Information recommendation method and device based on edge intelligence and storage medium

Abstract

The invention discloses an information recommendation method and device based on edge intelligence and a storage medium. The method comprises the following steps: firstly, training an information recommendation model which is uniformly used for recommending meals to each user; then, deploying the information recommendation model in the edge device of each user; furthermore, the edge device of the user performs meal recommendation through the information recommendation model, and acquires the updated health condition (updated health information) after the user has dinner according to the meal recommended by the information recommendation model. And updating parameters of at least part of network structures in an information recommendation model in the edge device based on the updated health condition. According to the method, the information recommendation model can be updated in a personalized manner according to the real-time health condition of the user, so that the information recommendation model performs meal recommendation aiming at enabling the health condition of the user to tend to be good, and compared with the prior art, meal recommendation can be performed in combination with the personal health condition of the user.

Description

Technical Field

[0001] This application relates to the field of intelligent recommendation technology for catering, and in particular to an information recommendation method, device and storage medium based on edge intelligence. Background Technology

[0002] Currently, in canteens or other dining scenarios, food service providers can recommend dishes to users, allowing them to make informed decisions when ordering.

[0003] In existing technologies, meal recommendations for users are usually based on their historical dining preferences and ordering behavior, while neglecting the impact on users' health. There is no effective technical solution in the current technology for making meal recommendations that take into account users' health conditions.

[0004] There is currently no effective solution to the technical problem that existing technologies ignore the impact on users' health when recommending meals. Summary of the Invention

[0005] The embodiments of this disclosure provide an information recommendation method, apparatus, and storage medium based on edge intelligence, to at least solve the technical problem in the prior art that the impact on users' health is ignored in meal recommendations.

[0006] According to one aspect of the present disclosure, an information recommendation method is provided. The method is used in a user-held edge device, in which an information recommendation model pre-trained by a server device is deployed. The method includes: acquiring first health information of the user within a first preset time period, the first health information representing the user's health status within the first preset time period; determining first recommendation information based on the first health information within the first preset time period and the information recommendation model deployed in the user's edge device, the first recommendation information being meal information recommended to the user, wherein the information recommendation model includes: a BERT network, a public network, a personalized network, and a prediction network; acquiring updated health information of the user when the user performs business based on the first recommendation information; determining reward information corresponding to the first recommendation information based on the updated health information; and updating the parameters of the personalized network in the information recommendation model through reinforcement learning based on the reward information, while freezing the parameters of the BERT network, the public network, and the prediction network in the information recommendation model, the updated information recommendation model being used to continue recommending information to the user.

[0007] According to another aspect of the present disclosure, a storage medium is also provided, the storage medium including a stored program, wherein, when the program is executed, a processor performs any of the methods described above.

[0008] According to another aspect of the present disclosure, an information recommendation device is also provided, comprising: an acquisition module, configured to acquire first health information of a user within a first preset time period, the first health information representing the user's health status within the first preset time period; a recommendation information determination module, configured to determine first recommendation information based on the first health information within the first preset time period and an information recommendation model deployed on the user's edge device, the first recommendation information being recommended meal information for the user, wherein the information recommendation model includes: a BERT network, a public network, a personalized network, and a prediction network, the information recommendation model being pre-trained by a server device; a reward information determination module, configured to acquire updated health information of the user when the user performs business operations based on the first recommendation information, and determine reward information corresponding to the first recommendation information based on the updated health information; and an update module, configured to update the parameters of the personalized network in the information recommendation model through reinforcement learning based on the reward information, while freezing the parameters of the BERT network, the public network, and the prediction network in the information recommendation model, the updated information recommendation model being used to continue recommending information to the user.

[0009] According to another aspect of the present disclosure, an information recommendation device is also provided, comprising: a processor; and a memory connected to the processor, configured to provide the processor with instructions for processing the following steps: acquiring first health information of a user within a first preset time period, the first health information representing the user's health status within the first preset time period; determining first recommendation information based on the first health information within the first preset time period and an information recommendation model deployed on the user's edge device, the first recommendation information being meal information recommended to the user, the information recommendation model being pre-trained by a server device, wherein the information recommendation model includes: a BERT network, a public network, a personalized network, and a prediction network; acquiring updated health information of the user when the user performs business operations based on the first recommendation information; determining reward information corresponding to the first recommendation information based on the updated health information; and updating the parameters of the personalized network in the information recommendation model through reinforcement learning based on the reward information, while freezing the parameters of the BERT network, the public network, and the prediction network in the information recommendation model, the updated information recommendation model being used to continue recommending information to the user.

[0010] In this embodiment, a unified information recommendation model is first trained to recommend meals to each user. Then, this information recommendation model is deployed on each user's edge device. Simultaneously, while recommending meals using the information recommendation model, the user's edge device obtains updated health information (updated health information) after consuming the recommended meals, and updates the parameters of the personalized network in the information recommendation model on the edge device based on this updated health information. Therefore, this method can personalize the information recommendation model according to the user's real-time health status, ensuring that the information recommendation model recommends meals with the aim of improving the user's health. Compared to existing technologies, this method can incorporate the user's individual health status into meal recommendations. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this disclosure and form part of this application, illustrate exemplary embodiments of the present disclosure and are used to explain the disclosure, but do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a hardware structure block diagram of a computing device for implementing the method described in Embodiment 1 of this disclosure; Figure 2 This is a flowchart illustrating the information recommendation method according to the first aspect of Embodiment 1 of this disclosure; Figure 3 This is a schematic diagram illustrating a process for updating the information recommendation model in the edge devices of each user, as provided in Embodiment 1 of this disclosure. Figure 4A This is a schematic diagram of the structure of an information recommendation model provided in Embodiment 1 of this disclosure; Figure 4B This is a schematic diagram of a process for evaluating the value information output by a network, provided in Embodiment 1 of this disclosure. Figure 4C This is a schematic diagram illustrating the interaction process between the edge device and the server device provided in Embodiment 1 of this disclosure; Figure 5 This is a schematic diagram of an information recommendation device according to the first aspect of Embodiment 2 of this disclosure; and Figure 6 This is a schematic diagram of an information recommendation device according to the first aspect of Embodiment 3 of this disclosure. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0013] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0014] Example 1 According to this embodiment, an information recommendation method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0015] The method embodiments provided in this example can be executed in a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computing device for implementing an information recommendation method is shown. Figure 1 As shown, a computing device may include one or more processors (processors may include, but are not limited to, microprocessors such as MCUs or programmable logic devices such as FPGAs), a memory for storing data, a transmission device for communication functions, and an input / output interface. The memory, transmission device, and input / output interface are connected to the processor via a bus. In addition, it may also include a display, keyboard, and cursor control device connected to the input / output interface. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, a computing device may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0016] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element in a computing device. As involved in the embodiments of this disclosure, the data processing circuits serve as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0017] The memory can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the information recommendation method in this embodiment of the present disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the information recommendation method of the aforementioned application. The memory 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 instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the computing device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0018] The transmission device is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the computing device's communication provider. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0019] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows users to interact with the user interface of the computing device.

[0020] It should be noted here that, in some optional embodiments, the above... Figure 1 The computing device shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 1 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computing devices.

[0021] According to a first aspect of this embodiment, an information recommendation method is provided. This method can be implemented by a user's edge device, and some technical features of the method can be implemented by a server device. The hardware architecture of both the edge device and the server device can refer to the hardware architecture of the computing devices mentioned above. The edge device is deployed with an information recommendation model pre-trained by the server device. Figure 2 A flowchart illustrating the method is shown below. (Refer to...) Figure 2 As shown, the method includes: S202: Obtain the user's first health information within a first preset time period, the first health information being used to represent the user's physical health status; S204: Based on the first health information within the first preset time period, determine the first recommended information based on the information recommendation model. The first recommended information is meal information recommended to the user. The information recommendation model includes: BERT network, public network, personalized network and prediction network. S206: When a user performs business operations based on the first recommendation information, obtain the user's updated health information, and determine the reward information corresponding to the first recommendation information based on the updated health information; and S208: Based on the reward information, update the parameters of the personalized network in the information recommendation model through reinforcement learning, while freezing the parameters of the BERT network, the public network, and the prediction network in the information recommendation model.

[0022] refer to Figure 3 As shown, the server device can pre-train an initial information recommendation model, which is used to recommend meal information to users. After the initial information recommendation model is trained by the server device, it can be deployed on each user's edge device. This specification does not limit the form of the edge device; for example, the edge device can be a mobile phone, tablet computer, smart wearable device, etc.

[0023] The edge device can obtain the user's first health information within a first preset time period (S202).

[0024] Then, the edge device can determine the first recommendation information (S204) based on the first health information and an information recommendation model. The first recommendation information is meal information recommended to the user. The edge device can then recommend the first recommendation information to the user.

[0025] The specific duration of the first preset time period can be set in advance according to business needs; for example, the duration of the first preset time period can be set to 7 days, 14 days, or 21 days. In this embodiment, health information (including first health information and updated health information and second health information, which will be mentioned later) can be used to represent the user's physical health status. The information recommendation model includes: a BERT network, a public network, a personalized network, and a prediction network. The BERT network is used to extract features from the sequence-based health information; the personalized network is used to learn the relationship between the user's own health status and the recipes recommended for that user; the public network is used to learn recipe recommendation knowledge at the level of all users; and the prediction network is used to provide the first recommendation information. The personalized network and the public network will be described in detail later.

[0026] When a user performs a service based on the first recommendation information, the edge device obtains the user's updated health information and determines the reward information corresponding to the first recommendation information based on the updated health information (S206).

[0027] Finally, based on the reward information, the edge device updates the parameters of the personalized network in the information recommendation model through reinforcement learning (S208), while freezing the parameters of the BERT network, the public network, and the prediction network in the information recommendation model. The updated information recommendation model is used to continue to recommend information to users who own the edge device.

[0028] The aforementioned user's execution of business based on the first recommendation information refers to the user choosing to dine on the meal recommended in the first recommendation information. In other words, when a user chooses a meal recommended in the first recommendation information, the edge device can obtain the user's updated health information. This updated health information is used to update the parameters of the personalized network in the information recommendation model. Specifically, the edge device determines the reward information corresponding to the first recommendation information based on the updated health information, and updates the parameters of the personalized network in the information recommendation model using reinforcement learning based on this reward information. However, if the user does not choose the meal corresponding to the first recommendation information, but instead dines on another meal according to their own preference, then this sample (and therefore the corresponding updated health information) will not be obtained to update the parameters of the personalized network in the information recommendation model.

[0029] Continue to refer to Figure 3 As shown, with Figure 3Taking user i as an example, after the edge device i held by user i deploys the initial information recommendation model, the edge device i simultaneously recommends information (meal information) to user i using its deployed information recommendation model and updates the information recommendation model based on user i's real-time health information feedback. Therefore, corresponding to step S206, after the edge device i generates the first recommendation information through the information recommendation model and recommends the first recommendation information to user i, the edge device i can determine whether user i has selected the meal recommended by the first recommendation information. If so, the edge device i can obtain user i's updated health information. For example, the edge device i can obtain user i's updated health information the next time it makes an information recommendation for user i.

[0030] Then, edge device i can determine the reward information corresponding to the first recommendation information based on the updated health information, and update the parameters of the personalized network in information recommendation model i based on the reward information through reinforcement learning.

[0031] The following example illustrates the process of updating the information recommendation model and the process of recommending information using the information recommendation model in this embodiment.

[0032] For example, the first health information mentioned above was obtained when making information recommendations to the user for the sth time, and this first health information is referred to as H. s H s Includes health information within a first preset time period (e.g., from day l to day k) [h l , ..., h k ]. h l This is the health information for day l.

[0033] Edge devices, based on an information recommendation model, determine the first health information H. s The corresponding first recommendation information R can be determined. s Edge devices will provide the first recommendation information R. s After being displayed to the user, it can be determined whether the user dined according to the first recommended information. If the user dined according to the first recommended information, updated health information can be obtained. The time period corresponding to the updated health information is the time period after the first preset time period. For example, updated health information H up It can be [h] k+1 , ..., h k+t (For example, health information from day k+1 to day k+t), the specific value of t can be preset, for example, the shortest health information after the update can be the user's health information within h days. k+1 Health information for the day.

[0034] Then, the edge device can adjust the updated health information H.up Determine the corresponding reward information r s And based on the reward information r s The parameters of the personalized network in the information recommendation model are updated. The determination of reward information and the method for updating the parameters of the personalized network in the information recommendation model based on the reward information will be explained in detail later.

[0035] refer to Figure 4A As shown, the information recommendation model can include a BERT network, a common network, a personalized network, a fully connected layer, and a Softmax layer, where the fully connected layer and the Softmax layer constitute the prediction network. The personalized network can be composed of a neural network (e.g., a feedforward neural network). Figure 4A The overall structure of the information recommendation model is shown. The overall structure of the information recommendation model is consistent during the training phase on the server device and the use and update phase on the edge device. When the edge device updates the information recommendation model deployed locally based on the reward information, it can update only the parameters of the personalized network, while the parameters of the BERT network, the public network, and the prediction network can remain unchanged.

[0036] Continue to refer to Figure 4A As shown, the edge device will send the first health information H s The information is input into the information recommendation model. The BERT network outputs a first semantic feature corresponding to the first health information. This first semantic feature is then input into a public network, which outputs a corresponding second semantic feature based on the first semantic feature. The second semantic feature is then input into a personalized network, which outputs a corresponding third semantic feature based on the second semantic feature. This third semantic feature is then input into a fully connected layer, followed by a Softmax layer. The Softmax layer outputs the probability corresponding to each option. Each option can be a meal combination (i.e., a recipe), and the option with the highest probability is selected as the first recommendation.

[0037] As described in the background section, existing technologies typically recommend meals to users based on their historical dining preferences and ordering behavior, neglecting to consider the impact on users' health. Currently, there is no effective technical solution for recommending meals to users by taking into account their health conditions.

[0038] Therefore, in this method, a unified information recommendation model is first trained to recommend meals to each user. Then, the information recommendation model is deployed on each user's edge device. Simultaneously, while recommending meals through the information recommendation model, the user's edge device obtains updated health information (updated health information) after consuming the recommended meals, and updates the parameters of the personalized network in the information recommendation model on the edge device based on this updated health information. Thus, this method can personalize the information recommendation model according to the user's real-time health status, making the information recommendation model recommend meals with the aim of improving the user's individual health condition. Compared to existing technologies, this method can better incorporate the user's individual health status into meal recommendations.

[0039] Optionally, the operation of obtaining the user's first health information within a first preset time period includes obtaining at least one of the user's emotional information, heart rate information, blood pressure information, blood sugar information, height information, and weight information within the first preset time period.

[0040] In other words, the dimensions of health information (including but not limited to first health information, second health information, and updated health information) in this embodiment can include emotional information, heart rate information, blood pressure information, blood sugar information, height information, and weight information. The edge device can acquire health information through user input each time it obtains it, or it can acquire user health information through real-time measurement. For example, the edge device can identify the user's facial image itself or through a server device to determine the user's current emotional information, and it can also acquire the user's heart rate information through a smart wearable device. For another example, this embodiment can recommend information to users in a cafeteria scenario. This allows for the setting up of measurement devices for heart rate, blood pressure, blood sugar, weight, and height, respectively. The edge device can acquire the heart rate, blood pressure, blood sugar, and weight information measured by these devices through the Internet of Things. For adults, height usually does not change, so height information only needs to be acquired once initially when the user is an adult. For minors, the edge device can acquire height information at regular intervals, rather than acquiring it every time health information is determined.

[0041] Optionally, the operation of determining the reward information corresponding to the first recommendation information based on the updated health information includes: Based on the updated health information, a first reward value is determined, and based on the difference between the first health information and the updated health information, a second reward value is determined; based on the first reward value and the second reward value, reward information corresponding to the first recommendation information is determined.

[0042] In other words, the reward information in this embodiment can be formulated in two ways: 1. If the updated health information indicates that the user's body is at a relatively healthy level, then the specific value of the reward information is higher. 2. If the physical condition corresponding to the updated health information is better than the physical condition corresponding to the first health information, then the specific value of the reward information is higher.

[0043] The following is an example of calculating reward information.

[0044] Continuing with the example above, the edge device can adjust its health information based on the updated health information H. up Determine the reward information r corresponding to the first recommendation. s Among these, the first reward value r corresponding to the first recommendation information can be determined. s-1 Second reward value r s-2 Then we can apply the first reward value r. s-1 Second reward value r s-2 Perform a summation, or a weighted summation, to obtain the reward information r. s .

[0045] The first reward value can be determined solely based on the updated health information, while the second reward value can be determined using both the updated and first health information. Since both the first and updated health information may contain health data from multiple days, the average or mode of the first and updated health information can be used to determine the first and second reward values, respectively.

[0046] For the first reward value r s-1 Based on the updated health information, the updated health information H can be determined. up The corresponding time period determines whether the user is in a healthy or unhealthy state. When the user is in a healthy state, r s-1 =+10, when the user is in an unhealthy state, r s-1 =-10.

[0047] For the second reward value r s-2 If it is determined that the physical condition corresponding to the updated health information is better than the physical condition corresponding to the updated health information, then r s-2 =+10. Conversely, if it is determined that the physical state corresponding to the updated health information is not better than the physical state corresponding to the first health information, then r s-2 =-10.

[0048] In this embodiment, the normal range for each dimension of health information can be predetermined. For example, heart rate between 60-100 beats / minute is considered normal; diastolic blood pressure between 91 mmHg and 139 mmHg is considered normal, and systolic blood pressure between 89 mmHg and 61 mmHg is considered normal. Blood glucose (in this example, only the user's fasting blood glucose is used) between 3.9 mmol / L and 6.1 mmol / L is considered normal. For height and weight information, the user's BMI can be determined based on these data. A BMI of 18.5 kg / m² is considered normal. 2 -24kg / m 2 The values ​​between these ranges are within the normal range. For emotional information, when the emotional information falls within a preset emotion, the emotion is considered to be within the normal range. Assuming emotional information can be categorized into several types (such as joy, calmness, anger, and sadness), the preset emotions can be set to joy and calmness.

[0049] If the edge device determines that the data in the first preset quantity dimension of the updated health information is not within the normal range, it can determine that the updated health information indicates that the user is in an unhealthy state. If it determines that the data in the first preset quantity dimension of the updated health information is within the normal range, it can determine that the health information indicates that the user is in a healthy state.

[0050] In addition, the edge device determines the second reward value r s-2 In some cases, the updated health information can indicate that the user is in a healthy state, while the first health information indicates either an unhealthy or healthy state. This is considered a case where the updated health information indicates a better physical state than the first health information. Conversely, if both the updated and first health information indicate an unhealthy state, and the updated health information shows a number of data points closer to the normal range than the first health information (a number exceeding a second preset threshold), then the edge device can also determine that the updated health information indicates a better physical state than the first health information. Both the first and second preset thresholds can be pre-set.

[0051] Optionally, based on the reward information, the parameters of the personalized network in the information recommendation model are updated through reinforcement learning. This includes: inputting first health information and first recommendation information into the evaluation network to obtain first value information; inputting second health information into the information recommendation model to obtain second recommendation information, wherein the second health information is health information obtained within a second preset time period when the user performs business based on the first recommendation information; inputting the second health information and second recommendation information into the evaluation network to obtain second value information; updating the evaluation network based on the reward information, the first value information, and the second value information; and updating the parameters of the personalized network in the information recommendation model based on the first value information, the first health information, and the first recommendation information.

[0052] Specifically, in this embodiment, the parameters of the personalized network in the information recommendation model are updated using the Actor-Critic approach in reinforcement learning. To achieve this update of the personalized network parameters via the Actor-Critic approach, an evaluation network for determining the value information Q also needs to be deployed in the edge device. (See reference...) Figure 4B As shown, this evaluation network is used to output value information corresponding to the corresponding recommendation information based on health information and recommendation information. That is, in this embodiment, health information is treated as a state, and recommendation information is treated as an action.

[0053] Continuing with the previous example, the edge device uses the first health information H... s Recommending the first recommendation information to users R s And determine the user based on the first recommendation information R s After the meal, updated health information H was obtained. up And based on the updated health information H up The reward information r corresponding to the first recommendation information has been determined. s .

[0054] Then, the edge device determines the second health information H. y Second health information H y This could refer to the user's health information obtained when the edge device makes its information recommendation for the yth time, where y > s, for example, y = s + 1. The second health information H... y It can contain health information within a second preset time period [h] p , ..., h q (Health information from day p to day q), where q can be k+t, meaning the last day of the second preset time period can coincide with the last day of the time period corresponding to the updated health information. On this day, the reward information corresponding to the first recommendation information can be determined, and the information recommendation model can be updated based on this reward information. The edge device will transmit the second health information H...y Inputting this information into the information recommendation model yields the second recommendation information R. y Then, the edge device can use the first recommendation information R s The input is fed into the evaluation network to obtain the first recommendation information R. s The corresponding first value information Q s and the second recommendation information R y The input is fed into the evaluation network to obtain the second recommendation information R. y The corresponding second value information Q y It should be noted that the second recommendation information R y This is for training purposes only and should not be recommended for users.

[0055] Then, the edge device can determine the timing difference error corresponding to the first recommendation information. The formula is shown below: (1) In formula (1) above, γ is a preset discount factor, and Q s To match the first recommended information R s The corresponding first value information, Q y To match the second recommendation information R y The corresponding second value information, To match the first recommended information R s The corresponding reward information. Edge devices can minimize this timing difference error. The evaluation network is trained to achieve the desired result. The specific formula for parameter updates is: (2) (3) in, This indicates the evaluation network. Let represent the parameters used to evaluate the network during the s-th information recommendation. α is the learning rate. Formula (2) represents the evaluation parameters based on the first health information H. s And the first recommended information R s The parameters of the evaluation network are differentiated. Equation (3) represents the updated parameters obtained by updating the parameters of the evaluation network using the gradient descent algorithm. .

[0056] Finally, the edge device uses the first health information H s First value information Q s And the first recommendation information R s The parameters of the personalized network in the information recommendation model are updated using the following formula: (4) (5) in, Let θ represent the information recommendation model, and let θ represent the parameters of the information recommendation model. s Let be the parameters of the information recommendation model when making information recommendations for the sth time. β is the learning rate. Formula (4) represents the parameters based on the first health information H. s And the first recommended information R s The derivative of Re() in the information recommendation model is obtained. Equation (5) represents the updating of the parameters of the personalized network in the information recommendation model using the gradient ascent algorithm. .

[0057] Optionally, it also includes: Training samples are obtained through a server device. These training samples include health information samples and corresponding recommendation information samples. The health information samples are then input into the information recommendation model to be trained through the server device to determine the predicted recommendation information output by the model. Finally, the initial information recommendation model is obtained by training the model to be trained through the server device with the goal of minimizing the difference between the predicted recommendation information and the recommended information samples.

[0058] It's important to note that before deploying the information recommendation model on edge devices, the initial model needs to be trained uniformly by the server device. The server device can acquire several training samples, which include health information samples from different users and corresponding recommendation information samples. These training samples can be obtained through manual annotation. Specifically, a large number of user health information samples can be acquired, and then, through manual annotation, suitable meal information (which is beneficial to the health of the corresponding population) can be determined, thus labeling the health information samples with corresponding recommendation information samples. Then, using these training samples, the information recommendation model to be trained can undergo supervised training to obtain the initial model.

[0059] Optionally, the method further includes: constructing an auxiliary training model based on the information recommendation model using a server device; the auxiliary training model includes a BERT network, a public network, and a prediction network; obtaining target recommendation information generated by each user's edge device with a reward value higher than a preset reward score, and target health information corresponding to the target recommendation information, using the server device; inputting the target health information into the auxiliary training model using the server device to obtain predicted recommendation information corresponding to the target health information, and updating the parameters of the public network in the auxiliary training model through supervised learning with the training objective of minimizing the difference between the predicted recommendation information corresponding to the target health information and the target recommendation information, while freezing the parameters of the BERT network and the prediction network in the auxiliary training model; and synchronizing the updated parameters of the public network to each user's edge device using the server device.

[0060] Specifically, from Figure 4A As can be seen, the execution order within the information recommendation model is as follows: First, the BERT network extracts features from the user's health information over a period of time. Then, the extracted features are input into a public network, which can be the encoder of a large model (LLM) or an open-source large model (LLM) that only outputs token features. Next, the server device inputs the features further extracted by the public network into a personalized network. Finally, the corresponding recommendation information is determined through fully connected layers and softmax layers. The public network possesses relatively complete information extraction and reasoning capabilities, while the personalized network is a simpler neural network compared to the public network. In this embodiment, the public network is responsible for learning more universally applicable recommendation information suitable for each user's health condition, while the personalized network is responsible for learning personalized recommendation information adapted to the specific user's health condition.

[0061] Before deploying the information recommendation model on users' edge devices, the server device pre-trains the entire model using supervised learning, enabling it to recommend meals based on users' health information. After deployment on the edge devices, the edge devices only update the parameters of the personalized network of the information recommendation model, allowing it to make personalized meal recommendations based on the user's individual health condition. To further enhance the comprehensive information recommendation capabilities of the edge device-side model, the server device periodically collects optimal recommendation information from each user's edge device and updates the public network based on this information, giving the public network broader and more universal information recommendation capabilities. The server then synchronizes the updated public network parameters to the edge devices.

[0062] refer to Figure 4CAs shown, the server device can obtain recommendation information with reward values ​​higher than a preset reward score from each user's edge device at predetermined intervals (e.g., every week / month). This target recommendation information is used to construct training samples for updating the parameters of the public network. The training samples contain the target recommendation information and the target health information corresponding to the target recommendation information. The server device can construct an auxiliary training model, which includes a BERT network, a public network, fully connected layers, and a Softmax layer. In other words, the auxiliary training model is the initial information recommendation model without the personalized network. This is because the parameters of the personalized network are constantly updated on the edge devices, and the public network plays a relatively dominant role in the information recommendation model. Therefore, the auxiliary training model does not need a personalized network; it mainly updates the parameters of the public network.

[0063] Furthermore, after the server device constructs training samples including target recommendation information and target health information, it performs supervised learning training on the auxiliary training model based on the corresponding training samples. During the training process, only the parameters of the public network are updated, while the parameters of the BERT network, fully connected layer and Softmax layer are frozen.

[0064] In this way, the edge device mainly updates the parameters of the personalized network, while the server device mainly updates the parameters of the public network. This enables the information recommendation model of the edge device to have both the ability to make personalized recommendations based on the user's own health status and a more comprehensive information recommendation capability.

[0065] In addition, refer to Figure 1 As shown, according to a second aspect of this embodiment, a storage medium is provided. The storage medium includes a stored program, wherein, when the program is executed, a processor performs any of the methods described above.

[0066] Therefore, the initial information recommendation model deployed on the edge device in this embodiment is a relatively universal model for recommending information (meal recommendations) to users based on their health information. In this embodiment, the edge device will subsequently perform local updates, continuously updating the information recommendation model by utilizing real-time changes in the user's health status on the edge device and through reinforcement learning. The goal of updating the information recommendation model is to improve the user's own health status. Thus, this method can personalize and optimize the information recommendation model according to the user's individual physical condition and health status, thereby providing more accurate meal recommendations based on the user's health condition.

[0067] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0068] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they 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, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0069] Example 2 Figure 5 An information recommendation apparatus according to a first aspect of this embodiment is shown, which corresponds to the method described according to the first aspect of Embodiment 1. (See reference...) Figure 5 As shown, the information recommendation device includes: an acquisition module 510, used to acquire first health information of a user within a first preset time period, the first health information representing the user's health status within the first preset time period; a recommendation information determination module 520, used to determine first recommendation information based on the first health information within the first preset time period and an information recommendation model deployed on the user's edge device, the first recommendation information being recommended meal information for the user, the information recommendation model being pre-trained by a server device, wherein the information recommendation model includes: a BERT network, a public network, a personalized network, and a prediction network; a reward information determination module 530, used to acquire the user's updated health information when the user performs business based on the first recommendation information, and determine the reward information corresponding to the first recommendation information based on the updated health information; and an update module 540, used to update the parameters of the personalized network in the information recommendation model through reinforcement learning based on the reward information, while freezing the parameters of the BERT network, the public network, and the prediction network in the information recommendation model, the updated information recommendation model being used to continue to recommend information to the user.

[0070] Optionally, the acquisition module 510 is used to acquire at least one of the following information during a first preset time period: the user's emotional information, heart rate information, blood pressure information, blood sugar information, height information, and weight information.

[0071] Optionally, the reward information determination module 530 is used to determine a first reward value based on the updated health information, and to determine a second reward value based on the difference between the first health information and the updated health information; and to determine reward information corresponding to the first recommendation information based on the first reward value and the second reward value.

[0072] Optionally, the update module 540 is used to input the first health information and the first recommendation information into the evaluation network to obtain the first value information; input the second health information into the information recommendation model to obtain the second recommendation information, wherein the second health information is the health information within a second preset time period obtained when the user performs business based on the first recommendation information; input the second health information and the second recommendation information into the evaluation network to obtain the second value information; update the evaluation network according to the reward information, the first value information and the second value information; and update the parameters of the personalized network in the information recommendation model according to the first value information, the first health information and the first recommendation information.

[0073] Optionally, the device further includes: a training module 550, used to acquire training samples through a server device, the training samples including: health information samples and recommendation information samples corresponding to the health information samples; inputting the health information samples into the information recommendation model to be trained through the server device, determining the predicted recommendation information output by the information recommendation model to be trained; and using the server device, with the training objective of minimizing the difference between the predicted recommendation information and the recommendation information samples, obtaining an initial information recommendation model for the information recommendation model to be trained.

[0074] Optionally, the training module 550 is further configured to: construct an auxiliary training model based on the information recommendation model via a server device; the auxiliary training model includes a BERT network, a common network, and a prediction network; obtain target recommendation information generated by each user's edge device with a reward value higher than a preset reward score, and target health information corresponding to the target recommendation information via the server device; input the target health information into the auxiliary training model via the server device to obtain predicted recommendation information corresponding to the target health information, and update the parameters of the common network in the auxiliary training model through supervised learning with the training objective of minimizing the difference between the predicted recommendation information corresponding to the target health information and the target recommendation information, while freezing the parameters of the BERT network and the prediction network in the auxiliary training model; and synchronize the updated parameters of the common network to each user's edge device via the server device.

[0075] In this embodiment, the initial information recommendation model deployed on the edge device is a relatively universal model used to recommend information (meal recommendations) to various users. Subsequently, the edge device will perform local updates, continuously updating the information recommendation model through reinforcement learning by utilizing real-time changes in the user's health status on the edge device. The direction of these updates is to improve the user's health status. Therefore, this embodiment can personalize the information recommendation model according to the user's individual physical condition and health status, thereby providing more accurate meal recommendations based on the user's health condition.

[0076] Example 3 Figure 6 An information recommendation device according to a first aspect of this embodiment is shown, which corresponds to the method described according to a first aspect of Embodiment 1. (Reference) Figure 6 As shown, the information recommendation device includes: a processor 610; and a memory 620 connected to the processor 610, used to provide the processor 610 with instructions to process the following steps: acquiring first health information of a user within a first preset time period, the first health information representing the user's health status within the first preset time period; determining first recommendation information based on the first health information within the first preset time period and an information recommendation model deployed in the user's edge device, the first recommendation information being meal information recommended to the user, the information recommendation model being pre-trained by a server device, wherein the information recommendation model includes: a BERT network, a public network, a personalized network, and a prediction network; acquiring updated health information of the user when the user performs business operations based on the first recommendation information; determining reward information corresponding to the first recommendation information based on the updated health information; and updating the parameters of at least some network structures in the information recommendation model through reinforcement learning based on the reward information, while freezing the parameters of the BERT network, the public network, and the prediction network in the information recommendation model, the updated information recommendation model being used to continue recommending information to the user.

[0077] Optionally, the operation of obtaining the user's first health information within a first preset time period includes: Obtain at least one of the following information from the user's first preset time period: emotional information, heart rate information, blood pressure information, blood sugar information, height information, and weight information.

[0078] Optionally, the operation of determining the reward information corresponding to the first recommendation information based on the updated health information includes: Based on the updated health information, a first reward value is determined, and based on the difference between the first health information and the updated health information, a second reward value is determined; based on the first reward value and the second reward value, reward information corresponding to the first recommendation information is determined.

[0079] Optionally, based on the reward information, the parameters of the personalized network in the information recommendation model are updated through reinforcement learning, including: First health information and first recommendation information are input into the evaluation network to obtain first value information; second health information is input into the information recommendation model to obtain second recommendation information, which is health information obtained within a second preset time period when the user performs business based on the first recommendation information; second health information and second recommendation information are input into the evaluation network to obtain second value information; the evaluation network is updated based on reward information, first value information and second value information; the parameters of the personalized network in the information recommendation model are updated based on the first value information, first health information and first recommendation information.

[0080] Optionally, the memory 620 is also configured to provide the processor 610 with instructions to perform the following processing steps: acquiring training samples via a server device, the training samples including: health information samples and recommendation information samples corresponding to the health information samples; inputting the health information samples into the information recommendation model to be trained via the server device, and determining the predicted recommendation information output by the information recommendation model to be trained; and obtaining an initial information recommendation model via the server device, with the training objective of minimizing the difference between the predicted recommendation information and the recommendation information samples.

[0081] Optionally, the memory 620 is also used to provide the processor 610 with instructions to process the following steps: constructing an auxiliary training model based on the information recommendation model via a server device, the auxiliary training model including a BERT network, a common network, and a prediction network; obtaining target recommendation information generated by each user's edge device with a reward value higher than a preset reward score, and target health information corresponding to the target recommendation information via the server device; inputting the target health information into the auxiliary training model via the server device to obtain predicted recommendation information corresponding to the target health information, and updating the parameters of the common network in the auxiliary training model through supervised learning with the training objective of minimizing the difference between the predicted recommendation information corresponding to the target health information and the target recommendation information, while freezing the parameters of the BERT network and the prediction network in the auxiliary training model; and synchronizing the updated parameters of the common network to each user's edge device via the server device.

[0082] In this embodiment, the initial information recommendation model deployed on the edge device is a relatively universal model used to recommend information (meal recommendations) to various users. Subsequently, the edge device will perform local updates, continuously updating the information recommendation model through reinforcement learning by utilizing real-time changes in the user's health status on the edge device. The direction of these updates is to improve the user's health status. Therefore, this embodiment can personalize the information recommendation model according to the user's individual physical condition and health status, thereby providing more accurate meal recommendations based on the user's health condition.

[0083] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0084] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0085] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0086] The units described as separate components may or may not be physically separate. 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 the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0087] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0088] If the integrated unit is implemented as 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, in essence, 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. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0089] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An information recommendation method, characterized in that, The method is used on a user-held edge device, in which an information recommendation model pre-trained by a server device is deployed, including: Obtain the user's first health information within a first preset time period, wherein the first health information is used to represent the user's physical health status within the first preset time period; Based on the first health information within the first preset time period, and based on the information recommendation model deployed on the user's edge device, the first recommendation information is determined. The first recommendation information is meal information recommended to the user. The information recommendation model includes: BERT network, public network, personalized network, and prediction network. When the user performs business operations based on the first recommendation information, the updated health information of the user is obtained, and reward information corresponding to the first recommendation information is determined based on the updated health information; and Based on the reward information, the parameters of the personalized network in the information recommendation model are updated through reinforcement learning, while the parameters of the BERT network, the public network, and the prediction network in the information recommendation model are frozen. The updated information recommendation model is then used to continue recommending information to the user.

2. The method according to claim 1, characterized in that, The operation of obtaining the user's first health information within a first preset time period includes: Obtain at least one of the following information from the user during a first preset time period: emotional information, heart rate information, blood pressure information, blood sugar information, height information, and weight information.

3. The method according to claim 1, characterized in that, The operation of determining the reward information corresponding to the first recommendation information based on the updated health information includes: Based on the updated health information, a first reward value is determined, and based on the difference between the first health information and the updated health information, a second reward value is determined; based on the first reward value and the second reward value, reward information corresponding to the first recommendation information is determined.

4. The method according to claim 1, characterized in that, Based on the reward information, the parameters of the personalized network in the information recommendation model are updated using reinforcement learning, including: The first health information and the first recommendation information are input into the evaluation network to obtain the first value information; The second health information is input into the information recommendation model to obtain the second recommendation information, where the second health information is the health information within a second preset time period. The second health information and the second recommendation information are input into the evaluation network to obtain the second value information; The evaluation network is updated based on the reward information, the first value information, and the second value information; The parameters of the personalized network are updated based on the first value information, the first health information, and the first recommendation information.

5. The method according to claim 1, characterized in that, Also includes: Training samples are obtained through server devices, and the training samples include: health information samples and recommendation information samples corresponding to the health information samples; The health information sample is input into the information recommendation model to be trained through the server device, and the predicted recommendation information output by the information recommendation model to be trained is determined. The information recommendation model is trained using the server device with the training objective of minimizing the difference between the predicted recommendation information and the recommendation information sample, thus obtaining an initial information recommendation model.

6. The method according to claim 5, characterized in that, The method also includes: The server device constructs an auxiliary training model based on the information recommendation model, and the auxiliary training model includes the BERT network, the public network, and the prediction network. The server device acquires target recommendation information where the reward information generated by each user's edge device is higher than the preset reward score, as well as target health information corresponding to the target recommendation information. The target health information is input into the auxiliary training model via the server device to obtain predicted recommendation information corresponding to the target health information. Then, through supervised learning, the parameters of the common network in the auxiliary training model are updated with the training objective of minimizing the difference between the predicted recommendation information and the target recommendation information, while the parameters of the BERT network and the prediction network in the auxiliary training model are frozen. The server device synchronizes the updated public network parameters to the edge devices of each user.

7. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, a processor performs the method according to any one of claims 1 to 6.

8. An information recommendation device, characterized in that, include: The acquisition module is used to acquire the user's first health information within a first preset time period, wherein the first health information is used to represent the user's physical health status within the first preset time period; The recommendation information determination module is used to determine the first recommendation information based on the first health information within the first preset time period and the information recommendation model deployed in the user's edge device. The first recommendation information is meal information recommended to the user. The information recommendation model is pre-trained by a server device and includes: a BERT network, a personalized network, a public network, and a prediction network. The reward information determination module is used to, when the user performs business operations based on the first recommendation information, obtain the user's updated health information, and determine the reward information corresponding to the first recommendation information based on the updated health information; and The update module is used to update the parameters of the personalized network in the information recommendation model through reinforcement learning based on the reward information, while freezing the parameters of the BERT network, the public network, and the prediction network in the information recommendation model. The updated information recommendation model is used to continue to recommend information to the user.

9. The apparatus according to claim 8, characterized in that, The acquisition module is used to acquire at least one of the following information during the first preset time period of the user: emotional information, heart rate information, blood pressure information, blood sugar information, height information, and weight information.

10. An information recommendation device, characterized in that, include: processor; as well as A memory, connected to the processor, for providing the processor with instructions to perform the following processing steps: Obtain first health information of the user within a first preset time period, wherein the first health information is used to represent the user's physical health status within the first preset time period; Based on the first health information within the first preset time period, and based on the information recommendation model deployed in the user's edge device, the first recommendation information is determined. The first recommendation information is meal information recommended to the user. The information recommendation model includes: BERT network, personalized network, public network and prediction network. The information recommendation model is pre-trained by server device. When the user performs business operations based on the first recommendation information, the updated health information of the user is obtained, and reward information corresponding to the first recommendation information is determined based on the updated health information; and Based on the reward information, the parameters of the personalized network in the information recommendation model are updated through reinforcement learning, while the parameters of the BERT network, the public network, and the prediction network in the information recommendation model are frozen. The updated information recommendation model is then used to continue recommending information to the user.

Images