CN114418613A

CN114418613A - Data prediction method, model generation method and equipment

Info

Publication number: CN114418613A
Application number: CN202111566743.1A
Authority: CN
Inventors: 赵阳; 曹雷; 庄丽学; 李利明; 韦阳; 叶嘉成; 张吉仲; 李贝贝; 邓玉明
Original assignee: Alibaba China Co Ltd
Current assignee: Alibaba China Co Ltd
Priority date: 2021-12-20
Filing date: 2021-12-20
Publication date: 2022-04-29

Abstract

The embodiment of the application provides a data prediction method, a model generation method and equipment. Extracting product characteristics from a plurality of information which corresponds to a target product and is associated with a target behavior; determining the target product level to which the target product belongs by combining the product characteristics and the product party attribute information; determining a target prediction model corresponding to the target product level from prediction models respectively constructed aiming at different product levels; predicting, with the target prediction model, a first number of actions to perform the target action for the target product based on the product features. The technical scheme provided by the embodiment of the application improves the data prediction accuracy.

Description

Data prediction method, model generation method and equipment

Technical Field

The embodiment of the application relates to the technical field of data processing, in particular to a data prediction method, a model generation method and equipment.

Background

With the development of computer technology and network technology, product processing systems that perform product transactions in an online manner are rapidly rising, and with the help of the product processing systems, a product party can publish products, and users can browse, collect, purchase products, and the like.

Still taking product sales quantity prediction as an example, at present, a time series analysis method is generally adopted, and sales quantity in a future period of time is obtained based on historical sales quantity analysis of products, but in this way, under the condition that the historical sales quantity is lack or the information amount is small, the accuracy of the prediction result is low.

Disclosure of Invention

The embodiment of the application provides a data prediction method, a model generation method and equipment, which are used for solving the technical problem of low data prediction accuracy in the prior art.

In a first aspect, an embodiment of the present application provides a data prediction method, including:

extracting product features from a plurality of information corresponding to the target product and associated with the target behavior;

determining the target product level to which the target product belongs by combining the product characteristics and the product party attribute information;

determining a target prediction model corresponding to the target product level from prediction models respectively constructed aiming at different product levels;

predicting, with the target prediction model, a first number of actions to perform the target action for the target product based on the product features.

In a second aspect, an embodiment of the present application provides a model generation method, including:

constructing prediction models respectively corresponding to different product levels;

for any product level, determining a sample product belonging to the product level;

extracting product sample characteristics from a plurality of information corresponding to the sample product and associated with the target behaviors;

and training a prediction model corresponding to the product level based on the product sample characteristics and the corresponding behavior prediction quantity.

In a third aspect, embodiments of the present application provide a computing device, comprising a storage component and a processing component, the storage component storing one or more computer instructions; the one or more computer instructions are for execution by the processing component to implement a data prediction method as described in the first aspect above or a model generation method as described in the second aspect above.

In the embodiment of the application, product features are extracted from a plurality of information related to a target product and a target behavior, product features and product side attribute information are combined, a target product level to which the target product belongs is determined firstly, then a target prediction model corresponding to the target product level is selected from prediction models respectively corresponding to different pre-constructed product levels, and then the target prediction model is used for predicting the number of first behaviors corresponding to the target behavior based on the product features. By extracting the product characteristics from the information associated with the target behaviors, the data information participating in prediction is enriched, the data prediction accuracy is improved, and the target products can be layered by combining the product characteristics and the product side attribute information, and prediction models are independently constructed for different product layers, so that the behavior quantity prediction is performed in a targeted manner based on the product characteristics by using the target prediction model corresponding to the target products, and the data prediction accuracy is further improved.

These and other aspects of the present application will be more readily apparent from the following description of the embodiments.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative efforts.

FIG. 1 illustrates an exemplary system architecture diagram to which the subject technology can be applied;

FIG. 2 is a flow chart illustrating one embodiment of a data prediction method provided herein;

FIG. 3 is a schematic diagram illustrating a data prediction process in an actual application according to an embodiment of the present application;

FIG. 4 illustrates a flow chart of one embodiment of a model generation method provided herein;

FIG. 5 is a schematic diagram illustrating an embodiment of a data prediction apparatus provided herein;

FIG. 6 illustrates a schematic structural diagram of one embodiment of a computing device provided herein;

fig. 7 is a schematic structural diagram illustrating an embodiment of a model generation apparatus provided in the present application.

Detailed Description

In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.

In some of the flows described in the specification and claims of this application and in the above-described figures, a number of operations are included that occur in a particular order, but it should be clearly understood that these operations may be performed out of order or in parallel as they occur herein, the number of operations, e.g., 101, 102, etc., merely being used to distinguish between various operations, and the number itself does not represent any order of performance. Additionally, the flows may include more or fewer operations, and the operations may be performed sequentially or in parallel. It should be noted that, the descriptions of "first", "second", etc. in this document are used for distinguishing different messages, devices, modules, etc., and do not represent a sequential order, nor limit the types of "first" and "second" to be different.

The technical scheme of the embodiment of the application can be applied to an online product processing scene and is used for predicting the behavior quantity of target behaviors executed by products provided online, such as product sales quantity, browsing quantity, collection quantity or purchase quantity generated by purchasing, browsing, collecting and purchase adding (adding a shopping cart) and the like.

It is understood that in the online product processing scenario, the products, shopping carts, etc. described in the embodiments of the present application are all virtual objects, and in the online processing system, they are usually represented in different data forms, etc.

Taking the predicted sales volume of a product (the sales volume is hereinafter often referred to as a sales volume) as an example, the current sales volume prediction method generally predicts the future sales volume based on the historical sales volume. One common method is to use a time series analysis method, such as EMA (Exponential Moving Average) method, but this method cannot make an accurate prediction for a product with a lack of historical sales or less information, for example, for a product that is newly released by a product party or a product with a short release time, such as less than a predetermined number of days (such a product is generally called a new product), there may be no historical sales or the historical sales is not regular obviously or the sales is discontinuous, and the like, which may affect the final prediction accuracy.

In the embodiment of the application, product characteristics are extracted from a plurality of pieces of information related to target behaviors corresponding to target products, the historical behavior quantity of the target behaviors is limited, data information participating in prediction is enriched, the data prediction accuracy is improved, the target products can be layered by combining the product characteristics and product side attribute information, prediction models are independently constructed for different product layers, the behavior quantity prediction is carried out on the basis of the product characteristics in a targeted mode by using the target prediction models corresponding to the target products, and the data prediction accuracy is further improved.

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

Fig. 1 is a diagram of an exemplary system architecture to which embodiments of the present application may be applied, which may include a client 101 and a server 102. The server 102 may perform construction and the like of the relevant model related in the embodiment of the present application, and the trained model may be deployed in the server 102 or in the client 101, so as to implement the data prediction method provided in the embodiment of the present application. In general, the server 102 may execute the data prediction method provided by the embodiment of the present application.

The client 101 may be generally understood as an application program deployed in an electronic device, for example, one or more of a smart phone, a tablet computer, and a portable computer, or certainly, a desktop computer, etc., and for ease of understanding, the client is mainly represented in fig. 1 in a device image. Various other types of applications may also be configured in the electronic device, such as a search type, an instant messaging type, and so on. Of course, the client 101 may also be a browser, or a web application such as H5(HyperText Markup Language5, 5 th edition) application, or a light application (also called an applet, a light application) or a cloud application,

the server 102 may be hardware or software. When the server is hardware, the server can be implemented as a distributed server cluster formed by a plurality of servers, or can be implemented as a single server. When the server is software, it may be implemented as multiple pieces of software or software modules (e.g., multiple pieces of software or software modules for providing distributed services), or as a single piece of software or software module. And is not particularly limited herein. The server can also be a server of a distributed system or a server combined with a blockchain. The server can also be a cloud server, or an intelligent cloud computing server or an intelligent cloud host with an artificial intelligence technology. In fig. 1, the server is mainly represented by a cloud server image. The model building process and the process of predicting by using the model may be implemented by different servers or different software modules in the server 102.

The details of implementation of the technical solution of the embodiments of the present application are set forth in the following.

Fig. 2 is a flowchart of an embodiment of a data prediction method provided in an embodiment of the present application, where the method may include the following steps:

201: and extracting product characteristics from a plurality of information which corresponds to the target product and is associated with the target behaviors.

The target product may refer to any product.

In addition, in an actual application, the target product may refer to a target type of product divided according to the release time, for example, the target product may refer to a product with a release time less than a predetermined time, and the target type is a new product type, which is a new product, such as a product with a release time less than 28 days in an actual application. Due to the fact that historical data of the new product is lack or less, effective and accurate prediction can be achieved by the aid of the technical scheme.

The target behavior may refer to, for example, a purchasing behavior, a browsing behavior, a collecting behavior, or an purchasing behavior, the purchasing behavior may bring product sales, the browsing behavior may bring user traffic, the collecting behavior may bring product collecting quantity, the purchasing behavior may bring purchasing quantity, and the like, therefore, the technical scheme of the embodiment of the application can be suitable for predicting the product sales volume, the user flow, the collection volume, the purchase adding volume and the like, the warehouse management of the product has instructive significance, the technical scheme of the application is mainly described by taking the product sales amount as an example in one or more of the following embodiments, of course, those skilled in the art can understand that the present disclosure is not limited to the prediction of the product sales, and any time-series behavior quantity brought by the behavior executed on the product can be predicted according to the present disclosure.

The plurality of information associated with the target behavior may refer to information that may affect the user to perform the target behavior, such as information that may help motivate the user to perform the target behavior, and in an alternative implementation, the plurality of information may include one or more of product text information, product attribute information, product conversion information generated during a historical first time period, associated product-related information, and product party-related information, for example. The first time period may be a first time period formed by a certain historical time point and a current predicted time, for example, a time period formed by 28 days before the current predicted time, and for a new product, the first time period may be a time period formed from a product release time to the current predicted time.

The product textual information may include, for example, product title text, and/or a plurality of product rating texts generated during the first time period, and/or the like. The product title text may be provided by a product party when the product is released in the product processing system, a user may obtain some related information of the product based on the product title text, the product title text is usually shown in a product aggregation page corresponding to some multiple products or a product description page corresponding to a single product, and the specific content of the product title text may affect the probability of executing a target behavior on the product.

In the product processing system, after a product is purchased on line, a product party can entrust a logistics party to mail a real product to a purchaser, a product evaluation text can refer to the purchaser of the product, and a user who arbitrarily browses the product can see the product evaluation text according to the information recorded in the system in combination with the received feelings of the product quality of the real product, the logistics distribution service quality, the service quality of the product party and the like, and the user can decide whether to execute a target behavior by referring to the product evaluation text, so that the product evaluation text can also influence the probability of executing the target behavior on the product.

The product attribute information may include, for example, product category, product brand, participating promotional campaign date, and the like.

The product conversion information may include behavior quantity time series data corresponding to the target behavior and at least one associated behavior generated in the first time period; the target behavior may refer to, for example, a purchasing behavior, and the corresponding at least one associated behavior may include, for example, browsing, collecting, purchasing, and the like, where the associated behavior may facilitate execution of the target behavior. Therefore, the product conversion information may specifically include, for example, product sales time series data, flow conversion rate time series data, product purchase quantity time series data, and the like. In addition, the product conversion information may also include the product transaction amount, and the like. The product sales time series data can be formed by arranging the product sales corresponding to each time point in the first time period according to the sequence of the time points; the user traffic conversion rate time series data can be formed by arranging user traffic conversion rates corresponding to each time point in the first time period according to the sequence of the time points; the product purchase quantity time sequence data can be formed by arranging the product purchase quantity corresponding to each time point in the first time period according to the sequence of the time points; the each time point may for example refer to each day, etc. But may also refer to monthly, quarterly, yearly, or some other specified period of time, etc.

The related product information may refer to related information of a related product of the target product, such as behavior quantity time series data of a target behavior corresponding to the related product in the first time period, and may further include attribute information of the related product. The associated product may be, for example, the same product as the target product category, etc., wherein the categories may be divided by product properties, such as jacket, pants, hat, etc.

The product side means a provider providing the target product, and the product side related information may include, for example, behavior quantity time series data of a target behavior and behavior quantity time series data of at least one associated behavior generated by a plurality of products of the same type as the target product in the first time period, respectively. For example, when the target product is a new product, the plurality of products are all new products. In addition, the product side related information may further include product side attribute information, such as an average sales number of the product side in a unit time or a total sales amount in a certain time period, and the product side sales number is obtained by counting sales numbers of a plurality of products provided by the product side.

From a plurality of information associated with the target behaviors, product features can be obtained through statistical analysis, data extraction, cleaning, normalization processing and the like.

The product feature also includes sub-features extracted from the plurality of pieces of information, and the specific extraction manner will be described in detail in the following embodiments.

202: and determining the target product hierarchy to which the target product belongs by combining the product characteristics and the corresponding product side attribute information.

203: and determining a target prediction model corresponding to the target product level from the prediction models respectively constructed aiming at different product levels.

Because the probabilities of target behaviors possibly generated by different products of different product parties are different, in the embodiment of the application, in order to perform data prediction more accurately, product layering can be performed according to product party attribute information and product characteristics, and prediction models for performing data prediction are respectively constructed for different product layers.

The product side attribute information may include, for example, a product side name, a product side sales amount, and the like, and the product side sales amount may refer to, for example, an average sales amount of all products in a unit time or a total sales amount of all products in a certain time period.

For example, the product heat corresponding to the target product can be identified according to the product characteristics, such as whether the target product is a explosive product or not. It is generally understood that explosives are products that are more likely to be popular with users and thus will perform a series of actions. The corresponding class type of the product party, such as whether the product party is a head seller or not, can be determined according to the attribute information of the product party. It is generally understood that products offered by a head seller are more likely to be popular with users to perform a range of activities. Accordingly, for example, four product levels can be distinguished: explosives + head vendor; non-explosive + head seller; explosives + non-head vendor; non-explosive + non-head seller. Of course, this is merely an example of how to perform product layering, and there are other alternative implementations in practical applications, and the application is not limited thereto.

By respectively constructing prediction models aiming at different product levels, more targeted prediction can be performed on different products, so that the data prediction accuracy can be further improved.

Therefore, through the determined product characteristics and the product party attribute information of the product party providing the target product, the target product hierarchy to which the target product belongs can be determined, and further the target prediction model corresponding to the target product hierarchy can be determined.

The prediction model corresponding to each product level may be obtained by pre-training, for example, the prediction model may be obtained by training in a supervised manner by using the product features of the sample product corresponding to each product level and the corresponding prediction behavior quantity, which will be described in more detail in the following corresponding embodiments.

204: a first quantity of actions to perform the target action for the target product is predicted based on the product features using the target prediction model.

And inputting the product characteristics into the target prediction model by using the determined target prediction model, so that the first action quantity of executing the target actions on the target product can be predicted. Specifically, the predicted first behavior amount may be a first behavior amount corresponding to a future second time period from the current predicted time, such as a future day, a week, or a month, the second time period may be a unit time corresponding to a time point in the first time period, such as a day, or a time length formed by a plurality of unit times, such as a week, and the first behavior amount may include a behavior amount of a target behavior that may be generated at each time point in the second time period.

In the embodiment, the product characteristics are extracted from the plurality of information, so that the historical behavior quantity of the target behavior is limited, the data information participating in prediction is enriched, the data prediction accuracy is improved, the target product can be layered by combining the product characteristics and the product side attribute information, the prediction model is independently constructed for different product layers, the behavior quantity prediction is pertinently performed on the basis of the product characteristics by using the target prediction model corresponding to the target product, and the data prediction accuracy is further improved.

As can be seen from the foregoing description, the target product hierarchy to which the target product belongs may be determined according to the heat type of the target product and the level type of the product party, where as an alternative, the level type of the product party may be determined according to the product party attribute information, the product party attribute information may include information such as the sales volume of the product party and the name of the product party as described above, the level type of the product party may be assumed to be classified into a head seller and a non-head seller, and whether the product party is a head seller may be determined according to whether the name of the product party belongs to the name of the product party corresponding to the specified level category or whether the sales volume of the product party is greater than a certain threshold. For example, if the name of the product party is the preset head seller name and/or the sales volume of the product party is greater than a certain threshold, it is determined that the product party is the head seller, otherwise, it is determined that the product party is a non-head seller.

In addition, as another optional mode, the level type to which the product side belongs can be determined by combining the attribute information of the product side and the behavior quantity trend information corresponding to the product side.

The behavior quantity trend information represents the variation situation of the behavior quantity of the product side corresponding to the target behavior, and the behavior quantity trend information can comprise an ascending trend, a descending trend, a smooth trend and the like.

Therefore, the attribute information of the product party and the behavior quantity trend information can be specifically combined, and the corresponding class type can be determined, for example, if at least one condition of the name of the product party being a preset head seller, the product party selling quantity being greater than a certain threshold value and the behavior quantity trend information being an ascending trend is met, the product party can be determined to be a head seller, otherwise, the product party is determined to be a non-head seller.

Wherein, the behavior quantity trend information may be implemented in a manner of:

counting the historical target behavior quantity of the target behavior corresponding to the target product to obtain behavior quantity trend information;

the historical target behavior amount may refer to behavior amount time series data of the target behavior generated by the target product in the first time period, for example, behavior amount time series data formed by target behavior amounts of each day in the previous 28 days. The historical target behavior quantity of the target behavior corresponding to the target product is counted, so that the fact that the behavior quantity of the target behavior corresponding to the target product is ascending, descending or stable and the like can be obtained, and the behavior quantity trend information of the target behavior corresponding to the target product can be used as the behavior quantity trend information of a product party. The trend judgment of the historical target behavior quantity can be obtained by adopting a cluster analysis mode or other statistical analysis methods, and this is not particularly limited in the present application.

In order to further improve the information accuracy, another implementation manner may be:

counting historical target behavior quantity of at least one product of the same type as the target product, which is provided by a product party, to obtain behavior quantity trend information;

wherein at least one product of the same type as the target product may or may not also include the target product.

The historical target behavior amount of the at least one product may be behavior amount time series data of target behaviors of the at least one product respectively generated in the first time period, for example, behavior amount time series data formed by target behavior amounts of each day in the previous 28 days respectively.

By counting the respective historical target behavior quantity of at least one product, behavior quantity trend information of each product can be obtained, for example, cluster analysis is carried out on the behavior quantity trend information of each product, and the behavior quantity trend information with the largest proportion can be used as the behavior quantity trend information of a product side; for another example, when the historical target behavior quantity is behavior quantity time series data, the average quantity of target behaviors or the total quantity of target behaviors corresponding to each time point (for example, each day) may be counted based on the behavior quantity time series data corresponding to each of the at least one product, so that the behavior quantity trend information corresponding to the product side may be obtained based on the average quantity of target behaviors or the total quantity of target behaviors at each time point.

As can be appreciated from the foregoing description, the plurality of information may include one or more of product textual information, product attribute information, product conversion information generated during a first time period, associated product-related information, and product party-related information, among others.

In some embodiments, the product feature may be composed of sub-features extracted from the product according to one or more of the following:

a: extracting target entity words belonging to the target entity type from the product title text; according to the evaluation data of the plurality of heat evaluation indexes respectively corresponding to the target entity words before and after the target entity words appear in the plurality of product title texts, counting to obtain gain data of the target entity words respectively corresponding to the plurality of heat evaluation indexes; based on the gain data, product features are generated.

Entity words and entity types corresponding to each entity word may be first identified from a product title file using a named entity identification model.

Then, entity calibration may be performed on the entity extraction result, and calibration may be performed according to a calibration rule, for example, for a certain entity type identification obtained entity word, a calibration word matched with the certain entity type identification obtained entity word in the calibration rule may be searched as an entity word corresponding to the entity type, and the like, for example, if the entity word corresponding to the "obvious" entity type obtained by identification is "liyi", and the "star" defined in the calibration rule is "liyi three", then the "liyi" obtained by identification may be calibrated as "liyi three".

In addition, the obtained entity types and entity words can be screened according to the entity screening rule, entity types and entity words which do not accord with the screening rule are screened, and finally target entity words and the like corresponding to the target entity types are obtained.

Then, for each heat evaluation index, according to the evaluation data of the heat evaluation index corresponding to each target entity word before and after the target entity word respectively appears in the plurality of product title texts, gain data of the target entity word corresponding to the heat evaluation index before and after each product title text appears is obtained through statistics, and the gain data corresponding to each product title text can be weighted and fused to obtain the final gain data corresponding to the heat evaluation index. Each product title text may correspond to a product, multiple product title texts may correspond to different products, and the multiple product title texts may be multiple products recorded in history in the product processing system.

The gain data may be represented as a difference or a ratio of two evaluation data of the same heat evaluation index.

Of course, the gain data may also be obtained by statistics in advance, that is, the gain data of the target entity word corresponding to the plurality of heat evaluation indexes respectively may be directly obtained.

The plurality of heat evaluation indicators may include, for example, product sales and customer flow conversion rates. The user traffic conversion rate can refer to the proportion of the number of purchasing behaviors brought by the user traffic in the user traffic, and the user traffic means the number of visits or browsing of products; for convenience of calculation, the product sales amount may refer to an average product sales amount per unit time, i.e., at each time point.

In addition, the target entity word may also refer to an entity word or the like in which a corresponding user traffic conversion rate or product sales amount in the product title file is greater than a certain value.

The gain data may be used as product sub-features, or may be subjected to normalization processing, feature conversion processing, and the like, and then converted into feature vectors and then used as product sub-features, and the like.

The foregoing illustrates an implementation manner of obtaining the product sub-features based on the product title text, and of course, other implementations may also be adopted, for example, keywords in the product title text may be extracted (for example, the keywords may be target entity words or extracted by using a pre-trained extraction model), and vectors corresponding to the keywords are directly expressed as the product sub-features, and the like. This is not overly restrictive in this application.

B: analyzing corresponding emotion polarities of the plurality of product evaluation texts in at least one target emotion aspect respectively; counting the occurrence number, the polarity number corresponding to different emotion polarities and the ratio value of at least one target emotion in a plurality of product evaluation texts to generate statistical data; based on the statistical data, product characteristics are generated.

The emotion polarities of the plurality of product evaluation texts respectively corresponding to at least one target emotion can be obtained by utilizing ABSA (aspect based sensory analysis).

Optionally, a plurality of emotional aspects corresponding to the plurality of product evaluation texts and the emotional polarity of each emotional aspect may be obtained by first parsing, where the emotional aspects may include, for example, taste, quality, package, validity period, and the emotional polarity may mainly include three polarities: positive, negative and neutral.

Then, the plurality of analyzed emotion aspects can be combined with the categories and industries to which the target product side belongs, and the like, and screening is carried out according to emotion screening rules, so that at least one target emotion aspect is obtained. The emotion screening rule specifies, for example, an emotion aspect or the like corresponding to a specific category or a specific industry or the like.

And then, counting the occurrence number of at least one target emotion aspect in a plurality of product evaluation texts, the polarity number corresponding to different emotion polarities and the ratio value to generate statistical data. For example, assuming that 7 target emotion directions are reserved, 56 data are used as statistical data, and the statistical data can be used as product sub-features or product sub-features after some necessary features such as normalization processing and feature conversion processing are converted into feature vector expressions.

C: respectively carrying out statistical processing on behavior quantity time sequence data of a plurality of behaviors corresponding to the target product; and generating product sub-characteristics based on data statistical results corresponding to the target product and the plurality of behaviors respectively.

The plurality of behaviors also include a target behavior and at least one associated behavior corresponding to the target behavior.

The data statistics for each behavior may include, for example, a mean, variance, median, mean and/or total, etc. obtained for the behavior quantity time series data statistics.

Finally, a plurality of obtained data statistical results can be used as product sub-features, or can be used as product sub-features after being converted into feature vector expressions through operations such as normalization processing, feature conversion processing and the like.

D: carrying out statistical processing on the behavior quantity time sequence data of the target behaviors generated by the related product in the first time period; generating product sub-characteristics based on data statistics results of target behaviors corresponding to the associated products;

for example, the data statistics result may be a mean, a variance, a median, a mean and/or a total obtained by the behavior quantity time series data statistics of the associated product.

E: performing statistical processing on the behavior quantity time sequence data of a plurality of behaviors of a plurality of products of the same type as the target product; and generating product sub-characteristics based on the data statistical results corresponding to the plurality of behaviors respectively corresponding to the plurality of products.

That is, the product sub-features may be generated based on the data statistics of the target product corresponding to the plurality of behaviors, or the product sub-features may be generated based on the data statistics of the target product corresponding to the plurality of behaviors.

The data statistics for each behavior for each product may include, for example, mean, variance, median, mean and/or total obtained from the behavior quantity time series data statistics, and the like.

Wherein, the product characteristics can be obtained by one or more product sub-characteristics extracted in the operations A to E.

As can be seen from the foregoing description, the time-series data related to a plurality of behavior quantities, for example, the product conversion information may include time-series data of behavior quantities corresponding to a target behavior generated in a first time period and at least one associated behavior of the target behavior, and the like, while some abnormal behavior quantities may occur due to some reasons in the historical behavior quantities, and therefore, in order to ensure the information accuracy, in some embodiments, the method may further include:

and aiming at any behavior, adjusting the behavior quantity generated at the time point which meets the abnormal condition in the corresponding behavior quantity time sequence data.

The time sequence data of the behavior quantity corresponding to the first time period can be detected by adopting a sliding window, the time point meeting the abnormal condition can be detected, the size of the sliding window can be set according to the actual condition, data statistics such as variance calculation can be carried out by carrying out data statistics on the corresponding behavior quantity corresponding to each time point in the sliding window, and if the data statistics result meets the abnormal condition, the corresponding behavior quantity corresponding to each time point in the current sliding window can be considered as abnormal data, and the data can be adjusted.

For example, the number of behaviors generated at the time point meeting the abnormal condition in the first time period may be directly deleted, that is, after deletion, the number of behaviors corresponding to the time point meeting the abnormal condition may be updated to 0.

Of course, as another adjustment method, the number of behaviors at the time point meeting the abnormal condition in the first time period may be updated to a predetermined number, for example, the predetermined number may be an average value of the numbers of behaviors at the time points of non-abnormality in the first time period, or the predetermined number may be a number of behaviors corresponding to a time point of non-abnormality closest to the time point of abnormality.

The first behavior amount predicted by the target prediction model may be used to perform a corresponding processing operation as the predicted behavior amount of the target product. Thus, in some embodiments, the method may further comprise:

generating product prompt information based on the first action number;

and informing the product side of product prompt information.

Wherein, the product prompt message is used to prompt the first action number, etc. The product side is informed of the product prompt information, so that the product side can be helped to know the future first action quantity of the target product, the warehousing management of the target product is facilitated, and for example, the replenishment of the target product can be carried out in time.

The product prompting information can be notified to the product side in various ways, for example, the product prompting information can be sent to a communication account corresponding to the product side, and the communication account can be, for example, a mobile communication account such as a mobile phone number, or an instant messaging account, or an email account.

Of course, the product prompt information may also be sent to a client account corresponding to the product side, and the product side may view the product prompt information through the client, for example, in the system architecture diagram in fig. 1, the product prompt information generated by the server 102 may be sent to the client 101, so as to achieve the purpose of notifying the product prompt information to the product side, and the like.

Furthermore, to further improve the prediction accuracy, in some embodiments, the method may further comprise:

predicting a second behavior quantity of the target behavior executed aiming at the target product by adopting at least one prediction mode according to behavior quantity time sequence data of the target behavior generated by the target product in a first time period;

and fusing to obtain the predicted behavior quantity of the target product based on the first behavior quantity and the at least one second behavior quantity.

According to the behavior quantity of each time point in the behavior quantity time series data corresponding to the target behavior, a time series analysis method, such as a trend prediction method like EMA, a single-sequence time average method, a weighted sequence time average method, a moving average method, a weighted moving average method, etc., may be adopted to obtain a second behavior quantity by prediction, for example, in the case of predicting the behavior quantity of the next time point, a weighted average of the behavior quantities of each time point may be used as the second behavior quantity, etc., and the weighting coefficient corresponding to each time point may be set according to the principle that the farther the time is, the smaller the weighting coefficient is, the closer the time is, the larger the weighting coefficient is, etc.

In addition, a time sequence prediction model, such as a deep neural network model, can be used for prediction, and the time sequence prediction model is obtained through pre-training, for example, in a scene that the time sequence prediction model is used for predicting the sales volume of the new product every 28 days in the future by using the sales volume of the new product every 28 days, and the training sample can be used for historical sales volume time sequence data of any new product. And adopting a rolling time window mode, taking the sales volume of the historical 28 time points in the rolling time window as input, and taking the sales volume of the future 28 time points as output to carry out model training to obtain the sales volume.

Wherein the first behavior quantity and the at least one second behavior quantity can be calculated as an average value or a weighted average value, and the average value is used as the predicted behavior quantity of the final target product. The product reminder information may thus be generated specifically based on the predicted number of actions.

In some embodiments, the method may further comprise:

judging whether the first behavior quantity is larger than a quantity threshold value or not;

if not, taking the first behavior quantity as the predicted behavior quantity of the target product;

and if so, taking the quantity threshold value as the predicted behavior quantity of the target product.

The first action number is restricted by adding a number threshold value, so that the stability of a prediction result and the like can be ensured.

The quantity threshold may be, for example, a quantity predicted by using time series data of a quantity of behaviors of the target product generated in the first time period, that is, the quantity threshold may be the second quantity of behaviors in the above embodiment.

Of course, the setting and the like may be performed in advance in accordance with actual conditions.

In an actual application, the target behavior may specifically refer to a purchasing behavior, and the first behavior quantity means a sales quantity. The technical solution of the embodiment of the present application is described below by taking prediction of sales quantity of new products as an example, and is shown in the processing diagram shown in fig. 3. For a new product to be predicted, a data input operation 301 is first included to determine a plurality of information associated with a purchase behavior, which may include product text information, including a product title text and a product evaluation text, for example; product attribute information; the product conversion information generated in the first time period is the sales time sequence data, the user flow conversion rate time sequence data, the purchase quantity time sequence data and the like; the related information of the related products comprises sales amount time series data of the related products in the first time period, and the related information of the product side comprises sales amount time series data of a plurality of new products in the first time period, user flow conversion rate time series data, purchase adding amount time series data and the like.

Then abnormal data detection 302 can be carried out on the time sequence data, and various time sequence data can be readjusted;

thereafter, the product sub-features may be extracted from the plurality of information and assembled into the product feature 303, which may include: extracting text characteristics from product text information including a product title text and a product evaluation text to serve as product sub-characteristics; the product attribute information can be directly used as the product sub-characteristics; extracting product sub-features from product conversion information such as sales time series data, user flow conversion rate time series data, purchase quantity time series data and the like; extracting product sub-features from related product information, such as sales time series data of the related product in a first time period; and extracting product sub-characteristics from product side related information, such as sales volume time sequence data, user flow conversion rate time sequence data, purchase quantity time sequence data and the like of a plurality of new products in a first time period.

Then, a target product hierarchy 304 to which the new product belongs can be determined according to the product characteristics and the product side attribute information; for example, four product levels can be divided according to different level types and heat types: explosives + head vendor; non-explosive + head seller; explosives + non-head vendor; non-explosive + non-head seller. The product characteristics and the product side attribute information of the new product are combined to determine the target product hierarchy

Then, a first behavior quantity 305 is obtained by using a target prediction model corresponding to the target product hierarchy.

In addition, the at least one second behavior quantity obtained by prediction may be combined with the quantity threshold and the at least one prediction mode to constrain and fuse 306 the predicted behavior quantity corresponding to the new product.

Based on the predicted behavior amount, product prompt information may be generated to notify a product party or the like.

The prediction models corresponding to different product levels in the above corresponding embodiments may be pre-constructed, as shown in fig. 4, and as a flowchart of another embodiment of the model generation method provided by the embodiment of the present application, the method may include the following steps:

401: and constructing prediction models respectively corresponding to different product levels.

The model architecture of the prediction model corresponding to different product levels and the initial model parameters can be the same, and different model parameters are finally corresponding to different training samples.

402: for any product level, a sample product belonging to that product level is determined.

The sample product may be a sample product of a target type selected from historical products, for example, the target type may be a new product, and the historical new product in the product processing system may be the sample product.

403: and extracting product sample characteristics from a plurality of information which corresponds to the sample product and is associated with the target behaviors.

The plurality of information associated with the target behavior corresponding to the sample product may include, for example: one or more of product text information, product attribute information, product conversion information generated during a first time period, associated product related information, and product party related information; the product text information comprises a product title text and/or a plurality of product evaluation texts produced in a first time period; the product conversion information comprises behavior quantity time sequence data respectively corresponding to the target behavior and the at least one associated behavior generated in the first time period; the related product related information comprises behavior quantity time sequence data of the related product of the sample product for generating the target behavior in the first time period; the product side related information comprises behavior quantity time sequence data of target behaviors and at least one associated behavior generated by a plurality of products of the same type as the sample product in a first time period;

the product sample features may be constituted by product sub-sample features extracted according to one or more of the following implementation manners:

extracting target entity words belonging to the target entity type from the product title text; according to the evaluation data of the target entity words respectively corresponding to the plurality of heat evaluation indexes before and after the target entity words appear in the plurality of product title texts, counting to obtain gain data of the target entity words respectively corresponding to the plurality of heat evaluation indexes; generating product subsample features based on the gain data;

analyzing corresponding emotion polarities of the plurality of product evaluation texts in at least one target emotion aspect respectively; counting the occurrence number, the polarity number corresponding to different emotion polarities and the ratio value of at least one target emotion in a plurality of product evaluation texts to generate statistical data; generating product sub-sample features based on the statistical data;

respectively carrying out statistical processing on the behavior quantity time sequence data of a plurality of behaviors corresponding to the sample product; generating product sub-sample characteristics based on data statistics results corresponding to a plurality of behaviors corresponding to the sample product respectively;

carrying out statistical processing on the behavior quantity time sequence data of the target behaviors generated by the associated product in the first time period; generating product sub-sample characteristics based on the data statistics result of the target behavior corresponding to the associated product;

and the number of the first and second groups,

performing statistical processing on the behavior quantity time sequence data of a plurality of behaviors of a plurality of products of the same type as the sample product; and generating product sub-sample characteristics based on data statistical results corresponding to a plurality of behaviors corresponding to a plurality of products respectively.

The manner of extracting the product sample features is the same as the manner of extracting the product features from the plurality of information of the target product, and the details are not repeated here. The types of the information have been described in detail above, and are not repeated herein. The related various behavior quantity time sequence data can also be subjected to abnormality detection, and the behavior quantity generated at the time point meeting the abnormal condition in the behavior quantity time sequence data can be adjusted.

In addition, the product side attribute information and the product characteristics can be combined to determine the sample product corresponding to the product hierarchy. Therefore, for the products in the historical product data set, the product features are respectively extracted from the corresponding information associated with the target behaviors, and the sample products and the like corresponding to the product hierarchy can be determined by combining the product side attribute information and the product features which respectively correspond to the products. Wherein the historical product data set may be comprised of target types of historical products.

404: and training a prediction model corresponding to the product level based on the product sample characteristics and the corresponding behavior prediction quantity.

The product sample features may be extracted from a plurality of information associated with the target behavior generated for the sample product for the first time period. The first time period may refer to a specified duration of the sample product from the release time, such as a time point in days, and may refer to a plurality of information associated with the target behavior occurring in the first 28 days from the release time. The behavior prediction amount may be an amount of behavior prediction statistically obtained one or more days after the 28 th day from the release time, and in the case of a plurality of days, the behavior prediction amount may be time series data of the amount of behavior corresponding to the plurality of days, or the like.

Then, based on the product sample characteristics and the corresponding behavior prediction quantity, the prediction model corresponding to the product level can be trained.

The prediction model obtained by training can be used for predicting the first behavior quantity of the target product belonging to the product level.

In practical applications, the first action number may specifically refer to a product sales number and the like.

Fig. 5 is a schematic structural diagram of an embodiment of a data prediction apparatus according to an embodiment of the present application, where the apparatus may include:

a feature extraction module 501, configured to extract product features from a plurality of pieces of information associated with a target behavior corresponding to a target product;

the hierarchical division module 502 is configured to determine a target product hierarchy to which a target product belongs, in combination with product characteristics and product side attribute information;

a model determining module 503, configured to determine a target prediction model corresponding to a target product hierarchy from prediction models respectively constructed for different product hierarchies;

a quantity prediction module 504 to predict a first quantity of actions to perform the target action for the target product based on the product characteristics using the target prediction model.

In some embodiments, the hierarchical classification module may be specifically configured to identify a heat type of the target product based on product features using a product classification model; determining the level type of a product party corresponding to the target product according to the attribute information of the product party; and determining a target product hierarchy consisting of the heat type and the level type.

In some embodiments, the apparatus may further comprise:

the trend counting module is used for counting the historical target behavior quantity of the target product to obtain behavior quantity trend information; or counting the historical target behavior quantity of at least one product with the same type as the target product, which is provided by a product party, to obtain behavior quantity trend information;

the hierarchical division module determines the corresponding product side level type according to the product side attribute information, and comprises the following steps: and determining the level type of the product party by combining the attribute information of the product party and the behavior quantity trend information.

In some embodiments, the plurality of information includes one or more of product textual information, product attribute information, product conversion information generated during the first time period, associated product related information, and product party related information; the product text information comprises a product title text and/or a plurality of product evaluation texts produced in a first time period; the product conversion information comprises behavior quantity time sequence data respectively corresponding to the target behavior and the at least one associated behavior generated in the first time period; the related product related information comprises behavior quantity time sequence data of the related product generating target behaviors in a first time period; the product side related information comprises behavior quantity time sequence data of target behaviors and at least one associated behavior generated by a plurality of products of the same type as the target products in a first time period;

the feature extraction module is used for specifically forming product sub-features extracted according to one or more of the following implementation modes, and the product features.

Extracting target entity words belonging to the target entity type from the product title text; according to the evaluation data of the target entity words respectively corresponding to the plurality of heat evaluation indexes before and after the target entity words appear in the plurality of product title texts, counting to obtain gain data of the target entity words respectively corresponding to the plurality of heat evaluation indexes; generating product sub-features based on the gain data;

analyzing corresponding emotion polarities of the product evaluation texts in at least one target emotion aspect respectively; counting the occurrence number, the polarity number corresponding to different emotion polarities and the ratio value of the at least one target emotion aspect in the product evaluation texts to generate statistical data; generating product sub-features based on the statistical data;

respectively carrying out statistical processing on the behavior quantity time sequence data of a plurality of behaviors corresponding to the target product; generating product sub-characteristics based on data statistics results corresponding to a plurality of behaviors corresponding to the target product respectively;

performing statistical processing on the behavior quantity time series data of the target behaviors generated by the associated product in the first time period; generating product sub-characteristics based on the data statistics result of the target behavior corresponding to the associated product;

and the number of the first and second groups,

performing statistical processing on the behavior quantity time sequence data of a plurality of behaviors of a plurality of products of the same type as the target product; and generating product sub-characteristics based on data statistical results corresponding to a plurality of behaviors corresponding to the products respectively.

In some embodiments, the product conversion information includes behavior quantity time series data corresponding to the target behavior generated in the first time period and at least one associated behavior of the target behavior respectively; the apparatus may further include:

the abnormality detection module is used for adjusting the quantity of the behaviors generated at the time point when the first time period meets the abnormality condition aiming at any behavior; and acquiring behavior quantity time sequence data based on the behavior quantities corresponding to different time points in the first time period.

In some embodiments, the quantity prediction module is further configured to predict, in at least one prediction manner, a second behavior quantity for executing the target behavior with respect to the target product according to behavior quantity time series data of the target behavior generated by the target product in a first time period; and fusing to obtain the predicted behavior quantity of the target product based on the first behavior quantity and at least one second behavior quantity.

In some embodiments, the quantity prediction module is further configured to determine whether the first quantity of activities is greater than a quantity threshold; if not, taking the first behavior quantity as the predicted behavior quantity of the target product; and if so, taking the quantity threshold value as the predicted behavior quantity of the target product.

In some embodiments, the apparatus may further comprise:

the notification module is used for generating product prompt information based on the first behavior quantity; and informing the product side of the product prompt information.

The data prediction apparatus shown in fig. 5 may execute the data prediction method shown in the embodiment shown in fig. 2, and the implementation principle and the technical effect are not repeated. The specific manner in which each module and unit of the data prediction apparatus in the above embodiments perform operations has been described in detail in the embodiments related to the method, and will not be described in detail here.

In one possible design, the data prediction apparatus of the embodiment shown in fig. 5 may be implemented as a computing device, which may be used as a server or a client as shown in fig. 1, and as shown in fig. 6, the computing device may include a storage component 601 and a processing component 602;

the storage component 601 stores one or more computer instructions for the processing component to call and execute to implement the data prediction method as shown in fig. 2.

Among other things, the processing component 602 may include one or more processors to execute computer instructions to perform all or some of the steps of the methods described above. Of course, the processing elements may also 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, micro-controllers, microprocessors or other electronic components configured to perform the above-described methods.

The storage component 601 is configured to store various types of data to support operations at the terminal. The memory components may be implemented by any type or combination of volatile or non-volatile memory devices 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 memory, flash memory, magnetic or optical disks.

Of course, a computing device may also necessarily include other components, such as input/output interfaces, communication components, and so forth.

The input/output interface provides an interface between the processing components and peripheral interface modules, which may be output devices, input devices, etc. The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, and the like.

When the computing device is used as a server device, the computing device may be a physical device or an elastic computing host provided by a cloud computing platform, and the computing device may be a cloud server, and the processing component, the storage component, and the like may be basic server resources rented or purchased from the cloud computing platform.

In addition, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a computer, the data prediction method of the embodiment shown in fig. 2 may be implemented. The computer-readable medium may be included in the electronic device described in the above embodiment; or may exist separately without being assembled into the electronic device. The computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing.

Embodiments of the present application further provide a computer program product, which includes a computer program carried on a computer-readable storage medium, and when the computer program is executed by a computer, the data prediction method as described in the embodiment shown in fig. 2 can be implemented.

In such embodiments, the computer program may be downloaded and installed from a network, and/or installed from a removable medium. The computer program, when executed by a processor, performs various functions defined in the system of the present application.

Fig. 7 is a schematic structural diagram of an embodiment of a model generation apparatus provided in an embodiment of the present application, where the apparatus may include:

a model construction module 701, configured to construct prediction models corresponding to different product levels respectively;

a sample determination module 702, configured to determine, for any product level, a sample product belonging to the product level;

a sample processing module 703, configured to extract product sample features from a plurality of pieces of information associated with target behaviors corresponding to the sample product;

and a model training module 704, configured to train a prediction model corresponding to the product hierarchy based on the product sample features and the corresponding behavior prediction quantity.

The model generating apparatus shown in fig. 7 may execute the model generating method shown in the embodiment shown in fig. 4, and the implementation principle and the technical effect are not repeated. The specific manner in which each module and unit of the model generation apparatus in the above embodiments perform operations has been described in detail in the embodiments related to the method, and will not be elaborated herein.

In one possible design, the model generation apparatus of the embodiment shown in fig. 7 may be implemented as a computing device, which may be a server as shown in fig. 1, and which may include a storage component and a processing component

The storage component stores one or more computer instructions for invocation and execution by the processing component to implement the model generation method shown in fig. 4.

The organizational structure of the computing device may be the same as the organizational structure of the computing device shown in fig. 6, and therefore, the drawing is not repeated here.

The computing device may be a physical device or an elastic computing host provided by a cloud computing platform, and the computing device may be a cloud server, and the processing component, the storage component, and the like may be basic server resources rented or purchased from the cloud computing platform.

In addition, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a computer, the computer program can implement the model generation method in the embodiment shown in fig. 4. The computer-readable medium may be included in the electronic device described in the above embodiment; or may exist separately without being assembled into the electronic device. The computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing.

Embodiments of the present application further provide a computer program product, which includes a computer program carried on a computer-readable storage medium, and when the computer program is executed by a computer, the method for generating a model as described in the embodiment shown in fig. 4 can be implemented.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims

1. A method of data prediction, comprising:

2. The method of claim 1, wherein the determining a target product hierarchy to which the target product belongs in combination with the product characteristics and product party attribute information comprises:

identifying a heat type of the target product based on the product features by using a product classification model;

determining the level type of a product party corresponding to the target product according to the product party attribute information;

and determining a target product hierarchy consisting of the heat type and the level type.

3. The method of claim 2, further comprising:

counting the historical target behavior quantity of the target product to obtain behavior quantity trend information; or counting the historical target behavior quantity of at least one product of the same type as the target product, which is provided by the product party, to obtain behavior quantity trend information;

the determining the corresponding product side level type according to the product side attribute information includes:

and determining the level type of the product party by combining the product party attribute information and the behavior quantity trend information.

4. The method of claim 1, wherein the plurality of information includes one or more of product text information, product attribute information, product conversion information generated over a first time period, associated product related information, and product party related information; the product text information comprises a product title text and/or a plurality of product evaluation texts produced in the first time period; the product conversion information comprises behavior quantity time sequence data respectively corresponding to the target behavior and the at least one associated behavior generated in the first time period; the related product related information comprises behavior quantity time series data of the related product generating the target behavior in the first time period; the product side related information comprises behavior quantity time sequence data of the target behaviors and at least one associated behavior generated by a plurality of products of the same type as the target products in the first time period;

the method comprises the following steps of forming product features from a plurality of information related to products, specifically product sub-features extracted according to one or more of the following implementation modes:

and the number of the first and second groups,

5. The method of claim 4, wherein the product conversion information comprises behavior quantity time series data corresponding to the target behavior generated in the first time period and at least one associated behavior of the target behavior; the method further comprises the following steps:

6. The method of claim 1, further comprising:

and fusing to obtain the predicted behavior quantity of the target product based on the first behavior quantity and at least one second behavior quantity.

7. The method of claim 1, further comprising:

8. The method of claim 1, further comprising:

generating product prompt information based on the first action number;

and informing the product side of the product prompt information.

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

10. A method of model generation, comprising:

11. A computing device comprising a storage component that stores one or more computer instructions and a processing component; the one or more computer instructions to be invoked for execution by the processing component to implement the data prediction method of any one of claims 1 to 9 or the model generation method of claim 10.