Federal learning modeling method, device and equipment and medium

Abstract

The invention provides a federal learning modeling method and device, equipment and a medium. The method comprises the following steps: acquiring a user data set about a target user behavior; obtaining first feature data, wherein the first feature data is output of a feature extraction model after model training is performed on the feature extraction model of the first participant by using the user data set; and using the first feature data and second feature data of the first participant to cooperate with a second participant in the plurality of participants to execute a federated learning modeling task to obtain a federated learning model for analyzing user behaviors.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technologies, and in particular to the field of federated learning technologies. Specifically, the present disclosure relates to a federated learning modeling method, an apparatus, a system, an electronic device, and a non-transitory computer-readable storage medium therefor. Background Art

[0002] Federated Learning, also known as Federated Machine Learning, is a machine learning framework applicable to multiple participating parties. Using the federated learning framework, machine learning model modeling, predictive analysis, data querying, data statistics, and other federated learning tasks can be performed while satisfying the user privacy protection and data security requirements of multiple participating parties.

[0003] The methods described in this section are not necessarily methods that have been previously conceived or adopted. Unless otherwise specified, no method described in this section should be considered prior art merely because it is included in this section. Similarly, unless otherwise specified, the problems mentioned in this section should not be considered to have been recognized in any prior art. Summary of the Invention

[0004] The present disclosure provides a federated learning modeling method, an apparatus, an electronic device, and a non-transitory computer-readable storage medium therefor.

[0005] According to one aspect of the present disclosure, there is provided a federated learning modeling method, which is applied to a first participating party among multiple participating parties performing a federated learning modeling task. The method includes: obtaining a user data set regarding target user behaviors; obtaining first feature data, where the first feature data is the output of a feature extraction model of the first participating party after training the feature extraction model using the user data set; and using the first feature data and second feature data of the first participating party to cooperate with a second participating party among the multiple participating parties to perform the federated learning modeling task to obtain a federated learning model for analyzing user behaviors.

[0006] According to one aspect of the present disclosure, there is provided a federated learning modeling device, which is applied to a first participant among multiple participants performing a federated learning modeling task. The device includes: an input data set acquisition unit configured to acquire a user data set regarding target user behavior; a feature data acquisition unit configured to acquire first feature data, where the first feature data is the output of the feature extraction model after training the feature extraction model of the first participant using the user data set; and a federated learning modeling unit configured to use the first feature data and second feature data of the first participant to cooperate with a second participant among the multiple participants to perform the federated learning modeling task to obtain a federated learning model for analyzing user behavior.

[0007] According to one aspect of the present disclosure, there is provided a method for analyzing user behavior based on federated learning, which is applied to a first participant among multiple participants performing an online prediction task. Wherein, the method includes: acquiring a user data set to be analyzed regarding user behavior; acquiring first feature data corresponding to the user data set to be analyzed, where the first feature data is obtained by a trained feature extraction model based on the user data set to be analyzed; based on the first feature data and second feature data of the first participant, the first participant and a second participant among the multiple participants cooperate to perform the analysis of the user behavior using a federated learning model to obtain a behavior label of the user data set to be analyzed, where the federated learning model is obtained according to the above-mentioned federated learning modeling method.

[0008] According to one aspect of the present disclosure, there is provided a device for analyzing user behavior based on federated learning, which is applied to a first participant among multiple participants performing an online prediction task. Wherein, the device includes: a first acquisition module configured to acquire a user data set to be analyzed regarding user behavior; a second acquisition module configured to acquire first feature data corresponding to the user data set to be analyzed, where the first feature data is obtained by a trained feature extraction model based on the user data set to be analyzed; a federated prediction module configured to, based on the first feature data and second feature data of the first participant, cooperate with a second participant among the multiple participants to perform the analysis of the user behavior using a federated learning model to obtain a behavior label of the user data set to be analyzed, where the federated learning model is obtained according to the above-mentioned federated learning modeling method.

[0009] According to one aspect of the present disclosure, there is provided an electronic device, including: a processor; and a memory storing a program, where the program includes instructions that, when executed by the processor, cause the processor to execute the method according to the above.

[0010] According to one aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing a computer program including instructions that, when executed by a processor of a computing device, cause the computing device to execute the method according to the above.

[0011] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood from the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings exemplarily illustrate embodiments and form part of the specification, and are used together with the written description of the specification to explain the exemplary embodiments. The illustrated embodiments are for illustrative purposes only and do not limit the scope of the claims. In all the drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0013] Figure 1 A flowchart showing a federated learning modeling method according to an exemplary embodiment of the present disclosure;

[0014] Figure 2 A schematic diagram showing data collection according to an exemplary embodiment of the present disclosure;

[0015] Figure 3 A schematic diagram showing an embedding layer and a feature extraction model architecture according to an exemplary embodiment of the present disclosure;

[0016] Figure 4 A flowchart showing a training method of a feature extraction model according to an exemplary embodiment of the present disclosure;

[0017] Figure 5 A flowchart showing a training method of a feature extraction model according to an exemplary embodiment of the present disclosure;

[0018] Figure 6 A flowchart showing a federated learning modeling method according to an exemplary embodiment of the present disclosure;

[0019] Figure 7 A schematic diagram showing a federated learning architecture according to an exemplary embodiment of the present disclosure;

[0020] Figure 8 A block diagram showing a federated learning modeling apparatus according to an exemplary embodiment of the present disclosure;

[0021] Figure 9 A block diagram showing an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The following describes exemplary embodiments of the present disclosure with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, descriptions of well-known functions and structures are omitted in the following description for clarity and conciseness.

[0023] In the present disclosure, unless otherwise specified, the terms "first", "second", etc. are used to describe various elements and are not intended to limit the positional relationship, temporal relationship, or relative importance of these elements. Such terms are only used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of the element, and in certain cases, based on the context description, they may also refer to different instances.

[0024] The terms used in the description of various examples in the present disclosure are only for the purpose of describing specific examples and are not intended to be limiting. Unless the context clearly indicates otherwise, if the number of elements is not specifically limited, the element may be one or more. In addition, the term "and / or" used in the present disclosure covers any one of the listed items and all possible combinations.

[0025] Federated learning is a distributed machine learning framework that can be applied to multiple participants, and it can help achieve data encryption and privacy protection among multiple participants when performing federated learning tasks. Specifically, the federated learning task can be a training task for a machine learning model, or a federated data query task, a federated data statistics task, etc.

[0026] Aggregate data is a collection of events that occur over a period of time. Aggregate data can be applied to machine learning modeling tasks such as federated learning modeling and is very important for the modeling process. Since aggregate data reflects the events that occur over a period of time, the number and form of events included in each aggregate data are not necessarily the same. Therefore, aggregate data is irregular data. For example, the aggregate lengths for different objects (e.g., users) may be different, even extremely different. In addition, even for the same object, the aggregate lengths in different time periods may also be different. For example, the number of events that a user has in a month may be different from the number of events in previous months. In addition to the irregularity of aggregate data, multiple events in aggregate data can be independent of each other, and there is not necessarily a chronological relationship or a dependency relationship between multiple events. Therefore, for aggregate data, changing the position of the data points corresponding to the events therein should not affect the prediction result. Specifically, for machine learning modeling tasks such as federated learning modeling that use aggregate data as input, the order of the data points corresponding to the events in the aggregate data should not affect the output result of the machine learning modeling task.

[0027] In practical applications, aggregate data can be regarded as the words in a language, so as to use the method of natural language processing (NLP) to process aggregate data. However, due to the limitation of computing resources, overly long sentences in natural language processing will be truncated, resulting in information loss. Since multiple events in aggregate data can be independent of each other and do not have a chronological dependency relationship, the truncated data may affect the results of machine learning modeling tasks such as federated learning modeling. In addition, in natural language processing, due to the natural sequential nature of language, the order of words is very important. Multiple natural language processing models process data based on the assumption of the order of words. Therefore, aggregate data with the same event content but different orders will generate different processing results when processed using a natural language model, which cannot meet the requirement that changing the position of the data points corresponding to the events in the aggregate data does not affect the processing result.

[0028] To solve the above problems, the present disclosure provides a federated learning modeling method, its device, system, electronic device, non-transitory computer-readable storage medium, and computer program product. Specifically, a model suitable for aggregate data and a method for applying this model in a federated learning framework are proposed. Through this specific model design, the disadvantages in the process of using a natural language model to process aggregate data can be avoided. At the same time, the features corresponding to the aggregate data are first "locally trained by one party" at the local end of the participating party, increasing the flexibility and efficiency of the model. The model data obtained through "local training by one party" is used for the federated learning tasks of two or more parties, increasing the number of features of one party and ensuring the training efficiency while enhancing the model complexity.

[0029] Embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0030] Figure 1 FIG. 4 shows a flowchart of a federated learning modeling method 100 for user behavior analysis according to an exemplary embodiment of the present disclosure. The federated learning modeling method 100 is applied to a first participant (e.g., a Guest participant with true data labels) among multiple participants performing a federated learning modeling task. The federated learning modeling method 100 includes: step S102, obtaining a user data set regarding target user behavior; step S104, obtaining first feature data, where the first feature data is the output of a feature extraction model after training the feature extraction model of the first participant using the user data set; and step S106, using the first feature data and second feature data of the first participant to cooperate with a second participant among the multiple participants to perform a federated learning modeling task to obtain a federated learning model for analyzing user behavior.

[0031] In some embodiments, the user data set as an input to the federated learning modeling task includes user behavior analysis data for a target user (e.g., consumption behavior analysis data, browsing behavior analysis data, etc.), and the data output by the federated learning modeling task is a federated learning model, which predicts user behavior for the target user (e.g., consumption behavior prediction, browsing behavior prediction, etc.).

[0032] Figure 1 In the federated learning modeling method 100, by using the first feature data output by the feature extraction model for feature extraction of the user data set local to the first participant, when the first participant cooperates with the second participant to perform the federated learning modeling task, additional feature data is added, which improves the complexity of the model trained by the first participant while ensuring the training efficiency.

[0033] In some embodiments, the feature extraction model is a machine learning model for processing set data, and the user data set includes behavior data. An embedding layer is provided at the input end of the feature extraction model, and the embedding layer converts the behavior data in the user data set into a vector representation, and the converted vector representation is input into the feature extraction model.

[0034] In some embodiments, the feature extraction model further includes a plurality of neural networks, and the plurality of neural networks process the converted vector representation corresponding to the behavior data to obtain feature data (e.g., the first feature data described in the method 100 above).

[0035] In some embodiments, the user dataset in the above-mentioned federated learning modeling method 100 is "set data", and each behavioral data included in the user dataset corresponds to a user's behavior. The number of behavioral data in different user datasets may be different. The following will be combined with Figure 2 describe the form of set data in detail.

[0036] Figure 2 Fig. shows a schematic diagram of set data according to an exemplary embodiment of the present disclosure. Specifically, Figure 2 shows two set data f x1 and f x2 as examples. Among them, the set data f x1 includes a total of k behavioral data x i , and the set data f x2 includes a total of h behavioral data x i , where each behavioral data x i corresponds to a user behavior, and in this embodiment, the number of behavioral data in the two set data f x1 and f x2 is not the same, that is, h > k.

[0037] The user dataset described herein may be in the form of set data, and each behavioral data included in the user dataset corresponds to a user's behavior.

[0038] In some embodiments, the first feature data in the above-mentioned federated learning modeling method 100 represents the features of the behavioral data in the user dataset in the form of set data, while the second feature data represents other features of the user. The following gives examples of two scenarios that can apply the above-mentioned federated learning modeling method 100. It can be understood that, in addition to the listed example scenarios, other federated modeling tasks related to user behavior can also apply the various methods described in this application.

[0039] According to some embodiments, the user dataset is a commodity dataset of the commodities purchased by the user, and the behavioral data x i corresponds to the commodity item purchased by the user. For example, the first feature data in the above-mentioned federated learning modeling method 100 represents the features of the commodities purchased by the user (such as commodity type, commodity price, sales area, etc.), and the second feature data represents other features of the user related to the purchased commodity (such as user age, user occupation, user gender, etc.).

[0040] According to some embodiments, the user dataset is an advertisement dataset of the advertisements browsed by the user, and the behavioral data x iCorresponding to the advertisement item browsed by the user, the first feature data represents the features of the advertisement browsed by the user (for example, advertisement type, advertisement placement platform, advertisement placement time, etc.), and the second feature data represents other features of the user related to browsing the advertisement (for example, user age, user occupation, user gender, etc.).

[0041] In some embodiments, the second feature data and the third feature data respectively depend on the data types owned by the first party and the second party, which may be user personal information, such as the above-mentioned user age, user gender, etc., or user tags, such as discount lovers, etc.

[0042] By processing the commodity data set and the advertisement data set in the above method, the behavior data in the user data set with a large number of behavior data and different quantities from each other can be processed and applied to the federated learning modeling task. It avoids the losses and defects caused by processing the user data set through natural language methods, improves the complexity of the federated learning model (for example, can cover more features), and also ensures the execution effect.

[0043] In some embodiments, for different users, the number of behavior data in the set data corresponding to the user data set can be different. For example, as Figure 2 shown, h≠k. In some embodiments, for the same user, the number of behavior data in the set data corresponding to the user data set at different time periods can also be different.

[0044] Inputting the user data set in the form of set data into the feature extraction model for the first party can obtain the corresponding first feature data. The structure of the feature extraction model and the method for obtaining the first feature data will be described in detail below in conjunction with Figure 3 the detailed description of the structure of the feature extraction model and the method for obtaining the first feature data.

[0045] Figure 3 Shows a schematic diagram of an embedding layer and a feature extraction model architecture according to an exemplary embodiment of the present disclosure.

[0046] Exemplarily, the input data for the feature extraction process is the user data set f x (for example, the user data set in the form of the above-mentioned "set data"). The user data set f x includes a total of k behavior data {x 1 , x 2 ,..., x k}, where each behavior data x i corresponds to an embedding layer respectively (that is, the embedding layers 311 to 31k shown in Figure 3 ) and a neural network (that is, the neural networks 321 to 32k shown in Figure 3 ). Each behavior data x iInput its corresponding embedding layer to convert the behavior data into a corresponding sub-vector representation. Input the sub-vector representation into separate neural networks, and an intermediate vector representation φ(x i ) of the same dimension can be obtained. Based on the intermediate vector representation φ(x i ) corresponding to each behavior data, a target vector representation 330 can be obtained. Input the target vector representation into neural network 340, and the finally output feature data 350 can be obtained as, for example, the first feature data mentioned above, and this first feature data represents the features of multiple behavior data x x in the user dataset f i .

[0047] In some embodiments, a summation operation is performed on the intermediate vector representations corresponding to each behavior data to obtain the target vector representation 330, ∑ i φ(x i ). In some embodiments, the summation operation can be a weighted summation operation to highlight the contribution of certain specific behavior data to the prediction result.

[0048] The embedding layers 311 to 31k can be generated in different ways. For example, the embedding layers 311 to 31k can be initialized using the embedding layers previously trained for similar or the same users. In some embodiments, multiple embedding layers can be determined by means of an embedding dictionary, and the embedding dictionary includes the sub-vector representations corresponding to multiple behavior data. Exemplarily, the embedding dictionary can be generated locally by random logic, that is, the correspondence between behavior data and intermediate vectors is randomly initialized.

[0049] According to some embodiments, the user dataset includes multiple behavior data, and each behavior data has a corresponding sub-vector representation. Among them, the federated learning modeling method further includes: generating an embedding dictionary, where the embedding dictionary includes the sub-vector representations corresponding to multiple behavior data respectively; and obtaining the sub-vector representation corresponding to each behavior data based on the embedding dictionary, where the first feature data is the output of the feature extraction model after model training by inputting the sub-vector representations corresponding to multiple behavior data respectively into the feature extraction model.

[0050] Thus, the conversion between behavior data and vector representation can be realized, and while processing the behavior data, defects such as information loss caused by using natural language processing methods are avoided. And by generating the embedding dictionary, the storage requirement can be reduced, and the generated embedding dictionary can be used for the initialization of subsequent training.

[0051] According to some embodiments, the federated learning modeling method further includes: obtaining the first feature data corresponding to the user dataset output by the trained feature extraction model and storing it locally, where obtaining the first feature data includes obtaining the first feature data from local storage.

[0052] Thus, the stored first feature data can be directly invoked, thereby improving the processing efficiency.

[0053] The feature extraction model and the corresponding embedding layer can be determined through training. The following combines Figure 4 describes the training process of the feature extraction model and the corresponding embedding layer.

[0054] Figure 4 FIG. 4 shows a flowchart of a training method 400 of a feature extraction model according to an exemplary embodiment of the present disclosure. The training method 400 of the feature extraction model includes: step S402, initializing the embedding layer; step S404, inputting the behavior data of the user dataset into the embedding layer, where the user dataset has real behavior labels; step S406, inputting the output of the embedding layer into the feature extraction model to obtain the feature data corresponding to the user dataset output by the feature extraction model; step S408, obtaining the predicted behavior labels of the user dataset based on the feature data; and step S410, adjusting the parameters of the embedding layer and the feature extraction model based on the real labels and the predicted labels, where generating the embedding dictionary includes storing the correspondence between the behavior data and its sub-vector representation, and the sub-vector representation corresponding to the behavior data is the output of the embedding layer obtained by inputting the behavior data into the embedding layer.

[0055] Thus, the training of the feature extraction model can be realized, and the generated embedding dictionary and the output of the trained feature extraction model are retained locally by the participating parties. The feature data output by the trained feature extraction model is directly used to participate in the federated learning modeling tasks of multiple participating parties, improving the model complexity while ensuring the training efficiency.

[0056] In some embodiments, the adjustment of the parameters of the embedding layer and the feature extraction model can be determined based on the loss function calculated between the predicted labels and the real labels.

[0057] According to some embodiments, generating the embedding dictionary includes: after each iteration in the training process of the feature extraction model, updating the embedding dictionary based on the output obtained by inputting the behavior data into the embedding layer with adjusted parameters, or, only after the feature extraction model is trained, updating and generating the embedding dictionary based on the output obtained by inputting the behavior data into the trained embedding layer.

[0058] According to some embodiments, the method for generating the embedding dictionary further includes: initializing the embedding dictionary; initializing the embedding layer includes: initializing the embedding layer based on the initialized embedding dictionary.

[0059] Thus, the embedding dictionary can be directly used to initialize the embedding layer during training and prediction, reducing the storage requirements of the model and ensuring the efficiency of the model.

[0060] For exampleFigure 3 The feature extraction model structure shown in [the figure] includes neural networks 321 to 32k corresponding to each piece of behavior data, and a neural network 340 for converting the target vector representation 330 into feature data 350. The neural network 340 does not correspond to a single piece of behavior data, but rather to the entire user data set. In some embodiments, some or all of the neural networks among neural networks 321 to 32k and neural network 340 can be trained separately. Below will be combined with Figure 5 Describe an embodiment of training all neural networks.

[0061] Figure 5 Shows a flowchart of a training method 500 of a feature extraction model according to an exemplary embodiment of the present disclosure.

[0062] Among them, the feature extraction model includes a first feature extraction sub-model and a second feature extraction sub-model corresponding to each piece of behavior data in the user data set, and inputs the output of the embedding layer into the feature extraction model, and obtains the feature data corresponding to the user data set based on the output of the feature extraction model, including: Step S502, inputting the sub-vector representation corresponding to each piece of behavior data output by the embedding layer into the first feature extraction sub-model corresponding to this piece of behavior data; Step S504, based on the result output by the first feature extraction sub-model corresponding to each piece of behavior data, obtaining the target vector representation corresponding to the user data set; and Step S506, inputting the target vector representation into the second feature extraction sub-model to obtain the first feature data corresponding to the user data set output by the second feature extraction sub-model.

[0063] Thus, the behavior data can be converted into vector representations with the same dimension, and a single target vector representation can be obtained based on multiple vector representations, thereby avoiding the adverse effects caused by inconsistent or excessive numbers of behavior data in the user data set on the federated learning process.

[0064] According to some embodiments, adjusting the parameters of the feature extraction model includes adjusting the parameters of the first feature extraction sub-model and the second feature extraction sub-model. Thus, the adjustment of the parameters of each link in the feature extraction model can be realized, increasing the flexibility of the feature extraction model and improving the overall performance of the feature extraction model.

[0065] It can be understood that adjusting the parameters of the feature extraction model can also only adjust the parameters of the first feature extraction sub-model or only adjust the parameters of the second feature extraction sub-model. In addition, in some embodiments, it is also possible to only adjust the parameters of a part of the sub-models in the first feature extraction sub-model to perform adjustment for a specific neural network.

[0066] According to some embodiments, the first participating party has true behavior labels of the user dataset, and the true behavior labels are used to represent the true analysis results corresponding to the user dataset. In other words, the above method is applied to the Guest party in the federated learning framework with true labels of the user dataset. The true labels of the Guest party are preferentially used to locally pre-train the set data features, which increases the flexibility of the model and improves the efficiency of the Guest party (e.g., the above first participating party) and the Host party (e.g., the above second participating party) in cooperating to execute the federated learning task.

[0067] The following will be combined with Figures 6 to 7 to describe in detail an example of the first participating party and the second participating party cooperating to execute the federated learning task.

[0068] Specifically, Figure 6 shows a flowchart of a federated learning modeling method according to an exemplary embodiment of the present disclosure. The federated learning model of the first participating party includes a first machine learning model and a second machine learning model. As in step S106 of the above method 100, using the first feature data and the second feature data of the first participating party to cooperate with the second participating party to execute the federated learning modeling task, including: step S1062, inputting the first feature data and the second feature data of the first participating party into the first machine learning model of the first participating party to obtain first intermediate data; step S1064, obtaining second intermediate data sent by the second participating party, where the second intermediate data is obtained by inputting third feature data of the second participating party into a third machine learning model of the second participating party; step S1066, combining the first intermediate data and the second intermediate data, and inputting the combined first intermediate data and second intermediate data into the second machine learning model of the first participating party; step S1068, adjusting the parameters of the first machine learning model and the second machine learning model based on the predicted behavior labels output by the second machine learning model and the true behavior labels of the user dataset.

[0069] It can be understood that the federated learning modeling method can also be executed simultaneously by more participating parties, which is not limited herein.

[0070] Thus, the output of the feature extraction model can be directly used to participate in machine learning tasks involving multiple participating parties, increasing the number of features of one party, improving the overall complexity and flexibility of the model, and ensuring the training efficiency at the same time.

[0071] Figure 7A schematic diagram of a federated learning architecture 700 according to an exemplary embodiment of the present disclosure is shown. In this schematic diagram, a first participant 710 and a second participant 720 are used to perform a federated learning task (e.g., a federated learning modeling method). It can be understood that the federated learning task can also be performed simultaneously by more participants, which is not limited herein.

[0072] In the federated learning architecture 700, the first participant 710 includes first feature data 711 and second feature data 712, where one of the first feature data 711 and the second feature data 712 can be a feature extraction model (e.g., the feature extraction model described above) obtained by pre-training locally at the first participant (e.g., the training method 400 described above), and the other can be other feature data at the first participant. The specific training steps and methods of this pre-training have been introduced in the description above in combination with, for example, Figure 4 and are not repeated here. Figure 3 The specific training steps and methods of this pre-training have been introduced in the description above in combination with, for example, Figure 4 and are not repeated here.

[0073] In some embodiments, the first feature data represents the features of the behavioral data in the user dataset, and the second feature data represents other features of the user.

[0074] At the first participant 710, the first feature data 711 and the second feature data 712 are input into the first model 713, where the first model 713 can also be referred to as the "bottom model" of the first participant 710. At the second participant 720, the third feature data 721 is input into the third model 722, where the third model 722 can also be referred to as the "bottom model" of the second participant 720.

[0075] The results output by the first model 713 and the third model 722 are input into the intermediate processing 730 for operations such as splicing and merging, where the intermediate processing 730 is also referred to as the "interaction layer". The data obtained through the intermediate processing 730 can reflect the features of both the first participant and the second participant, that is, the features combining the first feature data 711, the second feature data 712, and the third feature data 721. The data obtained through the intermediate processing 730 is input into the second model 714 of the first participant 710 to output the final result, where the second model 714 can also be referred to as the "top model" of the first participant 710.

[0076] Using the output prediction result and the true label of the second model 714 (i.e., the "top model" of the first participant 710), the parameters of the first model 713 (i.e., the "bottom model" of the first participant 710), the third model 722 (i.e., the "bottom model" of the second participant 720), and the second model 714 (i.e., the "top model" of the first participant 710) can be adjusted.

[0077] In some embodiments, the adjustment of the parameters of the above model can be determined based on a loss function calculated between the output prediction result and the true label.

[0078] In some embodiments, the federated learning modeling task can be various tasks such as the above-mentioned purchase commodity prediction task, advertisement browsing prediction task, etc.

[0079] Figure 8 The block diagram of a federated learning modeling device 800 for user behavior analysis according to an exemplary embodiment of the present disclosure is shown. The federated learning modeling device 800 is applied to the first participant among multiple participants performing the federated learning modeling task. The device 800 includes: an input data set acquisition unit 801 configured to acquire a user data set regarding the target user behavior; a feature data acquisition unit 802 configured to acquire first feature data, where the first feature data is the output of the feature extraction model after training the feature extraction model of the first participant using the user data set; and a federated learning modeling unit 803 configured to use the first feature data and the second feature data of the first participant to cooperate with a second participant among the multiple participants to perform the federated learning modeling task to obtain a federated learning model for analyzing user behavior.

[0080] The federated learning modeling device 800 can be configured to perform operations similar to those of the federated learning modeling method 100 described above.

[0081] According to some embodiments, wherein the user data set includes multiple behavior data, and each behavior data has a corresponding sub-vector representation. The federated learning modeling device further includes: an embedding dictionary generation unit configured to generate an embedding dictionary, where the embedding dictionary includes the sub-vector representations corresponding to the multiple behavior data respectively; and a query unit configured to obtain the sub-vector representation corresponding to each behavior data based on the embedding dictionary, where the first feature data is the output of the feature extraction model after training by inputting the sub-vector representations corresponding to the multiple behavior data into the feature extraction model.

[0082] According to some embodiments, the federated learning modeling device further includes: a feature data storage unit configured to acquire and store locally the first feature data corresponding to the user data set output by the trained feature extraction model, where acquiring the first feature data includes acquiring the first feature data from local storage.

[0083] According to some embodiments, the federated learning modeling device further includes: a model training unit configured for the training process of the feature extraction model, including: initializing an embedding layer; inputting the behavior data of the user dataset into the embedding layer, where the user dataset has real behavior labels; inputting the output of the embedding layer into the feature extraction model to obtain the feature data corresponding to the user dataset output by the feature extraction model; obtaining the predicted behavior labels of the user dataset based on the feature data; and adjusting the parameters of the embedding layer and the feature extraction model based on the real labels and the predicted labels, where generating the embedding dictionary includes storing the correspondence between the behavior data and its sub-vector representation, and the sub-vector representation corresponding to the behavior data is the output of the embedding layer obtained by inputting the behavior data into the embedding layer.

[0084] According to some embodiments, the embedding dictionary generation unit is further configured to: after each iteration in the training process of the feature extraction model, update the embedding dictionary based on the output obtained by inputting the behavior data into the embedding layer with adjusted parameters, or, only after the feature extraction model is trained, update and generate the embedding dictionary based on the output obtained by inputting the behavior data into the trained embedding layer.

[0085] According to some embodiments, the embedding dictionary generation unit is further configured to: initialize the embedding dictionary, and the model training unit is further configured to: initialize the embedding layer based on the initialized embedding dictionary.

[0086] According to some embodiments, the feature extraction model includes a first feature extraction sub-model and a second feature extraction sub-model corresponding to each behavior data of the user dataset, and the feature data acquisition unit is configured to: input the sub-vector representation corresponding to each behavior data output by the embedding layer into the first feature extraction sub-model corresponding to the behavior data; obtain the target vector representation corresponding to the user dataset based on the result output by the first feature extraction sub-model corresponding to each behavior data; and input the target vector representation into the second feature extraction sub-model to obtain the first feature data corresponding to the user dataset output by the second feature extraction sub-model.

[0087] According to some embodiments, adjusting the parameters of the feature extraction model includes adjusting the parameters of the first feature extraction sub-model and the second feature extraction sub-model.

[0088] According to some embodiments, the first party has the real behavior labels of the user dataset, and the real behavior labels are used to represent the real analysis results corresponding to the user dataset.

[0089] According to some embodiments, the federated learning model of the first participating party includes a first machine learning model and a second machine learning model, and the federated learning modeling unit 803 is configured to: input the first feature data and the second feature data of the first participating party into the first machine learning model of the first participating party to obtain first intermediate data; obtain the second intermediate data sent by the second participating party, where the second intermediate data is obtained by inputting the third feature data of the second participating party into the third machine learning model of the second participating party; combine the first intermediate data and the second intermediate data, and input the combined first intermediate data and second intermediate data into the second machine learning model of the first participating party; based on the predicted behavior label output by the second machine learning model and the true behavior label of the user dataset, adjust the parameters of the first machine learning model and the second machine learning model.

[0090] According to some embodiments, the user dataset is a product dataset of products purchased by the user, the behavior data corresponds to the product items purchased by the user, the first feature data represents the features of the products purchased by the user, and the second feature data represents other features of the user related to the purchased products.

[0091] According to some embodiments, the user dataset is an advertisement dataset of advertisements browsed by the user, the behavior data corresponds to the advertisement items browsed by the user, the first feature data represents the features of the advertisements browsed by the user, and the second feature data represents other features of the user related to the browsed advertisements.

[0092] According to some embodiments, a user behavior analysis method based on federated learning is provided, which is applied to the first participating party among multiple participating parties performing an online prediction task. The user behavior analysis method includes: obtaining a user dataset to be analyzed regarding user behavior; obtaining first feature data corresponding to the user dataset to be analyzed, where the first feature data is obtained by the trained feature extraction model based on the user dataset to be analyzed; based on the first feature data and the second feature data of the first participating party, the first participating party and the second participating party among the multiple participating parties cooperate to perform the analysis of the user behavior by using the federated learning model to obtain the behavior label of the user dataset to be analyzed, where the federated learning model is obtained according to the above-mentioned federated learning modeling method.

[0093] Thus, the federated learning model obtained by applying the above-mentioned federated learning modeling method can be used to cooperate with the first feature data to perform the user behavior analysis task.

[0094] According to some embodiments, the user dataset to be analyzed includes multiple behavior data, and each behavior data has a corresponding sub-vector representation. Obtaining the first feature data corresponding to the user dataset to be analyzed includes: obtaining the sub-vector representation corresponding to each behavior data based on the locally stored embedding dictionary; inputting the sub-vector representations corresponding to the multiple behavior data into the trained feature extraction model, and obtaining the first feature data corresponding to the user dataset to be analyzed output by the feature extraction model.

[0095] Thus, the user dataset can be processed by the feature extraction model, so that when performing user behavior analysis, additional feature data is added, the complexity of the model trained by the first party is increased, and the training efficiency is ensured.

[0096] According to some embodiments, a user behavior analysis device based on federated learning is provided, which is applied to the first party among multiple parties performing an online prediction task. The user behavior analysis device includes: a first acquisition module configured to acquire a user dataset to be analyzed regarding user behavior; a second acquisition module configured to acquire the first feature data corresponding to the user dataset to be analyzed, where the first feature data is obtained by the trained feature extraction model based on the user dataset to be analyzed; a federated prediction module configured to perform an analysis of user behavior in cooperation with a second party among multiple parties based on the first feature data and the second feature data of the first party by using a federated learning model, so as to obtain a behavior label of the user dataset to be analyzed, where the federated learning model is obtained according to the above-mentioned federated learning modeling method.

[0097] According to some embodiments, the user dataset to be analyzed includes multiple behavior data, and each behavior data has a corresponding sub-vector representation. The second acquisition module includes: a first acquisition sub-module configured to obtain the sub-vector representation corresponding to each behavior data based on the locally stored embedding dictionary; a first acquisition sub-module configured to input the sub-vector representations corresponding to the multiple behavior data into the trained feature extraction model, and obtain the first feature data corresponding to the user dataset to be analyzed output by the feature extraction model.

[0098] In some embodiments, the method involved in the present disclosure is applied to an application scenario of user commodity purchase. Among them, the behavior data corresponds to the commodity items purchased by the user, the first feature data represents the features of the commodities purchased by the user, and the second feature data represents other features related to the commodities purchased by the user.

[0099] In this application scenario, the federated learning modeling method includes:

[0100] Obtaining a user dataset regarding the purchase behavior of a target user, where the user dataset includes multiple commodity items that the user has purchased;

[0101] After obtaining the first feature data capable of representing the characteristics of the goods purchased by the user through the above-mentioned federated modeling method (e.g., the federated modeling method 100), specifically including: respectively inputting multiple commodity items purchased by the user into the embedding layer, inputting the output of the embedding layer into the feature extraction model to obtain the corresponding feature data; obtaining the predicted behavior labels corresponding to the user data set based on the feature data; adjusting the parameters of the embedding layer and the feature extraction model based on the predicted behavior labels and the true behavior labels corresponding to the user; and obtaining and storing the first feature data corresponding to the user data set output by the trained feature extraction model locally;

[0102] Use the first feature data capable of representing the characteristics of the goods purchased by the user, the second feature data capable of representing other characteristics of the user related to the purchased goods, and cooperate with its participating party (e.g., the above-mentioned second participating party) to perform a federated learning modeling task to obtain a federated learning model for analyzing user behavior.

[0103] Exemplarily, the federated learning model may have an architecture as shown in Figure 7 shown.

[0104] The federated learning model obtained by the above-mentioned federated learning modeling method can be used to analyze user behavior in the same application scenario. The analysis method includes:

[0105] Obtain a user data set to be analyzed regarding user behavior, where the user data set to be analyzed includes multiple commodity items that the user has purchased;

[0106] Obtain the first feature data corresponding to the user data set to be analyzed, which is obtained by the trained feature extraction model based on the user data set to be analyzed, and the first feature data can represent the characteristics of multiple commodity items that the user has purchased;

[0107] Use the first feature data capable of representing the characteristics of the goods purchased by the user, the second feature data capable of representing other characteristics of the user related to the purchased goods, and cooperate with its participating party (e.g., the above-mentioned second participating party) to perform the analysis of the user behavior to obtain the behavior label of the user data set to be analyzed (e.g., the label can represent the tendency of the user to purchase goods).

[0108] Similarly, in some embodiments, the method involved in the present disclosure is applied to the application scenario of user browsing advertisements, where the behavior data corresponds to the advertisement items browsed by the user, the first feature data represents the characteristics of the advertisements browsed by the user, and the second feature data represents other characteristics of the user related to browsing advertisements.

[0109] In this application scenario, the federated learning modeling method includes:

[0110] Obtain a user data set regarding the purchase behavior of a target user, where the user data set includes multiple advertisement items browsed by the user;

[0111] Obtain first feature data that can represent the features of the advertisements browsed by the user through the above-mentioned federated modeling method (for example, federated modeling method 100), specifically including: inputting multiple advertisement items browsed by the user into the embedding layer respectively, inputting the output of the embedding layer into the feature extraction model to obtain corresponding feature data; obtaining a predicted behavior label corresponding to the user data set based on the feature data; adjusting the parameters of the embedding layer and the feature extraction model based on the predicted behavior label and the true behavior label corresponding to the user; and obtaining and storing the first feature data corresponding to the user data set output by the trained feature extraction model locally;

[0112] Use the first feature data that can represent the features of the advertisements browsed by the user, the second feature data that can represent other features of the user related to the browsed advertisements, and cooperate with its participating party (for example, the second participating party mentioned above) to execute a federated learning modeling task to obtain a federated learning model for analyzing user behavior.

[0113] Exemplarily, the federated learning model may have an architecture as shown in Figure 7 shown.

[0114] The federated learning model obtained through the above-mentioned federated learning modeling method can be used to analyze the behavior of users in the same application scenario. The analysis method includes:

[0115] Obtain a user data set to be analyzed regarding user behavior, where the user data set to be analyzed includes multiple advertisement items that the user has browsed;

[0116] Obtain the first feature data corresponding to the user data set to be analyzed, where the first feature data is obtained by the trained feature extraction model based on the user data set to be analyzed, and the first feature data can represent the features of multiple advertisement items that the user has browsed;

[0117] Use the first feature data that can represent the features of the advertisements browsed by the user, the second feature data that can represent other features of the user related to the browsed advertisements, and cooperate with its participating party (for example, the second participating party mentioned above) to execute the analysis of the user behavior to obtain the behavior label of the user data set to be analyzed (for example, the label may be the tendency of the user to browse advertisements).

[0118] According to one aspect of the present disclosure, there is provided an electronic device, including: a processor; and a memory, where the memory stores a program, and the program includes instructions that, when executed by the processor, cause the processor to execute the above-mentioned method.

[0119] According to one aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing a computer program including instructions which, when executed by a processor of a computing device, cause the computing device to execute the above-described method.

[0120] According to one aspect of the present disclosure, there is provided a computer program product including a computer program, wherein the computer program, when executed by a processor, implements the above-described method.

[0121] See Figure 9 , a block diagram of an electronic device 900 that may be according to an embodiment of the present disclosure will now be described, which is an example of a hardware device applicable to various aspects of the present disclosure. The electronic device may be different types of computing devices, such as, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as, a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0122] Figure 9 A block diagram of an electronic device according to an embodiment of the present disclosure is shown. As Figure 9 shown, the electronic device 900 may include at least one processor 901, a working memory 902, an I / O device 904, a display device 905, a storage device 906, and a communication interface 907 that are capable of communicating with each other via a system bus 903.

[0123] The processor 901 may be a single processing unit or multiple processing units, and all processing units may include a single or multiple computing units or multiple cores. The processor 901 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuits, and / or any device that manipulates signals based on operation instructions. The processor 901 may be configured to obtain and execute computer-readable instructions stored in the working memory 902, the storage device 906, or other computer-readable media, such as program codes of an operating system 902a, program codes of an application 902b, and the like.

[0124] The working memory 902 and the storage device 906 are examples of computer-readable storage media for storing instructions that are executed by the processor 901 to implement the various functions described above. The working memory 902 may include both volatile and non-volatile memories (e.g., RAM, ROM, etc.). In addition, the storage device 906 may include a hard disk drive, a solid state drive, removable media, including external and removable drives, memory cards, flash memory, floppy disks, optical disks (e.g., CD, DVD), storage arrays, network-attached storage, storage area networks, and the like. The working memory 902 and the storage device 906 may both be collectively referred to herein as memory or computer-readable storage media, and may be non-transitory media capable of storing computer-readable, processor-executable program instructions as computer program code that can be executed by the processor 901 as a specific machine configured to implement the operations and functions described in the examples herein.

[0125] The I / O device 904 may include input devices and / or output devices. The input devices may be any type of device capable of inputting information to the electronic device 900, and may include, but are not limited to, a mouse, a keyboard, a touch screen, a trackpad, a trackball, a joystick, a microphone, and / or a remote control. The output devices may be any type of device capable of presenting information, and may include, but are not limited to, a video / audio output terminal, a vibrator, and / or a printer.

[0126] The communication interface 907 allows the electronic device 900 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks, and may include, but are not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, such as a BluetoothTM device, an 802.11 device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0127] The application 902b in the working register 902 may be loaded and executed to perform the various methods and processes described above, such as Figure 1 the steps S102 - S106 in. In some embodiments, part or all of the computer program may be loaded and / or installed onto the electronic device 900 via the storage device 906 and / or the communication interface 907. When the computer program is loaded and executed by the processor 901, one or more steps of the data processing method described above may be performed.

[0128] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0129] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine, or entirely on the remote machine or server.

[0130] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0131] To provide for interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide for interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0132] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), and the Internet.

[0133] A computing system can include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client - server relationship is created by computer programs that run on the respective computers and have a client - server relationship to each other.

[0134] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in this disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this is not limited herein.

[0135] Although embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it should be understood that the above methods, systems, and devices are merely exemplary embodiments or examples, and the scope of the present invention is not limited by these embodiments or examples, but is only defined by the authorized claims and their equivalent scope. Various elements in the embodiments or examples may be omitted or replaced by their equivalent elements. In addition, the steps may be executed in an order different from that described in the present disclosure. Further, the various elements in the embodiments or examples may be combined in various ways. Importantly, with the evolution of technology, many of the elements described herein may be replaced by equivalent elements that emerge after the present disclosure.

Claims

1. A federated learning modeling method, applied to a first participant among multiple participants performing a federated learning modeling task, the method comprises: Obtain a user data set regarding the behavior of a target user; Obtain first feature data, where the first feature data is the output of the feature extraction model after training the feature extraction model of the first participant using the user data set; and Use the first feature data and the second feature data of the first participant to cooperate with a second participant among the multiple participants to perform the federated learning modeling task to obtain a federated learning model for analyzing user behavior.

2. The method according to claim 1, wherein, The user data set includes multiple behavior data, and each behavior data has a corresponding sub-vector representation, wherein the method further comprises: Generate an embedding dictionary, where the embedding dictionary includes the sub-vector representations corresponding to the multiple behavior data respectively; and Based on the embedding dictionary, obtain the sub-vector representation corresponding to each behavior data, wherein the first feature data is the output of the feature extraction model after inputting the sub-vector representations corresponding to the multiple behavior data into the feature extraction model for model training.

3. The method according to claim 2, further comprises: Obtain and store locally the first feature data corresponding to the user data set output by the trained feature extraction model, wherein obtaining the first feature data includes obtaining the first feature data from local storage.

4. The method according to claim 2, wherein, The training process of the feature extraction model includes: Initialize the embedding layer; Input the behavior data of the user data set into the embedding layer, where the user data set has real behavior labels; Input the output of the embedding layer into the feature extraction model, and obtain the feature data corresponding to the user data set output by the feature extraction model; Obtain the predicted behavior labels of the user data set based on the feature data; and Based on the real behavior labels and the predicted behavior labels, adjust the parameters of the embedding layer and the feature extraction model, wherein generating the embedding dictionary includes storing the corresponding relationship between the behavior data and its sub-vector representation, and the sub-vector representation corresponding to the behavior data is the output of the embedding layer obtained by inputting the behavior data into the embedding layer.

5. The method according to claim 4, wherein, Generating the embedding dictionary includes: After each iteration in the training process of the feature extraction model, update the embedding dictionary based on the output obtained by inputting the behavior data into the embedding layer with adjusted parameters, or, only after the feature extraction model is trained, update and generate the embedding dictionary based on the output obtained by inputting the behavior data into the trained embedding layer.

6. The method for generating the embedding dictionary according to claim 4, the method for generating the embedding dictionary further comprises: Initialize the embedding dictionary; The initialization of the embedding layer includes: Initialize the embedding layer based on the initialized embedding dictionary.

7. The method according to claim 4, wherein, The feature extraction model includes a first feature extraction sub-model and a second feature extraction sub-model corresponding to each behavior data of the user data set, and inputting the output of the embedding layer into the feature extraction model, and obtaining the feature data corresponding to the user data set based on the output of the feature extraction model, including: inputting the sub-vector representation corresponding to each behavior data output by the embedding layer into the first feature extraction sub-model corresponding to this behavior data; obtaining the target vector representation corresponding to the user data set based on the result output by the first feature extraction sub-model corresponding to each behavior data; and inputting the target vector representation into the second feature extraction sub-model to obtain the first feature data corresponding to the user data set output by the second feature extraction sub-model.

8. The method according to claim 7, wherein, adjusting the parameters of the feature extraction model includes adjusting the parameters of the first feature extraction sub-model and the second feature extraction sub-model.

9. The method according to any one of claims 1 to 8, wherein, the first party has the true behavior labels of the user data set, and the true behavior labels are used to represent the true analysis results corresponding to the user data set.

10. The method according to any one of claims 1 to 8, wherein, the federated learning model of the first party includes a first machine learning model and a second machine learning model, and wherein, using the first feature data and the second feature data of the first party, and cooperating with the second party to perform the federated learning modeling task, including: inputting the first feature data and the second feature data of the first party into the first machine learning model of the first party to obtain first intermediate data; obtaining the second intermediate data sent by the second party, wherein the second intermediate data is obtained by inputting the third feature data of the second party into the third machine learning model of the second party; combining the first intermediate data and the second intermediate data, and inputting the combined first intermediate data and second intermediate data into the second machine learning model of the first party; adjusting the parameters of the first machine learning model and the second machine learning model based on the predicted behavior labels output by the second machine learning model and the true behavior labels of the user data set.

11. The method according to any one of claims 1 to 8, wherein, the user data set is a commodity data set of commodities purchased by the user, the behavior data corresponds to the commodity items purchased by the user, the first feature data represents the features of the commodities purchased by the user, and the second feature data represents other features of the user related to the purchased commodities.

12. The method according to any one of claims 1 to 8, wherein, the user data set is an advertisement data set of advertisements browsed by the user, the behavior data corresponds to the advertisement items browsed by the user, the first feature data represents the features of the advertisements browsed by the user, and the second feature data represents other features of the user related to the browsed advertisements.

13. A federated learning modeling device is applied to a first participant among multiple participants performing a federated learning modeling task. The device comprises: an input data set acquisition unit configured to acquire a user data set regarding target user behavior; a feature data acquisition unit configured to acquire first feature data, where the first feature data is the output of a feature extraction model of the first participant after the feature extraction model is trained using the user data set; and a federated learning modeling unit configured to use the first feature data and second feature data of the first participant, and cooperate with a second participant among the multiple participants to perform the federated learning modeling task to obtain a federated learning model for analyzing user behavior.

14. A method for analyzing user behavior based on federated learning is applied to a first participant among multiple participants performing an online prediction task. Wherein, the method comprises: acquiring a user data set to be analyzed regarding user behavior; acquiring first feature data corresponding to the user data set to be analyzed, where the first feature data is obtained by a trained feature extraction model based on the user data set to be analyzed; based on the first feature data and second feature data of the first participant, the first participant and a second participant among the multiple participants cooperate to perform the analysis of the user behavior using a federated learning model to obtain a behavior label of the user data set to be analyzed. Wherein, the federated learning model is obtained according to the method described in any one of claims 1 to 12.

15. According to the method of claim 14, wherein, the user data set to be analyzed includes multiple behavior data, and each behavior data has a corresponding sub-vector representation. Acquiring first feature data corresponding to the user data set to be analyzed includes: obtaining the sub-vector representation corresponding to each behavior data based on an embedding dictionary stored locally; inputting the sub-vector representations corresponding to the multiple behavior data into the trained feature extraction model, and acquiring the first feature data corresponding to the user data set to be analyzed output by the feature extraction model.

16. A device for analyzing user behavior based on federated learning is applied to a first participant among multiple participants performing an online prediction task. Wherein, the device comprises: a first acquisition module configured to acquire a user data set to be analyzed regarding user behavior; a second acquisition module configured to acquire first feature data corresponding to the user data set to be analyzed, where the first feature data is obtained by a trained feature extraction model based on the user data set to be analyzed; a federated prediction module configured to, based on the first feature data and second feature data of the first participant, cooperate with a second participant among the multiple participants to perform the analysis of the user behavior using a federated learning model to obtain a behavior label of the user data set to be analyzed. Wherein, the federated learning model is obtained according to the method described in any one of claims 1 to 12.

17. According to the device of claim 16, wherein, The user dataset to be analyzed includes a plurality of behavioral data, and each behavioral data has a corresponding sub-vector representation. The second acquisition module includes: A first acquisition sub-module, configured to obtain the sub-vector representation corresponding to each behavioral data based on the embedding dictionary stored locally; A first acquisition sub-module, configured to input the sub-vector representations corresponding to the plurality of behavioral data into the trained feature extraction model, and obtain the first feature data corresponding to the user dataset to be analyzed output by the feature extraction model.

18. An electronic device comprising: A processor; and A memory, the memory stores a program, the program includes instructions, and the instructions, when executed by the processor, cause the processor to execute the method according to any one of claims 1-12.

19. A non-transitory computer-readable storage medium, the non-transitory computer-readable storage medium stores a computer program, the computer program includes instructions, and the instructions, when executed by the processor of a computing device, cause the computing device to execute the method according to any one of claims 1-12.

Images