Information recommendation method, device, computer-readable storage medium, and electronic device

Abstract

Embodiments of the present application relate to information recommendation methods, devices, computer-readable storage media, and electronic devices. In the embodiments of the present invention, the probability estimation model is any one of the first model and the second model obtained by training through a joint learning training method. In the process of joint learning, the two training labels with noise are mutually corrected to finally obtain a pseudo-label with minimized noise, and the first model and the second model obtained by training based on the pseudo-label. In this way, the probability estimation model obtained by training based on the joint learning training method can obtain more accurate prediction results for the information to be displayed, thereby more accurately obtaining the probability that the user will perform a preset operation on the information to be recommended. The information to be recommended is determined based on the prediction result, so that the information to be recommended is more accurate, and the probability of the user performing a preset operation on the recommended information is increased.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data processing technology, and more specifically, to an information recommendation method, device, computer-readable storage medium, and electronic device. Background Art

[0002] With the development of the Internet, users can complete various businesses through various platforms. At present, most platforms will push candidate information to users to recommend content that is highly relevant to them, thereby increasing the user's stay time on the corresponding platform and the user's platform experience.

[0003] The commonly used information recommendation method is to train the machine learning model based on the user's historical behavior data, and use the trained model to score the candidate information to finally determine the recommended information. In the training process of the machine learning model, the successfully recommended information (that is, the data where the user has performed the preset operation) is used as a positive sample and marked with a score of 1, and the failed recommended information (that is, the data where the user has not performed the preset operation) is used as a negative sample and marked with a score of 0. However, in fact, the information that fails to be recommended is not necessarily completely irrelevant to the user, so it cannot be assumed that the score of this information must be 0. It can be seen that this way of marking samples will introduce a lot of noise, thereby affecting the accuracy of the machine learning model, and then affecting the accuracy of the recommended information, and ultimately leading to recommendation failures, wasting recommendation resources, and affecting user experience.

[0004] This shows that there is an urgent need for a more accurate information recommendation method. Summary of the invention

[0005] The information recommendation method, device, computer-readable storage medium, and electronic device provided in the embodiments of the present application are intended to provide a more accurate information recommendation method.

[0006] A first aspect of an embodiment of the present application provides an information recommendation method, the method comprising:

[0007] Get multiple pieces of information to be displayed;

[0008] Inputting the plurality of information to be displayed into a probability estimation model respectively, and obtaining probabilities that the user performs preset operations respectively on the plurality of information to be displayed;

[0009] Determining the information to be recommended from the multiple pieces of information to be displayed according to probabilities that the user performs preset operations respectively on the multiple pieces of information to be displayed;

[0010] Pushing the information to be recommended to the user terminal of the user;

[0011] Among them, the probability estimation model is any one of the first model and the second model of joint learning, the first model performs supervised training based on the first type of labels pre-labeled for the training samples, and the second model performs supervised training based on the second type of labels pre-labeled for the training samples, the value of the first type of labels is: a first value pre-labeled according to the user performing the preset operation on the training samples, or, a second value pre-labeled according to the user not performing the preset operation on the training samples, and the value of the second type of labels is: the probability of the pre-labeled user performing the preset operation on the training samples; in the process of joint learning, the prediction results output by the first model and the second model are corrected to the pre-labeled labels of the training samples.

[0012] Optionally, in the joint learning process, according to the prediction results outputted by the first model and the second model respectively, the labels pre-labeled for the training samples are corrected, including:

[0013] Determining a first loss value according to a prediction result of the first model on a training sample pre-labeled with a first class label;

[0014] Determining a second loss value according to a prediction result of the first model on the training sample pre-labeled with a second class label;

[0015] When the difference between the first loss value and the second loss value is greater than a preset threshold, the label pre-labeled for the training sample is corrected.

[0016] Optionally, when the difference between the first loss value and the second loss value is greater than a preset threshold, revising the label pre-labeled for the training sample includes:

[0017] When the first loss value is less than the second loss value, correcting the label pre-labeled for the training sample to the first category label;

[0018] When the second loss value is less than the first loss value, the label pre-labeled for the training sample is corrected to the second category label.

[0019] Optionally, during the joint learning process, for all training samples, if the difference between the loss value corresponding to the prediction result output by the first model and the loss value corresponding to the prediction result output by the second model is no greater than the preset threshold, stop correcting the labels pre-marked for the training samples, end the training, and obtain the probability estimation model.

[0020] Optionally, the method further comprises:

[0021] Inputting the samples to be labeled that have not been subjected to the preset operation into a pre-trained labeling model to obtain the value of the second type of label for the samples to be labeled, wherein the output of the labeling model represents the probability that the user performs the preset operation on the samples to be labeled;

[0022] According to the value of the second type label for the sample to be labeled, the sample to be labeled is labeled with the second type label to obtain a training sample with the second type label.

[0023] A second aspect of an embodiment of the present application provides an information recommendation device, the device comprising:

[0024] A module for acquiring information to be displayed, used for acquiring multiple information to be displayed;

[0025] A first input module, used to input the plurality of information to be displayed into a probability estimation model respectively, and obtain the probability that the user performs a preset operation on the plurality of information to be displayed respectively;

[0026] A module for determining information to be recommended, configured to determine information to be recommended from the plurality of information to be displayed according to probabilities that the user performs preset operations respectively on the plurality of information to be displayed;

[0027] A push module, used for pushing the information to be recommended to the user terminal of the user;

[0028] Among them, the probability estimation model is any one of the first model and the second model of joint learning, the first model performs supervised training based on the first type of labels pre-labeled for the training samples, and the second model performs supervised training based on the second type of labels pre-labeled for the training samples, the value of the first type of labels is: a first value pre-labeled according to the user performing the preset operation on the training samples, or, a second value pre-labeled according to the user not performing the preset operation on the training samples, and the value of the second type of labels is: the probability of the pre-labeled user performing the preset operation on the training samples; in the process of joint learning, the prediction results output by the first model and the second model are corrected to the pre-labeled labels of the training samples.

[0029] Optionally, in the joint learning process, according to the prediction results outputted by the first model and the second model respectively, the labels pre-labeled for the training samples are corrected, including:

[0030] Determining a first loss value according to a prediction result of the first model on a training sample pre-labeled with a first class label;

[0031] Determining a second loss value according to a prediction result of the first model on the training sample pre-labeled with a second class label;

[0032] When the difference between the first loss value and the second loss value is greater than a preset threshold, the label pre-labeled for the training sample is corrected.

[0033] Optionally, when the difference between the first loss value and the second loss value is greater than a preset threshold, revising the label pre-labeled for the training sample includes:

[0034] When the first loss value is less than the second loss value, correcting the label pre-labeled for the training sample to the first category label;

[0035] When the second loss value is less than the first loss value, the label pre-labeled for the training sample is corrected to the second category label.

[0036] Optionally, during the joint learning process, for all training samples, if the difference between the loss value corresponding to the prediction result output by the first model and the loss value corresponding to the prediction result output by the second model is no greater than the preset threshold, stop correcting the labels pre-marked for the training samples, end the training, and obtain the probability estimation model.

[0037] Optionally, the device further comprises:

[0038] A second input module is used to input the samples to be labeled that have not been subjected to the preset operation into a pre-trained labeling model to obtain the value of the second type of label for the samples to be labeled, and the output of the labeling model represents the probability that the user performs the preset operation on the samples to be labeled;

[0039] The labeling module is used to label the samples to be labeled with second-class labels according to the values ​​of the second-class labels for the samples to be labeled, so as to obtain training samples with second-class labels.

[0040] A third aspect of an embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the information recommendation method described in the first aspect of the present application are implemented.

[0041] A fourth aspect of an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps in the information recommendation method described in the first aspect of the present application are implemented.

[0042] In an embodiment of the present invention, the probability prediction model is any one of the first model and the second model obtained by training through a joint learning training method. In the process of joint learning, the two training labels with noise are mutually corrected, and finally a pseudo label with minimized noise and a first model and a second model obtained by training based on the pseudo label are obtained. In this way, the probability prediction model obtained by training based on the joint learning training method can obtain a more accurate prediction result for the information to be displayed, so as to more accurately obtain the probability that the user performs a preset operation on the information to be recommended. The information to be recommended is determined based on the prediction result, so that the information to be recommended is more accurate, and the probability that the user performs a preset operation on the recommended information is increased. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 is a flow chart of a joint learning model training method proposed in one embodiment of the present application;

[0045] Figure 2 is a flow chart of a joint learning model training method proposed in another embodiment of the present application;

[0046] Figure 3 is a schematic diagram of the joint learning model training method proposed in the embodiment of the present application;

[0047] Figure 4 is a flow chart of an information recommendation method proposed in another embodiment of the present application;

[0048] Figure 5 is a schematic diagram of an information recommendation device proposed in an embodiment of the present application; DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] First of all, it needs to be stated that the user data mentioned in the embodiments of the present invention are collected and processed in compliance with relevant policies and regulations and with the consent of the corresponding subjects. After processing, the data is used in big data application scenarios and cannot be identified to any natural person or have a specific association with their privacy.

[0051] CTR (Click-Through-Rate) estimation has always been an important part of the recommendation system. Simply put, CTR estimation is to predict the click situation of each exposure sample. In related technologies, during the training process of the CTR estimation model, 0 and 1 labels are usually used as labels for exposure non-click and exposure click data.

[0052] However, exposure click data is actually a kind of implicit feedback data. In simple terms, exposure but not clicked data often has a certain probability of being clicked by users, and it does not completely mean that users will not click it. For example, if 5 hot pot content is recommended to a user at the same time, and the user only clicks on one, it does not mean that the user is completely irrelevant to the remaining 4 unclicked content. If the remaining content is recommended to the user at another time, the user is likely to click it. Directly treating the 4 content that the user did not click as negative samples and using label 0 to train the model will lead to inaccurate model estimates.

[0053] Therefore, in the related art, there is a lot of noise in the setting of the training sample labels of the model, which will affect the accuracy of the model's estimation and further affect the accuracy of the recommended information.

[0054] Based on this, the present invention proposes a joint learning model training method that performs machine learning based on noise samples and can reduce the interference caused by noise and improve the accuracy of the model, thereby providing a more accurate information recommendation method.

[0055] refer to Figure 1 , Figure 1 : is a flow chart of the joint learning model training method proposed in one embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0056] 101. Perform supervised training on a first model based on first-category labels pre-labeled for training samples, and perform supervised training on a second model based on second-category labels pre-labeled for the training samples.

[0057] In an embodiment of the present invention, first, a large number of training samples can be collected, and the first category labels and the second category labels can be respectively labeled for these training samples to obtain two groups of training samples with different labels, so as to perform supervised training on the first model and the second model respectively, wherein the supervisory signal of the first model is the first category label, and the supervisory signal of the second model is the second category label.

[0058] The following uses whether a user browses recommended products as an example to explain various embodiments of the present invention:

[0059] For example, user behavior data on recommended products can be collected, recommended products browsed by users are marked as 1, recommended products not browsed by users are marked as 0, and labels 0 and 1 are used as first-category labels to perform supervised training on the first model.

[0060] In addition, the recommended products that the user has not browsed are labeled with the second category labels. The labeling basis can be the probability 0 to 1 that the user browses the recommended product, so that the labels of 0 to 1 are used as the second category labels to perform supervised training on the second model.

[0061] In an embodiment of the present invention, the probability of a user browsing the recommended product can be obtained based on relevant data statistics of similar users, or can be predicted based on a pre-trained deep learning model.

[0062] In the embodiment of the present invention, the first model and the second model may be two neural network models with the same structure, or may be two neural network models with different structures.

[0063] 102. In the process of joint learning, according to the prediction results outputted by the first model and the second model respectively, the labels pre-labeled for the training samples are corrected to obtain pseudo labels for the training samples.

[0064] In the embodiment of the present invention, the first model and the second model can be jointly learned. In the process of joint learning, the first model and the second model can output their own prediction results for the same training sample under different supervisory signals. Therefore, the pre-labeled first label or second label can be corrected according to the prediction result.

[0065] In the embodiment of the present invention, when the prediction results obtained by two models for the same training sample under different supervisory signals differ greatly, it can be determined that a certain label of the training sample is incorrect and needs to be corrected.

[0066] In the embodiment of the present invention, the label of the training sample of the model with the smaller loss value can be judged to be correct according to the size of the loss value of the prediction results output by the two models, and the label of the other training sample can be corrected with the correct label. Therefore, in the embodiment of the present invention, after one iterative training is completed, the labels of the training samples with different models can be corrected according to the prediction results output by the two models.

[0067] In the embodiment of the present invention, after the training labels are corrected, the two models can be trained for the next iteration using the corrected training labels until the prediction results output by the two models no longer differ, indicating that the joint learning is completed and similar first and second models are obtained.

[0068] In the embodiment of the present invention, two training labels with noise can be mutually corrected by joint learning of the first model and the second model, and finally a training label with minimized noise can be obtained. Thus, a more accurate prediction model can be obtained by training the training samples carrying the training label.

[0069] refer to Figure 2 , Figure 2 FIG. 1 is a flow chart of a joint learning model training method proposed in another embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0070] 201 , performing supervised training on a first model based on first-category labels pre-labeled for training samples, and performing supervised training on a second model based on second-category labels pre-labeled for the training samples.

[0071] This step is similar to the above step 101 and will not be described again here.

[0072] 202. Determine a first loss value according to the prediction result of the first model for the training samples pre-labeled with the first category label; determine a second loss value according to the prediction result of the first model for the training samples pre-labeled with the second category label.

[0073] In the embodiment of the present invention, during the training process, the two models will obtain corresponding prediction results for each training sample and obtain corresponding loss values.

[0074] For example, the first model can predict the training sample A labeled as 1 to obtain a first loss value; the second model can also predict the training sample A labeled as 0.7 to obtain a second loss value.

[0075] S203: When the difference between the first loss value and the second loss value is greater than a preset threshold, correct the label pre-labeled for the training sample.

[0076] In the embodiment of the present invention, after obtaining the first loss value and the second loss value, the first category label or the second category label pre-labeled for the training sample can be corrected according to the first loss value and the second loss value. In short, when the difference in the loss values ​​obtained by the two models for training samples with different labels is large, it can be considered that there is a large difference in the prediction results of the two models for the training sample, and the pre-labeled label of the training sample needs to be corrected.

[0077] In the embodiment of the present invention, the preset threshold value can be determined by technicians based on experience, and the present invention does not impose any special limitation on this.

[0078] In the embodiment of the present invention, when the difference between the first loss value and the second loss value is greater than a preset threshold, it is considered that the training label based on the prediction of the model with a smaller loss value is closer to the actual situation. Therefore, the label of the training sample can be determined as the training label based on the model with a smaller loss value.

[0079] For example, the first model can predict the training sample A labeled as 1 to obtain a first loss value; the second model can also predict the training sample A labeled as 0.7 to obtain a second loss value.

[0080] Among them, the difference between the first loss value and the second loss value is greater than the preset threshold, and the first loss value is greater than the second loss value. Then the label of 0.7 can be used as a label that is closer to the actual situation, the label of training sample A is determined to be 0.7, and the label of the training sample of the first model is updated, and the training sample with the updated label is used for the next iterative training.

[0081] In an embodiment of the present invention, when the difference between the first loss value and the second loss value is greater than a preset threshold, the pre-annotated labels of the training samples are corrected, which can avoid the labels of the training samples of the two models from quickly converging and affecting the effect of joint learning.

[0082] S204: For all training samples, if the difference between the loss value corresponding to the prediction result output by the first model and the loss value corresponding to the prediction result output by the second model is not greater than the preset threshold, stop correcting the labels pre-marked for the training samples, end the training, and obtain the probability estimation model.

[0083] In the embodiment of the present invention, when the two models make predictions for each training sample and the prediction results obtained do not have large differences, it indicates that the two models have completed joint learning, and the annotations of the training samples of the two models tend to be consistent. The training can be ended to obtain the trained first model and the second model. Either of the trained first model and the second model can be used as a probability estimation model to estimate the probability of users clicking on the recommended information.

[0084] Figure 3 A schematic diagram of the joint learning model training method proposed in the embodiment of the present application is shown, combined with Figure 3 , the following further explains the joint learning model training method proposed in the embodiment of the present application:

[0085] like Figure 3As shown, model A and model B are jointly trained, where model A uses training samples labeled with the first category of labels, and model B uses training samples labeled with the second category of labels. The training samples used by the two models are the same batch of training samples labeled with different labels.

[0086] After each round of training, the labels of the training samples to be corrected whose prediction results of the two models differ greatly can be corrected to obtain pseudo labels according to the prediction results. In this process, there may be a part of the training samples to be corrected with smaller loss values ​​in the prediction results output by model A, and another part of the training samples to be corrected with smaller loss values ​​in the prediction results output by model B. At this time, these training samples to be corrected can be corrected to each other, and the labels of the samples to be corrected can be corrected to the labels used by the model with smaller loss values.

[0087] In each round of training, both models will update their model parameters. Therefore, in the next round of training, the two models may have large differences in the training samples that had small differences in the previous round of training, thereby correcting the labels of these training samples.

[0088] After multiple iterations of training, the prediction results of the two models for all training samples tend to be consistent, and the joint learning training process can be terminated at this time.

[0089] In an optional implementation manner of the present invention, before step S201, the method further includes the following steps:

[0090] S1, inputting the samples to be labeled that have not been subjected to the preset operation into a pre-trained labeling model to obtain the value of the second type of label for the samples to be labeled, and the output of the labeling model represents the probability that the user performs the preset operation on the samples to be labeled.

[0091] S2: labeling the samples to be labeled with second-category labels according to the values ​​of the second-category labels for the samples to be labeled, so as to obtain training samples with second-category labels.

[0092] In the embodiment of the present invention, the training samples that have not been subjected to the preset operation are used as samples to be labeled and labeled with the second type of labels. In the embodiment of the present invention, the labeling model can be used to predict the probability of the user performing the preset operation on these samples to be labeled, thereby determining the second type of labels for the training samples.

[0093] In the embodiment of the present invention, the labeling model can be an exposure click rate prediction model obtained by training with traditional positive and negative samples. In the embodiment of the present invention, since the model is trained with traditional positive and negative samples, there is sample noise, and the prediction accuracy of the corresponding model is limited. Therefore, the accuracy of the value of the second category label of the sample to be labeled obtained by the labeling model is also limited. Correspondingly, the second category label of the training sample also has noise.

[0094] Based on this, in the embodiment of the present invention, the two labels with noise in the training samples can be corrected through the joint learning model training method to reduce noise interference and obtain a more accurate probability estimation model.

[0095] refer to Figure 4 , Figure 4 : is a flow chart of an information recommendation method proposed in an embodiment of the present application. Figure 4 As shown, the method comprises the following steps:

[0096] 401, obtain multiple pieces of information to be displayed.

[0097] In the embodiment of the present invention, a plurality of information to be displayed can be acquired in real time from the network platform.

[0098] 402 , input the plurality of information to be displayed into a probability estimation model respectively, and obtain the probability that the user performs a preset operation on the plurality of information to be displayed respectively.

[0099] In the embodiment of the present invention, the preset operation may be a preset user operation such as a click operation, an order placement operation, a browsing operation, etc.

[0100] In an embodiment of the present invention, the probability estimation model is any one of the first model and the second model trained using the aforementioned joint learning training method.

[0101] The first model performs supervised training based on first-category labels pre-labeled for training samples, and the second model performs supervised training based on second-category labels pre-labeled for training samples. The value of the first-category labels is: a first value pre-labeled based on the user performing the preset operation on the training samples, or a second value pre-labeled based on the user not performing the preset operation on the training samples. The value of the second-category labels is: the probability of the pre-labeled user performing the preset operation on the training samples.

[0102] In the embodiment of the present invention, during the joint learning process, the labels pre-labeled for the training samples are corrected according to the prediction results outputted by the first model and the second model respectively.

[0103] 403 , determining the information to be recommended from the multiple pieces of information to be displayed according to the probabilities that the user performs preset operations respectively on the multiple pieces of information to be displayed.

[0104] In the embodiment of the present invention, the probability estimation model can output the probability that the user will perform a preset operation on each piece of information to be displayed, so that the information to be recommended can be determined based on the probability.

[0105] In the embodiment of the present invention, the data to be recommended may be sorted according to the probability of the user performing a preset operation, so that a preset number of information to be displayed at the top of the sorting is used as the information to be recommended.

[0106] In the embodiment of the present invention, the preset number can be preset in advance by a technician according to actual needs, and the embodiment of the present invention does not impose any special limitation on this.

[0107] 404: Push the information to be recommended to the user terminal of the user.

[0108] In an embodiment of the present invention, the probability prediction model is any one of the first model and the second model obtained by training through a joint learning training method. In the process of joint learning, the two training labels with noise are mutually corrected, and finally a pseudo label with minimized noise and a first model and a second model obtained by training based on the pseudo label are obtained. In this way, the probability prediction model obtained by training based on the joint learning training method can obtain a more accurate prediction result for the information to be displayed, so as to more accurately obtain the probability that the user performs a preset operation on the information to be recommended. The information to be recommended is determined based on the prediction result, so that the information to be recommended is more accurate, and the probability that the user performs a preset operation on the recommended information is increased.

[0109] Based on the same inventive concept, an embodiment of the present application provides an information recommendation device, referring to Figure 5 , Figure 5 is a schematic diagram of an information recommendation device 500 proposed in an embodiment of the present application. The device includes:

[0110] The to-be-displayed information acquisition module 501 is used to acquire a plurality of to-be-displayed information;

[0111] A first input module 502 is used to input the plurality of information to be displayed into a probability estimation model respectively, and obtain the probability that the user performs a preset operation on the plurality of information to be displayed respectively;

[0112] The module 503 for determining information to be recommended is used to determine the information to be recommended from the plurality of information to be displayed according to the probability that the user performs preset operations respectively on the plurality of information to be displayed;

[0113] A push module 504, configured to push the information to be recommended to the user terminal of the user;

[0114] Among them, the probability estimation model is any one of the first model and the second model of joint learning, the first model performs supervised training based on the first type of labels pre-labeled for the training samples, and the second model performs supervised training based on the second type of labels pre-labeled for the training samples, the value of the first type of labels is: a first value pre-labeled according to the user performing the preset operation on the training samples, or, a second value pre-labeled according to the user not performing the preset operation on the training samples, and the value of the second type of labels is: the probability of the pre-labeled user performing the preset operation on the training samples; in the process of joint learning, the prediction results output by the first model and the second model are corrected to the pre-labeled labels of the training samples.

[0115] Optionally, in the joint learning process, according to the prediction results outputted by the first model and the second model respectively, the labels pre-labeled for the training samples are corrected, including:

[0116] Determining a first loss value according to a prediction result of the first model on a training sample pre-labeled with a first class label;

[0117] Determining a second loss value according to a prediction result of the first model on the training sample pre-labeled with a second class label;

[0118] When the difference between the first loss value and the second loss value is greater than a preset threshold, the label pre-labeled for the training sample is corrected.

[0119] Optionally, when the difference between the first loss value and the second loss value is greater than a preset threshold, revising the label pre-labeled for the training sample includes:

[0120] When the first loss value is less than the second loss value, correcting the label pre-labeled for the training sample to the first category label;

[0121] When the second loss value is less than the first loss value, the label pre-labeled for the training sample is corrected to the second category label.

[0122] Optionally, during the joint learning process, for all training samples, if the difference between the loss value corresponding to the prediction result output by the first model and the loss value corresponding to the prediction result output by the second model is no greater than the preset threshold, stop correcting the labels pre-marked for the training samples, end the training, and obtain the probability estimation model.

[0123] Optionally, the device further comprises:

[0124] A second input module is used to input the samples to be labeled that have not been subjected to the preset operation into a pre-trained labeling model to obtain the value of the second type of label for the samples to be labeled, and the output of the labeling model represents the probability that the user performs the preset operation on the samples to be labeled;

[0125] The labeling module is used to label the samples to be labeled with second-class labels according to the values ​​of the second-class labels for the samples to be labeled, so as to obtain training samples with second-class labels.

[0126] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0127] Based on the same inventive concept, another embodiment of the present application provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the information recommendation method as described in any of the above embodiments of the present application.

[0128] Based on the same inventive concept, another embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the steps of the information recommendation method described in any of the above embodiments of the present application.

[0129] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0130] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the embodiments of the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0131] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0132] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0134] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0135] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0136] The information recommendation method, device, equipment and storage medium provided by the present application are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. An information recommendation method, characterized in that: The method comprises: obtaining a plurality of information to be displayed; inputting the plurality of information to be displayed into a probability prediction model respectively, and obtaining the probability that the user performs preset operations on the plurality of information to be displayed respectively; determining the information to be recommended from the plurality of information to be displayed according to the probability that the user performs preset operations on the plurality of information to be displayed respectively; and pushing the information to be recommended to the user terminal of the user; wherein the probability prediction model is any one of a first model and a second model of joint learning, the first model performs supervised training based on a first type of label pre-labeled for a training sample, and the second model performs supervised training based on a second type of label pre-labeled for the training sample, and the value of the first type of label is: based on the probability that the user performs the preset operations on the training sample A first value pre-marked for a preset operation, or a second value pre-marked based on the user not performing the preset operation on the training sample, the value of the second type of label is: the probability that the pre-marked user performs the preset operation on the training sample; a first loss value is determined based on the prediction results of the first model for the training samples pre-marked with the first type of labels; a second loss value is determined based on the prediction results of the first model for the training samples pre-marked with the second type of labels; when the first loss value is less than the second loss value, the label pre-marked for the training sample is corrected to the first type of label; when the second loss value is less than the first loss value, the label pre-marked for the training sample is corrected to the second type of label.

2. The information recommendation method according to claim 1, characterized in that: During the joint learning process, for all training samples, if the difference between the loss value corresponding to the prediction result output by the first model and the loss value corresponding to the prediction result output by the second model is no greater than a preset threshold, stop correcting the labels pre-marked for the training samples, end the training, and obtain the probability estimation model.

3. The information recommendation method according to claim 1, characterized in that: The method also includes: inputting the samples to be labeled that have not been subjected to the preset operation into a pre-trained labeling model to obtain the value of the second category label for the samples to be labeled, and the output of the labeling model represents the probability that the user performs the preset operation on the samples to be labeled; and labeling the samples to be labeled with the second category label according to the value of the second category label for the samples to be labeled, to obtain training samples with the second category label.

4. An information recommendation device, characterized in that: The device comprises: an information acquisition module for acquiring a plurality of information to be displayed; a first input module for inputting the plurality of information to be displayed into a probability prediction model respectively, and obtaining the probability that the user performs preset operations on the plurality of information to be displayed respectively; a module for determining information to be recommended, for determining the information to be recommended from the plurality of information to be displayed according to the probability that the user performs preset operations on the plurality of information to be displayed respectively; and a push module for pushing the information to be recommended to the user terminal of the user; wherein the probability prediction model is any one of the first model and the second model of joint learning, the first model is supervised training based on the first type of labels pre-labeled for the training samples, the second model is supervised training based on the second type of labels pre-labeled for the training samples, and the first type The value of the label is: a first value pre-marked based on the user performing the preset operation on the training sample, or a second value pre-marked based on the user not performing the preset operation on the training sample, and the value of the second type of label is: the probability of the pre-marked user performing the preset operation on the training sample; a first loss value is determined based on the prediction results of the first model for the training samples pre-marked with the first type of labels; a second loss value is determined based on the prediction results of the first model for the training samples pre-marked with the second type of labels; when the first loss value is less than the second loss value, the label pre-marked for the training sample is corrected to the first type of label; when the second loss value is less than the first loss value, the label pre-marked for the training sample is corrected to the second type of label.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the information recommendation method according to any one of claims 1 to 3 are implemented.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps in the information recommendation method according to any one of claims 1 to 3 are implemented.

Images