Method, device and electronic equipment for training a recommendation model

Abstract

Embodiments of the present disclosure provide a method, apparatus and electronic device for training a recommendation model. The method can include, in response to receiving a data deletion request of a user, determining data requested to be deleted in original data. The method can also include obtaining remaining data for the user based on the data requested to be deleted. Furthermore, the method can further include training the recommendation model using the remaining data. Through the technical solutions of the embodiments of the present disclosure, the long model training time and large amount of computing overhead caused by retraining the recommendation model can be avoided, and the model updating time is shortened. In addition, since the training of the recommendation model forgets the data expected to be deleted by the user, the performance of the recommendation model is improved, the recommendation result is more likely to hit the user's preference, and thus the user experience is improved.

Description

Technical Field

[0001] Embodiments of this disclosure relate to the field of data processing, and more specifically, to methods, apparatus, and electronic devices for training recommendation models. Background Technology

[0002] Recommendation models rely on machine learning techniques such as deep neural networks to simulate the complex interactions between users and items of interest (e.g., products, videos, movies, news stories), thereby predicting what users are interested in. However, recommendation models may need to or wish to intentionally forget some training data. When there is a need to remove some of the user's historical behavioral data from the recommendation system, traditional recommendation systems typically retrain the model based on the remaining data after deletion, resulting in significant computational overhead and long model training times. Summary of the Invention

[0003] Embodiments of this disclosure provide methods, apparatus, and electronic devices for training recommendation models.

[0004] In a first aspect, embodiments of this disclosure provide a method for training a recommendation model. The method may include, in response to receiving a user's data deletion request, determining the data in the original data to be deleted. The method may further include, based on the requested deleted data, obtaining remaining data for the user. Furthermore, the method may further include, using the remaining data, training the recommendation model.

[0005] Secondly, embodiments of this disclosure provide an apparatus for training a recommendation model. The apparatus may include: a data deletion determination module configured to determine, in response to receiving a user's data deletion request, the data to be deleted from the original data; a remaining data acquisition module configured to acquire remaining data for the user based on the data to be deleted; and a recommendation model training module configured to train the recommendation model using the remaining data.

[0006] Thirdly, embodiments of this disclosure provide an electronic device, including: a processor; and a memory coupled to the processor, the memory having instructions stored therein, the instructions causing the electronic device to perform actions when executed by the processor, the actions including: in response to receiving a user's data deletion request, determining the data to be deleted in the original data; obtaining remaining data for the user based on the data to be deleted; and training the recommendation model using the remaining data.

[0007] Fourthly, embodiments of this disclosure provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any step of the method according to the first aspect.

[0008] The technical solution of this disclosure avoids the long model training time and large computational overhead caused by retraining the recommendation model, thus shortening the model update time. Furthermore, since the recommendation model's training forgets the data that users expect to be deleted, the model's performance is improved, and the recommendation results are more likely to match user preferences, thereby enhancing the user experience.

[0009] This content section is provided for the purpose of presenting a simplified form of the chosen concepts, which will be further described in the detailed embodiments below. This content section is not intended to identify key or major features of this disclosure, nor is it intended to limit the scope of this disclosure. Attached Figure Description

[0010] The above and other objects, features, and advantages of this disclosure will become more apparent from the accompanying drawings, in which the same or similar reference numerals generally represent the same or similar parts. In the drawings:

[0011] Figure 1 A schematic diagram of an example environment in which several embodiments of the present disclosure can be implemented is shown;

[0012] Figure 2 A schematic diagram of a detailed example environment for training and applying a model according to embodiments of the present disclosure is shown;

[0013] Figure 3 A flowchart illustrating a process for training a recommendation model according to an embodiment of the present disclosure is shown;

[0014] Figure 4 A schematic diagram of raw data of multiple users maintained on the server side according to an embodiment of the present disclosure is shown;

[0015] Figure 5 A flowchart illustrating a process for training a recommendation model according to an embodiment of the present disclosure is shown;

[0016] Figure 6 A schematic diagram of an apparatus for training a recommendation model according to an embodiment of the present disclosure is shown; and

[0017] Figure 7 A schematic block diagram of an example device that can be used to implement embodiments of the present disclosure is shown. Detailed Implementation

[0018] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0019] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the software or hardware, such as the electronic device, application, server, or storage medium performing the operations of this disclosed technical solution, based on the prompt message.

[0020] As an optional but non-limiting implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0021] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0022] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0023] The principles of this disclosure will now be described with reference to several exemplary embodiments shown in the accompanying drawings.

[0024] In the description of embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0025] In the description of embodiments of this disclosure, the term "raw data" can refer to historical data related to the user stored on the server side, including but not limited to historical behavioral data generated by the user when using the application (APP) corresponding to the server. Furthermore, the term "feature data" generally refers to features extracted from the data by the machine learning module.

[0026] As described above, with the continuous development of computer technology, machine learning technology has been widely applied to all aspects of people's lives. To better perform recommendation tasks, the training process of traditional recommendation models needs to be optimized. In traditional recommendation systems, there is often a need for controlled forgetting of user data maintained in the system, motivated by two aspects. First, recommendation models may leak historical behavioral information or other information generated by users while browsing web pages or using apps. Therefore, users expect to delete at least some of this information to mitigate this privacy risk. Second, the performance of recommendation models may rapidly degrade due to noise during training. For example, training data may contain outdated instances, outliers, or instances contaminated by virus attacks. Therefore, users expect to remove the influence of this data from the trained model to improve the recommendation system experience.

[0027] Therefore, traditional recommender systems typically retrain the recommendation model based on the remaining data after deletion. However, this approach is very time-consuming and incurs significant computational overhead, making it impractical for large-scale real-world recommender systems.

[0028] In addition, another traditional recommendation approach, RecEraser, attempts to divide the training data into disjoint subsets during the initial training process. After data sharding, a sub-model can be trained for each shard, and the parameters of the overall model can be determined based on these sub-models. When data needs to be deleted, the RecEraser system can determine the parameters of the overall model without considering the shard containing the data to be deleted. The drawback of this system is that when only a single or small amount of data needs to be deleted, the recommendation accuracy of the RecEraser system decreases. Furthermore, when deleting data that did not arrive in sequence, the RecEraser system's accurate algorithm may be forced to degrade to retraining based on all remaining data, as described above. Therefore, the RecEraser system is not widely applicable to various application scenarios of recommendation systems.

[0029] According to embodiments of this disclosure, a recommendation model training scheme based on user deletion requests is proposed. Upon receiving an instruction or request from one or more users to delete private or noisy data, this scheme first uses the remaining data of those users (i.e., data other than the data requested for deletion) to train or update the recommendation model. Therefore, this scheme avoids the long model training time and significant computational overhead caused by retraining the recommendation model, significantly shortens the model update time, and can provide timely updated recommendation results to users requesting data deletion, thereby solving the aforementioned problems and / or other potential problems.

[0030] The embodiments of this disclosure will be described in detail below with reference to example scenarios. It should be understood that this is for illustrative purposes only and is not intended to limit the scope of this disclosure in any way.

[0031] Figure 1 A block diagram of an example system 100 for training a recommendation model according to an embodiment of the present disclosure is shown. It should be understood that... Figure 1 The system 100 shown is merely one example of an embodiment that can be implemented according to this disclosure and is not intended to limit the scope of this disclosure. The embodiments of this disclosure are equally applicable to other systems or architectures.

[0032] like Figure 1 As shown, system 100 may include a computing device 120 located on the server side. The computing device 120 may be configured to receive user data from client 110. Furthermore, the computing device 120 may determine recommendations that the user of client 110 might be interested in using a recommendation model 130 disposed therein, and feed back the recommendation results to client 110.

[0033] In some embodiments, the user data acquired by the computing device 120 may be historical behavioral data generated by the user when browsing web pages or using an APP, and the recommendation model 130 may determine the recommendation results that the user may be interested in based on a large amount of user data (the server maintains sufficient user data for the user).

[0034] In this disclosure, the recommendation model 130 can be designed to perform recommendation tasks. As an example, a video recommendation model corresponding to a video app can be used to recommend videos of interest to a user based on their video viewing history. Examples of recommendation models include, but are not limited to, various deep neural networks (DNNs), convolutional neural networks (CNNs), support vector machines (SVMs), decision trees, random forest models, and so on. In the implementation of this disclosure, the recommendation model may also be referred to as a "neural network," "learning model," "learning network," "model," and "network," which are used interchangeably.

[0035] In some embodiments, the computing device 120 may include, but is not limited to, personal computers, server computers, handheld or laptop devices, mobile devices (such as mobile phones, personal digital assistants (PDAs), media players, etc.), consumer electronics, minicomputers, mainframe computers, cloud computing resources, etc.

[0036] It should be understood that the devices and / or units included in system 100 are exemplary only and are not intended to limit the scope of this disclosure. It should be understood that system 100 may also include additional devices and / or units not shown. For example, in some embodiments, the computing device 120 of system 100 may further include a storage unit (not shown) for storing pre-input hyperparameters, etc.

[0037] The following will refer to Figure 2 The training and use of the model in computing device 120 are described.

[0038] Figure 2 A schematic diagram of a detailed example environment 200 according to an embodiment of the present disclosure is shown. Figure 1 Similarly, example environment 200 may include computing device 220, user data 210 input to computing device 220, and recommendation results 230 output from computing device 220. The difference is that example environment 200 can generally include model training system 260 and model application system 270. As an example, model training system 260 and / or model application system 270 can be, for example... Figure 1 The computing device 120 shown or such Figure 2 The examples are implemented in the computing device 220 shown. It should be understood that the description of the structure and functionality of the example environment 200 is for illustrative purposes only and is not intended to limit the scope of the topics described herein. The topics described herein may be implemented in different structures and / or functionalities.

[0039] As mentioned earlier, the process of processing user data from client 110 to determine identification results such as the user's items of interest can be divided into two stages: the model training stage and the model application stage. As an example, such as... Figure 2 As shown, during the model training phase, the model training system 260 can use the user dataset 250 to train the recommendation model 240. It should be understood that the user dataset 250 can be historical behavioral data generated by a large number of users while using the app or browsing web pages. During the model application phase, the model application system 270 can receive the trained recommendation model 240. Thus, the recommendation model 240 loaded into the computing device 220 of the model application system 270 can determine the recommendation result 230 based on the user data 210.

[0040] In other embodiments, the recommendation model 240 can be constructed as a learning network. In some embodiments, the learning network may include multiple networks, each of which may be a multi-layer neural network composed of a large number of neurons. Through a training process, the corresponding parameters of the neurons in each network can be determined. The parameters of the neurons in these networks are collectively referred to as the parameters of the recommendation model 240.

[0041] The training process of recommendation model 240 can be performed iteratively until at least some of the parameters of recommendation model 240 converge or until a predetermined number of iterations is reached, thereby obtaining the final model parameters.

[0042] The technical solutions described above are for illustrative purposes only and are not intended to limit this disclosure. It should be understood that other networks can also be arranged in different ways and with different connections. To more clearly explain the principles of the above solutions, references will be made below. Figure 3 Let's describe the recommendation model training process in more detail.

[0043] Figure 3 A flowchart of a process 300 for training a recommendation model according to an embodiment of the present disclosure is shown. In some embodiments, process 300 may be performed at... Figure 1 The computing device 120 and Figure 2 This is implemented in computing device 220. Now refer to... Figure 3 The process 300 for training a recommendation model according to an embodiment of this disclosure is described below. For ease of understanding, the specific examples mentioned in the following description are exemplary and are not intended to limit the scope of protection of this disclosure.

[0044] In step 302, the computing device 120 may, in response to receiving a user's data deletion request, determine the data in the original data that is requested to be deleted. In some embodiments, the data corresponding to the data deletion request is historical data or noisy data generated by the user while browsing web pages or using applications.

[0045] To clearly describe the process for determining the "requested data to be deleted in the original data" mentioned in this disclosure, reference is now made to... Figure 4 This will be described in detail. Figure 4 A schematic diagram of raw data 400 of multiple users maintained on the server side according to an embodiment of the present disclosure is shown. Figure 4 As shown, the server side can typically maintain or store user data 410, user data 420, user data 430, user data 440, user data 450, and user data 460, and each user data corresponds to only one user. It should be understood that... Figure 4 The specific number of users shown is merely an example; the server side can maintain user data for many more users.

[0046] When the user corresponding to user data 410 sends a data deletion request to the server through a given interface, such as an APP, the computing device 120 on the server side can determine, based on the data deletion request, which parts of the user's original data are the data to be deleted. Figure 4 In this process, computing device 120 can determine that data 411 is the data to be deleted based on the data deletion request.

[0047] It should be understood that in the field of video recommendation, the raw data described herein typically includes multiple users and tag information of the videos they have each watched, and the video recommendation model to be trained is configured to identify other videos associated with that tag information that at least one of these users has not watched. It should also be understood that in the field of news recommendation, the raw data described herein typically includes multiple users and tag information of the news articles they have read, and the news recommendation model to be trained is configured to identify other news articles associated with that tag information that at least one of these users has not read. Furthermore, it should be understood that in the field of product recommendation, the raw data described herein typically includes multiple users and tag information of the products they have searched, and the product recommendation model to be trained is configured to identify other products associated with that tag information that at least one of these users has not searched.

[0048] Back Figure 3 In step 304, the computing device 120 can obtain the remaining data for the user based on the data requested to be deleted. In other words, in Figure 4 In this process, computing device 120 can obtain the remaining data 412 in user data 410 based on the requested deletion data 411. In some embodiments, to obtain the remaining data 412, computing device 120 can determine the remaining data 412 by removing the requested deletion data 411 from user data 410. In some embodiments, when a user instructs the deletion of all user data 410, computing device 120 can determine the remaining data 412 through random initialization. In this way, the impact of the requested deletion data on the model can be minimized as much as possible during subsequent training of the recommendation model.

[0049] Back to Figure 3In step 306, the computing device 120 can train the recommendation model 130 using the remaining data 412. In some embodiments, to train the recommendation model 130, the computing device 120 can obtain the parameters of the original recommendation model determined by training with the original data 400, and determine these parameters as the initial parameters of the recommendation model 130. Then, the computing device 120 can update the initial parameters of the recommendation model 130 using the remaining data 412. As an example, the recommendation model 130 can be trained once using the remaining data 412. In this way, the training process of the recommendation model 130 can be warmed up using the parameters of the original recommendation model, thereby shortening the training time. Furthermore, since the remaining data 412 better reflects the preferences of the users corresponding to the user data 410, directly using the remaining data 412 to train the recommendation model 130 for one or more rounds can quickly and effectively adjust the model's parameters. Moreover, since the training of the recommendation model forgets the data that the user expects to delete, the recommendation results of the recommendation model are more likely to match the user's preferences, thereby improving the user experience.

[0050] In some embodiments, process 300 may further include, for example, the computing device 120 may obtain additional data from the original data for users other than those who requested data deletion, and use this additional data to train the recommendation model 130 that has already been trained using the remaining data. In other words, as Figure 4 As shown, computing device 120 can obtain additional data 413 from the original data 400 and use the additional data 413 to train the recommendation model 130 that has already been trained using the remaining data 412. In this way, since the additional data 413 is training data that was previously used to train the original recommendation model, it is more stable and more conducive to the convergence of the recommendation model 130, thereby shortening the training time of the recommendation model 130.

[0051] In some embodiments, the raw data includes at least multiple users and corresponding items of interest. As an example, in... Figure 4 In the original data 400, user data 410 can contain corresponding users and their interests, and so on. The term "interest" can refer to personal preferences reflected in a user's historical behavioral data. As an example, in the application scenario of video recommendation, when a user frequently watches videos related to current affairs, their interest can be identified as current affairs. Of course, interest can have more detailed categories. As an example, in the application scenario of product recommendation, when a user frequently searches for "43 size leather shoes_men" in the app, their interest can be identified as "men's shoes," "leather shoes," "size 43," etc.

[0052] Figure 5A flowchart illustrating a portion of the detailed process 500 for training a recommendation model according to embodiments of the present disclosure is shown. In some embodiments, process 500 may be performed at... Figure 1 The computing device 120 and Figure 2 This is implemented in computing device 220. Now refer to... Figure 5 The present disclosure describes a portion of the detailed process 500 for training a recommendation model according to an embodiment of the present disclosure. For ease of understanding, the specific examples mentioned in the following description are exemplary and are not intended to limit the scope of protection of this disclosure.

[0053] In step 501, computing device 120 may keep at least a first feature data in recommendation model 130 unchanged, which corresponds to items of interest for a group of users among a plurality of users. In some embodiments, computing device 120 may fix the feature data of items of interest in recommendation model 130 as well as other model parameters in recommendation model 130 other than the feature data of users and items of interest. It should be understood that the group of users may be a single user or a small group of users.

[0054] In step 502, computing device 120 may update the second feature data of the group of users to minimize the loss function value of recommendation model 130. In some embodiments, computing device 120 may use a second-order optimizer to update the feature data of the group of users to minimize the loss of recommendation model 130 on the remaining data.

[0055] In step 503, the computing device 120 can determine whether all users have been traversed. If not, it continues the operation of step 501; if so, it proceeds to step 504.

[0056] In step 504, computing device 120 may maintain at least one set of second feature data of users in recommendation model 130 unchanged. In some embodiments, computing device 120 may fix the feature data of users in recommendation model 130 as well as other model parameters in recommendation model 130 other than the feature data of users and items of interest.

[0057] In step 505, computing device 120 may update the first feature data to minimize the loss function value. In some embodiments, computing device 120 may use a second-order optimizer to update the feature data of the item of interest to minimize the loss of recommendation model 130 on the remaining data.

[0058] In step 506, the computing device 120 can determine whether all items of interest have been traversed. If not, it continues the operation of step 504; if so, it proceeds to step 507.

[0059] In step 507, computing device 120 can return feature data of the user and items of interest, thereby training recommendation model 130.

[0060] Therefore, the model training process can be accelerated through the aforementioned alternating optimization method, thereby shortening the model training time. Furthermore, the model training method utilizes a second-order optimizer, which, combined with the aforementioned warm-start implementation, can significantly improve the efficiency and effectiveness of model training.

[0061] Through the above embodiments, this disclosure avoids the long model training time and significant computational overhead caused by retraining the recommendation model (forgotten model). Furthermore, using only the remaining data with a warm-start training method can significantly accelerate the training process of the recommendation model. Furthermore, by utilizing other data with alternating optimization and second-order optimization methods, the model training time can be shortened and model performance improved. Moreover, since the recommendation model's training forgets the data that the user expects to delete, the recommendation model's performance is improved, and the recommendation results are more likely to match the user's preferences.

[0062] This disclosure also provides a model training apparatus. Specifically, Figure 6 A schematic diagram of an apparatus 600 for training a recommendation model according to an embodiment of the present disclosure is shown. Figure 6 As shown, the device 600 may include at least a data deletion determination module 602, a remaining data acquisition module 604, and a recommendation model training module 606. The data deletion determination module 602 can determine the data to be deleted from the original data upon receiving a user's data deletion request. The remaining data acquisition module 604 can acquire the remaining data for the user based on the requested data. Furthermore, the recommendation model training module 606 can train a recommendation model using the remaining data.

[0063] In some embodiments, the recommendation model training module 606 may include: an original parameter acquisition module configured to acquire parameters of an original recommendation model determined by training with the original data; an initial parameter determination module configured to determine the parameters as initial parameters of the recommendation model; and a recommendation model update module configured to update the initial parameters of the recommendation model with the remaining data.

[0064] In some embodiments, the apparatus 600 may further include: an additional data acquisition module configured to acquire additional data of other users besides the aforementioned users from the original data; and a recommendation module retraining module configured to train a recommendation model trained with the remaining data using the additional data.

[0065] In some embodiments, the remaining data acquisition module 604 may be configured to determine the remaining data by removing the data to be deleted from the original data.

[0066] In some embodiments, the original data may include at least multiple users and corresponding items of interest, and the recommendation model training module 606 may be configured to: keep at least the feature data in the recommendation model corresponding to the items of interest of a group of users unchanged; and update the feature data of a group of users to minimize the loss function value of the recommendation model.

[0067] In some embodiments, the recommendation model training module 604 may also be configured to: keep the feature data of at least one set of users in the recommendation model unchanged; and update the feature data corresponding to the items of interest of a set of users to minimize the loss function value.

[0068] In some embodiments, updating feature data for a group of users and updating feature data corresponding to items of interest for a group of users can both be performed using a second-order optimizer.

[0069] In some embodiments, the data corresponding to the data deletion request may be historical data or noisy data generated by the user while browsing web pages or using applications.

[0070] In some embodiments, the raw data may include multiple users and tag information of videos watched by each user, and the recommendation model is configured to identify other videos associated with the tag information that at least one of the users has not watched.

[0071] In some embodiments, the raw data may include multiple users and tag information of items of interest that each user has browsed or visited, and the recommendation model is configured to determine other items of interest associated with the tag information that at least one of the multiple users has not browsed or visited.

[0072] In some embodiments, the item of interest can be a product, a themed article, an image, a social media account, etc. A product could correspond to a product page displayed on a shopping website. A themed article could correspond to news articles, short stories, or novels. An image could correspond to a photograph, screenshot, or vector graphic.

[0073] Figure 7 A schematic block diagram of an example device 700 that can be used to implement embodiments of the present disclosure is shown. For example, such as Figure 1 The computing device 120 shown and Figure 2The computing device 220 shown can be implemented by device 700. As shown, device 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) 702 or loaded from storage unit 708 into random access memory (RAM) 703. The RAM 703 can also store various programs and data required for the operation of device 700. The CPU 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0074] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks. It should be understood that this disclosure can utilize output unit 707 to display real-time dynamic changes in user satisfaction, key factor identification information for group or individual users of satisfaction, optimization strategy information, and strategy implementation effectiveness evaluation information, etc.

[0075] Processing unit 701 may be implemented by one or more processing circuits. Processing unit 701 may be configured to perform the various processes and procedures described above, such as processes 300 and 500. For example, in some embodiments, processes 300 and 500 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by CPU 701, one or more steps of processes 300 and 500 described above may be performed.

[0076] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of this disclosure.

[0077] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0078] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0079] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0080] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0081] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processing unit of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0082] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0083] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0084] According to one or more embodiments of this disclosure. Example 1. A method for training a recommendation model, comprising: in response to receiving a user's data deletion request, determining data to be deleted in original data; obtaining remaining data for the user based on the requested data; and training the recommendation model using the remaining data.

[0085] Example 2. According to the method in Example 1, training the recommendation model includes: obtaining parameters of an original recommendation model determined by training with the original data; determining the parameters as initial parameters of the recommendation model; and updating the initial parameters of the recommendation model with the remaining data.

[0086] Example 3. The method according to Example 1 further includes: obtaining additional data of other users besides the user from the original data; and using the additional data to train the recommendation model trained using the remaining data.

[0087] Example 4. According to the method of Example 1, obtaining the remaining data includes: determining the remaining data by removing the data to be deleted from the original data.

[0088] Example 5. The method according to any one of Examples 1-4, wherein the original data includes at least a plurality of users and corresponding items of interest, and wherein training the recommendation model includes: keeping at least a first feature data in the recommendation model unchanged, the first feature data corresponding to items of interest of a group of users among the plurality of users; and updating a second feature data of the group of users to minimize the loss function value of the recommendation model.

[0089] Example 6. According to the method of Example 5, training the recommendation model further includes: keeping at least the second feature data in the recommendation model unchanged; and updating the first feature data to minimize the loss function value.

[0090] Example 7. The method described in Example 6, wherein updating the feature data of the group of users and updating the feature data corresponding to the items of interest of the group of users are both performed using a second-order optimizer.

[0091] Example 8. The method according to any one of Examples 1-4, wherein the data corresponding to the data deletion request is historical data or noise data generated by the user while browsing web pages or using applications.

[0092] Example 9. The method according to any one of Examples 1-4, wherein the raw data includes multiple users and tag information of videos watched by each of the multiple users, and the recommendation model is configured to determine other videos associated with the tag information that at least one of the multiple users has not watched.

[0093] Example 10. The method according to any one of Examples 1-4, wherein the raw data includes multiple users and tag information of items of interest that each of the multiple users has browsed or visited, and the recommendation model is configured to determine other items of interest associated with the tag information that at least one of the multiple users has not browsed or visited.

[0094] Example 11. According to the method of Example 10, the item of interest includes at least one of the following: product; topic article; image; and social media account.

[0095] According to one or more embodiments of this disclosure, Example 12. An apparatus for training a recommendation model, comprising: a data deletion determination module configured to determine data to be deleted from original data in response to receiving a data deletion request from a user; a remaining data acquisition module configured to acquire remaining data for the user based on the data to be deleted; and a recommendation model training module configured to train the recommendation model using the remaining data.

[0096] Example 13. The apparatus according to Example 12, wherein the recommendation model training module comprises: an original parameter acquisition module configured to acquire parameters of an original recommendation model determined by training with the original data; an initial parameter determination module configured to determine the parameters as initial parameters of the recommendation model; and a recommendation model update module configured to update the initial parameters of the recommendation model with the remaining data.

[0097] Example 14. The apparatus according to Example 12 further includes: an additional data acquisition module configured to acquire additional data of other users besides the user from the original data; and a recommendation modular retraining module configured to train the recommendation model trained with the remaining data using the additional data.

[0098] Example 15. The apparatus according to Example 12, wherein the remaining data acquisition module is configured to determine the remaining data by removing the data to be deleted from the original data.

[0099] Example 16. An apparatus according to any one of Examples 12-15, wherein the original data includes at least a plurality of users and corresponding items of interest, and wherein the recommendation model training module is configured to: keep at least a first feature data in the recommendation model unchanged, the first feature data corresponding to items of interest of a group of users among the plurality of users; and update a second feature data of the group of users to minimize the loss function value of the recommendation model.

[0100] Example 17. The apparatus according to Example 16, wherein the recommendation model training module is further configured to: keep at least the second feature data in the recommendation model unchanged; and update the first feature data to minimize the loss function value.

[0101] Example 18. The apparatus according to Example 17, wherein updating the feature data of the group of users and updating the feature data corresponding to the items of interest of the group of users are both performed using a second-order optimizer.

[0102] Example 19. The apparatus according to any one of Examples 12-15, wherein the data corresponding to the data deletion request is historical data or noise data generated by the user while browsing web pages or using applications.

[0103] Example 20. An apparatus according to any one of Examples 12-15, wherein the raw data includes a plurality of users and tag information of videos watched by each of the plurality of users, and the recommendation model is configured to determine other videos associated with the tag information that have not been watched by at least one of the plurality of users.

[0104] Example 21. An apparatus according to any one of Examples 12-15, wherein the raw data includes a plurality of users and tag information of items of interest that each of the plurality of users has browsed or visited, and the recommendation model is configured to determine other items of interest associated with the tag information that at least one of the plurality of users has not browsed or visited.

[0105] Example 22. The apparatus according to Example 21, wherein the item of interest includes at least one of the following: a product; a topical article; an image; and a social media account.

[0106] According to one or more embodiments of this disclosure, Example 23. An electronic device includes: a processor; and a memory coupled to the processor, the memory having instructions stored therein, the instructions causing the electronic device to perform actions when executed by the processor, the actions including: in response to receiving a user's data deletion request, determining data to be deleted from original data; obtaining remaining data for the user based on the requested data deletion; and training the recommendation model using the remaining data.

[0107] Example 24. The device according to Example 23, wherein training the recommendation model includes: obtaining parameters of an original recommendation model determined by training with the original data; determining the parameters as initial parameters of the recommendation model; and updating the initial parameters of the recommendation model with the remaining data.

[0108] Example 25. The device according to Example 23 further includes: obtaining additional data of other users besides the user from the original data; and using the additional data to train the recommendation model trained using the remaining data.

[0109] Example 26. The device according to Example 23, wherein obtaining the remaining data includes: determining the remaining data by removing the data to be deleted from the original data.

[0110] Example 27. A device according to any one of Examples 23-26, wherein the original data includes at least a plurality of users and corresponding items of interest, and wherein training the recommendation model comprises: keeping at least the feature data in the recommendation model corresponding to the items of interest of a set of the plurality of users unchanged; and updating the feature data of the set of users to minimize the loss function value of the recommendation model.

[0111] Example 28. The device according to Example 27, wherein training the recommendation model further includes: keeping feature data of at least the set of users in the recommendation model unchanged; and updating feature data corresponding to the items of interest of the set of users to minimize the value of the loss function.

[0112] Example 29. The device according to Example 28, wherein updating the feature data of the group of users and updating the feature data corresponding to the items of interest of the group of users are both performed using a second-order optimizer.

[0113] Example 30. The device according to any one of Examples 23-26, wherein the data corresponding to the data deletion request is historical data or noise data generated by the user while browsing web pages or using applications.

[0114] Example 31. A device according to any one of Examples 23-26, wherein the raw data includes multiple users and tag information of videos watched by each of the multiple users, and the recommendation model is configured to determine other videos associated with the tag information that at least one of the multiple users has not watched.

[0115] Example 32. A device according to any one of Examples 23-26, wherein the raw data includes multiple users and tag information of items of interest that each of the multiple users has browsed or visited, and the recommendation model is configured to determine other items of interest associated with the tag information that at least one of the multiple users has not browsed or visited.

[0116] Example 33. The device according to Example 32, wherein the item of interest includes at least one of the following: a product; a topical article; an image; and a social media account.

[0117] According to one or more embodiments of the present disclosure, Example 34. A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method as described in any one of Examples 1-11.

[0118] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, and are not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical applications, or technical improvements to the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for training a recommendation model, comprising: In response to receiving a user's data deletion request, determine the data to be deleted from the original data; Based on the data that was requested to be deleted, obtain the remaining data for the user; as well as The recommendation model is trained using the remaining data. The raw data mentioned therein includes at least multiple users and corresponding items of interest, and training the recommendation model includes: Keep at least the first feature data in the recommendation model unchanged, the first feature data corresponding to the interest items of a group of users among the plurality of users; as well as Update the second feature data of the group of users to minimize the loss function value of the recommendation model.

2. The method according to claim 1, wherein training the recommendation model comprises: Obtain the parameters of the original recommendation model determined by training using the original data; The parameters are determined as the initial parameters of the recommendation model; as well as The initial parameters of the recommendation model are updated using the remaining data.

3. The method according to claim 1, further comprising: Obtain additional data from users other than the user mentioned above from the original data; as well as The recommendation model, trained using the remaining data, is then trained using the other data.

4. The method according to claim 1, wherein obtaining the remaining data comprises: Remove the requested data from the original data to determine the remaining data.

5. The method according to claim 1, wherein training the recommendation model further comprises: Keep at least the second feature data in the recommendation model unchanged; as well as Update the first feature data to minimize the loss function value.

6. The method according to claim 5, wherein updating the feature data of the group of users and updating the feature data corresponding to the items of interest of the group of users are both performed using a second-order optimizer.

7. The method according to any one of claims 1-4, wherein the data corresponding to the data deletion request is historical data or noise data generated by the user while browsing web pages or using applications.

8. The method according to any one of claims 1-4, wherein the raw data includes a plurality of users and tag information of videos watched by each of the plurality of users, and the recommendation model is configured to determine other videos associated with the tag information that have not been watched by at least one of the plurality of users.

9. The method according to any one of claims 1-4, wherein the raw data includes multiple users and tag information of items of interest that each of the multiple users has browsed or visited, and the recommendation model is configured to determine other items of interest associated with the tag information that at least one of the multiple users has not browsed or visited.

10. The method of claim 9, wherein the term of interest includes at least one of the following: product; Thematic articles; Images; and Self-media account.

11. An apparatus for training a recommendation model, comprising: The data deletion determination module is configured to, in response to receiving a user's data deletion request, determine the data to be deleted from the original data. The remaining data acquisition module is configured to acquire remaining data for the user based on the data requested to be deleted; and The recommendation model training module is configured to train the recommendation model using the remaining data. The raw data includes at least multiple users and corresponding items of interest, and the recommendation model training module is further configured to: Keeping at least the first feature data in the recommendation model unchanged, the first feature data corresponding to the interest items of a group of users among the plurality of users; and Update the second feature data of the group of users to minimize the loss function value of the recommendation model.

12. An electronic device, comprising: processor; as well as A memory coupled to the processor, the memory having instructions stored therein, which, when executed by the processor, cause the electronic device to perform the method as described in any one of claims 1-10.

13. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of claims 1-10.

Images