User mining method, user mining device, and electronic device

Abstract

The application relates to the technical field of big data processing, and provides a user mining method, a user mining device and electronic equipment. The method comprises the following steps: determining a target feature type corresponding to a to-be-recommended object, wherein the target feature type is used for indicating features in a target time range; acquiring first features of the to-be-recommended object in the target time range based on the target feature type; and determining a target user of the to-be-recommended object from to-be-selected users based on the first features. The user mining method provided in the application can improve the efficiency of user mining.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data processing, and in particular to a user mining method, a user mining device and an electronic device. BACKGROUND

[0002] With the continuous development of computer technology, more and more services and products can be provided to users, such as video entertainment, e-commerce, learning and office, food and travel services. Different services have different business scenarios and are easy to distinguish. However, the audiences of different products can be very close or very different, so it is very important to mine potential users of the product.

[0003] When mining potential users of a product, a large number of features are usually extracted from the historical data of the product to find the closest users. However, this way requires processing more historical data, consumes more computing power, and is less efficient. SUMMARY

[0004] The embodiments of the present application provide a user mining method, a user mining device and an electronic device to solve the problem of low efficiency in the mining process of potential users.

[0005] In a first aspect, the embodiments of the present application provide a user mining method, comprising:

[0006] determining a target feature type corresponding to a to-be-recommended object, the target feature type being used to indicate features in a target time range;

[0007] based on the target feature type, obtaining a first feature of the to-be-recommended object in the target time range;

[0008] based on the first feature, determining a target user of the to-be-recommended object from the to-be-recommended object.

[0009] In one embodiment, the determining of the target feature type corresponding to the to-be-recommended object comprises:

[0010] based on different time periods, obtaining sample features in each of the time periods;

[0011] determining a cycle type to which each candidate object belongs, the candidate objects including the to-be-recommended object;

[0012] determining the correlation of the sample features in each of the time periods with each of the cycle types;

[0013] based on the correlation and the cycle type to which the to-be-recommended object belongs, determining a target sample feature corresponding to the to-be-recommended object;

[0014] According to the target sample feature, a target feature type corresponding to the to-be-recommended object is determined.

[0015] In one embodiment, the determining of the correlation between the sample feature in each of the time periods and each of the period types respectively comprises:

[0016] The sample feature in each of the time periods is converted into a sample vector, and the sample feature is a multi-dimensional feature.

[0017] The first correlation degree between the sample vector of each of the time periods and each of the period types is determined.

[0018] The second correlation degree between the feature vector of each of the dimensional features in the sample vector and each of the period types is determined.

[0019] The correlation between the sample feature in each of the time periods and each of the period types is determined in combination with the first correlation degree and the second correlation degree.

[0020] In one embodiment, the method further comprises:

[0021] Obtaining historical data of a to-be-classified object;

[0022] Extracting period features of different time periods from the historical data;

[0023] Determining a candidate correlation between the period features and the to-be-classified object;

[0024] Based on the candidate correlation, determining a period type to which the to-be-classified object belongs.

[0025] In one embodiment, the determining of the target feature type corresponding to the to-be-recommended object according to the target sample feature comprises:

[0026] According to the time period of the target sample feature, a target time range of the to-be-recommended object is determined.

[0027] In one embodiment, the determining of the target user of the to-be-recommended object from the to-be-selected users based on the first feature comprises:

[0028] According to the period type of the to-be-recommended object, a feature combination corresponding to the to-be-recommended object is obtained, and the feature combination includes a target feature type and a non-target feature type.

[0029] Based on the non-target feature type, a second feature of the to-be-selected user in the target time range is obtained.

[0030] Based on the first feature and the second feature, a target user of the to-be-recommended object is determined from the to-be-selected users.

[0031] In an embodiment, the determining the target user of the to-be-recommended object from the candidate users based on the first feature and the second feature comprises:

[0032] dimensionally reducing the first feature and the second feature, and determining the target user of the to-be-recommended object based on the dimensionally reduced first feature and second feature.

[0033] In a second aspect, an embodiment of the present application provides a user mining apparatus, comprising:

[0034] a time determining module configured to determine a target feature type corresponding to a to-be-recommended object, the target feature type being used to indicate a feature in a target time range;

[0035] a data extracting module configured to acquire a first feature of a candidate user in the target time range based on the target feature type;

[0036] a data mining module configured to determine a target user of the to-be-recommended object from the candidate users based on the first feature.

[0037] In an embodiment, the time determining module comprises: a sample acquiring module configured to acquire sample features in different time periods based on different time periods; a type determining module configured to determine a period type to which each candidate object belongs, the candidate objects including the to-be-recommended object; a correlation calculating module configured to determine a correlation between the sample features in each time period and each period type; a sample determining module configured to determine a target sample feature corresponding to the to-be-recommended object based on the correlation and a period type to which the to-be-recommended object belongs; and a feature type determining module configured to determine a target feature type corresponding to the to-be-recommended object according to the target sample feature.

[0038] In an embodiment, the correlation calculating module comprises: a vector converting module configured to convert the sample features in each time period into sample vectors, the sample features being multi-dimensional features; a period correlation calculating module configured to determine a first correlation degree between the sample vectors of each time period and each period type; a dimension correlation calculating module configured to determine a second correlation degree between feature vectors of each dimensional feature in the sample vectors and each period type; and a correlation determining module configured to determine the correlation between the sample features in each time period and each period type by combining the first correlation degree and the second correlation degree.

[0039] In an embodiment, the user mining apparatus further comprises: a historical data acquisition module, configured to acquire historical data of the object to be classified; a period determination module, configured to extract period features of different time periods from the historical data; a candidate relevance determination module, configured to determine candidate relevance of the period features and the object to be classified; and a classification module, configured to determine a period type to which the object to be classified belongs based on the candidate relevance.

[0040] In an embodiment, the feature type determination module is configured to determine a target time range of the object to be recommended according to a time period of the target sample feature.

[0041] In an embodiment, the data mining module comprises: a feature combination module, configured to acquire a feature combination corresponding to the object to be recommended according to the period type of the object to be recommended, the feature combination comprising a target feature type and a non-target feature type; a feature extraction module, configured to acquire a second feature of the candidate user within the target time range based on the non-target feature type; and a data screening module, configured to determine a target user of the object to be recommended from the candidate user based on the first feature and the second feature.

[0042] In an embodiment, the data screening module is configured to perform dimension reduction on the first feature and the second feature, and determine the target user of the object to be recommended based on the first feature and the second feature after the dimension reduction.

[0043] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory storing a computer program, and the processor implements the user mining method of the first aspect when executing the program.

[0044] In a fourth aspect, an embodiment of the present application provides a computer program product, comprising a computer program, and the computer program implements the user mining method of the first aspect when executed by a processor.

[0045] The user mining method, the user mining apparatus and the electronic device provided by the embodiments of the present application can indicate a time period in which data is to be acquired through a target feature type of an object to be recommended, thereby acquiring features of a candidate user within a specified time period, which can greatly reduce the amount of data to be processed and improve mining efficiency. Moreover, for different objects to be recommended, corresponding target feature types can be acquired respectively for each object to be recommended, the target feature types and the objects to be recommended are targeted, data of different time periods can be determined for different objects, and the accuracy of mining can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0047] Figure 1 is one of the flowcharts of the user mining method provided by the embodiments of the application;

[0048] Figure 2 is the second flowchart of the user mining method provided by the embodiments of the application;

[0049] Figure 3 is the third flowchart of the user mining method provided by the embodiments of the application;

[0050] Figure 4 is the structural schematic diagram of the user mining device provided by the embodiments of the application;

[0051] Figure 5 is the structural schematic diagram of the electronic device provided by the embodiments of the application. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the application clearer, the technical solutions in the application will be described clearly and completely in the following with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are some of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0053] In the embodiments of the application, a user mining method is first provided. Exemplarily, the user mining method can be applied to electronic devices with display function such as mobile phones, tablet computers, personal computers (PCs), wearable electronic devices (such as smart watches), augmented reality (AR) / virtual reality (VR) devices, and vehicle-mounted devices, and the embodiments of the application do not make any limitation on this.

[0054] Figure 1 is a flowchart of the user mining method in the embodiments of the application. With reference to Figure 1 The user mining method provided by the embodiments of the application can include:

[0055] Step 100: determining a target feature type corresponding to a to-be-recommended object, the target feature type being used to indicate a feature in a target time range.

[0056] The object to be recommended refers to various objects that can be recommended to a user on an electronic device, such as a product or a service. For example, the object to be recommended can specifically include a traffic package, a call package, a broadband package, and the like, or can include a card service, a broadband service, and the like, and the present embodiment is not particularly limited in this regard.

[0057] The following takes the object to be recommended as a product to be recommended as an example.

[0058] The target feature type is used to indicate a feature in a target time range. The feature can include an interaction feature of a product and a feature of an interaction object of the product. For example, the target feature type can be a gender of a user who orders a product within 1 day, a profession of a user who orders a product within 1 day, a browsing time of a user who browses a product within 1 day, and the like.

[0059] According to the attributes of the product, each product can be pre-classified to obtain a plurality of classes. Each class corresponds to a target feature type, and the target feature type of the object to be recommended (product) can be determined according to the class in which the object to be recommended is located.

[0060] Figure 2 A flowchart for determining the target feature type corresponding to the object to be recommended is shown. As shown in Figure 2 The method can include the following steps:

[0061] Step 201: Based on different time periods, sample features in each of the time periods are obtained. A plurality of time periods can be pre-divided, such as a day, a week, two weeks, a month, three months, and the like. Alternatively, the time periods can be divided in other ways, such as one day, three days, five days, ten days, and the like, and the present embodiment is not limited in any way to the division of the time periods.

[0062] According to different time periods, historical data can be obtained, and then features are extracted from the historical data as sample features. The historical data can be information recorded by a platform, which can specifically include basic information of a user, such as the gender, profession, age, and online time of the user; basic information of a product, such as the attributes and type of the product; and interaction information of the user and the product, such as the number of times, time, and duration of browsing a product, and the like. Each piece of historical data can be a sample feature, and the sample feature is a multi-dimensional feature, such as a sample feature composed of dimensions such as the gender of the user, the profession of the user, the type of the product purchased by the user, the number of times the product is purchased by the user, and the amount of money. Alternatively, the historical data in each time period is statistically processed to obtain a sample feature corresponding to the time period.

[0063] For example, the time period is divided into day, week, two weeks, month, and then the historical data of a day, a week, two weeks, and a month is obtained according to each time period. The historical data of each time period is counted respectively to obtain the characteristics of the user logged in within a day, such as the traffic package used by the user, the traffic usage, the traffic remaining amount, the interactive characteristics of the user, such as the products browsed by the user, the products ordered, and the characteristics of the corresponding products, such as the browsing amount and the ordering amount of each product. The characteristics obtained by counting each time period are used as the sample characteristics of the time period. Similarly, the sample characteristics also include multiple dimensions.

[0064] Step 202: Determine the cycle type to which each candidate object belongs, wherein the candidate object includes the to-be-recommended object. The candidate object can refer to all products. The candidate object can include all products or services, and also includes the to-be-recommended product. According to experience, the corresponding cycle type of each candidate object can be determined in advance. The cycle type is the type information of the candidate object. For example, the cycle type can be divided into day, month, and the like, and can also include other types, such as week, two weeks, half a year, and the like, and the present embodiment does not make any limitation thereon. For example, the daily traffic package can be divided into day, and the monthly traffic package can be month.

[0065] Step 203: Determine the correlation between the sample characteristics in each time period and each cycle type. The sample characteristics in each time period are converted into a vector to obtain the sample vector of each cycle, and then the correlation between the sample vector and the product of each cycle type is calculated. For example, the word2vec method can be used to encode the sample characteristics, convert them into a vector, and convert the characteristics of the candidate object corresponding to each cycle type into a vector. For convenience of description, the vector converted from the sample characteristics is referred to as the first sample vector, and the vector converted from the candidate object corresponding to the cycle type is referred to as the second sample vector.

[0066] For example, the sample characteristics of each time period include multiple, each sample characteristic can be converted into a first sample vector, each time period can include multiple vectors, and the set of all first sample vectors can be represented as: S = (W1, W2, W3, …, Wn). Wherein, n represents the nth time period, and n can be a positive integer. The sample characteristics are multi-dimensional characteristics, for example, each sample characteristic can include data type, data table to which it belongs, script used for reading data, feature description, Chinese name of the feature, English name of the feature, and multiple dimensions. Each dimension can be converted into a vector, for example, each sample characteristic includes m dimensions, and the first sample vector can also be represented as: S = (X1, X2, X3, …, Xm). n 1,1 1,2 1,3 ​​​, ..., X n,m Similarly, the features of candidate objects for each period type, such as product name, type, and product description, are transformed into a second sample vector. Then, the correlation between each first sample vector and the second sample vector is calculated, which is the correlation between the sample features of each time period and the period type.

[0067] For example, correlation can be calculated in various ways, such as by using a pre-trained model to calculate the correlation between each period type and sample features. Specifically, a certain amount of historical data is acquired, which can include user features, product features, and user-product interaction features. The acquired historical data is used as training samples and converted into vectors. The training objective can be determined through the user and product interaction features. The training objective can include products and labels, where the label represents whether a user has ordered a product. For example, if a user ordered product A in the historical data, then the training objective for that historical data is to order product A. Product information can be encoded and converted into features or vectors. Then, the vectors of the training samples and the vectors corresponding to the products are input into the model. The model fits the association between the input historical data vectors and the product vectors, outputting the probability that the product was ordered, and adjusting the model based on the loss between the output probability and the label. The model can be repeatedly trained using a large amount of historical data and the training objective corresponding to each historical data point. Training is complete when the model's prediction result for the historical data is the same as the historical data label, or when the loss is less than a preset value. The trained model is used as a correlation prediction model. The sample features (first sample vector) and the features of the candidate object (second sample vector) of each time period are input into the correlation prediction model. The model can predict the probability of the candidate object being ordered. The output probability can be used as the correlation between the sample features and the period type to which the candidate object belongs.

[0068] In this embodiment, a correlation prediction model is used to determine the correlation between sample features and periodicity type. This method is simple to operate and highly efficient. Furthermore, the model can identify implicit associations between features and has high accuracy in determining correlations.

[0069] For example, Figure 3 A flowchart illustrating the correlation between sample characteristics and period type is shown. Figure 3 As shown, the method may include:

[0070] Step 301: Convert the sample features within each time period into a sample vector, wherein the sample features are multi-dimensional features. For example, the word2vec method is used to convert the sample features within each time period into a sample vector. This sample vector includes vectors of multiple dimensions.

[0071] Step 302: determining a first correlation degree between each sample vector of the time period and each cycle type. The probability between each sample vector of the time period and each cycle type can be determined by the above-mentioned model as the first correlation degree. For example, the sample vector W i is input into the above-mentioned correlation prediction model, the probability P(W i ) between the sample vector W j and the cycle type Y i is determined by the model, and the obtained probability P(W i ) is taken as the first correlation degree. Wherein i represents the i-th time period, j represents the j-th cycle type, Y j represents the vector of the candidate object of the j-th cycle type. The sample vectors of the n time periods are sequentially input into the correlation prediction model, and the first correlation degrees between the n time periods and the i-th cycle type are obtained.

[0072] Step 303: determining a second correlation degree between each feature vector of the feature of the sample vector and each cycle type. Each sample vector is composed of m feature vectors of features, and each feature vector of each dimension of the same sample vector is input into the above-mentioned correlation prediction model, so that the second correlation degree between each feature vector of each dimension and each cycle type can be determined. For example, the feature vector X n,m is the vector of the m-th dimension in the n-th time period, and the second correlation degree P(X n,m ) between the vector X n,m and the j-th cycle type can be obtained by the correlation prediction model.

[0073] Step 304: combining the first correlation degree and the second correlation degree to determine the correlation between the sample feature in each time period and each cycle type. For example, the second correlation degrees of each dimension of the i-th time period and the j-th cycle type are added to obtain the total second correlation degree of the i-th time period and the j-th cycle type, and then the total second correlation degree is multiplied by the first correlation degree of the i-th time period and the j-th cycle type to obtain the correlation between the i-th time period and the j-th cycle type. For example, the correlation between the i-th time period and the j-th cycle type can be represented as: K=P(W i )(P(X i,1 )+P(X i,2 )+P(X i,3 )+…+P(X i,m )). Wherein m represents the dimension of the sample feature of the i-th time period, and m>0.

[0074] For example, the correlation can also be determined in combination with the Pearson correlation coefficient. Specifically, the sum of the second correlation degrees of each dimension in the same time period is multiplied by the first correlation degree, and then multiplied by the Pearson correlation coefficient. The result is taken as the correlation between the time period and the cycle type. That is, K = P(W i )(P(X i,1 )+P(X i,2 )+P(X i,3 )+…+P(X i,m ))t. Where t is the Pearson correlation coefficient, which can be calculated by the Pearson correlation coefficient formula. The Pearson correlation coefficient can measure the degree of linear correlation between the sample features and the cycle type. The first correlation degree and the second correlation degree output by the model can measure the degree of non-linear correlation between the sample features and the cycle type. Measuring the correlation degree from the perspective of linearity and non-linearity can improve the accuracy of the correlation.

[0075] Step 204: Based on the correlation and the cycle type to which the object to be recommended belongs, determine the target sample feature corresponding to the object to be recommended. For example, the correlation of each time period with the same cycle type can be sorted to determine the time period with the highest correlation with the cycle type, and the correspondence between the cycle type and the time period with the highest correlation is established. The sample feature of the time period corresponding to the cycle type is the sample feature corresponding to the cycle type. For example, for the first cycle type, the sample feature of the day class has the highest correlation with the first cycle type, and the sample feature of the day class is taken as the sample feature corresponding to the first cycle type. Therefore, for the jth cycle type, the ith time period associated with the jth cycle type can be determined, and the sample feature of the ith time period is the sample feature corresponding to the jth cycle type.

[0076] After determining the correspondence between the cycle type and the time period, the target sample feature corresponding to the product to be recommended can be determined. For example, if the cycle type to which the product to be recommended belongs is the day class, the time period corresponding to the day class can be obtained, and the sample feature in the time period is taken as the target sample feature corresponding to the product to be recommended.

[0077] Step 205: determining a target feature type corresponding to the to-be-recommended object according to the target sample feature. A target time range of the target feature type of the to-be-recommended object can be determined according to the time period of the target sample feature. For example, the time period of the target sample feature can be used as the target time range of the data set or the feature of the to-be-recommended product. For example, the time period of the target sample feature is a day, and the target time range of the to-be-recommended product can be a day. The data of the to-be-recommended product within a day is obtained, and the target user is determined according to the data within a day.

[0078] In an exemplary embodiment, the time range of the feature of each time period can be determined in advance. For example, the time range of the day is 3 days, the time range of the week is 7 days, the time range of the bi-week is 14 days, and the time range of the month is 30 days. Further, the time range of the feature of the to-be-recommended product can be determined according to the time period of the target sample feature.

[0079] In addition, the target feature type can include multiple features. The features included in the target sample feature can be used as the features of the target feature type. For example, the target sample feature includes user gender, user occupation, used traffic, remaining traffic, and the like. The feature set composed of these features is used as the target feature type.

[0080] For example, the target sample feature can be reduced dimensionally, and the target feature type can be determined based on the reduced target sample feature. For example, the target sample feature is reduced dimensionally by a principal component analysis (PCA) method or the like. By reducing the dimension, the global feature or the unimportant feature can be reduced, the dimension of the feature is reduced, the purpose of removing the noise in the feature is achieved, and the resource required for processing the target sample feature can also be reduced.

[0081] Next, step 200: obtaining a first feature of the target time range of the to-be-selected user based on the target feature type.

[0082] The first feature is a feature in the data related to the user. For example, attribute information of the user, historical data of the user logging into the platform, order data of the user ordering a product, and the like. After determining the target feature type of the to-be-recommended object, the features of the to-be-selected users can be obtained according to the feature types included in the target feature type and the target time range. Specifically, according to the target time range in the target feature type, the historical data of each to-be-selected user in the target time range is obtained, and then the corresponding features are extracted from the historical data according to the feature types included in the target feature type, as the first features of the to-be-selected users. For example, the target feature type is daily features: user gender, user occupation, used traffic, remaining traffic, and the like. According to these feature types, “male, programmer, 500MB, 100MB” and the like can be extracted from the historical data of the to-be-selected user in a day as the first features of the to-be-selected user. Similarly, the first features are multi-dimensional features.

[0083] In this embodiment, by calculating the correlation between the data of different time periods and each period type, the time period most related to each period type is determined, so that the data set for feature extraction can be reduced, the calculation amount can be reduced, the calculation resources can be saved, and the calculation efficiency can be improved.

[0084] Step 300: determining the target user of the to-be-recommended object from the to-be-selected users based on the first features.

[0085] According to the obtained first features, it is predicted whether the to-be-selected user orders the to-be-recommended object, and the target user is selected according to the probability obtained by the prediction. For example, the first features are converted into vectors, and the information of the to-be-recommended object is also converted into vectors, the vectors corresponding to the first features and the to-be-recommended object are input into the above correlation prediction model, and the to-be-selected users are selected according to the probability output by the model. For example, the to-be-selected users whose probability output by the model is greater than 0.6 are selected as the target users. In addition, the to-be-selected users can also be selected according to other manners, for example, the Pearson correlation coefficient between the to-be-selected users and the to-be-recommended object is calculated, and the to-be-selected users whose calculation result is greater than a preset value are selected as the target users, and the like.

[0086] The target users are the users most likely to order or click the to-be-recommended product, which are selected from all the to-be-selected users. The to-be-recommended product can be pushed to the selected target users, for example, the information of the to-be-recommended product is displayed in the display interface of the target user, which can improve the success rate of recommendation.

[0087] In this embodiment, the data set is selected by the target feature type, so that all data can be avoided to be processed, and the calculation amount can be saved. Moreover, the features corresponding to the target feature type have the highest correlation with the to-be-recommended object, and the success rate of user mining can be improved.

[0088] In an exemplary embodiment, a feature combination strategy corresponding to each cycle type can be determined in advance, and the target feature type corresponding to the cycle type is combined with other features to determine the target user of the product of the cycle type. Specifically, according to the cycle type of the product to be recommended, a feature combination of the product pair to be recommended is obtained, which includes the target feature type and the non-target feature type corresponding to the product pair to be recommended; based on the non-target feature type, the second feature of the candidate user is obtained within the target time range; and based on the second feature and the first feature, the target user is determined from the candidate user.

[0089] For example, the feature combination can be as shown in Table 1:

[0090] Table 1

[0091]

[0092]

[0093] The cycle type of the product to be recommended can be divided into daily, weekly, biweekly, and monthly. The feature combination strategy of the daily type can be determined in advance as: 80% target feature type, 20% non-target feature type; the feature combination strategy of the weekly type is: 70% target feature type, 30% non-target feature type; the feature combination strategy of the biweekly type is: 65% target feature type, 35% non-target feature type; and the feature combination strategy of the monthly type is: 60% target feature type, 40% non-target feature type. The target feature type of the product to be recommended can be determined according to the cycle type of the product to be recommended. If the cycle type of the product to be recommended is daily, the time range of the corresponding target feature type is 7 days, and the data of the candidate user within 7 days is obtained, and the features are extracted from the data within 7 days. If the target feature type is: daily feature A, daily feature B, and daily feature C, daily feature A, daily feature B, daily feature C, and daily feature D can be extracted from the data of each candidate user within 7 days, and features a and b are randomly extracted. Features a and b can be various features that are not included in the target feature type, i.e., non-target feature type.

[0094] According to the feature combination strategy, the first feature corresponding to the target feature type is combined with the second feature of the non-target feature type as the feature of the candidate user, and then the feature of the candidate user and the feature of the product to be recommended are input into the relevance prediction model to determine the relevance probability between the candidate user and the product to be recommended. A certain number of users with a larger probability are selected from the candidate users as target users.

[0095] For example, the first feature and the second feature can be reduced in dimension according to the PCA method to obtain reduced first features and second features, and the reduced first features and second features are combined as the features of the candidate user. The features of the candidate user obtained after the reduction are input into the correlation prediction model to determine the target user. The reduction can reduce the number of features of the candidate user, thereby improving the prediction efficiency.

[0096] In this embodiment, the target user is screened by adding features of non-target feature types, which can improve the recall rate of the model and improve the accuracy of user mining. Moreover, inputting random non-target features into the model can avoid overfitting of the model and enhance the robustness of the model.

[0097] In the example embodiment, if the object to be recommended belongs to the object to be classified, the object to be classified can be classified according to the correspondence between the classified period types and the time periods, i.e., the period type of the object to be classified is determined. The object to be classified can be a new product that has not been classified in advance. For the new product, the new product can be randomly recommended to some users, and all historical data of the new product can be collected. Period features of different time periods are extracted from the historical data. For example, the historical data is classified according to the time periods of day, week, bi-week, and month to obtain data of each time period, and then features are extracted from the data to obtain period features of each time period. For example, from the data of a day, features such as user gender, user age, and user occupation are extracted as period features of the day type time period. Then, the candidate correlation of each time period of the period features of the new product is determined. For example, the period features of each time period and the features of the new product are input into the correlation prediction model to determine the candidate correlation of each time period in the historical data of the new product. The period type corresponding to the time period with the highest candidate correlation is taken as the period type to which the new product belongs, thereby realizing automatic classification of the new product. Since the data of the new product is less, the new product can also be directly classified manually to determine the period type of the new product.

[0098] The user mining device provided in the embodiments of the present application is described below. The user mining device described below can be correspondingly referred to the user mining method described above.

[0099] Figure 4 The structure diagram of the user mining device provided in the embodiment is shown. As shown in Figure 4As shown, the user mining apparatus 400 can include a time determining module 410 configured to determine a target feature type corresponding to the to-be-recommended object, the target feature type being used to indicate a feature in a target time range; a data extracting module 420 configured to acquire a first feature of the to-be-recommended object in the target time range based on the target feature type; and a data mining module 430 configured to determine a target user of the to-be-recommended object from the candidate users based on the first feature.

[0100] In an example embodiment, the time determining module 410 specifically includes: a sample acquiring module configured to acquire sample features in different time periods based on the different time periods; a type determining module configured to determine a period type to which each candidate object belongs, the candidate objects including the to-be-recommended object; a correlation calculating module configured to determine a correlation between the sample features in each time period and each period type; a sample determining module configured to determine a target sample feature corresponding to the to-be-recommended object based on the correlation and a period type to which the to-be-recommended object belongs; and a feature type determining module configured to determine a target feature type corresponding to the to-be-recommended object according to the target sample feature.

[0101] In an example embodiment, the correlation calculating module specifically includes: a vector converting module configured to convert the sample features in each time period into sample vectors, the sample features being multi-dimensional features; a period correlation calculating module configured to determine a first correlation degree between the sample vectors of each time period and each period type; a dimension correlation calculating module configured to determine a second correlation degree between feature vectors of each dimensional feature in the sample vectors and each period type; and a correlation determining module configured to determine a correlation between the sample features in each time period and each period type by combining the first correlation degree and the second correlation degree.

[0102] In an example embodiment, the user mining apparatus further includes a historical data acquiring module configured to acquire historical data of a to-be-classified object; a period determining module configured to extract period features of different time periods from the historical data; a candidate correlation determining module configured to determine a candidate correlation between the period features and the to-be-classified object; and a classification module configured to determine a period type to which the to-be-classified object belongs based on the candidate correlation.

[0103] In an example embodiment, the feature type determining module is specifically configured to determine a target time range of the to-be-recommended object according to a time period of the target sample feature.

[0104] In the example embodiment, the data mining module 430 comprises a feature combination module configured to obtain a feature combination corresponding to the to-be-recommended object according to a period type of the to-be-recommended object, the feature combination comprising a target feature type and a non-target feature type; a feature extraction module configured to obtain a second feature of the to-be-selected user within the target time range based on the non-target feature type; and a data screening module configured to determine a target user of the to-be-recommended object from the to-be-selected user based on the first feature and the second feature.

[0105] In the example embodiment, the data screening module is configured to reduce dimensions of the first feature and the second feature, and determine the target user of the to-be-recommended object based on the reduced first feature and the second feature.

[0106] It should be noted that the user mining apparatus provided by the embodiments of the present application can implement all the method steps achieved by the method embodiments and achieve the same technical effects. Therefore, the same parts and beneficial effects of the embodiments of the present application as the method embodiments will not be described in detail.

[0107] Figure 5 An example of a schematic diagram of an entity structure of an electronic device is shown in FIG. 5. Figure 5 As shown in FIG. 5, the electronic device can include a processor 510, a communication interface 520, a memory 530, and a communication bus 540. The processor 510, the communication interface 520, and the memory 530 can communicate with each other through the communication bus 540. The processor 510 can invoke a computer program in the memory 530 to execute the steps of the user mining method, for example, comprising: step 100 of determining a target feature type corresponding to a to-be-recommended object, the target feature type being used to indicate a feature within a target time range; step 200 of obtaining a first feature of a to-be-selected user within the target time range based on the target feature type; and step 300 of determining a target user of the to-be-recommended object from the to-be-selected user based on the first feature.

[0108] Further, the logic instructions in the memory 530 described above can be implemented in the form of software functional units and sold or used as independent products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0109] In another aspect, the embodiments of the present application also provide a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program is executed by a processor, so that the computer can perform the steps of the user mining method provided by the above-mentioned embodiments, for example, comprising: step 100: determining a target feature type corresponding to a to-be-recommended object, the target feature type is used to indicate a feature in a target time range; step 200: based on the target feature type, acquiring a first feature of a to-be-selected user in the target time range; step 300: based on the first feature, determining a target user of the to-be-recommended object from the to-be-selected user.

[0110] In another aspect, the embodiments of the present application also provide a processor readable storage medium, which stores a computer program, and the computer program is used to enable a processor to perform the steps of the user mining method provided by the above-mentioned embodiments, for example, comprising: step 100: determining a target feature type corresponding to a to-be-recommended object, the target feature type is used to indicate a feature in a target time range; step 200: based on the target feature type, acquiring a first feature of a to-be-selected user in the target time range; step 300: based on the first feature, determining a target user of the to-be-recommended object from the to-be-selected user.

[0111] The processor readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to a magnetic storage (such as a floppy disk, a hard disk, a magnetic tape, a magneto-optical disk (MO), etc.), an optical storage (such as a CD, a DVD, a BD, a HVD, etc.), and a semiconductor memory (such as a ROM, an EPROM, an EEPROM, a NAND FLASH, a solid state disk (SSD)), etc.

[0112] The apparatus embodiments described above are merely illustrative, wherein the units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0113] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, and the computer software products can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and include a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0114] 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 them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A user mining method, characterized in that, include: Determine the target feature type corresponding to the object to be recommended, wherein the target feature type is used to indicate features within the target time range; Based on the target feature type, obtain the first feature of the candidate user within the target time range; Based on the first feature, the target user of the recommended object is determined from the candidate users; Determining the target feature type corresponding to the object to be recommended includes: Based on different time periods, obtain sample features within each time period; Determine the cycle type to which each candidate object belongs, the candidate objects including the object to be recommended; Determine the correlation between the sample features within each of the time periods and each of the period types; Based on the correlation and the period type to which the object to be recommended belongs, the target sample features corresponding to the object to be recommended are determined; Based on the target sample features, determine the target feature type corresponding to the object to be recommended; Determining the correlation between sample features within each time period and each period type includes: The sample features within each time period are converted into sample vectors, and the sample features are multi-dimensional features. Determine the first correlation between the sample vector of each time period and each period type; Determine the second correlation between the feature vectors of each dimension of the sample vector and the periodicity type; By combining the first correlation and the second correlation, the correlation between sample features and each period type within each time period is determined.

2. The user mining method according to claim 1, characterized in that, The method further includes: Obtain historical data for the object to be classified; Periodic characteristics of different time periods are extracted from the historical data; Determine the candidate correlation between the periodic features and the object to be classified; Based on the candidate correlations, the periodicity type to which the object to be classified belongs is determined.

3. The user mining method according to claim 1, characterized in that, The step of determining the target feature type corresponding to the object to be recommended based on the target sample features includes: Based on the time period of the target sample features, the target time range of the object to be recommended is determined.

4. The user mining method according to claim 1, characterized in that, The step of determining the target user of the recommended object from the candidate users based on the first feature includes: Based on the period type of the object to be recommended, obtain the feature combination corresponding to the object to be recommended, wherein the feature combination includes target feature type and non-target feature type; Based on the non-target feature type, the second feature of the candidate user is obtained within the target time range; Based on the first feature and the second feature, the target user of the recommended object is determined from the candidate users.

5. The user mining method according to claim 4, characterized in that, The step of determining the target user of the recommended object from the candidate users based on the first feature and the second feature includes: The first feature and the second feature are dimensionality reduced, and the target users of the object to be recommended are determined based on the dimensionality-reduced first feature and second feature.

6. A user excavation device, characterized in that, include: The time determination module is used to determine the target feature type corresponding to the object to be recommended, wherein the target feature type is used to indicate the features within the target time range; The data extraction module is used to obtain the first feature of the candidate user within the target time range based on the target feature type; The data mining module is used to determine the target user of the object to be recommended from the candidate users based on the first feature; Determining the target feature type corresponding to the object to be recommended includes: Based on different time periods, obtain sample features within each time period; Determine the cycle type to which each candidate object belongs, the candidate objects including the object to be recommended; Determine the correlation between the sample features within each of the time periods and each of the period types; Based on the correlation and the period type to which the object to be recommended belongs, the target sample features corresponding to the object to be recommended are determined; Based on the target sample features, determine the target feature type corresponding to the object to be recommended; Determining the correlation between sample features within each time period and each period type includes: The sample features within each time period are converted into sample vectors, and the sample features are multi-dimensional features. Determine the first correlation between the sample vector of each time period and each period type; Determine the second correlation between the feature vectors of each dimension of the sample vector and the periodicity type; By combining the first correlation and the second correlation, the correlation between sample features and each period type within each time period is determined.

7. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the user mining method according to any one of claims 1 to 5.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the user mining method according to any one of claims 1 to 5.

Images