Recommendation model training method, information recommendation method, device, equipment and medium

Abstract

The present disclosure relates to a training method and device for a recommendation model, an information recommendation method and device, an electronic device and a computer-readable medium, and belongs to the field of machine learning technology. The training method of the recommendation model includes: dividing all terminals into terminal pairs and generating keys; calculating the statistical information of the features after binning the features of each terminal; obtaining the encrypted statistical information of each feature in the terminal and the overall statistical information of each feature according to the feature statistical information and the key of the terminal; using the training sample set as a node in the decision tree model, and determining the target splitting point according to the overall statistical information; after splitting the decision tree according to the target splitting point, recalculating the statistical information of each feature in the training sample set and determining the next target splitting point, until the iteration is stopped when the preset conditions are met, and obtaining the recommendation model of the terminal. The present disclosure can protect user privacy by encrypting the feature statistical information of the terminal during the model training process.

Description

Technical Field

[0001] The present disclosure relates to the technical field of machine learning, and in particular to a training method and device for a recommendation model, an information recommendation method and device, an electronic device, and a computer-readable medium. Background Art

[0002] Personalized recommendations for users are an important means to improve user experience. Currently, personalized recommendation methods are based on user behavior data or social relationship network information.

[0003] In the current user personalized recommendation method, since it involves the acquisition and transmission of sensitive data such as user privacy, it is very likely to cause security issues such as user privacy leakage.

[0004] In view of this, the field urgently needs a recommendation model training method and an information recommendation method that can ensure data security and user privacy.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of the present disclosure is to provide a training method and device for a recommendation model, an information recommendation method and device, an electronic device, and a computer-readable medium, thereby protecting data security and user privacy at least to a certain extent.

[0007] According to a first aspect of the present disclosure, a training method for a recommendation model is provided, comprising:

[0008] Divide all terminals into terminal pairs, and generate keys corresponding to each terminal, wherein the keys of two terminals in the terminal pair are opposite numbers to each other;

[0009] Obtaining user behavior data of each of the terminals, obtaining features in each of the terminals according to the user behavior data, using the features in the terminals as a training sample set corresponding to the terminals, and calculating statistical information of the features after binning the features;

[0010] Obtaining encrypted statistical information of each feature in the terminal according to the statistical information of each feature in the terminal and the key of the terminal, and obtaining overall statistical information corresponding to each feature according to the encrypted statistical information of each feature in all the terminals;

[0011] The training sample set is used as a node in a decision tree model, and a target splitting point of the decision tree model is determined according to the overall statistical information, wherein the target splitting point is a feature used for this splitting and a corresponding feature value;

[0012] After splitting the nodes in the decision tree model according to the target splitting point, recalculate the statistical information of each feature in the training sample set and determine the next target splitting point, and stop iteration when the feature in the terminal meets the preset conditions to obtain the recommendation model of the terminal.

[0013] In an exemplary embodiment of the present disclosure, the generating of keys corresponding to the terminals, wherein the keys of the two terminals in the terminal pair are opposite numbers of each other, includes:

[0014] Generate tuple data corresponding to two terminals in the terminal pair, wherein the tuple data includes positive and negative identifiers of keys of the two terminals in the terminal pair;

[0015] An initial key corresponding to the terminal is generated according to the tuple data corresponding to the terminal, and a key corresponding to the terminal is obtained according to the initial key and the positive and negative identifiers of the key.

[0016] In an exemplary embodiment of the present disclosure, the step of calculating the statistical information of the feature after binning the feature includes:

[0017] Obtaining a feature value of each feature in each of the terminals, and determining a comprehensive maximum value and a comprehensive minimum value of each of the features according to the feature value of each feature in all of the terminals;

[0018] Binning the features according to the comprehensive maximum and comprehensive minimum values ​​of the features to obtain comprehensive binning results of the features;

[0019] The terminal binning results of each of the terminals are obtained according to the comprehensive binning results of each of the features, and the statistical information of the features in the terminal is calculated according to the terminal binning results.

[0020] In an exemplary embodiment of the present disclosure, determining the comprehensive maximum value and the comprehensive minimum value of each feature according to the feature value of each feature in all the terminals includes:

[0021] Acquire a feature value of each feature in each of the terminals, and obtain a feature value pair corresponding to each feature in the terminal according to the feature value;

[0022] According to the feature value pairs corresponding to each of the features in all the terminals, the comprehensive maximum value and the comprehensive minimum value of each of the features are determined.

[0023] In an exemplary embodiment of the present disclosure, obtaining a feature-value pair corresponding to each feature in the terminal according to the feature value includes:

[0024] Obtaining a terminal maximum value and a terminal minimum value of each of the features in the terminal according to the feature value of each of the features in the terminal;

[0025] An interference parameter of the feature is obtained according to the value interval of the feature, and a feature value pair corresponding to the feature is obtained according to the terminal maximum value and the terminal minimum value of the feature and the interference parameter.

[0026] According to a second aspect of the present disclosure, there is provided an information recommendation method, comprising:

[0027] Acquire user behavior data of a terminal, and input the user behavior data of the terminal into a recommendation model of the terminal, wherein the recommendation model is obtained by any of the training methods of the recommendation model described above;

[0028] According to the output result of the recommendation model of the terminal, the recommendation information corresponding to the terminal is obtained.

[0029] According to a third aspect of the present disclosure, a training device for a recommendation model is provided, comprising:

[0030] A terminal key generation module, used to divide all terminals into two pairs of terminals, and generate a key corresponding to each terminal, wherein the keys of the two terminals in the terminal pair are opposite numbers to each other;

[0031] A statistical information determination module, used to obtain user behavior data of each of the terminals, obtain features in each of the terminals according to the user behavior data, use the features in the terminals as a training sample set corresponding to the terminals, and calculate statistical information of the features after binning the features;

[0032] A statistical information encryption module, used to obtain encrypted statistical information of each feature in the terminal according to the statistical information of each feature in the terminal and the key of the terminal, and obtain overall statistical information corresponding to each feature according to the encrypted statistical information of each feature in all the terminals;

[0033] A split node determination module, used to use the training sample set as a node in a decision tree model, and determine a target split point of the decision tree model according to the overall statistical information, wherein the target split point is a feature used for this split and a corresponding feature value;

[0034] The recommendation model iteration module is used to split the nodes in the decision tree model according to the target splitting point, recalculate the statistical information of each feature in the training sample set and determine the next target splitting point, and stop the iteration when the feature in the terminal meets the preset conditions to obtain the recommendation model of the terminal.

[0035] According to a fourth aspect of the present disclosure, there is provided an information recommendation device, comprising:

[0036] A user information acquisition module, used to acquire user behavior data of a terminal, and input the user behavior data of the terminal into a recommendation model of the terminal, wherein the recommendation model is obtained by the training device of the recommendation model;

[0037] The recommendation information determination module is used to obtain the recommendation information corresponding to the terminal according to the output result of the recommendation model of the terminal.

[0038] According to a fifth aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-described recommendation model training methods or information recommendation methods by executing the executable instructions.

[0039] According to a sixth aspect of the present disclosure, a computer-readable medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the training method of the recommendation model or the information recommendation method described in any one of the above is implemented.

[0040] The exemplary embodiments of the present disclosure may have the following beneficial effects:

[0041] In the training method of the recommendation model of the example implementation method of the present disclosure, the corresponding key is generated in the terminal in the form of a terminal pair, and when the statistical information of each terminal is obtained, the statistical information of the terminal is encrypted by the key, and then the recommendation model corresponding to each terminal is trained based on the encrypted statistical information of all terminals. The training method of the recommendation model in the example implementation method of the present disclosure, on the one hand, by processing all data related to user privacy in the terminal, and encrypting the statistical information when obtaining the statistical information of each terminal, the user behavior data is not directly transmitted between the networks, thereby maximizing the protection of user privacy, enhancing the data security in the model training process, and reducing the risk of privacy leakage. At the same time, since the keys of the two terminals in the terminal pair are opposite numbers to each other, the correct statistical results can still be obtained when the statistical information of each terminal is integrated, thereby ensuring the accuracy of modeling without leaking user privacy data; on the other hand, by integrating the multi-party data features of each terminal to build a model, the user preference information is deeply mined, which can improve the accuracy of the recommendation model, thereby improving the user experience.

[0042] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0044] Figure 1 A schematic diagram showing a flow chart of a training method for a recommendation model according to an exemplary embodiment of the present disclosure;

[0045] Figure 2 A schematic diagram of a process of generating keys corresponding to each terminal according to an exemplary embodiment of the present disclosure is shown;

[0046] Figure 3 A schematic diagram showing a process of binning features and calculating statistical information of features according to an exemplary embodiment of the present disclosure;

[0047] Figure 4 A schematic diagram of a process for determining the comprehensive maximum value and the comprehensive minimum value of each feature according to an exemplary embodiment of the present disclosure is shown;

[0048] Figure 5 A schematic diagram showing a flow chart of an information recommendation method according to an exemplary embodiment of the present disclosure;

[0049] Figure 6 A system architecture diagram of a training method for a recommendation model according to a specific embodiment of the present disclosure is shown;

[0050] Figure 7 A block diagram of a training device for a recommendation model according to an exemplary embodiment of the present disclosure is shown;

[0051] Figure 8 A block diagram showing an information recommendation device according to an exemplary embodiment of the present disclosure;

[0052] Fig. 9 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0053] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0054] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0055] Personalized recommendations for users are an important means to improve user experience. The technology of implementing recommendation algorithms based on big data analysis is the current mainstream method, including collaborative filtering, graph-based algorithms, etc., but these methods all use user behavior data or social relationship networks as a basis, which involves the acquisition and transmission of sensitive data such as user privacy. In an environment where personal privacy is increasingly valued and the country is stepping up efforts to protect data security, it may cause major legal issues.

[0056] In some related embodiments, information recommendation technology needs to collect user behavior data at the terminal, send it to the cloud server for data aggregation, and perform modeling. Therefore, user privacy infringement and leakage problems may occur during data transmission, and data security cannot be guaranteed.

[0057] Based on the above problems, this example implementation first provides a training method for a recommendation model. Figure 1 As shown, the training method of the above recommendation model may include the following steps:

[0058] Step S110: Divide all terminals into terminal pairs, and generate a key corresponding to each terminal, wherein the keys of two terminals in a terminal pair are opposite numbers.

[0059] Step S120: Obtain user behavior data of each terminal, obtain features in each terminal according to the user behavior data, use the features in the terminal as a training sample set corresponding to the terminal, and calculate statistical information of the features after binning the features.

[0060] Step S130. Obtain encrypted statistical information of each feature in the terminal according to the statistical information of each feature in the terminal and the key of the terminal, and obtain overall statistical information corresponding to each feature according to the encrypted statistical information of each feature in all terminals.

[0061] Step S140: Use the training sample set as a node in the decision tree model, and determine the target splitting point of the decision tree model according to the overall statistical information, wherein the target splitting point is the feature used for this splitting and the corresponding feature value.

[0062] Step S150. After splitting the nodes in the decision tree model according to the target splitting point, recalculate the statistical information of each feature in the training sample set and determine the next target splitting point, and stop iterating when the features in the terminal meet the preset conditions to obtain the recommendation model of the terminal.

[0063] In the training method of the recommendation model of the example implementation method of the present disclosure, the corresponding key is generated in the terminal in the form of a terminal pair, and when the statistical information of each terminal is obtained, the statistical information of the terminal is encrypted by the key, and then the recommendation model corresponding to each terminal is trained based on the encrypted statistical information of all terminals. The training method of the recommendation model in the example implementation method of the present disclosure, on the one hand, by processing all data related to user privacy in the terminal, and encrypting the statistical information when obtaining the statistical information of each terminal, the user behavior data is not directly transmitted between the networks, thereby maximizing the protection of user privacy, enhancing the data security in the model training process, and reducing the risk of privacy leakage. At the same time, since the keys of the two terminals in the terminal pair are opposite numbers to each other, the correct statistical results can still be obtained when the statistical information of each terminal is integrated, thereby ensuring the accuracy of modeling without leaking user privacy data; on the other hand, by integrating the multi-party data features of each terminal to build a model, the user preference information is deeply mined, which can improve the accuracy of the recommendation model, thereby improving the user experience.

[0064] The training method of the recommendation model of the example implementation mode of the present disclosure is a universal method for encrypted transmission and multi-party modeling of user data, which can be used in scenarios such as joint modeling of multiple enterprises, and can also be applied to modeling of various enterprise value-added services that need to protect user privacy, such as video recommendation and user personalized analysis services, etc., which can improve the accuracy of modeling and increase the operating income of the enterprise.

[0065] Next, combine Figures 2 to 4 The above steps of this exemplary embodiment are described in more detail.

[0066] In step S110, all terminals are divided into terminal pairs, and keys corresponding to each terminal are generated, wherein the keys of two terminals in a terminal pair are opposite numbers.

[0067] In this example implementation, the terminal may refer to various electronic devices having a processor, including but not limited to a smart phone, a tablet computer, a portable computer, and a set-top box of a television.

[0068] All terminals involved in modeling can be randomly divided into pairs of terminals, and corresponding keys are generated at each terminal, wherein the keys of the two terminals in a terminal pair are opposite numbers to each other.

[0069] In this example implementation, Figure 2 As shown, generating the key corresponding to each terminal may include the following steps:

[0070] Step S210: Generate tuple data corresponding to two terminals in the terminal pair, where the tuple data includes the positive and negative identifiers of the keys of the two terminals in the terminal pair.

[0071] In this example implementation, corresponding tuple data may be generated for each pair of terminals by the server, such as a tuple (p, q, f), where p is a large prime number and may be set to 128 bits or 256 bits, etc., according to the strength of the password. q is the primitive root of p, and f is the positive and negative identifier of the key of the terminal. After the tuple data corresponding to the two terminals in each terminal pair is generated, it is sent to each terminal to generate the corresponding key.

[0072] Step S220: Generate an initial key corresponding to the terminal according to the tuple data corresponding to the terminal, and obtain the key corresponding to the terminal according to the initial key and the positive and negative identifiers of the key.

[0073] After receiving the corresponding tuple data, each terminal can generate the corresponding key according to the preset key generation algorithm, such as the DH (Diffie-Hellman key exchange protocol) algorithm, and take the positive and negative values ​​of the key according to the positive and negative identifiers of the key in the tuple data to prepare for the encryption of the statistical data in the next stage.

[0074] In step S120, user behavior data of each terminal is obtained, features in each terminal are obtained according to the user behavior data, the features in the terminal are used as a training sample set corresponding to the terminal, and statistical information of the features is calculated after the features are binned.

[0075] In this example implementation, the features in each terminal can be obtained by collecting user behavior data at each terminal and preprocessing the user behavior data, such as data cleaning and data normalization. The features in the terminal are then used as the training sample set corresponding to the terminal, and statistical information of each feature is calculated based on the features in each terminal.

[0076] In this example implementation, Figure 3 As shown in the figure, after binning the features, the statistical information of the features is calculated, which can specifically include the following steps:

[0077] Step S310: Obtain the feature value of each feature in each terminal, and determine the comprehensive maximum value and the comprehensive minimum value of each feature according to the feature value of each feature in all terminals.

[0078] First, the value of each feature in each terminal is obtained, and then the comprehensive maximum value and comprehensive minimum value of each feature after integration of all terminals are obtained according to the value of each feature in each terminal.

[0079] In this example implementation, Figure 4As shown, according to the characteristic value of each characteristic in all terminals, the comprehensive maximum value and the comprehensive minimum value of each characteristic are determined, which may specifically include the following steps:

[0080] Step S410: Acquire the feature value of each feature in each terminal, and obtain the feature value pair corresponding to each feature in the terminal according to the feature value.

[0081] In this example implementation, the terminal maximum value and terminal minimum value of each feature in the terminal can be obtained based on the feature value of each feature in the terminal, and then the interference parameter of the feature can be obtained based on the value range of the feature, and the feature value pair corresponding to the feature can be obtained based on the terminal maximum value and terminal minimum value of the feature and the interference parameter.

[0082] Specifically, for all values ​​of each feature in the terminal, calculate its terminal maximum value max and terminal minimum value min, then randomly generate two interference parameters r1 and r2, and obtain the feature value pair (min-r1, max+r2) corresponding to the feature according to the terminal maximum value max and terminal minimum value min, as well as the interference parameters r1 and r2. The value of the interference parameter can be determined according to the value interval between the terminal maximum value max and the terminal minimum value min of the feature. For example, if the value range of a feature is [1, 10], the interference parameter can be about 10% of its value interval 9, for example, the value of the interference parameter can be 0.9 or 1. In this way, the terminal feature value can be scrambled by the interference parameter, which can protect data security during data transmission.

[0083] Step S420: Determine the comprehensive maximum value and the comprehensive minimum value of each feature according to the feature value pairs corresponding to each feature in all terminals.

[0084] After each terminal generates its own corresponding feature value pair, the feature value pairs corresponding to each feature in all terminals can be aggregated to obtain the comprehensive maximum value and comprehensive minimum value of each feature. Specifically, a minimum value can be obtained from all terminal minimum values ​​as the comprehensive minimum value of the feature, and a maximum value can be obtained from all terminal maximum values ​​as the comprehensive maximum value of the feature.

[0085] Step S320: bin the features according to the comprehensive maximum and comprehensive minimum values ​​of each feature to obtain the comprehensive binning results of each feature.

[0086] After obtaining the comprehensive maximum and minimum values ​​of each feature of all terminals, the features are binned according to the comprehensive maximum and minimum values ​​of each feature to obtain the comprehensive binning results of each feature. For example, if the comprehensive value range of a feature is [1, 20], the feature can be binned according to the four intervals of [1, 5], [5, 10], [10, 15], and [15, 20].

[0087] Step S330: Obtain the terminal binning result of each terminal according to the comprehensive binning result of each feature, and calculate the statistical information of the features in the terminal according to the terminal binning result.

[0088] After determining the comprehensive binning results of each feature, the comprehensive binning results of each feature are sent to each terminal, and then each terminal bins the local features of the terminal according to each interval segment in the comprehensive binning results of the feature. After binning its local features, each terminal uses the local features to calculate the statistical information of each feature. The statistical information may include statistical gradients, second-order derivatives, and histograms of quantities.

[0089] In step S130, encrypted statistical information of each feature in the terminal is obtained according to the statistical information of each feature in the terminal and the key of the terminal, and overall statistical information corresponding to each feature is obtained according to the encrypted statistical information of each feature in all terminals.

[0090] In this example implementation, after each terminal obtains the statistical information of each feature in the terminal, the statistical information of each feature in the terminal, such as a histogram, is added to the key of the terminal to obtain the encrypted statistical information of each feature in the terminal. The server obtains the encrypted statistical information of each feature in all terminals and accumulates them to obtain the overall statistical information corresponding to each feature. Among them, since the keys of the two terminals in each pair of terminals are opposite numbers to each other, when the encrypted statistical information of each terminal is accumulated, the keys of the two terminals in each pair of terminals can cancel each other after addition, so the overall statistical information corresponding to each feature obtained in the end is accurate, and the security of the data is also guaranteed during the transmission of the statistical information.

[0091] In step S140, the training sample set is used as a node in the decision tree model, and the target splitting point of the decision tree model is determined according to the overall statistical information, wherein the target splitting point is the feature used for this splitting and the corresponding feature value.

[0092] In this example implementation, a decision tree model can be used as the trained recommendation model. When performing model training, the training sample set can be used as a node in the decision tree model corresponding to the terminal, and then the target splitting point of the decision tree model in this round of iteration is determined based on the overall statistical information corresponding to each feature. Among them, the target splitting point is the best splitting point in this round of iteration, which is the feature used for this split and the corresponding feature value. Specifically, in each split, a suitable target feature is determined from the training sample set, and then a target feature value is determined within the value range of this feature. According to this target feature and the corresponding target feature value, the training sample set in the node is divided into two sub-nodes, and then the next round of iteration is performed.

[0093] In step S150, after splitting the nodes in the decision tree model according to the target splitting point, the statistical information of each feature in the training sample set is recalculated and the next target splitting point is determined. The iteration is stopped until the features in the terminal meet the preset conditions to obtain the recommendation model of the terminal.

[0094] During each iteration, each terminal splits the nodes in the decision tree model according to the target splitting point, returns to the previous step, recalculates the statistical information of each feature in the training sample set according to the split features, and determines the target splitting point of the next round again, until the features in the terminal meet the preset conditions, stops the iteration, completes the construction of the decision tree, and finally obtains the recommendation model of the terminal according to the decision tree model at the end of the iteration. The decision tree model obtained after the training is completed has nodes that only contain the selected features and the corresponding feature values. Among them, the preset conditions for the end of the iteration can be determined based on parameters such as the information entropy or information gain of the features, which are not specifically limited in this example implementation.

[0095] The training method of the recommendation model of the exemplary implementation method of the present disclosure provides a method for combining the behavior data of all users to conduct distributed joint modeling, under the premise that the user behavior data does not leave the terminal. The user privacy data is not transmitted between the terminal and the cloud server, and only the statistical information required for the modeling process is transmitted, and the statistical information is also encrypted, so that the cloud server cannot restore the user related information. Therefore, it can protect the user privacy while improving the user experience.

[0096] The training process in the training method of the recommendation model of the exemplary embodiment of the present disclosure can be set to be executed once a month or longer, or to start updating at any time as needed. It can also be set to run during off-peak hours to reduce the impact on system efficiency.

[0097] In addition, this exemplary embodiment also provides an information recommendation method. Figure 5 As shown, the above information recommendation method may include the following steps:

[0098] Step S510: Acquire user behavior data of the terminal, and input the user behavior data of the terminal into the recommendation model of the terminal.

[0099] Among them, the recommendation model is as mentioned above Figures 1 to 4 The training method of the recommended model in is obtained.

[0100] Step S520: Obtain recommendation information corresponding to the terminal according to the output result of the recommendation model of the terminal.

[0101] The information recommendation method in this exemplary embodiment uses the above Figures 1 to 4 The recommendation model is obtained by the training method of the recommendation model in , and the recommendation information corresponding to the terminal is obtained according to the output result of the recommendation model of the terminal, so as to realize personalized recommendation of network videos, etc.

[0102] like Figure 6 The system architecture diagram of the training method of the recommendation model in a specific embodiment of the present disclosure is shown, which is an example of the above steps in this example embodiment. The system architecture diagram specifically includes the following parts:

[0103] The terminal module 601 is deployed on the client and is used to perform local model training and upload necessary parameters for joint modeling, such as statistical information.

[0104] The key coordination module 602 is responsible for determining terminal pairs and serving as a channel for key negotiation between terminal pairs.

[0105] The model coordination module 603 is responsible for collecting and integrating the modeling parameters uploaded by each terminal module, and updating the parameters and then sending them to each terminal module.

[0106] The training method of the recommendation model in the example implementation of the present disclosure can be implemented in the form of functional expansion through the above-mentioned algorithm module, and the data modeling and analysis process in the data operation system is expanded. Among them, the key collaboration module 602 and the model collaboration module 603 are newly added modules, which will not have any impact on the existing production process and can be implemented by calling the corresponding algorithm processing module. The user behavior data contained in the terminal module 601 is the user's private data, which is only processed in the terminal and not transmitted between networks, thereby protecting the user's privacy.

[0107] It should be noted that although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0108] Furthermore, the present disclosure also provides a training device for a recommendation model. Figure 7 As shown, the training device of the recommendation model may include a terminal key generation module 710, a statistical information determination module 720, a statistical information encryption module 730, a split node determination module 740 and a recommendation model iteration module 750. Among them:

[0109] The terminal key generation module 710 may be used to divide all terminals into two pairs of terminals, and generate keys corresponding to each terminal, wherein the keys of the two terminals in the terminal pair are opposite numbers to each other;

[0110] The statistical information determination module 720 may be used to obtain user behavior data of each terminal, obtain features in each terminal according to the user behavior data, use the features in the terminal as a training sample set corresponding to the terminal, and calculate statistical information of the features after binning the features;

[0111] The statistical information encryption module 730 may be used to obtain encrypted statistical information of each feature in the terminal according to the statistical information of each feature in the terminal and the key of the terminal, and obtain overall statistical information corresponding to each feature according to the encrypted statistical information of each feature in all terminals;

[0112] The split node determination module 740 can be used to use the training sample set as a node in the decision tree model, and determine the target split point of the decision tree model according to the overall statistical information, wherein the target split point is the feature used for this split and the corresponding feature value;

[0113] The recommendation model iteration module 750 can be used to split the nodes in the decision tree model according to the target splitting point, recalculate the statistical information of each feature in the training sample set and determine the next target splitting point, and stop the iteration when the features in the terminal meet the preset conditions to obtain the recommendation model of the terminal.

[0114] In some exemplary embodiments of the present disclosure, the terminal key generation module 710 may include a tuple data generation unit and a terminal key generation unit.

[0115] The tuple data generating unit may be used to generate tuple data corresponding to two terminals in the terminal pair, wherein the tuple data includes positive and negative key identifiers of the two terminals in the terminal pair;

[0116] The terminal key generation unit can be used to generate an initial key corresponding to the terminal according to the tuple data corresponding to the terminal, and obtain the key corresponding to the terminal according to the initial key and the positive and negative identifiers of the key.

[0117] In some exemplary embodiments of the present disclosure, the statistical information determination module 720 may include a comprehensive maximum value determination unit, a comprehensive binning result determination unit, and a statistical information determination unit.

[0118] The comprehensive maximum value determination unit may be used to obtain the characteristic value of each characteristic in each terminal, and determine the comprehensive maximum value and the comprehensive minimum value of each characteristic according to the characteristic value of each characteristic in all terminals;

[0119] The comprehensive binning result determination unit can be used to bin the features according to the comprehensive maximum value and the comprehensive minimum value of each feature to obtain the comprehensive binning result of each feature;

[0120] The statistical information determination unit may be configured to obtain a terminal binning result for each terminal according to the comprehensive binning result for each feature, and calculate statistical information of the features in the terminal according to the terminal binning result.

[0121] In some exemplary embodiments of the present disclosure, the comprehensive maximum value determining unit may include a feature value pair determining unit and a comprehensive maximum value calculating unit.

[0122] The feature value pair determination unit may be used to obtain the feature value of each feature in each terminal, and obtain the feature value pair corresponding to each feature in the terminal according to the feature value;

[0123] The comprehensive maximum value calculation unit can be used to determine the comprehensive maximum value and the comprehensive minimum value of each feature according to the feature value pairs corresponding to each feature in all terminals.

[0124] In some exemplary embodiments of the present disclosure, the feature value pair determination unit may include a terminal maximum value determination unit and an interference parameter determination unit.

[0125] The terminal maximum value determination unit may be used to obtain a terminal maximum value and a terminal minimum value of each feature in the terminal according to the feature value of each feature in the terminal;

[0126] The interference parameter determination unit can be used to obtain the interference parameter of the feature according to the value interval of the feature, and obtain the feature value pair corresponding to the feature according to the terminal maximum value and terminal minimum value of the feature and the interference parameter.

[0127] Furthermore, the present disclosure also provides an information recommendation device. Figure 8 As shown, the information recommendation device may include a user information acquisition module 810 and a recommended information determination module 820. Among them:

[0128] The user information acquisition module 810 may be used to acquire user behavior data of the terminal and input the user behavior data of the terminal into a recommendation model of the terminal, wherein the recommendation model is obtained by the training device of the recommendation model;

[0129] The recommendation information determination module 820 may be configured to obtain recommendation information corresponding to the terminal according to the output result of the recommendation model of the terminal.

[0130] The specific details of each module / unit in the training device of the recommendation model and the information recommendation device have been described in detail in the corresponding method embodiment part, and will not be repeated here.

[0131] Fig. 9 A schematic diagram of the structure of a computer system suitable for implementing an electronic device of an embodiment of the present invention is shown.

[0132] It should be noted that Fig. 9 The computer system 900 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0133] like Fig. 9 As shown, the computer system 900 includes a central processing unit (CPU) 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 into a random access memory (RAM) 903. In the RAM 903, various programs and data required for system operation are also stored. The CPU 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0134] The following components are connected to the I / O interface 905: an input section 906 including a keyboard, a mouse, etc.; an output section 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, a modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read therefrom is installed into the storage section 908 as needed.

[0135] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication section 909, and / or installed from a removable medium 911. When the computer program is executed by a central processing unit (CPU) 901, various functions defined in the system of the present application are executed.

[0136] It should be noted that the computer-readable medium shown in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0137] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0138] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the above embodiment.

[0139] It should be noted that, although several modules of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules described above can be concretized in one module. Conversely, the features and functions of one module described above can be further divided into multiple modules to be concretized.

[0140] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modifications, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure.

[0141] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A training method for a recommendation model, characterized in that: include: Divide all terminals into terminal pairs, and generate keys corresponding to each terminal, wherein the keys of two terminals in the terminal pair are opposite numbers to each other; Obtaining user behavior data of each of the terminals, obtaining features in each of the terminals according to the user behavior data, using the features in the terminals as a training sample set corresponding to the terminals, and calculating statistical information of the features after binning the features; Obtaining encrypted statistical information of each feature in the terminal according to the statistical information of each feature in the terminal and the key of the terminal, and obtaining overall statistical information corresponding to each feature according to the encrypted statistical information of each feature in all the terminals; The training sample set is used as a node in a decision tree model, and a target splitting point of the decision tree model is determined according to the overall statistical information, wherein the target splitting point is a feature used for this splitting and a corresponding feature value; After splitting the nodes in the decision tree model according to the target splitting point, recalculate the statistical information of each feature in the training sample set and determine the next target splitting point, and stop iteration when the feature in the terminal meets the preset conditions to obtain the recommendation model of the terminal.

2. The training method of the recommendation model according to claim 1, characterized in that: The generating of keys corresponding to the terminals, wherein the keys of the two terminals in the terminal pair are opposite numbers of each other, comprises: Generate tuple data corresponding to two terminals in the terminal pair, wherein the tuple data includes positive and negative identifiers of keys of the two terminals in the terminal pair; An initial key corresponding to the terminal is generated according to the tuple data corresponding to the terminal, and a key corresponding to the terminal is obtained according to the initial key and the positive and negative identifiers of the key.

3. The training method of the recommendation model according to claim 1, characterized in that: The step of calculating the statistical information of the feature after binning the feature includes: Obtaining a feature value of each feature in each of the terminals, and determining a comprehensive maximum value and a comprehensive minimum value of each of the features according to the feature value of each feature in all of the terminals; Binning the features according to the comprehensive maximum and comprehensive minimum values ​​of the features to obtain comprehensive binning results of the features; The terminal binning results of each of the terminals are obtained according to the comprehensive binning results of each of the features, and the statistical information of the features in the terminal is calculated according to the terminal binning results.

4. The training method of the recommendation model according to claim 3, characterized in that: The step of determining the comprehensive maximum value and the comprehensive minimum value of each feature according to the feature value of each feature in all the terminals includes: Acquire a feature value of each feature in each of the terminals, and obtain a feature value pair corresponding to each feature in the terminal according to the feature value; According to the feature value pairs corresponding to each of the features in all the terminals, the comprehensive maximum value and the comprehensive minimum value of each of the features are determined.

5. The training method of the recommendation model according to claim 4, characterized in that: The obtaining, according to the feature value, a feature value pair corresponding to each feature in the terminal includes: Obtaining a terminal maximum value and a terminal minimum value of each of the features in the terminal according to the feature value of each of the features in the terminal; An interference parameter of the feature is obtained according to the value interval of the feature, and a feature value pair corresponding to the feature is obtained according to the terminal maximum value and the terminal minimum value of the feature and the interference parameter.

6. An information recommendation method, characterized in that: include: Acquire user behavior data of a terminal, and input the user behavior data of the terminal into a recommendation model of the terminal, wherein the recommendation model is obtained by the training method of the recommendation model according to any one of claims 1 to 5; According to the output result of the recommendation model of the terminal, the recommendation information corresponding to the terminal is obtained.

7. A training device for a recommendation model, characterized in that: include: A terminal key generation module, used to divide all terminals into two pairs of terminals, and generate a key corresponding to each terminal, wherein the keys of the two terminals in the terminal pair are opposite numbers to each other; A statistical information determination module, used to obtain user behavior data of each of the terminals, obtain features in each of the terminals according to the user behavior data, use the features in the terminals as a training sample set corresponding to the terminals, and calculate statistical information of the features after binning the features; A statistical information encryption module, used to obtain encrypted statistical information of each feature in the terminal according to the statistical information of each feature in the terminal and the key of the terminal, and obtain overall statistical information corresponding to each feature according to the encrypted statistical information of each feature in all the terminals; A split node determination module, used to use the training sample set as a node in a decision tree model, and determine a target split point of the decision tree model according to the overall statistical information, wherein the target split point is a feature used for this split and a corresponding feature value; The recommendation model iteration module is used to split the nodes in the decision tree model according to the target splitting point, recalculate the statistical information of each feature in the training sample set and determine the next target splitting point, and stop the iteration when the feature in the terminal meets the preset conditions to obtain the recommendation model of the terminal.

8. An information recommendation device, characterized in that: include: A user information acquisition module, used to acquire user behavior data of a terminal, and input the user behavior data of the terminal into a recommendation model of the terminal, wherein the recommendation model is obtained by the training device of the recommendation model as claimed in claim 7; The recommendation information determination module is used to obtain the recommendation information corresponding to the terminal according to the output result of the recommendation model of the terminal.

9. An electronic device, characterized in that: include: processor; as well as A memory for storing one or more programs, which, when executed by the processor, enables the processor to implement the training method of the recommendation model described in any one of claims 1 to 5 or the information recommendation method described in claim 6.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the training method of the recommendation model according to any one of claims 1 to 5 or the information recommendation method according to claim 6 is implemented.

Images