Recommendation information generation method, device, storage medium and electronic device

Abstract

The present invention discloses a method, device, storage medium, and electronic device for generating recommendation information. The method comprises: responding to a recommendation request from a user device, selecting a target patch model from a plurality of patch models based on the recommendation request; and sending the target patch model to the user device, wherein the user device is configured to use a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and use the target recommendation model to generate recommendation information for the user. The present invention solves the technical problem that, due to the prior art of generating recommendation information primarily based on specific user behavior data, some user recommendation information may be biased, resulting in a low user experience.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, in particular to the field of information recommendation and model prediction technology, and specifically to a method, device, storage medium and electronic device for generating recommendation information. Background Art

[0002] The current mainstream approach to generating recommendation information is to train a single recommendation model based on the historical behavior of all platform users. After launch, this model infers user interests based on their historical behavior and generates personalized recommendations. However, because a small number of highly active users often contribute the majority of behavioral data, the recommendation system tends to favor highly active users and perform poorly for medium- and low-activity users. This compromises the effectiveness of recommendations for a large number of long-tail users, resulting in a lack of fairness and a negative impact on user experience.

[0003] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0004] Embodiments of the present invention provide a method, device, storage medium, and electronic device for generating recommendation information, so as to at least solve the technical problem that, due to the fact that the existing technology mainly generates recommendation information based on specific user behavior data, some user recommendation information may be biased, resulting in a low user experience.

[0005] According to one aspect of an embodiment of the present invention, a method for generating recommendation information is provided, comprising: responding to a recommendation request from a user device, selecting a target patch model from a plurality of patch models according to the recommendation request; sending the target patch model to the user device, wherein the user device is configured to use a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and use the target recommendation model to generate recommendation information for the user.

[0006] According to another aspect of an embodiment of the present invention, a method for generating recommendation information is also provided, including: sending a recommendation request to a server, wherein the server is used to select a target patch model from multiple patch models based on the recommendation request; receiving the target patch model issued by the server; using a predetermined initial recommendation model and the target patch model to generate a target recommendation model adapted to the user device; and using the target recommendation model to generate recommendation information for the user.

[0007] According to another aspect of an embodiment of the present invention, a recommendation information generating device is also provided, including: a response module, used to respond to a recommendation request from a user device, and select a target patch model from multiple patch models according to the above recommendation request; a first sending module, used to send the above target patch model to the above user device, wherein the above user device is used to adopt a predetermined initial recommendation model and the above target patch model to generate an adapted target recommendation model, and adopt the above target recommendation model to generate recommendation information for the user.

[0008] According to another aspect of an embodiment of the present invention, a recommendation information generating device is also provided, including: a second sending module, used to send a recommendation request to a server, wherein the server is used to select a target patch model from multiple patch models according to the recommendation request; a receiving module, used to receive the target patch model sent by the server; a first generating module, used to use a predetermined initial recommendation model and the target patch model to generate a target recommendation model adapted to the user device; and a second generating module, used to use the target recommendation model to generate recommendation information for the user.

[0009] According to another aspect of an embodiment of the present invention, a non-volatile storage medium is further provided. The non-volatile storage medium includes a stored program, wherein when the program is executed, the device where the non-volatile storage medium is located is controlled to execute any one of the above-mentioned recommendation information generation methods.

[0010] According to another aspect of an embodiment of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any one of the above-mentioned recommendation information generating methods.

[0011] In an embodiment of the present invention, by responding to a recommendation request from a user device, a target patch model is selected from multiple patch models according to the recommendation request and the target patch model is sent to the user device, wherein the user device is used to adopt a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and adopt the target recommendation model to generate recommendation information for the user, thereby achieving the purpose of generating personalized recommendation information according to the actual needs of the user, thereby achieving the technical effect of generating more accurate recommendation information according to the actual needs of the user and improving the user experience, and further solving the technical problem that the recommendation information of some users is biased and the user experience is low due to the fact that the recommendation information is mainly generated based on specific user behavior data in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0013] Figure 1 This is a flowchart of the first recommendation method in the recommendation scenario of the related art;

[0014] Figure 2 This is a flowchart of the second recommendation method in the recommendation scenario of the related art;

[0015] Figure 3 is a flow chart of a recommendation method provided according to an embodiment of the present invention;

[0016] Figure 4 is a flow chart of a method for generating recommendation information according to an embodiment of the present invention;

[0017] Figure 5 is a flowchart of an optional method for generating recommendation information according to an embodiment of the present invention;

[0018] Figure 6 is a flowchart of another optional method for generating recommendation information according to an embodiment of the present invention;

[0019] Figure 7 is a flow chart of another method for generating recommendation information according to an embodiment of the present invention;

[0020] Figure 8 is a structural diagram of a device for generating recommendation information according to an embodiment of the present invention;

[0021] Figure 9 is a schematic structural diagram of another device for generating recommendation information according to an embodiment of the present invention;

[0022] Figure 10 It is a structural block diagram of a computer terminal according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] Example 1

[0026] According to an embodiment of the present invention, an embodiment of a method for generating recommendation information is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0027] In the recommendation scenarios of related technologies, the following two methods are generally used for recommendation.

[0028] Figure 1 This is a flowchart of the first recommendation method in the recommendation scenario of the related technology, such as Figure 1 As shown, the process of the recommended method includes the following steps:

[0029] S102, the server generates a single recommendation model based on all platform user data (i.e., all sample data);

[0030] S104: The user triggers a recommendation request, which may be a search request, and the server subsequently responds to the recommendation corresponding to the feedback based on the search request;

[0031] S106, the client receives the user's recommendation request and sends the recommendation request to the server;

[0032] S108: After receiving the recommendation request from the client, the server obtains the user's historical behavior data, infers the user's interests based on the historical behavior data, uses the trained single recommendation model, determines a recommendation result based on the user's interests, and feeds the recommendation result back to the client;

[0033] S110: After receiving the recommendation results, the client displays the recommendation results on a display interface, for example, a list of search objects corresponding to the search request.

[0034] In the first recommendation method described above, a single recommendation model is trained on the server. Users send recommendation requests through the client, and the server determines the recommendation results based on the user's historical behavior data. The server then feeds the recommendation results back to the client for display to the user. During this recommendation process, because the single recommendation model is trained on the server using all sample data, which is primarily generated from the behavioral data of a small number of highly active users, the single recommendation model is biased towards these small number of highly active users, while the recommendation effect on medium- and low-activity users is relatively mediocre. To improve the recommendation effect for these medium- and low-activity users, the single recommendation model is generally designed to be large to expand model capacity. Therefore, such a single recommendation model needs to be deployed on a high-performance server. However, when using a large single recommendation model on the server for recommendations, the server takes a long time to respond to each recommendation request, resulting in poor recommendation efficiency. Furthermore, a large single recommendation model cannot promptly respond to users' changing operational needs, resulting in a poor user experience.

[0035] To address the problems of poor recommendation effects for medium and low-activity users, and the inability of a single recommendation model on the server to respond to users' needs for timely operations, a method for deploying personalized recommendation models on the client side is provided, namely the "Thousand Machines, Thousand Models" solution.

[0036] Figure 2 This is a flowchart of the second recommendation method in the recommendation scenario of the related technology, such as Figure 2 As shown, the process of the recommended method includes the following steps:

[0037] S202: The client performs training based on the user's historical behavior data to obtain an on-device model for the user.

[0038] S204, the user triggers a recommendation request;

[0039] S206: After receiving the recommendation request, the client uses the trained on-device model to generate a recommendation result based on the recommendation request;

[0040] S208: The client displays the recommendation results on the display interface.

[0041] In the second recommendation method mentioned above, that is, in the above-mentioned thousand-machine, thousand-model solution, an on-end model for the user is trained on each client based on the user's historical data. Although the on-end model can achieve extremely personalized recommendation effects and meet the recommendation needs of medium and low-activity users, and can also respond to the user's timely changing operational needs when deployed on the client, there are some problems with the on-end model: the on-end model is trained on the client, which requires the client's computing resources, and the client's computing resources are limited. If on-end training is used, it will consume more computing power and power, affecting the user experience; in addition, the on-end model is trained based on the user's historical data. Since a single user has fewer behaviors and his interests are usually unstable, it will cause overfitting of the on-end model training to a certain extent, and the effect will be poor.

[0042] In order to solve the problems existing in the first recommendation method and the second recommendation method mentioned above, a new recommendation method is provided in an embodiment of the present application.

[0043] Figure 3 is a flow chart of a recommendation method provided according to an embodiment of the present invention. Figure 3 As shown, the process of the recommended method includes the following steps:

[0044] S302: Based on the user behavior logs recorded by the server, construct model features and positive and negative labels to obtain an initial model and the full sample data for training. The initial model is trained based on the full sample data to obtain a baseline model.

[0045] S304: The server sends the trained benchmark model to each client.

[0046] In this step, the baseline model, trained on the full sample data, is distributed to each client (i.e., all user devices) as the foundation for the personalized recommendation model. It should be noted that the baseline model can be several MB in size, but all client-side reordering uses a single copy of the foundation. After receiving the model, there's no need for repeated downloads, thus minimizing data usage.

[0047] S306: The server groups the full amount of sample data obtained in S302. The basis and requirement for grouping may be: the sample data of the same group have similar behavior patterns and are significantly different from the behavior patterns of other groups.

[0048] For example, you can group samples based on fixed user attributes, which can include a variety of information, such as user activity, gender, and age group. When grouping sample data, you can group all sample data based on one or more of these fixed user attributes. Sample data in different groups will have different fixed user attributes, while sample data in the same group will have the same fixed user attributes.

[0049] It should be noted that fixed user attributes: such as user activity, user gender, age group and other portrait tags, reflect the most basic information of users and are often strongly correlated with user behavior patterns. They need to be used as the basis for grouping with the user's permission.

[0050] For another example, user groups can be formed based on user interests, where user interests can be derived based on the user's historical behavior data. For example, a user's historical behavior data can be multiple types of data or a series of discrete data. By aggregating and classifying a series of multiple types of data, the user's interest classification can be obtained. When grouping sample data, the entire sample data can be grouped based on the user's interest classification. Sample data in different groups have different user interest classifications, while sample data in the same group have the same user interest classification.

[0051] It should be noted that grouping based on user interests is usually more effective because the user's historical operation behavior (for example, historical click behavior) contains user interests. However, the user's operation behavior (for example, the user's click sequence) is a string of discrete data and is not easy to group directly. To this end, you can first use a model such as (Self-Attentive Sequential Recommendation, abbreviated as SASRec) to convert the click behavior sequence into a representation vector (embedding), and then use a clustering method such as (K-means clustering algorithm, abbreviated as KMeans) to group in the embedding space. In addition, any reasonable user sequence encoder and clustering method can be used.

[0052] For example, grouping can be done based on the context at the time of recommendation, where the context at the time of recommendation means that the behavior patterns of the same user may be very different in different states. For example, the payment services they want to use may be very different next to the subway station in the morning and at home in the evening. Therefore, contextual features such as the position of interest (POI location for short), time, and even weather when the user initiates the recommendation request can be regarded as the basis for grouping. When grouping sample data, the entire amount of sample data can be grouped based on the context at the time of recommendation. The sample data in different groups have different contexts for recommendation, and the sample data in the same group have the same contexts for recommendation.

[0053] When grouping the full amount of sample data, generally speaking, the number of groups can be controlled below 100, because too many groups may lead to sparse data within the group, making it difficult to train an effective patch model later.

[0054] S308: The server performs training based on samples of each group to obtain a patch model (Patch model) corresponding to each group.

[0055] The server independently trains a patch model for each group of data to effectively alleviate the recommendation bias caused by highly active users. However, the amount of data in each group is relatively limited, which is often not enough to obtain a good model. In an embodiment of the present invention, the server uses the baseline model trained with the full amount of data as a base, and trains a lightweight small Patch model for each group on it. Specifically, the Patch model is a residual network structure that takes the output of the last (Fully Connected Layer, FC for short) layer of the baseline model as input, predicts a correction value and returns it to the baseline model. During training, the parameters of the baseline model are fixed, and the patch model parameters are fine-tuned using the data in the group. The size of the Patch model is usually only on the order of tens of KB and can be transmitted with RPC calls.

[0056] S310, after the service on the server is launched, the user triggers a recommendation request;

[0057] S312: The client receives the recommendation request triggered by the user and sends the recommendation request to the server;

[0058] S314, after receiving the recommendation request sent by the client, the server dynamically calculates the target patch model for the recommendation request based on the user status. That is, the server determines the target patch model corresponding to the recommendation request from the patch models corresponding to multiple groups based on the recommendation request. Among them, when the server determines the target patch model corresponding to the recommendation request from the patch models corresponding to multiple groups based on the recommendation request, a variety of methods can be adopted. For example, the server determines the user interest classification based on the recommendation request, and determines the target patch model corresponding to the target group based on the target group corresponding to the user interest classification. Therefore, the process of determining the target patch model can be similar to the above-mentioned grouping of the full amount of sample data.

[0059] S316, the server sends the determined target patch model to the client. When sending the target patch model to the client, the target patch model may be compressed before sending it to the client.

[0060] S318, the client generates a target recommendation model based on the baseline model and target patch model sent by the server. For example, the previously sent baseline model and target patch model can be directly assembled to form the final personalized end rearrangement model (ie, target recommendation model).

[0061] S320: The client generates a recommendation result for the recommendation request based on the target recommendation model, and displays the recommendation result on a display interface.

[0062] Through the above embodiment, a baseline model is first trained based on the full sample data, and the resulting baseline model is deployed on all user devices. Afterwards, the full sample data is grouped according to certain rules, and a small patch model is learned for each group of sample data on top of the baseline model. The baseline model plus the corresponding patch model generates a target recommendation model for the corresponding grouped sample data, alleviating the recommendation bias problem. After online deployment, when a user initiates a recommendation request to the server, the target patch model that is optimal for this recommendation request will be dynamically selected and sent to the client, realizing a personalized model.

[0063] Compared to the first and second recommendation methods in related technologies, the recommendation method of the present invention can effectively reduce the recommendation bias caused by highly active users, allowing users to enjoy the personalized end-to-end re-ranking model of their group. In addition, it also has the following beneficial effects:

[0064] Because the recommendation method in this embodiment does not involve on-device training, all model training occurs on the server side. This is invisible to the user, consumes no extra battery, and does not slow down the operation. Furthermore, the patch model is very small, and the resulting data usage is almost negligible.

[0065] In addition, the recommendation method of this embodiment uses group training, with each group containing a sufficient number of samples. Individual behavior is often unstable, but groups of similar people tend to behave more regularly, which can be effectively captured by the Patch model, thus avoiding the overfitting problem in training thousands of people with thousands of models.

[0066] Therefore, the recommendation method of this embodiment utilizes a novel end-to-end re-ranking recommendation model with group training and dynamic delivery. This leverages the system advantages of end-to-end re-ranking to generate personalized recommendation models for each user group, alleviating the bias problem of the recommendation system. Each group only needs to learn a lightweight patch model based on the baseline model, effectively reducing the data requirements for each group, effectively controlling the model size, significantly reducing data consumption, and achieving the effect of dynamic delivery to the client.

[0067] Based on the recommendation method provided in the above embodiment, the processing flow executed by the server and the client will be described below.

[0068] This application provides Figure 4 A method for generating recommendation information is shown. Figure 4 is a flow chart of a method for generating recommendation information according to an embodiment of the present invention. Figure 4 As shown, the method includes the following steps:

[0069] Step S402, responding to a recommendation request from a user device, and selecting a target patch model from a plurality of patch models according to the recommendation request;

[0070] Step S404: sending the target patch model to the user device, wherein the user device is configured to use a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and use the target recommendation model to generate recommendation information for the user.

[0071] It should be noted that the actual application scenarios of the embodiments of the present application can be, but are not limited to, online shopping, for example, shopping or selecting items on an e-commerce app installed on a user's device.

[0072] Optionally, the execution subject of the above steps S402 to S404 is a server, which is used to determine the user behavior pattern corresponding to the above user device (user on the client) according to the above recommendation request; and select the above target patch model from the multiple patch models according to the user behavior pattern. That is, the embodiment of the present application combines the business scenario advantages of real-time recommendation of online network products, and generates a target patch model for each group of users according to the user behavior pattern corresponding to the user device. Since the role of the patch model is to make up for the difference between different user behavior patterns and the initial recommendation model (i.e., the above benchmark model), in the embodiment of the present application, by designing a patch model mechanism, each group of users only needs to learn a lightweight patch model based on the initial recommendation model. By combining different patch models on the basis of the same initial recommendation model to form the final target recommendation model (personalized recommendation model), the recommendation deviation problem in the current recommendation scenario can be alleviated. Since the initial recommendation model has been predetermined and uniformly configured, it only needs to dynamically send the patch model to the user device in real time, so it can effectively reduce the data demand of each group of users, and can also effectively control the model volume of the target recommendation model, greatly reducing traffic consumption.

[0073] Optionally, in an embodiment of the present application, after the user device initiates a recommendation request, the recommendation request carries the user behavior pattern corresponding to the recommendation request, which is used to request the server (i.e., the cloud) to dynamically calculate and select a target patch model for the recommendation request from multiple patch models based on the user behavior pattern, and compress it and send it to the user device. The user device assembles the target patch model with a predetermined initial recommendation model to form a target recommendation model.

[0074] It should be noted that the above-mentioned initial recommendation model is deployed on all the above-mentioned user devices as a benchmark model for obtaining the target recommendation model. The volume of the above-mentioned initial recommendation model can reach several MB level, but all user devices share an initial recommendation model. After receiving the target patch model, there is no need to repeatedly download the initial recommendation model, so it will not cause excessive traffic consumption.

[0075] It should also be noted that this embodiment of the present invention does not involve model training on the user device side. All model training occurs on the server side, which does not consume additional user power or slow down the user device. Furthermore, the patch model is very small, and the resulting data usage for users is almost negligible.

[0076] In an embodiment of the present invention, by responding to a recommendation request from a user device, a target patch model is selected from multiple patch models according to the recommendation request and the target patch model is sent to the user device, wherein the user device is used to adopt a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and adopt the target recommendation model to generate recommendation information for the user, thereby achieving the purpose of generating personalized recommendation information according to the actual needs of the user, thereby achieving the technical effect of generating more accurate recommendation information according to the actual needs of the user and improving the user experience, and further solving the technical problem that the recommendation information of some users is biased and the user experience is low due to the fact that the recommendation information is mainly generated based on specific user behavior data in the existing technology.

[0077] As an optional embodiment, the step of selecting a target patch model from a plurality of patch models according to the recommendation request includes:

[0078] Determining a user behavior pattern corresponding to the user device according to the recommendation request;

[0079] The target patch model is selected from the plurality of patch models according to the user behavior pattern corresponding to the user equipment.

[0080] Optionally, after the grouped users initiate a recommendation request to the server through the user device, the server (i.e., the cloud) can parse the recommendation request to obtain the user behavior pattern corresponding to the above-mentioned user device, and then dynamically calculate the target patch model for the recommendation request from multiple patch models based on the user behavior pattern, that is, dynamically select the target patch model that is most suitable for the recommendation request or has the best recommendation result and send it to the user device.

[0081] Optionally, low-activity users and high-activity users have different user behavior patterns, where the user behavior patterns are used to reflect the user activity states of different user devices, for example, low-activity states and high-activity states. After initiating a recommendation request on the user device, the server can dynamically calculate and select the best patch model corresponding to the recommendation request from multiple candidate patch models based on the user activity state of the user device determined by the user behavior pattern, compress the best patch model, and send it to the user device. The user device directly uses the best patch model and assembles it with a predetermined initial recommendation model to generate the final personalized recommendation model.

[0082] As an optional embodiment, Figure 5 is a flow chart of an optional method for generating recommendation information according to an embodiment of the present invention. Figure 5 As shown, before selecting a target patch model from multiple patch models according to the recommendation request, the method further includes:

[0083] Step S502: Using pre-acquired training sample data to train an initial recommendation model, and deploying the initial recommendation model to all the user devices;

[0084] Step S504: grouping the training sample data to obtain multiple groups of sample data, wherein different groups of sample data correspond to different user behavior patterns;

[0085] Step S506 , training each set of the sample data according to the user behavior pattern to obtain a plurality of the patch models corresponding to the plurality of sets of the sample data.

[0086] Optionally, the above-mentioned training sample data are grouped according to predetermined grouping rules to obtain multiple groups of the above-mentioned sample data, wherein the above-mentioned predetermined grouping rules include at least one of the following: grouping rules formulated based on user attribute information, grouping rules formulated based on user interest information, and grouping rules formulated based on context feature information; the above-mentioned user attribute information is determined based on the user basic information associated with the above-mentioned user behavior pattern, the above-mentioned user interest information is determined based on the user's historical click sequence, and the above-mentioned context feature information is determined based on the time state and / or spatial state of the above-mentioned user.

[0087] As an optional embodiment, the processing process of the server can be divided into a cloud training stage and an online service stage. In the above-mentioned cloud training stage, the historical behavior feature information and positive and negative sample labels are determined according to the user behavior log recorded by the server, and the above-mentioned training sample data are generated according to the above-mentioned historical behavior feature information and the above-mentioned positive and negative sample labels; the initial recommendation model is obtained by training based on the above-mentioned training sample data, and the above-mentioned initial recommendation model is used as the baseline model; the above-mentioned training sample data are grouped and processed, and each group of sample data is trained to obtain multiple patch models. In the above-mentioned online service stage, after the user terminal initiates a recommendation request, the server (i.e., the cloud) will dynamically calculate the target patch model for the recommendation request from multiple patch models based on the user behavior pattern, and compress it and send it to the user device. The user device uses the predetermined initial recommendation model and target patch model to assemble a target recommendation model, and uses the above-mentioned target recommendation model to generate recommendation information for the user.

[0088] Through the above optional embodiments, it can be seen that in terms of implementation, as the hardware performance of user devices (for example, smart phones) continues to improve, it is possible to deploy different exclusive recommendation models (i.e., the target recommendation model in the embodiment of the present application) on different user phones. The embodiment of the present application adopts a low-cost approach to design and develop personalized target patch models for users, and uses a predetermined initial recommendation model and the above target patch model on the user device side to generate an adapted target recommendation model, that is, it is possible to deploy different target recommendation models on different user phones, and then use the above target recommendation model to generate recommendation information for users. This can reduce the recommendation bias problem caused by highly active users and improve the user experience of a large number of medium-active users, and even low-active users.

[0089] In an optional embodiment, grouping the training sample data to obtain multiple groups of sample data includes: grouping the training sample data according to a predetermined grouping rule to obtain multiple groups of sample data.

[0090] Optionally, when grouping the training sample data, the grouping basis is: similar behavior patterns exist in the recommendation requests of the same group, and are significantly different from the patterns of other groups. The predetermined grouping rules include at least one of the following: grouping rules based on user attribute information, grouping rules based on user interest information, and grouping rules based on contextual feature information; the user attribute information is determined based on user basic information associated with the user behavior pattern, the user interest information is determined based on the user's historical click sequence, and the contextual feature information is determined based on the user's current time state and / or spatial state.

[0091] Optionally, the above-mentioned user attribute information includes at least one of the following: user activity, user gender, and user age group.

[0092] In an optional embodiment, the above-mentioned training using pre-acquired training sample data to obtain the initial recommendation model includes: using a sequence recommendation algorithm to train the above-mentioned training sample data to obtain the above-mentioned initial recommendation model, wherein the above-mentioned initial recommendation model is a baseline model for generating the above-mentioned target recommendation model.

[0093] Optionally, in an embodiment of the present application, a sequence recommendation algorithm (such as the DIN model algorithm, the DeepFM model algorithm, etc.) can be used, but is not limited to, to train the above-mentioned training sample data to obtain the above-mentioned initial recommendation model, and the above-mentioned initial recommendation model is used as a benchmark model for generating the above-mentioned target recommendation model.

[0094] Optionally, the initial recommendation model is trained using the pre-acquired training sample data. This can be understood as training the initial recommendation model using the full set of training sample data. The initial recommendation model is then deployed on all of the user devices. The training sample data is then grouped and a lightweight patch model is trained for each group of data, resulting in multiple patch models. The patch model is typically only tens of KB in size and can be transmitted with RPC calls.

[0095] As an optional embodiment, Figure 6 is a flowchart of another optional method for generating recommendation information according to an embodiment of the present invention. Figure 6 As shown, before using the pre-acquired training sample data to train the initial recommendation model, the above method further includes:

[0096] Step S602: Obtain the sample user behavior log recorded locally on the server;

[0097] Step S604: determining historical behavior feature information and positive and negative sample labels based on the user behavior log;

[0098] Step S606: Generate the training sample data based on the historical behavior feature information and the positive and negative sample labels.

[0099] Optionally, the above-mentioned sample user behavior log includes: material sequence information and material attribute information that the sample user has clicked and / or exposed in a historical time period, wherein the above-mentioned material attribute information may include but is not limited to material description text, material category, etc.

[0100] Optionally, the above historical behavior feature information may be whether it is clicked, whether it is exposed, etc., and the positive or negative label of the sample is determined based on whether it is clicked or exposed.

[0101] Optionally, historical behavior feature information and positive and negative sample labels are determined based on the user behavior log recorded by the server, where typical historical behavior feature information includes material sequence information and material attribute information (such as material description text, material category, etc.) that the user has clicked or exposed in the past period of time, and the above-mentioned training sample data is generated based on the above-mentioned historical behavior feature information and the above-mentioned positive and negative sample labels.

[0102] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0103] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0104] Example 2

[0105] According to an embodiment of the present invention, another method embodiment for generating recommendation information is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0106] According to an embodiment of the present invention, Figure 7 is a flow chart of another method for generating recommendation information according to an embodiment of the present invention. Figure 7 As shown, the method includes the following steps:

[0107] Step S702: sending a recommendation request to a server, wherein the server is configured to select a target patch model from a plurality of patch models according to the recommendation request;

[0108] Step S704: receiving the target patch model sent by the server;

[0109] Step S706: using the predetermined initial recommendation model and the target patch model to generate a target recommendation model adapted to the user equipment;

[0110] Step S708: Generate recommendation information for the user using the above-mentioned target recommendation model.

[0111] It should be noted that the actual application scenarios of the embodiments of the present application can be, but are not limited to, online shopping, for example, shopping or selecting items on an e-commerce app installed on a user's device.

[0112] Optionally, the execution entity of the above steps S702 to S708 is the user device end, and the user device sends a recommendation request to the server end, wherein the above server end is used to determine the user behavior pattern corresponding to the above user device based on the above recommendation request; and select the above target patch model from the multiple above patch models based on the user behavior pattern.

[0113] The recommendation information generation method provided in the embodiment of the present application can be used, but is not limited to, to improve the recommendation experience of low-activity users in the recommendation system. Since the learning samples of the recommendation system mainly come from a small number of highly active users, the learned recommendation model often does not work well on medium and low-activity users and lacks fairness. In the embodiment of the present application, multiple patch models are learned on the server side to optimize the recommendation experience of different groups of people in different states, and personalized model deployment is achieved on the terminal with the help of smart user devices, so that different users have different recommendation systems, thereby alleviating recommendation bias.

[0114] As an optional embodiment, the user device receives the target patch model sent by the server, and uses the predetermined initial recommendation model and the target patch model to generate a target recommendation model adapted to the user device, and then uses the target recommendation model to generate recommendation information for the user. That is, the embodiment of the present application combines the business scenario advantage of real-time recommendation of online network products, and generates a target patch model for each group of users according to the user behavior pattern corresponding to the user device. Since the role of the patch model is to make up for the difference between different user behavior patterns and the initial recommendation model, in the embodiment of the present application, by designing a patch model mechanism, each group of users only needs to learn a lightweight patch model based on the initial recommendation model. By combining different patch models on the basis of the same initial recommendation model to form the final target recommendation model (personalized recommendation model), the recommendation deviation problem in the current recommendation scenario can be alleviated. Since the initial recommendation model has been predetermined and uniformly configured, it only needs to dynamically send the patch model to the user device in real time, so it can effectively reduce the data demand of each group of users, and can also effectively control the model volume of the target recommendation model, greatly reducing traffic consumption.

[0115] Optionally, in an embodiment of the present application, after the user device initiates a recommendation request, the recommendation request carries the user behavior pattern corresponding to the user device, which is used to request the server (i.e., the cloud) to dynamically calculate and select a target patch model for the recommendation request from multiple patch models based on the user behavior pattern, and compress it and send it to the user device. The user device assembles the target patch model with a predetermined initial recommendation model to form a target recommendation model.

[0116] It should be noted that the above-mentioned initial recommendation model is deployed on all the above-mentioned user devices as the basic model for obtaining the target recommendation model. The volume of the above-mentioned initial recommendation model can reach several MB level, but all user devices share an initial recommendation model. After receiving the target patch model, there is no need to repeatedly download the initial recommendation model, so it will not cause excessive traffic consumption.

[0117] It should also be noted that this embodiment of the present invention does not involve model training on the user device side. All model training occurs on the server side, which does not consume additional user power or slow down the user device. Furthermore, the patch model is very small, and the resulting data usage for users is almost negligible.

[0118] In an embodiment of the present invention, a recommendation request is sent to a server, wherein the server is used to select a target patch model from multiple patch models according to the recommendation request; the target patch model issued by the server is received; a predetermined initial recommendation model and the target patch model are used to generate a target recommendation model adapted to the user device; and the target recommendation model is used to generate recommendation information for the user, thereby achieving the purpose of generating personalized recommendation information according to the actual needs of the user, thereby achieving the technical effect of generating more accurate recommendation information according to the actual needs of the user and improving the user experience, thereby solving the technical problem that the recommendation information of some users is biased and the user experience is low due to the fact that the recommendation information is mainly generated based on specific user behavior data in the existing technology.

[0119] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0120] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0121] Example 3

[0122] According to an embodiment of the present invention, there is also provided an embodiment of a device for implementing the above-mentioned method for generating recommendation information. Figure 8 FIG. 1 is a schematic diagram of a structure of a device for generating recommendation information according to an embodiment of the present invention. Figure 8 As shown, the recommendation information generating device includes: a response module 800 and a first sending module 802, wherein:

[0123] The response module 800 is configured to respond to a recommendation request from a user device and select a target patch model from a plurality of patch models according to the recommendation request;

[0124] The first sending module 802 is used to send the target patch model to the user device, wherein the user device is used to use a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and use the target recommendation model to generate recommendation information for the user.

[0125] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0126] It should be noted that the response module 800 and the first sending module 802 correspond to steps S402 to S404 in Example 1. The examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in Example 1. It should be noted that the above modules, as part of the device, can be run in the computer terminal provided in Example 1.

[0127] Example 4

[0128] According to an embodiment of the present invention, there is also provided an embodiment of a device for implementing the above-mentioned method for generating recommendation information. Figure 9 is a structural diagram of another device for generating recommendation information according to an embodiment of the present invention. Figure 9 As shown, the recommendation information generating device includes: a second sending module 900, a receiving module 902, a first generating module 904, and a second generating module 906, wherein:

[0129] The second sending module 900 is configured to send a recommendation request to a server, wherein the server is configured to select a target patch model from a plurality of patch models according to the recommendation request;

[0130] The receiving module 902 is configured to receive the target patch model sent by the server;

[0131] The first generating module 904 is configured to generate a target recommendation model adapted to the user equipment using a predetermined initial recommendation model and the target patch model;

[0132] The second generating module 906 is configured to generate recommendation information for the user using the target recommendation model.

[0133] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0134] It should be noted that the second sending module 900, receiving module 902, first generating module 904, and second generating module 906 correspond to steps S702 to S708 in Example 2. The examples and application scenarios implemented by the four modules and the corresponding steps are the same, but are not limited to the contents disclosed in Example 2. It should be noted that the above modules, as part of the device, can be run in the computer terminal provided in Example 2.

[0135] In this embodiment, a recommendation information generation embodiment is provided to implement the above-mentioned embodiments and preferred implementations, and details that have been described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

[0136] Example 5

[0137] The embodiment of the present invention can provide a computer terminal, which can be any computer terminal device in a computer terminal group. Optionally, in this embodiment, the computer terminal can also be replaced by a terminal device such as a mobile terminal.

[0138] Optionally, in this embodiment, the computer terminal may be located in at least one network device among a plurality of network devices of a computer network.

[0139] In this embodiment, the above-mentioned computer terminal can execute the program code of the following steps in the recommendation information generation method: responding to a recommendation request from a user device, selecting a target patch model from multiple patch models according to the above-mentioned recommendation request and sending the above-mentioned target patch model to the above-mentioned user device, wherein the above-mentioned user device is used to adopt a predetermined initial recommendation model and the above-mentioned target patch model to generate an adapted target recommendation model, and adopt the above-mentioned target recommendation model to generate recommendation information for the user.

[0140] In this embodiment, the computer terminal can execute the program code of the following steps in the recommendation information generation method: sending a recommendation request to the server, wherein the server is used to select a target patch model from multiple patch models according to the recommendation request; receiving the target patch model issued by the server; using a predetermined initial recommendation model and the target patch model to generate a target recommendation model adapted to the user device; and using the target recommendation model to generate recommendation information for the user.

[0141] Optionally, Figure 10 1 is a block diagram of a computer terminal according to an embodiment of the present invention. Figure 10 As shown, the computer terminal may include: one or more (only one is shown in the figure) processors 102 and a memory 104.

[0142] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry". The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuitry may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal (or mobile device). As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0143] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the recommendation information generation method in the embodiment of the present invention. The processor executes the software programs and modules stored in the memory 104 to perform various functional applications and data processing, thereby implementing the above-mentioned recommendation information generation method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include memory remotely located relative to the processor, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0144] The display screen may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of a computer terminal (or mobile device).

[0145] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the recommendation information generation method and device in the embodiments of the present invention. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the above-mentioned recommendation information generation method. The memory may include a high-speed random access memory and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely located relative to the processor, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0146] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: respond to the recommendation request from the user device, select the target patch model from multiple patch models according to the above recommendation request, and send the above target patch model to the above user device, wherein the above user device is used to adopt a predetermined initial recommendation model and the above target patch model to generate an adapted target recommendation model, and adopt the above target recommendation model to generate recommendation information for the user.

[0147] Optionally, the processor may further execute program code of the following steps: determining a user behavior pattern corresponding to the user device according to the recommendation request; and selecting the target patch model from the plurality of patch models according to the user behavior pattern corresponding to the user device.

[0148] Optionally, the processor may also execute the program code of the following steps: using pre-acquired training sample data to train an initial recommendation model, and deploying the initial recommendation model to all of the user devices; grouping the training sample data to obtain multiple groups of sample data, wherein different groups of sample data correspond to different user behavior patterns; training each group of sample data according to the user behavior pattern to obtain multiple patch models corresponding to the multiple groups of sample data.

[0149] Optionally, the processor may further execute the program code of the following steps: using a sequence recommendation algorithm to train the training sample data to obtain the initial recommendation model, wherein the initial recommendation model is a baseline model for generating the target recommendation model.

[0150] Optionally, the processor may also execute the program code of the following steps: grouping the training sample data according to predetermined grouping rules to obtain multiple groups of sample data, wherein the predetermined grouping rules include at least one of the following: grouping rules formulated based on user attribute information, grouping rules formulated based on user interest information, and grouping rules formulated based on context feature information; the user attribute information is determined based on user basic information associated with the user behavior pattern, the user interest information is determined based on the user's historical click sequence, and the context feature information is determined based on the user's current time state and / or spatial state.

[0151] Optionally, the processor may also execute the program code of the following steps: obtaining a sample user behavior log recorded locally on the server, wherein the sample user behavior log includes: material sequence information and material attribute information that the sample user has clicked and / or exposed during a historical time period; determining historical behavior feature information and positive and negative sample labels based on the user behavior log; and generating the training sample data based on the historical behavior feature information and the positive and negative sample labels.

[0152] Optionally, the processor may also execute the program code of the following steps: sending a recommendation request to the server, wherein the server is used to select a target patch model from multiple patch models according to the recommendation request; receiving the target patch model issued by the server; using a predetermined initial recommendation model and the target patch model to generate a target recommendation model adapted to the user device; and using the target recommendation model to generate recommendation information for the user.

[0153] An embodiment of the present invention provides a solution for generating recommendation information. By responding to a recommendation request from a user device, a target patch model is selected from multiple patch models according to the recommendation request and sent to the user device. The user device is configured to use a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and use the target recommendation model to generate recommendation information for the user. This achieves the purpose of generating personalized recommendation information based on the actual needs of the user, thereby achieving the technical effect of generating more accurate recommendation information based on the actual needs of the user and improving the user experience. This further solves the technical problem that, due to the prior art of generating recommendation information mainly based on specific user behavior data, some user recommendation information has deviations and the user experience is low.

[0154] It can be understood by those skilled in the art that Figure 10The structure shown is for illustration only, and the computer terminal may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a handheld computer, a mobile Internet device (MID), a PAD, or other terminal devices. Figure 10 It does not limit the structure of the above electronic device. For example, the computer terminal may also include Figure 10 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 10 Different configurations shown.

[0155] A person skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0156] The embodiment of the present invention further provides a non-volatile storage medium. Optionally, in this embodiment, the non-volatile storage medium can be used to store the program code executed by the recommendation information generation method provided in the first embodiment.

[0157] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0158] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: responding to a recommendation request from a user device, selecting a target patch model from multiple patch models according to the recommendation request and sending the target patch model to the user device, wherein the user device is used to adopt a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and adopt the target recommendation model to generate recommendation information for the user.

[0159] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for executing the following steps: sending a recommendation request to a server, wherein the server is configured to select a target patch model from multiple patch models based on the recommendation request; receiving the target patch model issued by the server; using a predetermined initial recommendation model and the target patch model to generate a target recommendation model adapted to the user device; and using the target recommendation model to generate recommendation information for the user.

[0160] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0161] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0162] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0163] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0164] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0165] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program codes.

[0166] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for generating recommendation information, characterized in that: include: The server responds to a recommendation request from a user device and selects a target patch model from a plurality of patch models according to the recommendation request, wherein the plurality of patch models are trained by the server based on respective group samples, and user interests of different group samples are different; The server sends the target patch model to the user device, wherein the user device is used to adopt a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and adopt the target recommendation model to generate recommendation information for the user, wherein the initial recommendation model is obtained by the server using a sequential recommendation algorithm for training based on the full amount of sample data, and the initial recommendation model is a baseline model for generating the target recommendation model, and the baseline model is a base shared by all user devices.

2. The method according to claim 1, characterized in that The selecting a target patch model from a plurality of patch models according to the recommendation request includes: determining a user behavior pattern corresponding to the user device according to the recommendation request; The target patch model is selected from the plurality of patch models according to a user behavior pattern corresponding to the user equipment.

3. The method according to claim 1, characterized in that Before selecting a target patch model from a plurality of patch models according to the recommendation request, the method further includes: Using pre-acquired training sample data to train an initial recommendation model, and deploying the initial recommendation model to all of the user devices; The training sample data is grouped to obtain multiple groups of sample data, wherein different groups of sample data correspond to different user behavior patterns; Each group of the sample data is trained according to the user behavior pattern to obtain a plurality of patch models corresponding to the plurality of groups of the sample data.

4. The method according to claim 3, characterized in that The grouping of the training sample data to obtain multiple groups of sample data includes: The training sample data are grouped according to predetermined grouping rules to obtain multiple groups of sample data, wherein the predetermined grouping rules include at least one of the following: grouping rules formulated based on user attribute information, grouping rules formulated based on user interest information, and grouping rules formulated based on context feature information; the user attribute information is determined based on user basic information associated with the user behavior pattern, the user interest information is determined based on the user's historical click sequence, and the context feature information is determined based on the user's current time state and / or spatial state.

5. The method according to claim 1, wherein Before using the pre-acquired training sample data to train an initial recommendation model, the method further includes: Obtaining a sample user behavior log locally recorded on the server, wherein the sample user behavior log includes: material sequence information and material attribute information that the sample user has clicked and / or viewed within a historical time period; Determining historical behavior feature information and positive and negative sample labels based on the user behavior log; The training sample data is generated according to the historical behavior feature information and the positive and negative sample labels.

6. A method for generating recommendation information, characterized in that: include: Sending a recommendation request to a server, wherein the server is configured to select a target patch model from a plurality of patch models according to the recommendation request, wherein the plurality of patch models are trained by the server based on respective group samples, and user interests of different group samples are different; Receiving the target patch model sent by the server; Generate a target recommendation model adapted to the user device using a predetermined initial recommendation model and the target patch model, wherein the initial recommendation model is trained by the server using a sequential recommendation algorithm based on the full sample data. The initial recommendation model is a baseline model for generating the target recommendation model, and the baseline model is a base shared by all user devices. The target recommendation model is used to generate recommendation information for the user.

7. A device for generating recommendation information, characterized in that: include: A response module, configured for the server to respond to a recommendation request from a user device and select a target patch model from a plurality of patch models according to the recommendation request, wherein the plurality of patch models are trained by the server based on respective group samples, and the user interests of different group samples are different; The first sending module is used for the server to send the target patch model to the user device, wherein the user device is used to adopt a predetermined initial recommendation model and the target patch model to generate an adapted target recommendation model, and adopt the target recommendation model to generate recommendation information for the user, wherein the initial recommendation model is obtained by the server using a sequential recommendation algorithm trained based on the full amount of sample data, the initial recommendation model is a baseline model for generating the target recommendation model, and the baseline model is a base shared by all user devices.

8. A device for generating recommendation information, characterized in that: include: a second sending module, configured to send a recommendation request to a server, wherein the server is configured to select a target patch model from a plurality of patch models according to the recommendation request, wherein the plurality of patch models are trained by the server based on respective group samples, and user interests of different group samples are different; A receiving module, configured to receive the target patch model sent by the server; A first generation module is configured to generate a target recommendation model adapted to the user device using a predetermined initial recommendation model and the target patch model, wherein the initial recommendation model is trained by the server using a sequential recommendation algorithm based on the full amount of sample data. The initial recommendation model is a baseline model for generating the target recommendation model, and the baseline model is a base shared by all user devices; The second generating module is used to generate recommendation information for the user by using the target recommendation model.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is executed, the device where the non-volatile storage medium is located is controlled to execute the recommendation information generating method according to any one of claims 1 to 6.

10. An electronic device comprising a memory and a processor, characterized in that: The memory stores a computer program, and the processor is configured to run the computer program to execute the recommendation information generating method according to any one of claims 1 to 6.

Images