CN111353815A - A method and system for predicting potential users - Google Patents

Abstract

The invention discloses a potential user prediction method and system. The potential user prediction method of the present invention includes: acquiring and storing feature weight sets of user prediction models of all commodity categories; The feature weight of the commodity category to be predicted; the potential users of the commodity category to be predicted are obtained by combining the user prediction model and the feature weight. The predictions of all commodity categories in the present invention share the same user prediction model and can share the same set of codes. When making predictions, only the feature weights of the commodity categories to be predicted are substituted. If a new commodity category is added, only the commodity needs to be added. The characteristic parameters of the prediction model of the category do not need to be modified in the code of the prediction model, which reduces the development workload, reduces the amount of code, and reduces the difficulty of subsequent maintenance.

Description

A method and system for predicting potential users

technical field

The invention relates to the technical field of data analysis, in particular to a potential user prediction method and system.

Background technique

The traditional method of obtaining potential buyers is to analyze statistical data through statistical analysis methods, and draw some rules to determine which users have a high probability of purchasing products. This method is inefficient, and it takes a lot of time and energy to analyze statistical data to obtain rules. Moreover, the amount of data processed by this method is limited, and the accuracy of the rules derived from the limited amount of data is also low.

At present, the big data mining method is used to predict potential customers, and a prediction model is constructed based on a large amount of historical data. A prediction model is constructed for a commodity category, and the prediction model of the corresponding commodity category is called for prediction. Specifically, the Python language is used to write the prediction process. One product category corresponds to a set of Python codes, and the development workload is heavy. If the prediction model of a certain product category changes, the corresponding Python code needs to be changed, which is difficult to maintain. Moreover, distributed parallel computing cannot be realized during deployment and execution, and the execution efficiency is greatly dependent on hardware resources.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a potential user prediction method and system, which solves the technical problems of large development workload and difficult maintenance of existing prediction methods.

To achieve the above object, the embodiments of the present invention adopt the following technical solutions:

In a first aspect, an embodiment of the present invention provides a potential user prediction method, including:

Obtain and store the feature weight set of the user prediction model for all commodity categories;

From the feature weight set, find the feature weight of the commodity category to be predicted;

Combined with the user prediction model and the feature weight, the potential users of the commodity category to be predicted are obtained.

With reference to the first aspect, in a first possible implementation manner of the first aspect, the acquiring and storing the feature weight set of the user prediction model for all commodity categories further includes:

M features are acquired, and a user prediction model is constructed based on the M features; wherein, M is an integer greater than 1.

With reference to the first aspect, in a second possible implementation manner of the first aspect, the acquiring and storing the feature weight set of the user prediction model for all commodity categories specifically includes:

Using a machine learning algorithm, the feature weights of the M features of the user prediction model of the commodity category are obtained;

establishing the correspondence between the commodity category and the feature weight, and storing the feature weight;

The feature weights of all commodity categories are obtained to form a feature weight set.

With reference to the second possible implementation manner of the first aspect, in the third possible implementation manner of the first aspect, the machine learning algorithm is used to obtain the features of the M features of the user prediction model of the commodity category weights, including:

From the user information data and user behavior data of the commodity category, filter to obtain the user feature data corresponding to the M features of the user prediction model of the commodity category;

Preprocessing the user feature data to obtain feature training data;

A machine learning algorithm is used to train the feature training data to obtain feature weights of M features of the user prediction model of the commodity category.

With reference to the third possible implementation manner of the first aspect, in the fourth possible implementation manner of the first aspect, the preprocessing of the user feature data to obtain feature training data specifically includes:

cleaning the user feature data to obtain cleaned data;

The cleaned data is normalized using the z-score method to obtain feature training data.

With reference to the first aspect, in a fifth possible implementation manner of the first aspect, the method further includes:

A grid search method is used to adjust the feature weights of each commodity category, and the feature weight set is updated.

In a second aspect, an embodiment of the present invention provides a potential user prediction system, including:

The acquisition module is used to acquire and store the feature weight set of the user prediction model of all commodity categories;

a search module, configured to search and obtain the feature weight of the commodity category to be predicted from the feature weight set;

A prediction module, configured to combine the user prediction model and the feature weight to obtain potential users of the commodity category to be predicted.

With reference to the second aspect, in a first possible implementation manner of the second aspect, the system further includes:

A construction module, configured to acquire M features, and construct a user prediction model based on the M features; wherein, M is an integer greater than 1.

With reference to the first possible implementation manner of the second aspect, in the second possible implementation manner of the second aspect, the obtaining module includes:

a learning unit, used for using a machine learning algorithm to obtain the feature weights of the M features of the user prediction model of the commodity category;

a storage unit, configured to establish a correspondence between the commodity category and the feature weight, and store the feature weight;

The set unit is used to obtain the feature weights of all commodity categories to form a feature weight set.

With reference to the second aspect, in a third possible implementation manner of the second aspect, the system further includes:

The optimization module is used to adjust the feature weights of each commodity category by using the grid search method, and update the feature weight set.

The potential user prediction method and system provided by the embodiments of the present invention solve the technical problems of large development workload and difficult maintenance of the existing prediction method. The potential user prediction method of the present embodiment first acquires and stores the feature weight set of the user prediction model of all commodity categories; then searches and obtains the feature weight value of the commodity category to be predicted from the feature weight set; finally Combined with the user prediction model and the feature weight, the potential users of the commodity category to be predicted are obtained. Compared with the prior art, in this embodiment, the feature weights of the user prediction models of all commodity categories are obtained by training, and the feature weights corresponding to the commodity categories to be predicted are extracted during prediction, and are substituted into the user prediction model for calculation, so as to obtain the to-be-predicted feature weights. The potential users of the predicted commodity category share the same user prediction model for all commodity category predictions, so they can share the same set of prediction codes. If you add a commodity category, you only need to add the feature weight of the commodity category There is no need to modify the prediction code, reducing the development workload, reducing the amount of code, and reducing the difficulty of subsequent maintenance.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings that need to be used in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

1 is a flowchart of a potential user prediction method provided by an embodiment of the present invention;

2 is a structural block diagram of a potential user prediction system provided by an embodiment of the present invention;

FIG. 3 is a structural block diagram of a potential user prediction system provided by another embodiment of the present invention.

Detailed ways

In order to make those skilled in the art better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, the embodiments obtained by those of ordinary skill in the art without making creative efforts in advance shall all belong to the protection scope of the present invention.

It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as herein, are not to be taken in an idealized or overly formal sense. explain.

An embodiment of the present invention provides a potential user prediction method, as shown in FIG. 1 , including:

Step S10, acquiring and storing the feature weight sets of the user prediction models of all commodity categories;

Step S20, from the feature weight set, search to obtain the feature weight of the commodity category to be predicted;

Step S30, combining the user prediction model and the feature weight to obtain the potential users of the commodity category to be predicted.

The embodiment of the present invention first obtains the feature weight set of the user prediction model, the feature weight set is composed of the feature weights of the user prediction model of all commodity categories, and the commodity category to be predicted is extracted from the feature weight set during prediction. Feature weights, aggregate user prediction models, and obtain potential users of the commodity category to be predicted. User predictions of all commodity categories share the same user prediction model and the same set of user prediction model codes. If a new commodity category is added, only the feature weight of the commodity category needs to be added, and the user prediction model code does not need to be modified. Development workload, reduce the amount of code, and reduce the difficulty of subsequent maintenance.

According to an embodiment of the present invention, before step S10, the method further includes: acquiring M features, and constructing a user prediction model based on the M features; wherein, M is an integer greater than 1. In this embodiment, M features are selected and obtained by manually analyzing the relationship between user information, user behavior and purchase. In this embodiment, a user prediction model is constructed and obtained based on the logistic regression model and M features.

Preferably, M is 54. The 54 features include: whether it is a new user, whether it is a new user of the product category, the difference between the last purchase, the difference between the last purchase of the product category, the historical consumption information has the maximum payment amount, payment Number of times, average payment per payment, maximum payment in 6 months, number of payments in 6 months, average payment per payment in 6 months, maximum payment in 3 months, number of payments in 3 months, number of payments in 3 months Average value of each payment, maximum payment amount in 1 month, number of payments in 1 month, average value of each payment in 1 month, maximum payment amount in 15 days, number of payments in 15 days, average value of each payment in 15 days, the product category The maximum payment amount for the item, the number of payments for this item, the average value of each payment for this item, the maximum payment for this item within 6 months, the number of payments for this item within 6 months, and 6 items for this item The average value of each payment within a month, the maximum payment amount for the product category within 3 months, the number of payments for the product category within 3 months, the average value of each payment for the product category within 3 months, the product category for 1 month The maximum payment amount within 1 month, the number of payments for the product category within 1 month, the average value of each payment for the product category within 1 month, the maximum payment amount for the product category within 15 days, the number of payments for the product category within 15 days, the product category Category average value of each payment within 15 days, time difference since last visit, number of visits within 7 days, number of visits within 7 days, number of visits within 14 days, length of visit within 14 days, time difference since last search, number of searches within 7 days, within 14 days Search times, time difference since the last collection, collection times within 7 days, collection times within 14 days, time difference since the last car purchase 2, additional car purchases within 7 days, additional car purchases 2 times within 14 days, last additional car purchase 1 since then Time difference: 1 additional car purchase within 7 days, 1 additional car purchase within 14 days, the time difference since the last order was not paid, the number of times the order was not paid within 7 days, and the number of times the order was not paid within 14 days. Selecting the above 54 features can describe user information and user behavior more comprehensively, and the user prediction model constructed using the 54 features is more accurate and improves the accuracy of prediction. Selecting 54 features instead of more features reduces the complexity of obtaining feature weights and the complexity of user prediction calculations, and improves the prediction speed.

In step S10, it is preferred to obtain and store the feature weights of the user prediction models for each commodity category respectively. Wherein, according to an embodiment of the present invention, obtaining the feature weight of the prediction model of a certain commodity category specifically includes:

Step S101, using a machine learning algorithm to obtain the feature weights of the above M features of the user prediction model of the commodity category;

Step S102, establishing the correspondence between the commodity category and the feature weight, and storing the feature weight.

In step S103, the feature weights of all commodity categories are obtained to form a feature weight set, and the user prediction model is associated with the feature weight set.

Wherein, step S101 specifically includes:

Step S1011, from the user information data and user behavior data of the commodity category, filter to obtain the user feature data corresponding to the M features of the user prediction model of the commodity category;

Step S1012, preprocessing the user feature data to obtain feature training data;

Step S1013, using a machine learning algorithm to train the feature training data to obtain feature weights of M features of the prediction model of the commodity category.

In step S1011, user information data and user behavior data are extracted from traffic tables, search tables, add-on and order tables, etc., and user feature data corresponding to M features is obtained from the user information data and user behavior data. Among them, user information data includes gender, age, membership, membership years, etc., user behavior data includes the time since the last purchase, the maximum payment amount within 1 month, the number of payments within 1 month, the number of visits within 7 days, and the target Commodity category payment times, etc. Different commodity categories have different user information and user behavior, so user feature data corresponding to M features of each commodity category are obtained from the user information data and user behavior data of each commodity category.

Step S1012 specifically includes: cleaning the user feature data to obtain cleaned data; normalizing the cleaned data by using a z-score method to obtain feature training data. Among them, cleaning includes zero-filling or filling the maximum value for the null values in the user characteristic data, and covering the abnormal values with the 98th percentile quantile. The user feature data is processed to obtain feature training data, which facilitates the use of machine learning algorithms to train it to obtain feature weights.

In step S1013, a machine learning algorithm is used to train the feature training data obtained in step S1012 to obtain feature weights. The user prediction model in this embodiment is based on a logistic regression model, so a logistic regression algorithm is used to train the feature training data to obtain feature weights of M features in the user prediction model.

According to an embodiment of the present invention, the feature weights are stored in the hive table, and the user prediction model is associated with the hive table. One hive table can store the feature weights corresponding to one commodity category, or one hive table can store the feature weights corresponding to multiple or even all commodity categories. The set of all hive tables constitutes the feature weight set.

In the embodiment of the present invention, one commodity category corresponds to a group (M) of feature weights, and a corresponding relationship is established between each commodity category and its feature weights, so as to facilitate the quick search of the feature weights of the commodity category to be predicted in subsequent predictions . Store the feature weights in the hive table, use the HiveSq language to write a user prediction model, and associate the hive table, so that the feature weights of the commodity category to be predicted are extracted from the associated hive table during prediction, and passed into the user prediction model. To perform calculations, convert from hive tasks to MapReduce task execution, which can realize distributed parallel computing, fast prediction calculation speed, and high reliability. The HiveSql code of the same set of user prediction models can be reused for prediction of different commodity categories, and the development workload is small, reducing the difficulty of post-maintenance. It is only necessary to update the feature weights of each commodity category to optimize the prediction ability and improve the prediction accuracy without changing the code, which simplifies the update optimization and maintains the generalization ability of the prediction method.

In the embodiment of the present invention, considering that different commodity categories have different user information and user behaviors, user feature data corresponding to M features of each commodity category are obtained from the user information data and user behavior data of each commodity category, respectively. , after preprocessing, the feature training data of each commodity category is obtained, the feature training data of each commodity category is trained by machine learning algorithm, and finally the feature weight of each commodity category is obtained. The feature weight of the commodity category to be predicted is substituted into the user prediction model to predict the potential users of the commodity category to be predicted, and the prediction accuracy is high.

In step S20, during prediction, according to the corresponding relationship between the commodity category and the feature weight, the feature weight of the commodity category to be predicted is extracted from the feature weight set, which may be an associated hive table.

In step S40, the feature weights are input into the user prediction model for calculation to obtain potential users of the commodity category to be predicted.

According to an embodiment of the present invention, the method further includes: using a grid search method to adjust the feature weights of each commodity category, and update the feature weight set. Preferably, the grid search method is used to adjust the feature weights of each commodity category according to the preset time, so as to update the feature weight set. Optimize the feature weights of each commodity category, and improve the prediction accuracy after substituting it into the user prediction model.

The potential user prediction method provided by the embodiment of the present invention solves the technical problems of large development workload and difficult maintenance of the existing prediction method. First, obtain and store the feature weight sets of the user prediction models of all commodity categories; then, from the feature weight sets, find the feature weights of the commodity categories to be predicted; finally combine the user prediction model and the The feature weight is used to obtain the potential users of the commodity category to be predicted. Compared with the prior art, in this embodiment, the feature weights of the user prediction models of all commodity categories are obtained by training, and the feature weights corresponding to the commodity categories to be predicted are extracted during prediction, and are substituted into the user prediction model for calculation, so as to obtain the to-be-predicted feature weights. The potential users of the predicted commodity category share the same user prediction model for all commodity category predictions, so they can share the same set of prediction codes. If you add a commodity category, you only need to add the feature weight of the commodity category There is no need to modify the prediction code, reducing the development workload, reducing the amount of code, and reducing the difficulty of subsequent maintenance.

An embodiment of the present invention also provides a potential user prediction system, as shown in FIG. 2 , including:

The acquisition module is used to acquire and store the feature weight set of the user prediction model of all commodity categories;

a search module, configured to search and obtain the feature weight of the commodity category to be predicted from the feature weight set;

A prediction module, configured to combine the user prediction model and the feature weight to obtain potential users of the commodity category to be predicted.

According to an embodiment of the present invention, as shown in FIG. 3 , the system further includes:

A construction module, configured to acquire M features, and construct a user prediction model based on the M features; wherein, M is an integer greater than 1.

According to an embodiment of the present invention, the obtaining module includes:

a learning unit, used for using a machine learning algorithm to obtain the feature weights of the M features of the user prediction model of the commodity category;

a storage unit, configured to establish a correspondence between the commodity category and the feature weight, and store the feature weight;

The set unit is used to obtain the feature weights of all commodity categories to form a feature weight set.

According to an embodiment of the present invention, the learning unit specifically includes:

A screening subunit, configured to obtain user feature data corresponding to M features of the user prediction model of the commodity category by screening from the user information data and user behavior data of the commodity category;

a preprocessing subunit for preprocessing the user feature data to obtain feature training data;

A training subunit, configured to use a machine learning algorithm to train the feature training data to obtain feature weights of M features of the user prediction model of the commodity category.

According to an embodiment of the present invention, the preprocessing subunit is further used for:

The user feature data is cleaned to obtain cleaned data; the cleaned data is normalized by using the z-score method to obtain feature training data.

According to an embodiment of the present invention, the system further includes:

The optimization module is used to adjust the feature weights of each commodity category by using the grid search method, and update the feature weight set.

The potential user prediction system provided by the embodiment of the present invention solves the technical problems of large development workload and difficult maintenance of the existing prediction method. In the potential user prediction system of this embodiment, the acquisition module acquires and stores the feature weight sets of the user prediction models of all commodity categories; the search module searches and obtains the feature weights of the commodity categories to be predicted from the feature weight set. ; The prediction module combines the user prediction model and the feature weight to obtain the potential users of the commodity category to be predicted. Compared with the prior art, in this embodiment, the feature weights of the user prediction models of all commodity categories are obtained by training, and the feature weights corresponding to the commodity categories to be predicted are extracted during prediction, and are substituted into the user prediction model for calculation, so as to obtain the to-be-predicted feature weights. The potential users of the predicted commodity category share the same user prediction model for all commodity category predictions, so they can share the same set of prediction codes. If you add a commodity category, you only need to add the feature weight of the commodity category There is no need to modify the prediction code, reducing the development workload, reducing the amount of code, and reducing the difficulty of subsequent maintenance.

Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the apparatus embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and reference may be made to some descriptions of the method embodiments for related parts. Those skilled in the art can understand that the modules in the apparatus in the embodiment can be adaptively changed and arranged in one or more apparatuses different from the embodiment. The modules or units or components in the embodiments may be combined into one module or unit or component, and further they may be divided into multiple sub-modules or sub-units or sub-assemblies. All features disclosed in this specification (including accompanying claims, abstract and drawings) and any method so disclosed may be employed in any combination unless at least some of such features and/or procedures or elements are mutually exclusive. All processes or units of equipment are combined. Each feature disclosed in this specification (including accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. During execution, the processes of the embodiments of the above-mentioned methods may be included. The storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM) or the like.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical scope disclosed by the present invention can easily think of changes or substitutions. All should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (10)

1. A method for predicting potential users, comprising: Obtain and store the feature weight set of the user prediction model for all commodity categories; From the feature weight set, find the feature weight of the commodity category to be predicted; Combined with the user prediction model and the feature weight, the potential users of the commodity category to be predicted are obtained. 2. The method according to claim 1, wherein, acquiring and storing the feature weight set of the user prediction model of all commodity categories, further comprising: M features are acquired, and a user prediction model is constructed based on the M features; wherein, M is an integer greater than 1. 3. The method according to claim 2, wherein the acquiring and storing the feature weight set of the user prediction model of all commodity categories specifically includes: Using a machine learning algorithm, the feature weights of the M features of the user prediction model of the commodity category are obtained; establishing the correspondence between the commodity category and the feature weight, and storing the feature weight; The feature weights of all commodity categories are obtained to form a feature weight set. 4. The method according to claim 3, characterized in that, by using a machine learning algorithm, the feature weights of M features of the user prediction model of the commodity category are obtained, specifically comprising: From the user information data and user behavior data of the commodity category, filter to obtain the user feature data corresponding to the M features of the user prediction model of the commodity category; Preprocessing the user feature data to obtain feature training data; A machine learning algorithm is used to train the feature training data to obtain feature weights of M features of the user prediction model of the commodity category. 5. The method according to claim 4, wherein the preprocessing of the user characteristic data to obtain characteristic training data specifically comprises: cleaning the user feature data to obtain cleaned data; The cleaned data is normalized using the z-score method to obtain feature training data. 6. The method of claim 1, wherein the method further comprises: A grid search method is used to adjust the feature weights of each commodity category, and the feature weight set is updated. 7. A potential user prediction system, comprising: The acquisition module is used to acquire and store the feature weight set of the user prediction model of all commodity categories; a search module, configured to search and obtain the feature weight of the commodity category to be predicted from the feature weight set; A prediction module, configured to combine the user prediction model and the feature weight to obtain potential users of the commodity category to be predicted. 8. The system of claim 7, wherein the system further comprises: A construction module, configured to acquire M features, and construct a user prediction model based on the M features; wherein, M is an integer greater than 1. 9. The system according to claim 8, wherein the acquisition module comprises: a learning unit, used for using a machine learning algorithm to obtain the feature weights of the M features of the user prediction model of the commodity category; a storage unit, configured to establish a correspondence between the commodity category and the feature weight, and store the feature weight; The set unit is used to obtain the feature weights of all commodity categories to form a feature weight set. 10. The system of claim 7, wherein the system further comprises: The optimization module is used to adjust the feature weights of each commodity category by using the grid search method, and update the feature weight set.

Images