Data pushing method and system of online platform based on user response and medium

Abstract

The invention relates to the technical field of online platform data pushing, and particularly discloses a user response-based online platform data pushing method and system and a medium, and the method mainly comprises the steps of obtaining feature information and behavior data of a user; completing the training of the random forest model; obtaining an importance score v of each user feature c; calculating a weight w of each user feature c, obtaining a feature score s, and calculating a response score r of the user; calculating a preset pushing amount P in unit time according to the response score r; and pushing data to the user according to the preset pushing amount P. According to the method, the user characteristics with the highest importance score v are obtained by training the random forest model, then the response scores r of different customers are calculated, and data pushing is implemented according to the response scores r, so that the conversion rate of data pushing of an online platform is effectively improved, and the user experience is improved. And the brand image and the user satisfaction can be improved, and meanwhile, greater long-term value can be brought.

Description

Technical Field

[0001] The present invention relates to the technical field of online platform data push technology, and in particular to an online platform data push method based on user response, an online platform data push system based on user response, and a readable storage medium. Background Art

[0002] Smart data push on online platforms refers to the process of pushing the smart recommendation data generated by the recommendation system to users through online channels such as the Internet. The push data can be commodities, services, content, etc., and the push method can be SMS, email, message push, recommendation list, etc. By pushing smart data on online platforms, insurance companies can obtain the latest data information in a timely manner.

[0003] When pushing insurance data, it is necessary to comprehensively consider the user's personal situation and needs, collect a large amount of customer-related data, including personal information, health status, financial status, family status, etc., as well as their historical behavior data on insurance products, and then conduct data analysis to obtain customers with potential needs, and then make data push decisions and data push. Customers with potential needs often receive data push with larger traffic.

[0004] However, in the process of implementing relevant technical solutions, it was found that at least the following technical problems exist: users with potential needs can receive more push data, but due to individual differences, some users do not or rarely browse or read the pushed data, and these users have a low response rate to the pushed data. Therefore, when online platforms push data, they often push some data to people with a low response rate. When the amount of data pushed is large and the number of pushes is high, more push costs will be wasted, and the conversion rate of pushed data will be low. At the same time, it is easy to cause customers to dislike pushed data, reducing brand image and customer satisfaction. Summary of the invention

[0005] The present invention solves the problem in the prior art that data push to users with low response rates easily leads to high push costs and low conversion rates by providing an online platform data push method, system and medium based on user response. By achieving accurate data push based on user response rates, the conversion rate of data pushed by online platforms is effectively improved, which is beneficial to improving brand image and user satisfaction while bringing greater long-term value.

[0006] In a first aspect, the present invention provides a method for pushing data on an online platform based on user response, which comprises the following steps:

[0007] S1. Obtaining the user's characteristic information and behavior data, the characteristic information includes the user's characteristic c, and the behavior data includes the click behavior on the pushed data;

[0008] S2. Establish a random forest model, use user feature c as input sample, click behavior as output label, and complete the training of the random forest model;

[0009] S3. Obtain the importance score v of each user feature c according to the random forest model;

[0010] S4. Calculate the weight w of each user feature c according to the importance score v, obtain the feature score s, and calculate the user's response score r according to the weight w and the feature score s;

[0011] S5. Calculate the preset push volume P per unit time according to the response score r;

[0012] S6. Use the online platform to push data to users according to the preset push volume P.

[0013] By training the random forest model to obtain the user feature with the highest importance score v, and then calculating the response scores r of different customers, the data is pushed according to the response scores r. This can achieve accurate data push based on the user's response rate, effectively improving the conversion rate of data pushed by the online platform, which is beneficial to improving the brand image and user satisfaction while bringing greater long-term value.

[0014] Furthermore, in S1, the user is a user who has historically received data pushed by the online platform.

[0015] Furthermore, user characteristics c include: gender, age range, occupation and region.

[0016] Furthermore, in S1, it also includes: obtaining a preset feature type set C1, C1 is: C1 = {c i}, where c i Represents the i-th user feature c, user feature c∈feature category set C1.

[0017] Furthermore, taking user feature c as input sample and click behavior as output label, the training of random forest model is completed including:

[0018] S21, build a data set;

[0019] S22, divide the data set into a training set and a test set;

[0020] S23, completing the training of the random forest model according to the training set and the test set;

[0021] The data set includes input column X and output column Y, where X is:i}, Y is: Y = {0,1}, 1 represents a click behavior, and 0 represents no click behavior.

[0022] Furthermore, the response score r includes: 1 ; S4 includes:

[0023] S41, sort the importance score v of each user feature c from high to low, extract the top n user features c with the highest importance score v and construct a key feature sequence C2, C2 is: C2=

[0024] {c1,c2,...,c n}, construct the key feature importance score sequence V, V = {v1,v2,...,v n}, where c n Represents the nth user feature c, v n Represents the importance score v of the nth user feature c; calculates the weight w of each user feature c in the key feature sequence C2 n , w n for: Among them, n max represents the total number of user feature c types, w n represents the weight w of the nth user feature c; obtain the feature score s of the user feature c from the pre-constructed feature score set S1, where S1 is: S1 = {s ij},s ij represents the j-th feature score s of the i-th user feature c;

[0025] S42, calculate the user feature score s and the feature weight w after normalization ′ and w ′ ,s ′ for: w ′ for:

[0026] S43, build a column based on users, ′ The matrix S2 with users as columns and w ′ is the matrix W of rows; S2 is: Among them, s ′ mn represents the nth user feature score s of the mth user ′ ; W is: Among them, w ′ mn represents the weight w of the nth user feature of the mth user ′ ;

[0027] S44, calculate the response score r 1 : Among them, W T represents the transpose of the matrix W, represents the response score of the mth user 1 .

[0028] Furthermore, the behavior data also includes: the average reading time t of each push data; the response score r also includes: the response score r 2 ;

[0029] S45, construct a reading score set D, D is: D = {d k}, where d k represents the reading score d of the kth reading time t; obtain the user's reading score d in the reading score set D, and calculate the response score r 2 , r 2 For: r 2 =r 1 +d.

[0030] Furthermore, P is: P=p+s*r, where p represents the basic push amount, and s represents the influence coefficient of the response score r on the preset push amount P.

[0031] In a second aspect, the present invention further discloses a data push system for an online platform, which uses the data push method for an online platform of the first aspect.

[0032] A data push system for an online platform, comprising:

[0033] A data acquisition module is used to acquire the user's characteristic information and behavior data, wherein the characteristic information includes the user's characteristic c, and the behavior data includes the click behavior on the pushed data;

[0034] The model training module is used to establish a random forest model, taking user feature c as input sample and click behavior as output label to complete the training of the random forest model;

[0035] An importance score acquisition module is used to obtain the importance score v of each user feature c according to the random forest model;

[0036] A response score calculation module is used to calculate the weight w of each user feature c according to the importance score v, obtain the feature score s, and calculate the user's response score r according to the weight w and the feature score s;

[0037] A preset push volume calculation module, which is used to calculate the preset push volume P per unit time according to the response score r;

[0038] The data push module is used to push data to users according to a preset push volume P using an online platform.

[0039] In a third aspect, the present invention further discloses a readable storage medium, in which computer program instructions are stored. When the computer program instructions are read and executed by a processor, the steps of the data push method for an online platform of the first aspect are executed.

[0040] The technical solution provided by the present invention has at least the following technical effects or advantages:

[0041] 1. By training the random forest model to obtain the user features with the highest importance score v, and then calculating the response scores r of different customers and the preset push volume P per unit time, data push is implemented according to the preset push volume P. This effectively solves the problem of high push costs and low conversion rates when pushing data to users with low response rates. It can achieve accurate data push based on the user's response rate, effectively improve the conversion rate of data pushed by the online platform, which is beneficial to improving the brand image and user satisfaction while bringing greater long-term value.

[0042] 2. When calculating the user's response score r, the user's average historical reading time t for each pushed data is also taken into account. For users who have a historical reading record of the pushed data, the longer the average historical reading time t, the higher the response rate, which further improves the accuracy of calculating the preset push volume P. The calculation method of the response score r can be selected based on the user information obtained, which improves the flexibility of the online platform when pushing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of the data push method of the online platform based on user response in the present invention;

[0044] Figure 2 It is a module diagram of the data push system of the online platform based on user response in the present invention. DETAILED DESCRIPTION

[0045] As mentioned in the background technology, when online platforms push data, they often push some data to people with low response rates. When the amount of data pushed is large and the number of pushes is large, more push costs will be wasted, reducing the conversion rate of the pushed data. It is also easy to cause customers to dislike the pushed data, reduce the brand image and customer satisfaction, and thus reduce the push cost. The present application can achieve accurate data push based on the user's response rate, thereby effectively improving the conversion rate of data pushed by the online platform, which is beneficial to improving the brand image and user satisfaction while bringing greater long-term value.

[0046] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0047] Embodiment 1

[0048] Please refer to Figure 1 , a data push method for an online platform based on user response, comprising the following steps:

[0049] S1. Obtaining the user's characteristic information and behavior data, the characteristic information includes the user's characteristic c, and the behavior data includes the click behavior on the pushed data;

[0050] S2. Establish a random forest model, use user feature c as input sample, click behavior as output label, and complete the training of the random forest model;

[0051] S3. Obtain the importance score v of each user feature c according to the random forest model;

[0052] S4. Calculate the weight w of each user feature c according to the importance score v, obtain the feature score s, and calculate the user's response score r according to the weight w and the feature score s;

[0053] S5. Calculate the preset push volume P per unit time according to the response score r;

[0054] S6. Use the online platform to push data to users according to the preset push volume P.

[0055] The following is a detailed description of each step in turn:

[0056] S1. Obtain the user's characteristic information and behavior data, where the characteristic information includes user characteristics c, and the behavior data includes click behavior on pushed data.

[0057] The scheme of this embodiment mainly implements data push based on the user's response. Therefore, the users are all users who have historically received data pushed by the online platform. Different users can be given ID numbers, and the ID numbers can be used as a basis for identifying user feature information and behavior data of different users. Each user has a unique ID number to distinguish users with the same name to avoid confusion and errors.

[0058] The characteristic information is all the attribute characteristics of the user. Since there are many types of attribute characteristics, some main attribute characteristics are selected in this embodiment and these main attribute characteristics are used as user characteristics c. User characteristics c include but are not limited to: gender, age range, occupation and region.

[0059] The behavior data may be the user's click data, browsing data, search data, purchase data, and evaluation data. In this embodiment, the user's click behavior is mainly obtained to determine whether the user has responded to the data push of the online platform.

[0060] In S1, it also includes obtaining a preset feature type set C1, C1 is: C1 = {c i}, where c i Represents the i-th user feature c, user feature c∈feature category set C1.

[0061] The feature category set C1 contains different types of user features c. If a user has two user features c, 28 years old and 2 historical click behaviors, then the two user features c, 28 years old and 2 historical click behaviors, belong to the age category and historical click behavior category in the feature category set C1 respectively. The user features c obtained for each user belong to the feature category set C1. If a certain attribute feature is not in the feature category set C1, the attribute feature is not obtained when obtaining user feature information. For example, if the feature category set C1 includes age and historical click behaviors, the user feature c can only be age or historical click behaviors, and cannot be other features.

[0062] S2. Establish a random forest model, use user feature c as input sample, click behavior as output label, and complete the training of the random forest model.

[0063] Random forest is an ensemble learning method based on decision trees. It makes predictions by constructing multiple decision trees and finally aggregating the prediction results of multiple decision trees to obtain the final prediction results. Random forest mainly generates prediction results through random sampling, random feature selection, decision tree construction and prediction result aggregation. In this embodiment, the random forest model is mainly trained by taking user features c as input samples and click behaviors as output labels, so as to obtain user features c that affect different click behaviors respectively.

[0064] In S2, user feature c is used as the input sample and click behavior is used as the output label to complete the training of the random forest model, including:

[0065] S21. Construct a data set, which includes an input column X and an output column Y. X is: i}, Y is: Y={0,1}, where 1 represents a click behavior and 0 represents no click behavior.

[0066] Before constructing the data set, the user feature c needs to be cleaned and preprocessed, including missing value processing, outlier processing, data balancing processing, etc., and then feature engineering processing is performed on the user feature c, including feature selection, feature extraction, feature transformation, etc., to improve the prediction ability of the model.

[0067] S22, divide the data set into a training set and a test set;

[0068] The cross-validation method can be used to divide the training set and the test set to avoid overfitting and underfitting problems. When using the cross-validation method to divide the training set and the test set, the data set can be randomly divided into A subsets b of similar size. For each subset b, it is used as the test set in turn, and the remaining A-1 subsets are used as training sets. The random forest model is trained A times.

[0069] S23. Complete the training of the random forest model based on the training set and the test set.

[0070] During each training, the prediction error ε is calculated based on the difference between the predicted structure and the test set. a , ε a Represents the error during the ath training, and finally obtains A prediction errors ε a , calculate A prediction errors ε a The average value is used as the performance indicator of the random forest model.

[0071] The random forest model is trained through the training set, and the model is tuned to improve the accuracy and generalization ability of the model. The trained random forest model is then evaluated using the test set, including indicators such as accuracy, recall rate, and F1 value, to evaluate the performance of the model until the model meets the standards.

[0072] S3. Obtain the importance score v of each user feature c according to the random forest model.

[0073] The importance score v of each user feature c is calculated using the average reduced impurity of each decision tree in the random forest model. It is mainly used to determine which user features c have the greatest impact on user click behavior, so as to perform feature selection on user features c, thereby retaining user features c that make important contributions to the model's prediction ability, while removing redundant and useless user features c.

[0074] S4. Calculate the weight w of each user feature c according to the importance score v, obtain the feature score s, and calculate the user's response score r according to the weight w and the feature score s.

[0075] In this embodiment, the response score r includes: 1 ; S4 is mainly used to calculate the response score 1 , S4 comprises the following steps:

[0076] S41, sort the importance score v of each user feature c from high to low, extract the top n user features c with the highest importance score v and construct a key feature sequence C2, C2 is: C2=

[0077] {c1,c2,...,c n}, construct the key feature importance score sequence V, V is: V = {v1,v2,...,v n}, where c n Represents the nth user feature c, v n Represents the importance score v of the nth user feature c; calculates the weight w of each user feature c in the key feature sequence C2 n , w n for: Among them, n max represents the total number of user feature c types, w n represents the weight w of the nth user feature c; obtain the feature score s of the user feature c from the pre-constructed feature score set S1, where S1 is: S1 = {s ij},s ij represents the j-th feature score s of the i-th user feature c;

[0078] When predicting whether a user will click on the pushed data, since different types of user features c have different effects on clicks, we should first obtain the effect of each user feature c on clicks. For different types of user features c, those with greater effects on clicks should have a larger weight w, while those with less effects on clicks should have a smaller weight w.

[0079] For example: based on the two user features c of "historical click behavior" and "region", when predicting click behavior, the user feature c of "historical click behavior" has a greater impact on predicting user click behavior, so it has a larger weight w, while the user feature c of "region" has a smaller impact on predicting user click behavior, so the corresponding weight w is also smaller.

[0080] After calculating the weight w of each user feature c in the key feature sequence C2, the influence of different users on the predicted click behavior in the same user feature c should be considered. For user features c that are conducive to users' click behavior, they should have a larger feature score s, and user features c that are not conducive to users' click behavior should have a smaller corresponding feature score s.

[0081] For example, based on the user feature c of "historical click behavior", if the user's historical click rate is high, then this user feature c is conducive to the user's future click behavior on the pushed data, and if the user's historical click rate is low, it is not conducive to the user's future click behavior on the pushed data. In the user's "historical click behavior" feature, the click rate value can be used as the feature score s. The full score of the feature score s of this user feature c is 1 point. If the user's click rate is 0.3, then the user's feature score s of this user feature c is 0.3 points.

[0082] In this embodiment, if the characteristic score s of the user is divided into L segments and the L segments are sorted according to size, then s ij Represents the j-th feature score s in L segments of the i-th user feature c.

[0083] S42, calculate the user feature score s and the feature weight w after normalization ′ and w ′ ,s ′ for: w ′ for:

[0084] The user feature scores s and the feature weights w are normalized so that the data values ​​of the user feature scores s and the feature weights w can be converted to the same scale to avoid the differences between the user feature scores s and the imbalance of the feature weights w.

[0085] S43, build a column based on users, ′ The matrix S2 with users as columns and w ′ is the matrix W of rows; S2 is: Among them, s ′ mn represents the nth user feature score s of the mth user ′ ; W is: Among them, w ′ mn represents the weight w of the nth user feature of the mth user ′ .

[0086] By constructing the matrix S2 and the matrix W, parameters such as the user feature score s and the feature weight w can be flexibly adjusted to adapt to different business needs and user behavior patterns. At the same time, the model can also be easily expanded and new features can be added.

[0087] S44, calculate the response score r 1 : Among them, W Trepresents the transpose of the matrix W, represents the response score of the mth user 1 .

[0088] In order to obtain the user's response score 1 After that, the response rate of the user to the pushed data can be determined, and the response is divided into r 1 The higher the user, the greater the response rate to the push data.

[0089] S5. Calculate the preset push volume P per unit time according to the response score r.

[0090] After calculating the user response score r, a data recommendation strategy can be designed based on the response score r. For customers with a low response score r, data recommendations can be reduced or not made, thereby saving push costs. In addition, the saved push costs can be spent on users with a high response score r.

[0091] P is: P=p+s*r, p represents the basic push amount, and s represents the influence coefficient of the response score r on the preset push amount P.

[0092] The response score r can be divided into several high and low response segments, and different users can be divided into corresponding response segments. At the same time, different recommendation strategies can be designed for each response segment. For customers in high response segments, more data push can be implemented.

[0093] S6. Use the online platform to push data to users according to the preset push volume P.

[0094] Finally, the online platform can be used to implement data push with different recommendation strategies based on the calculated P. For users with a larger response score r, more data recommendations can be made, and more personalized and accurate content can be recommended. For customers with a smaller response score r, the recommendation volume can be reduced, and more general and popular content can be recommended. Ultimately, the conversion rate of pushed data can be improved, which is beneficial to improving brand image and user satisfaction while bringing greater long-term value.

[0095] Embodiment 2

[0096] This second embodiment provides another method for calculating the response score r. Based on the first embodiment, the response score r also includes: 2 ; Behavioral data also includes: average reading time t of each push data;

[0097] S45, construct a reading score set D, D is: D = {d k}, where d k represents the reading score d of the kth reading time t; obtain the user's reading score d in the reading score set D, and calculate the response score r2 , r 2 For: r 2 =r 1 +d.

[0098] In this embodiment, the user's historical reading records can be obtained to calculate the average historical reading time of each pushed data, and the reading time can be graded. For users with longer reading time, their reading score d is higher, and the reading score set D is constructed based on this.

[0099] In calculating r 1 When the matrix S2 and matrix W are both n-dimensional column vectors, representing the user feature score s and the feature weight w, d is a scalar, representing the reading score, and then the response is divided into two r 2 The calculation formula is a linear regression model, in which the user feature score s and the feature weight w, the reading score d is used as a constant term, and the response is divided into two 2 As the dependent variable. Since the average historical reading time of each pushed data is taken into account, for users who have a history of reading the pushed data, based on their average historical reading time, the longer the average historical reading time, the higher the response rate, and it is suitable to recommend more data to this user.

[0100] Embodiment 3

[0101] Please refer to Figure 2 , this embodiment three discloses a data push system for an online platform, which uses the data push method for an online platform of embodiment one or embodiment two.

[0102] A data push system for an online platform, comprising:

[0103] A data acquisition module is used to acquire the user's characteristic information and behavior data, wherein the characteristic information includes the user's characteristic c, and the behavior data includes the click behavior on the pushed data;

[0104] The model training module is used to establish a random forest model, taking user feature c as input sample and click behavior as output label to complete the training of the random forest model;

[0105] An importance score acquisition module is used to obtain the importance score v of each user feature c according to the random forest model;

[0106] A response score calculation module is used to calculate the weight w of each user feature c according to the importance score v, obtain the feature score s, and calculate the user's response score r according to the weight w and the feature score s;

[0107] A preset push volume calculation module, which is used to calculate the preset push volume P per unit time according to the response score r;

[0108] The data push module is used to push data to users according to a preset push volume P using an online platform.

[0109] Embodiment 4

[0110] The fourth embodiment discloses a readable storage medium, in which computer program instructions are stored. When the computer program instructions are read and executed by a processor, the steps of the data push method for an online platform of the first or second embodiment are executed.

[0111] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps of the functions specified in a box or multiple boxes. Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they know the basic creative concept. Therefore, the attached claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0113] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

[0114] What has been described above are only preferred specific implementations of the embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, can make equivalent replacements or changes based on the technical solutions and concepts of the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A data push method for an online platform based on user response, characterized in that: It includes the following steps: S1. Obtaining the user's characteristic information and behavior data, the characteristic information includes the user's characteristic c, and the behavior data includes the click behavior on the pushed data; S2. Establish a random forest model, use user feature c as input sample, click behavior as output label, and complete the training of the random forest model; S3. Obtain the importance score v of each user feature c according to the random forest model; S4. Calculate the weight w of each user feature c according to the importance score v, obtain the feature score s, and calculate the user's response score r according to the weight w and the feature score s; S5. Calculate the preset push volume P per unit time according to the response score r. S6. Use the online platform to push data to users according to the preset push volume P.

2. The data push method for an online platform according to claim 1, characterized in that: In S1, users are users who have historically received data pushed by online platforms.

3. The data push method based on the online platform of user response according to claim 1, characterized in that: User characteristics c include: gender, age range, occupation and region.

4. The method for pushing data on an online platform based on user response according to claim 1, characterized in that: In S1, it also includes: obtaining a preset feature type set C1, C1 is: C1 = {c i }, where c i Represents the i-th user feature c, user feature c∈feature category set C1.

5. The method for pushing data on an online platform based on user response according to claim 1, characterized in that: Taking user feature c as input sample and click behavior as output label, the training of random forest model includes: S21, build a data set; S22, divide the data set into a training set and a test set; S23, completing the training of the random forest model according to the training set and the test set; The data set includes input column X and output column Y, where X is: i }, Y is: Y = {0,1}, 1 represents a click behavior, and 0 represents no click behavior.

6. The method for pushing data on an online platform based on user response according to claim 1, characterized in that: The response score r includes: 1 ; S4 includes: S41, sort the importance score v of each user feature c from high to low, extract the top n user features c with the highest importance score v and construct a key feature sequence C2, C2 is: C2= {c1,c2,...,c n }, construct the key feature importance score sequence V, V = {v1,v2,...,v n }, where c n Represents the nth user feature c, v n Represents the importance score v of the nth user feature c; calculates the weight w of each user feature c in the key feature sequence C2 n , w n for: Among them, n max represents the total number of user feature c types, w n represents the weight w of the nth user feature c; obtain the feature score s of the user feature c from the pre-constructed feature score set S1, where S1 is: S1 = {s ij },s ij represents the j-th feature score s of the i-th user feature c; S42, calculate the user feature score s and the feature weight w after normalization ′ and w ′ ,s ′ for: w ′ for: S43, build a column based on users, ′ The matrix S2 with users as columns and w ′ is the matrix W of rows; S2 is: Among them, s ′ mn represents the nth user feature score s of the mth user ′ ; W is: Among them, w ′ mn represents the weight w of the nth user feature of the mth user ′ ; S44, calculate the response score r 1 : Among them, W T represents the transpose of matrix W, r m 1 represents the response score of the mth user 1 .

7. The method for pushing data on an online platform based on user response according to claim 6, characterized in that: Behavior data also includes: average reading time t for each push data; response score r also includes: response score r 2 ; S45, construct a reading score set D, D is: D = {d k }, where d k represents the reading score d of the kth reading time t; obtain the user's reading score d in the reading score set D, and calculate the response score r 2 , r 2 For: r 2 =r 1 +d.

8. The method for pushing data on an online platform based on user response according to claim 1, characterized in that: P is: P=p+s*r, p represents the basic push amount, and s represents the influence coefficient of the response score r on the preset push amount P.

9. A data push system for an online platform based on user response, characterized in that: It uses the data push method of an online platform based on user response as described in any one of claims 1 to 8, which includes: A data acquisition module is used to acquire the user's characteristic information and behavior data, wherein the characteristic information includes the user's characteristic c, and the behavior data includes the click behavior on the pushed data; The model training module is used to establish a random forest model, taking user feature c as input sample and click behavior as output label to complete the training of the random forest model; An importance score acquisition module is used to obtain the importance score v of each user feature c according to the random forest model; A response score calculation module is used to calculate the weight w of each user feature c according to the importance score v, obtain the feature score s, and calculate the user's response score r according to the weight w and the feature score s; A preset push volume calculation module, which is used to calculate the preset push volume P per unit time according to the response score r; The data push module is used to push data to users according to a preset push volume P using an online platform.

10. A readable storage medium, characterized in that: The readable storage medium stores computer program instructions, and when the computer program instructions are read and executed by a processor, the steps of the data push method for an online platform as described in any one of claims 1 to 8 are executed.

Images