences of each corresponding sentiment polarity, and the percentage of occurrences. For example, assuming seven target sentiment directions are retained, the statistical data consists of 56 data points: the occurrence count of each of the seven sentiment aspects, the number of occurrences of each of the seven sentiment aspects corresponding to three sentiment polarities, and the percentage of occurrences of each of the seven sentiment aspects. This statistical data can be used as a product sub-feature, or it can be converted into a feature vector expression after undergoing necessary processing such as normalization and feature transformation before being used as a product sub-feature.

[0092] C: Perform statistical processing on the time-series data of the number of behaviors corresponding to the target product for each of the multiple behaviors; generate product sub-features based on the statistical results of the data corresponding to the multiple behaviors of the target product.

[0093] These multiple behaviors include the target behavior and at least one associated behavior corresponding to the target behavior.

[0094] The statistical results for each behavior may include, for example, the mean, variance, median, average, and / or total number obtained from the time-series data of the number of behaviors.

[0095] The final statistical results can be used as product sub-features, or after normalization, feature transformation and other operations, they can be converted into feature vector expressions and then used as product sub-features.

[0096] D: Perform statistical processing on the time-series data of the number of target behaviors generated by related products in the first time period; generate product sub-features based on the statistical results of the target behaviors corresponding to related products;

[0097] For example, the mean, variance, median, average, and / or total number obtained from the time-series data of the behavior quantity of related products can be used as the statistical results.

[0098] The final statistical results can be used as product sub-features, or after normalization, feature transformation and other operations, they can be converted into feature vector expressions and then used as product sub-features.

[0099] E: Perform statistical processing on the time-series data of the number of behaviors of multiple products of the same type as the target product; generate product sub-features based on the statistical results of the data corresponding to the multiple behaviors of each product.

[0100] These multiple behaviors include the target behavior and at least one associated behavior corresponding to the target behavior.

[0101] In other words, in addition to generating product sub-features based on the statistical results of multiple behaviors corresponding to the target product, product sub-features can also be generated based on the statistical results of multiple behaviors corresponding to multiple products of the same type as the target product.

[0102] The statistical results for each product corresponding to each behavior can include, for example, the mean, variance, median, average, and / or total number obtained from time-series data on the number of behaviors.

[0103] The final statistical results can be used as product sub-features, or after normalization, feature transformation and other operations, they can be converted into feature vector expressions and then used as product sub-features.

[0104] The product feature can be obtained by constructing one or more product sub-features extracted from the above A to E operations.

[0105] As described above, the data involves various types of behavioral quantity time series data. For example, product conversion information may include time series data of the number of target behaviors generated in the first time period and at least one related behavior of the target behavior. Since some abnormal behavior quantities may appear in the historical behavior quantities for various reasons, in order to ensure the accuracy of the information, in some embodiments, the method may further include:

[0106] For any given behavior, adjust the number of behaviors generated at time points in the time series data that meet the abnormal conditions.

[0107] One approach is to use a sliding window to analyze the time series data of the number of behaviors in the first time period and detect time points that meet the abnormal conditions. The size of the sliding window can be set according to the actual situation. Data statistics can be performed on the number of behaviors corresponding to each time point in the sliding window, such as calculating the variance. If the data statistics result meets the abnormal conditions, the number of behaviors corresponding to each time point in the current sliding window can be considered as abnormal data and can be adjusted accordingly.

[0108] One adjustment method is to directly delete the number of behaviors generated at time points that meet the abnormal conditions in the first time period. That is, after deletion, the number of behaviors corresponding to time points that meet the abnormal conditions can be updated to 0.

[0109] Of course, as another adjustment method, the number of behaviors at time points that meet the abnormal conditions in the first time period can also be updated to a predetermined number. For example, the predetermined number can be the average value of the number of behaviors at each non-abnormal time point in the first time period, or the predetermined number can be the number of behaviors corresponding to the non-abnormal time point closest to the abnormal time point, etc.

[0110] The first number of actions predicted using the target prediction model can be used as the predicted number of actions for the target product to perform corresponding processing operations. Therefore, in some embodiments, the method may further include:

[0111] Generate product prompt information based on the quantity in the first line;

[0112] Notify the product provider of product alerts.

[0113] The product notification information is used to indicate the quantity of the first item. Notifying the product supplier of this information helps them understand the future quantity of the target product in the first item, which is beneficial for warehouse management, such as timely replenishment.

[0114] There are several ways to notify the product provider of product alerts. For example, the product alert can be sent to the product provider's corresponding communication account, which could be a mobile communication account such as a phone number, an instant messaging account, or an email account.

[0115] Alternatively, the product notification message can be sent to the product provider's corresponding client account, allowing the product provider to view the notification message through the client. Figure 1 In the system architecture diagram, the product prompt information generated by the server 102 can be sent to the client 101 to achieve the purpose of notifying the product provider of the product prompt information.

[0116] Furthermore, to further improve prediction accuracy, in some embodiments, the method may also include:

[0117] Based on the time-series data of the number of target behaviors generated by the target product in the first time period, at least one prediction method is used to predict the number of second behaviors that perform target behaviors on the target product.

[0118] Based on the number of first behaviors and at least one number of second behaviors, the predicted number of behaviors for the target product is obtained by fusion.

[0119] In this process, based on the number of behaviors at each time point in the time series data corresponding to the target behavior, time series analysis methods, such as EMA, single time series average, weighted time series average, moving average, and weighted moving average, can be used to predict a second behavior quantity. For example, when predicting the behavior quantity at the next time point, the weighted average of the behavior quantities at each time point can be used as the second behavior quantity. The weight coefficients corresponding to each time point can be set according to the principle that the further back in time, the smaller the weight coefficient, and the closer the time, the larger the weight coefficient.

[0120] In addition, time-series prediction models, such as deep neural network models, can be used for prediction. These models are pre-trained. For example, in scenarios where a time-series prediction model is used to predict daily sales for the next 28 days based on daily sales in the first 28 days of a new product, the training samples can be historical sales time-series data for any new product. A rolling time window approach is used, taking the sales figures for the past 28 time points within the rolling time window as input and the sales figures for the next 28 time points as output for model training.

[0121] Specifically, the number of first behaviors and at least one number of second behaviors can be averaged or weighted to obtain the predicted number of behaviors for the final target product. Therefore, the product prompt information can be generated specifically based on this predicted number of behaviors.

[0122] In some embodiments, the method may further include:

[0123] Determine if the quantity in the first row is greater than the quantity threshold;

[0124] If not, the number of the first action will be used as the predicted number of actions for the target product;

[0125] If so, the quantity threshold will be used as the predicted quantity of the target product.

[0126] By adding a quantity threshold, the quantity of the first row can be constrained, thereby ensuring the stability of the prediction results.

[0127] The quantity threshold can be, for example, the quantity of behavior predicted based on the time series data of the quantity of target behavior generated by the target product in the first time period. In other words, the quantity threshold can be the second quantity of behavior in the above embodiment.

[0128] Of course, preset settings can also be made based on the actual situation.

[0129] In a practical application, the target behavior in this application embodiment can specifically refer to the purchasing behavior, and the first behavior quantity means the sales quantity. The technical solution of this application embodiment can be specifically used to predict the sales volume of new products, so as to achieve the accuracy of new product sales prediction when new product sales data is scarce or limited. The technical solution of this application embodiment is described below using the prediction of new product sales quantity as an example. Figure 3The processing diagram shown illustrates the process. For a new product to be predicted, the process first includes data input operation 301 to determine multiple pieces of information associated with the purchase behavior. These may include product text information, such as product title text and product review text; product attribute information; product conversion information generated in the first time period, such as sales time-series data, user traffic conversion rate time-series data, and add-to-cart quantity time-series data; related product information, such as sales time-series data of related products in the first time period; and product-related information, such as sales time-series data, user traffic conversion rate time-series data, and add-to-cart quantity time-series data of multiple new products in the first time period.

[0130] Afterwards, anomaly detection (302) can be performed on the time series data, and various types of time series data can be readjusted, etc.

[0131] Subsequently, product sub-features can be extracted from multiple pieces of information and assembled into product feature 303. These can include: extracting text features from product text information, including product title text and product review text, as product sub-features; directly using product attribute information as product sub-features; extracting product sub-features from product conversion information, such as sales time-series data, user traffic conversion rate time-series data, and add-to-cart quantity time-series data; extracting product sub-features from related product information, such as sales time-series data of related products within a first time period; and extracting product sub-features from product-related information, such as sales time-series data, user traffic conversion rate time-series data, and add-to-cart quantity time-series data of multiple new products within a first time period.

[0132] Next, based on product characteristics and seller attributes, the target product tier (304) to which the new product belongs can be determined. For example, according to different levels and popularity types, it can be divided into four product tiers: best-selling product + top sellers; non-best-selling product + top sellers; best-selling product + non-top sellers; and non-best-selling product + non-top sellers. Combining the new product's characteristics and seller attributes, the target product tier to which it belongs can be determined.

[0133] Then, the target prediction model corresponding to the target product level is used to predict the quantity of the first row, which is 305.

[0134] Furthermore, the number of at least one second behavior obtained by combining a quantity threshold and at least one prediction method can be used to constrain and fuse 306 to obtain the final predicted behavior number corresponding to the new product.

[0135] Based on the predicted number of behaviors, product notification messages can be generated to notify product providers, etc.

[0136] The prediction models corresponding to different product levels involved in the above embodiments can be pre-built, such as... Figure 4The flowchart described herein is for another embodiment of a model generation method provided in this application. The method may include the following steps:

[0137] 401: Construct prediction models corresponding to different product levels.

[0138] The prediction models for different product levels can have the same model architecture and initial model parameters, but the final model parameters will be different due to different training samples.

[0139] 402: For any product level, identify the sample products belonging to that product level.

[0140] The sample product can be selected from historical products that belong to the target type. For example, if the target type is a new product, it can be a historical new product from the product processing system.

[0141] 403: Extract product sample features from multiple pieces of information related to the target behavior corresponding to the sample product.

[0142] The sample product's information related to the target behavior may include, for example, one or more of the following: product text information, product attribute information, product conversion information generated in the first time period, related product information, and product-related information; product text information includes product title text and / or multiple product review texts generated in the first time period; product conversion information includes time-series data of the number of behaviors corresponding to the target behavior and at least one related behavior generated in the first time period; related product information includes time-series data of the number of behaviors of the sample product's related products that generated the target behavior in the first time period; and product-related information includes time-series data of the number of behaviors of multiple products of the same type as the sample product that generated the target behavior and at least one related behavior in the first time period.

[0143] The product sample features can be composed of product sub-sample features extracted using one or more of the following methods:

[0144] Extract target entity words belonging to the target entity type from the product title text; based on the evaluation data of multiple popularity evaluation indicators corresponding to the target entity words before and after their appearance in multiple product title texts, respectively, obtain the gain data of the target entity words corresponding to multiple popularity evaluation indicators; based on the gain data, generate product subsample features;

[0145] Analyze the sentiment polarity of multiple product review texts for at least one target sentiment aspect; count the occurrence, corresponding polarity, and percentage of each target sentiment aspect in the multiple product review texts to generate statistical data; and generate product subsample features based on the statistical data.

[0146] Statistical processing is performed on the time-series data of the number of behaviors corresponding to multiple behaviors of the sample products; based on the statistical results of the data corresponding to multiple behaviors of the sample products, product sub-sample features are generated.

[0147] The time-series data of the number of target behaviors generated by related products in the first time period are statistically processed; based on the statistical results of the target behaviors of related products, product sub-sample features are generated.

[0148] as well as,

[0149] Statistical processing is performed on the time-series data of the number of behaviors of multiple products of the same type as the sample product; based on the statistical results of the data corresponding to the multiple behaviors of multiple products, product sub-sample features are generated.

[0150] The method for extracting product sample features is the same as the method for extracting product features from multiple pieces of information about the target product described earlier, and will not be repeated here. The types of this multiple pieces of information have also been described in detail earlier, and will not be repeated here. Furthermore, the various types of behavioral quantity time-series data involved can also be used for anomaly detection, adjusting the number of behaviors generated at time points that meet abnormal conditions in the behavioral quantity time-series data, etc.

[0151] Furthermore, by combining product attribute information and product characteristics, the sample products corresponding to a given product level can be determined. Therefore, for each product in the historical product dataset, product features can be extracted from multiple pieces of information associated with the target behavior, and combined with their respective product attribute information and product features to determine the sample products corresponding to that product level. This historical product dataset can consist of historical products of the target type.

[0152] 404: Based on the product sample features and the corresponding number of behavior predictions, train the prediction model corresponding to the product level.

[0153] This can involve extracting product sample features from multiple pieces of information related to the target behavior generated within a first time period of the sample product. This first time period can refer to a specified duration starting from the product's release date, such as a point in time measured in days. It could also refer to multiple pieces of information related to the target behavior that occurred within the first 28 days of the release date. The predicted behavior quantity can refer to the predicted behavior quantity obtained from one or more days after the 28th day from the release date. If it is for multiple days, the predicted behavior quantity can refer to the time-series data of the behavior quantity corresponding to those multiple days.

[0154] Then, based on the product sample characteristics and the corresponding number of behavior predictions, the prediction model corresponding to that product level can be trained.

[0155] Once trained, the prediction model can be used to predict the quantity of the first behavior of target products belonging to that product level.

[0156] In practical applications, the first line, "quantity," can specifically refer to the number of products sold, etc.

[0157] Figure 5 This application provides a schematic diagram of the structure of a data prediction device according to one embodiment. The device may include:

[0158] The feature extraction module 501 is used to extract product features from multiple pieces of information related to the target product and the target behavior.

[0159] The hierarchical classification module 502 is used to determine the target product hierarchy to which the target product belongs by combining product characteristics and product attribute information;

[0160] The model determination module 503 is used to determine the target prediction model corresponding to the target product level from the prediction models built for different product levels.

[0161] The quantity prediction module 504 is used to predict the quantity of the first action to be performed on the target product based on the product characteristics using the target prediction model.

[0162] In some embodiments, the hierarchical segmentation module can be specifically used to utilize a product classification model to identify the popularity type of the target product based on product characteristics; determine the level type of the product to which the target product belongs based on product attribute information; and determine the target product hierarchy composed of the popularity type and the level type.

[0163] In some embodiments, the device may further include:

[0164] The trend statistics module is used to count the number of historical target behaviors of the target product and obtain trend information on the number of behaviors; or, to count the number of historical target behaviors of at least one product of the same type as the target product provided by the product provider and obtain trend information on the number of behaviors.

[0165] This hierarchical classification module determines the corresponding product level type based on the product attribute information, including: combining product attribute information and behavioral quantity trend information to determine the level type to which the product belongs.

[0166] In some embodiments, the multiple information includes one or more of the following: product text information, product attribute information, product conversion information generated in the first time period, related product information, and product supplier information; the product text information includes product title text and / or multiple product review texts generated in the first time period; the product conversion information includes time-series data of the number of behaviors corresponding to the target behavior and at least one related behavior generated in the first time period; the related product information includes time-series data of the number of behaviors of related products that generate the target behavior in the first time period; and the product supplier information includes time-series data of the number of behaviors of multiple products of the same type as the target product that generate the target behavior and at least one related behavior in the first time period.

[0167] The feature extraction module will specifically extract product sub-features according to one or more of the following implementation methods, which constitute the product features.

[0168] Extract target entity words belonging to the target entity type from the product title text; based on the evaluation data of multiple popularity evaluation indicators corresponding to the target entity words before and after their appearance in multiple product title texts, respectively, obtain the gain data of the target entity words corresponding to multiple popularity evaluation indicators; generate product sub-features based on the gain data;

[0169] The process involves analyzing the emotional polarity of the multiple product review texts for at least one target emotional aspect; statistically analyzing the occurrence count, corresponding polarity count, and percentage of each target emotional aspect in the multiple product review texts to generate statistical data; and generating product sub-features based on the statistical data.

[0170] Statistical processing is performed on the time-series data of the number of behaviors corresponding to the target product for each behavior; based on the statistical results of the data corresponding to the multiple behaviors of the target product, product sub-features are generated.

[0171] The time-series data of the number of behaviors of the associated products that generate the target behavior in the first time period are statistically processed; based on the statistical results of the data of the associated products corresponding to the target behavior, product sub-features are generated.

[0172] as well as,

[0173] Statistical processing is performed on the time-series data of the number of behaviors of multiple products of the same type as the target product; based on the statistical results of the data corresponding to the multiple behaviors of the multiple products, product sub-features are generated.

[0174] In some embodiments, the product conversion information includes time-series data of the number of behaviors corresponding to the target behavior generated in a first time period and at least one associated behavior of the target behavior; the device may further include:

[0175] The anomaly detection module is used to adjust the number of behaviors that meet the anomaly conditions at the time points in the first time period for any type of behavior; and to obtain time-series data of behavior quantity based on the number of behaviors corresponding to different time points in the first time period.

[0176] In some embodiments, the quantity prediction module is further configured to predict the number of second behaviors performed on the target product based on the time-series data of the number of target behaviors generated by the target product in a first time period, using at least one prediction method; and to obtain the predicted number of behaviors of the target product by fusing the first behavior number and at least one second behavior number.

[0177] In some embodiments, the quantity prediction module is further configured to determine whether the first behavior quantity is greater than a quantity threshold; if not, the first behavior quantity is used as the predicted behavior quantity of the target product; if so, the quantity threshold is used as the predicted behavior quantity of the target product.

[0178] In some embodiments, the device may further include:

[0179] The notification module is used to generate product prompt information based on the number of the first actions; and to notify the product provider of the product prompt information.

[0180] Figure 5 The data prediction device can perform Figure 2 The implementation principle and technical effects of the data prediction method described in the illustrated embodiments will not be repeated here. The specific methods by which each module and unit of the data prediction device in the above embodiments performs its operations have been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0181] In one possible design, Figure 5 The data prediction apparatus of the illustrated embodiment can be implemented as a computing device, which can serve as such as Figure 1 The server or client shown, such as Figure 6 As shown, the computing device may include a storage component 601 and a processing component 602;

[0182] The storage component 601 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component to achieve, for example... Figure 2 The data prediction method shown.

[0183] The processing component 602 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0184] Storage component 601 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0185] Of course, computing devices may also include other components, such as input / output interfaces and communication components.

[0186] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc. Communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0187] When the computing device is used as a server-side device, it can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0188] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 2 The data prediction method of the illustrated embodiment. The computer-readable medium may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof.

[0189] This application also provides a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program, when executed by a computer, can perform the above-described functions. Figure 2 The data prediction method of the embodiment shown.

[0190] In such embodiments, the computer program may be downloaded and installed from a network, and / or installed from a removable medium. When the computer program is executed by a processor, it performs the various functions defined in the system of this application.

[0191] Figure 7 This application provides a schematic diagram of the structure of a model generation apparatus according to one embodiment. The apparatus may include:

[0192] Model building module 701 is used to build prediction models corresponding to different product levels;

[0193] The sample determination module 702 is used to determine the sample products belonging to any given product level.

[0194] The sample processing module 703 is used to extract product sample features from multiple pieces of information related to the target behavior corresponding to the sample product;

[0195] The model training module 704 is used to train the prediction model corresponding to the product level based on the product sample features and the corresponding number of behavior predictions.

[0196] Figure 7 The model generation device described above can perform Figure 4 The implementation principle and technical effects of the model generation method described in the illustrated embodiments will not be repeated here. The specific methods by which each module and unit of the model generation apparatus in the above embodiments performs its operations have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0197] In one possible design, Figure 7 The model generation apparatus of the illustrated embodiment can be implemented as a computing device, which can be used as such Figure 1 The server shown may include storage components and processing components.

[0198] The storage component stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component to achieve, for example... Figure 4 The model generation method shown.

[0199] Among them, the organizational structure of the computing device and Figure 6 The organizational structure of the computing devices shown can be the same, therefore, it will not be drawn again here.

[0200] Of course, computing devices may also include other components, such as input / output interfaces and communication components.

[0201] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc. Communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0202] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0203] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 4 The illustrated embodiment describes a model generation method. The computer-readable medium may be included in the electronic device described in the above embodiments; alternatively, it may exist independently and not assembled into the electronic device. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof.

[0204] This application also provides a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program, when executed by a computer, can perform the above-described functions. Figure 4 The model generation method of the embodiment shown.

[0205] In such embodiments, the computer program may be downloaded and installed from a network, and / or installed from a removable medium. When the computer program is executed by a processor, it performs the various functions defined in the system of this application.

[0206] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0207] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0208] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, 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 can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0209] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A data prediction method characterized by, The method comprises the following steps: extracting product features from a plurality of information associated with target behavior corresponding to the target product; determining the target product level to which the target product belongs by combining the product features and product party attribute information; determining the target prediction model corresponding to the target product level from the prediction models respectively constructed for different product levels; predicting the first behavior quantity of the target behavior performed on the target product based on the product features by using the target prediction model.

2. The method of claim 1, wherein, The step of determining the target product level to which the target product belongs by combining the product features and product party attribute information comprises the following steps: identifying the heat type of the target product based on the product features by using a product classification model; determining the level type to which the product party corresponding to the target product belongs according to product party attribute information; determining the target product level composed of the heat type and the level type.

3. The method of claim 2, wherein, The method further comprises the following steps: statistically obtaining behavior quantity trend information by counting the historical target behavior quantity of the target product; or, statistically obtaining behavior quantity trend information by counting the historical target behavior quantity of at least one product of the same type as the target product provided by the product party; The step of determining the level type to which the product party corresponding to the target product belongs according to product party attribute information comprises the following step: determining the level type to which the product party belongs by combining the product party attribute information and the behavior quantity trend information.

4. The method of claim 1, wherein, The plurality of information comprises one or more of product text information, product attribute information, product conversion information generated in a first time period, associated product related information, and product party related information; the product text information comprises product title text and / or a plurality of product evaluation texts generated in the first time period; the product conversion information comprises behavior quantity time series data corresponding to the target behavior and at least one associated behavior respectively generated in the first time period; the associated product related information comprises behavior quantity time series data of the target behavior of an associated product generated in the first time period; and the product party related information comprises behavior quantity time series data of the target behavior and at least one associated behavior of a plurality of products of the same type as the target product generated in the first time period; The product features are specifically composed of product sub-features extracted according to one or more of the following implementation manners from a plurality of information associated with the product: extracting a target entity word belonging to a target entity type from the product title text; statistically obtaining gain data of the target entity word corresponding to a plurality of heat evaluation indexes according to evaluation data of the target entity word before and after appearing in a plurality of product title texts respectively corresponding to the plurality of heat evaluation indexes; generating product sub-features based on the gain data; analyzing the sentiment polarity of the plurality of product evaluation texts in at least one target sentiment aspect respectively; statistically obtaining the number of occurrences of the at least one target sentiment aspect in the plurality of product evaluation texts, the polarity number and the proportion value corresponding to different sentiment polarities, and generating statistical data; generating product sub-features based on the statistical data; Respectively statistically process behavior quantity time series data of a plurality of behaviors corresponding to the target product; Generate product sub-features based on data statistical results respectively corresponding to a plurality of behaviors corresponding to the target product; Statistically process behavior quantity time series data of the target behavior generated by the associated product in the first time period; Generate product sub-features based on data statistical results of the target behavior corresponding to the associated product; And, Statistically process behavior quantity time series data of a plurality of behaviors of a plurality of products of the same type as the target product; Generate product sub-features based on data statistical results respectively corresponding to a plurality of behaviors corresponding to the plurality of products.

5. The method of claim 4, wherein, The product conversion information includes behavior quantity time series data respectively corresponding to the target behavior and at least one associated behavior of the target behavior generated in the first time period; the method further comprises: For any behavior, adjust the behavior quantity generated at the time point meeting the abnormal condition in the corresponding behavior quantity time series data.

6. The method of claim 1, wherein, Further comprising: According to the behavior quantity time series data of the target behavior generated by the target product in the first time period, at least one prediction method is used to predict a second behavior quantity of performing the target behavior for the target product; Based on the first behavior quantity and at least one second behavior quantity, the predicted behavior quantity of the target product is obtained by fusion.

7. The method of claim 1, wherein, Further comprising: Determine whether the first behavior quantity is greater than a quantity threshold; If not, the first behavior quantity is taken as the predicted behavior quantity of the target product; If yes, the quantity threshold is taken as the predicted behavior quantity of the target product.

8. The method of claim 1, wherein, Further comprising: Based on the first behavior quantity, generate product prompt information; Notify the product party of the product prompt information.

9. The method of claim 1, wherein, The target behavior is a purchase behavior, and the first behavior quantity is a first sales quantity.

10. A model generation method characterized by comprising: Comprising: Construct a prediction model respectively corresponding to different product levels; For any product level, determine a sample product belonging to the product level; From a plurality of information corresponding to the sample product and associated with the target behavior, extract product sample features; Based on the product sample features and the corresponding behavior prediction quantity, train the prediction model corresponding to the product level.

11. A computing device, comprising: Comprise a storage component and a processing component, the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to realize the data prediction method as claimed in any one of claims 1-9 or the model generation method as claimed in claim 10.

Images