Target recommendation method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of computer models, and provides a target recommendation method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining associated data of an object; the associated data is input into a target recommendation model, the target recommendation model is generated based on common training of a sample training set and an initial prediction baseline, and the initial prediction baseline is determined by performing out-of-fold prediction on the sample training set; recommendation data output by the target recommendation model based on the associated data are obtained, target recommendation is carried out based on the recommendation data, and most of initial prediction baselines, which are determined by carrying out excursion prediction on the sample training set and reserve original data features and baseline features, are information; according to the method, the target recommendation model in the training stage can learn multi-level data identifiers at the same time, the features of the sample training set are enhanced, the learning ability and prediction precision of the target recommendation model are improved, the target recommendation model can perform accurate prediction based on the associated data, and accurate recommendation data are obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer models, and particularly relates to a target recommendation method and device, electronic equipment and a storage medium. BACKGROUND

[0002] At present, many network applications recommend works to users. For example, a video application website recommends multimedia works of the same or related categories to users according to the operation history information of the users clicking videos. For example, if a user often watches car-related videos, the electronic equipment can recommend car-related videos or advertisement information to the user; for another example, a shopping website recommends interesting goods to users according to the purchase history, click data and interested goods of the users.

[0003] In the related art, the target of interest of the user cannot be accurately recommended, resulting in low recommendation accuracy. SUMMARY

[0004] Therefore, the embodiments of the present application provide a target recommendation method and device, electronic equipment and a storage medium to solve the problem that the target of interest of the user cannot be accurately recommended in the prior art, resulting in low recommendation accuracy.

[0005] The first aspect of the embodiments of the present application provides a target recommendation method, which comprises: acquiring associated data of an object; inputting the associated data into a target recommendation model, the target recommendation model being generated by jointly training based on a sample training set and an initial prediction baseline, the initial prediction baseline being determined by performing out-of-fold prediction on the sample training set; acquiring recommendation data output by the target recommendation model based on the associated data, and performing target recommendation based on the recommendation data.

[0006] The second aspect of the embodiments of the present application provides a target recommendation device, which comprises: an acquisition module configured to acquire associated data of an object; a processing module configured to input the associated data into a target recommendation model, the target recommendation model being generated by jointly training based on a sample training set and an initial prediction baseline, the initial prediction baseline being determined by performing out-of-fold prediction on the sample training set; and a recommendation module configured to acquire recommendation data output by the target recommendation model based on the associated data, and perform target recommendation based on the recommendation data.

[0007] The third aspect of the embodiments of the present application provides electronic equipment, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0008] In a fourth aspect, the present application provides a computer readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above method.

[0009] Compared with the prior art, the present application has the beneficial effects that: in the present application, the associated data of the object is obtained; the associated data is input into the target recommendation model, the target recommendation model is generated based on the sample training set and the initial prediction baseline, and the initial prediction baseline is determined by performing out-of-sample prediction on the sample training set; the recommendation data output by the target recommendation model based on the associated data is obtained, and target recommendation is performed based on the recommendation data, wherein the initial prediction baseline determined by performing out-of-sample prediction on the sample training set retains most of the information of the original data features and baseline features, so that the target recommendation model in the training stage can learn multiple levels of data identification at the same time, the features of the sample training set are enhanced, the learning ability and prediction accuracy of the target recommendation model are improved, and then the target recommendation model can accurately predict based on the associated data to obtain accurate recommendation data, thereby avoiding the problem that in related technologies, the target of interest cannot be accurately recommended to the user, resulting in low recommendation accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0011] Figure 1 is a flowchart of a target recommendation method provided by the present application;

[0012] Figure 2 is a flowchart of another target recommendation method provided by the present application;

[0013] Figure 3 is a flowchart of another target recommendation method provided by the present application;

[0014] Figure 4 is a flowchart of another target recommendation method provided by the present application;

[0015] Figure 5 is a flowchart of another target recommendation method provided by the present application;

[0016] Figure 6 is a flowchart of another target recommendation method provided by the present application;

[0017] Figure 7 is a basic flowchart of a target recommendation method provided by an embodiment of the application;

[0018] Figure 8 is a basic flowchart of data processing provided by an embodiment of the application;

[0019] Figure 9 is a basic flowchart of determining an initial prediction baseline provided by an embodiment of the application;

[0020] Figure 10 is a basic flowchart of determining a target value provided by an embodiment of the application;

[0021] Figure 11 is a basic flowchart of determining an error of a model provided by an embodiment of the application;

[0022] Figure 12 is a structural diagram of a target recommendation apparatus provided by an embodiment of the application;

[0023] Figure 13 is a structural diagram of an electronic device provided by an embodiment of the application. DETAILED DESCRIPTION

[0024] In the following description, specific details are set forth, such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the present application. However, persons skilled in the art will understand that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0025] A target recommendation method and apparatus according to an embodiment of the application will be described in detail below with reference to the accompanying drawings.

[0026] Figure 1 is a flowchart of a target recommendation method provided by an embodiment of the application, as shown in Figure 1 the target recommendation method comprises:

[0027] S101, obtaining associated data of an object;

[0028] S102, inputting the associated data to a target recommendation model, the target recommendation model being generated by joint training based on a sample training set and an initial prediction baseline, the initial prediction baseline being determined by out-of-sample prediction for the sample training set;

[0029] S103, obtaining recommendation data output by the target recommendation model based on the associated data, and performing target recommendation based on the recommendation data.

[0030] It can be understood that the association data of the above object changes according to the change of the specific use scene. For example, in the use scene of product recommendation, the association data of the object includes at least one of browsing data, click data, purchase data, collection data, and age data; in the use scene of medical diagnosis, the association data of the object includes at least one of age data, physical examination data, medical history data, and clinical data; in the use scene of financial risk control, the association data of the object includes at least one of age data, gender data, education data, income data, industry data, occupation data, post, marriage data, child data, and asset data.

[0031] In some examples, after the application obtains the above association data, the association data is input into a target recommendation model. The target recommendation model generates corresponding recommendation data based on the association data. It can be understood that the target recommendation model is generated based on the joint training of the sample training set and the initial prediction baseline. The initial prediction baseline is determined by out-of-sample prediction on the sample training set. The initial prediction baseline determined by out-of-sample prediction on the sample training set retains the multi-dimensional information of the original data features and baseline features, so that the target recommendation model in the training stage can learn multiple levels of data identification at the same time, thereby enhancing the features of the sample training set, and thereby avoiding compression or information loss in the prediction process of the target recommendation model. The target recommendation model in the application is generated based on the joint training of the sample training set and the initial prediction baseline, which improves the learning ability and prediction accuracy of the target recommendation model.

[0032] The application obtains the recommendation data output by the target recommendation model based on the association data, and performs target recommendation based on the recommendation data. Since the target recommendation model is generated based on the joint training of the sample training set and the initial prediction baseline, it has good prediction accuracy, so the output recommendation data has good accuracy.

[0033] According to the technical scheme provided in the embodiments of the present application, the associated data of the object is acquired; the associated data is input into a target recommendation model, the target recommendation model is generated based on a sample training set and an initial prediction baseline, the initial prediction baseline is determined by performing out-of-fold prediction on the sample training set; recommendation data output by the target recommendation model based on the associated data is acquired, and target recommendation is performed based on the recommendation data, wherein the initial prediction baseline determined by performing out-of-fold prediction on the sample training set retains most of the information of the original data features and baseline features, so that the target recommendation model in the training stage can learn multiple levels of data identification at the same time, the features of the sample training set are enhanced, the learning ability and prediction accuracy of the target recommendation model are improved, and then the target recommendation model can accurately predict based on the associated data to obtain accurate recommendation data, thereby avoiding the problem that in the related art, a target of interest cannot be accurately recommended to a user, resulting in low recommendation accuracy.

[0034] In some embodiments, as shown in Figure 2 Before the associated data is input into the target recommendation model, the method further includes:

[0035] S201, acquiring a sample training set;

[0036] S202, performing multiple sample divisions on the sample training set to obtain multiple sample combinations, each sample combination including multiple groups of training samples and one group of test samples;

[0037] S203, training a weak model according to the multiple groups of training samples in each sample combination, and testing the trained weak model based on the test samples to obtain a prediction value corresponding to each sample combination;

[0038] S204, determining an initial prediction baseline according to the prediction value corresponding to each sample combination.

[0039] It can be understood that the present application will acquire a sample training set, the samples in the sample training set are split into two parts of features (input) and labels (output), it can be understood that the features are used to describe the attributes of the classification object, such as user behavior data, device features, environment information, etc. The label is used to describe the target variable representing the classification result corresponding to the feature, such as whether to pass the audit, whether to purchase goods, disease category, etc.

[0040] It can be understood that in order to reduce the non-dimensional quantity of the samples in the sample training set and cause adverse effects on each model to be trained (such as a weak model and / or an initial recommendation) in the subsequent model training process, the present application will perform normalization processing on the sample data in the sample training set, thereby improving the convergence speed and generalization performance of each model, and improving the robustness and accuracy of each model.

[0041] It can be understood that out-of-fold prediction is a prediction result generated by cross-validation, which is a method used to maximize data utilization. In order to achieve out-of-fold prediction of the sample training set, this application will divide the sample training set into several folds (5 folds will be used as an example later), and use 4 folds of data as training samples (each fold of data is a group of training samples) to train the model, and use the remaining 1 fold of data for prediction.

[0042] In some examples, the present application will repeat the process of dividing the sample training set into several folds multiple times, thereby obtaining multiple sample combinations, each sample combination including multiple groups of training samples and one group of test samples.

[0043] It can be understood that the above-mentioned weak model includes but is not limited to one of a decision tree model, a linear regression model, and a ridge regression model. Preferably, the weak model is a ridge regression model.

[0044] In some examples, the present application trains the weak model based on multiple groups of training samples in each sample combination, and tests the trained weak model based on the test samples to obtain the prediction value corresponding to each sample combination; it can be understood that the prediction value corresponding to each sample combination reflects the weak model's understanding of the data distribution. Based on this, the prediction value corresponding to each sample combination can be used as the initial prediction baseline for the subsequent initial recommendation model.

[0045] Specifically, this application takes the ridge regression model as an example of a weak model. Each sample combination contains 4 groups of training samples and one group of test samples. This application performs 5-fold cross-validation on the ridge regression model through each sample combination. Each training model only uses 4 groups of training samples. Then, the trained ridge regression model is tested with the test samples to generate the prediction value corresponding to the sample combination. After completing the prediction of the test samples in all sample data in turn, the prediction value corresponding to each complete sample combination is obtained. Finally, based on the prediction value corresponding to each sample combination, the initial prediction baseline is determined, so that the initial prediction baseline retains the multi-dimensional information of the original data features and the baseline features, so that the target recommendation model in the subsequent training stage can simultaneously learn multi-level data identification, thereby enhancing the characteristics of the sample training set, thereby avoiding compression or information loss of the target recommendation model during the prediction process.

[0046] According to the technical scheme provided in the embodiments of the present application, a sample training set is obtained; the sample training set is divided into multiple sample combinations through multiple sample divisions, each sample combination including multiple training samples and one test sample; a weak model is trained according to the multiple training samples in each sample combination, and the trained weak model is tested based on the test sample to obtain a prediction value corresponding to each sample combination; and an initial prediction baseline is determined according to the prediction value corresponding to each sample combination, so that the original data features and the multi-dimensional information of the baseline features are retained in the initial prediction baseline, so that the target recommendation model in the subsequent training stage can learn the multi-level data identification, thereby enhancing the features of the sample training set, and avoiding the compression or information loss of the target recommendation model in the prediction process.

[0047] In some embodiments, as shown in Figure 3 According to the prediction value corresponding to each sample combination, the initial prediction baseline is determined, including:

[0048] S301, determining an average prediction value according to the prediction values corresponding to the multiple sample combinations;

[0049] S302, determining an initial prediction baseline according to the average prediction value.

[0050] It can be understood that, in order to make the initial prediction baseline more accurate, the present application will determine an average prediction value according to the prediction values corresponding to the multiple sample combinations, and the average prediction value can more accurately reflect the understanding of the weak model to the data distribution.

[0051] It can be understood that, in order to make the initial baseline baseline be able to retain more information, the present application will also convert the average prediction value into a logical value (logits value), and take the logical value as the initial prediction baseline. Specifically, the conversion method is as follows:

[0052]

[0053] Wherein, logits(P) represents the logical value converted by the average prediction value P, and P represents the average prediction value. The logits value represents the log-odds, which retains more information than the direct average prediction value and is suitable as the initial prediction baseline.

[0054] In some examples, the average prediction value can also be directly taken as the initial prediction baseline.

[0055] According to the technical scheme provided in the embodiment of the application, the average prediction value is determined according to the prediction values corresponding to the plurality of sample combinations, the initial prediction baseline is determined according to the average prediction value, and thus the initial prediction baseline contains more information, the multi-dimensional information of the original data features and the baseline features is retained in the initial prediction baseline, the target recommendation model in the subsequent training stage can learn the multi-level data identification at the same time, and thus the features of the sample training set are enhanced, and the compression or information loss of the target recommendation model in the prediction process is avoided.

[0056] In some embodiments, as shown in Figure 4 After the initial prediction baseline is determined according to the prediction values corresponding to each sample combination, the method further includes:

[0057] S401, determining a target prediction baseline of an initial recommendation model according to the initial prediction baseline;

[0058] S402, training the initial recommendation model according to the target prediction baseline and the sample training set;

[0059] S403, determining a target prediction value of the initial recommendation model trained of the sample test set, and determining an error value of the initial recommendation model trained according to the standard value corresponding to the sample test set and the target prediction value;

[0060] S404, if the error value of the initial recommendation model trained is lower than a pre-set error threshold, the initial recommendation model trained is taken as the target recommendation model.

[0061] It can be understood that the initial prediction baseline is used to determine the target prediction baseline of the initial recommendation model in the application, instead of simply using the original target value (original prediction baseline), and thus the model performance of the initial recommendation model is optimized by learning the deviation between the initial prediction baseline and the original prediction baseline.

[0062] Specifically, the original prediction baseline is subtracted from the initial prediction baseline to form a new training target, and the new training target is taken as the target prediction baseline.

[0063] In some examples, the initial recommendation model is further trained according to the target prediction baseline and the sample training set, and it can be understood that the type of the initial recommendation model is not limited, and the initial recommendation model can adopt a gradient boosting model (such as a CatBoost gradient boosting model or an XGBoost gradient boosting model), and thus the initial prediction baseline features predicted by the weak model are fully utilized.

[0064] After the initial recommendation model is trained according to the target prediction baseline and the sample training set, the model performance of the trained initial recommendation model is determined. Specifically, the sample test set is obtained, and then the target prediction value of the trained initial recommendation model is determined according to the sample test set. Then, the target prediction value is compared with the standard value corresponding to the sample test set, so as to determine the error value of the trained initial recommendation model.

[0065] Specifically, the model performance is evaluated by 5-fold cross-validation, and the mean square error (MSE) of the initial recommendation model is calculated. The mean square error is taken as the error value of the trained initial recommendation model, and the specific process is as follows:

[0066]

[0067] Subsequently, the error value of the trained initial recommendation model is compared with the pre-set error threshold. If the error value of the trained initial recommendation model is lower than the pre-set error threshold, the trained initial recommendation model is taken as the target recommendation model. If the error value of the trained initial recommendation model is not lower than the pre-set error threshold, the trained initial recommendation model is retrained.

[0068] According to the technical scheme provided in the embodiments of the present application, the target prediction baseline of the initial recommendation model is determined according to the initial prediction baseline. The initial recommendation model is trained according to the target prediction baseline and the sample training set. The target prediction value of the trained initial recommendation model is determined according to the sample test set, and the error value of the trained initial recommendation model is determined based on the standard value corresponding to the sample test set and the target prediction value. If the error value of the trained initial recommendation model is lower than the pre-set error threshold, the trained initial recommendation model is taken as the target recommendation model, thereby ensuring the accuracy of the target recommendation model.

[0069] In some embodiments, as shown in Figure 5 determining the target prediction value of the trained initial recommendation model according to the sample test set includes:

[0070] S501, obtaining a sample test set;

[0071] S502, testing the trained weak model based on the sample test set to obtain a first prediction value corresponding to the sample test set;

[0072] S502, testing the trained initial recommendation model according to the sample test set to obtain a second prediction value corresponding to the sample test set;

[0073] S503, determining a target prediction value according to the first prediction value and the second prediction value.

[0074] It can be understood that, since the initial recommendation model is determined by using the initial prediction baseline instead of simply using the original target value, the initial recommendation model learns the deviation between the initial prediction baseline and the original prediction baseline to optimize the model performance of the initial recommendation model, so the influence of the initial prediction baseline of the weak model on the initial recommendation model also needs to be considered in the process of determining the prediction value by testing the trained initial recommendation model according to the sample test set.

[0075] Based on the above reasons, after obtaining the sample test set, the trained weak model is tested based on the sample test set to obtain the first prediction value corresponding to the sample test set; the trained initial recommendation model is tested according to the sample test set to obtain the second prediction value corresponding to the sample test set; and then the target prediction value is determined according to the first prediction value and the second prediction value.

[0076] According to the technical scheme provided in the embodiments of the present application, the sample test set is obtained; the trained weak model is tested based on the sample test set to obtain the first prediction value corresponding to the sample test set; the trained initial recommendation model is tested according to the sample test set to obtain the second prediction value corresponding to the sample test set; and the target prediction value is determined according to the first prediction value and the second prediction value. The present application considers the influence of the initial prediction baseline on the initial recommendation model, and also considers the first prediction value of the sample test set in the weak model in the process of determining the prediction value of the trained initial recommendation model, thereby ensuring the accuracy of the obtained target prediction value.

[0077] In some embodiments, as shown in Figure 6 determining the target prediction value according to the first prediction value and the second prediction value includes:

[0078] S601, converting the first prediction value into a first logic value and converting the second prediction value into a second logic value;

[0079] S602, adding the first logic value and the second logic value to obtain the target prediction value.

[0080] It can be understood that, in order to make the first prediction value and the second prediction value retain more information, the present application will convert the first prediction value into a first logic value and convert the second prediction value into a second logic value, and the specific conversion method is the same as the method of "converting the average prediction value into a logic value", which will not be repeated here.

[0081] After converting the first prediction value into a first logic value and converting the second prediction value into a second logic value, the present application will also add the first logic value and the second logic value to obtain the target prediction value.

[0082] According to the technical scheme provided in the embodiment of the application, the first prediction value is converted into a first logic value, and the second prediction value is converted into a second logic value; the first logic value and the second logic value are added to obtain a target prediction value. The application takes into account the influence of the initial prediction baseline on the initial recommendation model, and in the process of determining the prediction value of the trained initial recommendation model, the first prediction value of the sample test set in the weak model is also taken into account, thereby ensuring the accuracy of the obtained target prediction value.

[0083] In order to better understand the application, a more specific embodiment is provided for illustration. The application provides a target recommendation method, as shown in Figure 7 The method first determines an initial prediction baseline through a weak model, and the initial prediction baseline is input as a baseline of a gradient boosting model, thereby significantly improving the training efficiency and prediction performance of the gradient boosting model.

[0084] As shown in Figure 7 The application first performs data loading and preprocessing, specifically by extracting a sample training set and a sample test set from an original data set, and then performing feature separation and standardization processing on the sample training set and the sample test set. Then, baseline generation is performed, specifically by using a weak model (Ridge regression) to generate out-of-fold prediction values (OOF) of the sample training set and prediction values of the sample test set, and converting them into logits form as baseline input of the gradient boosting model. Then, baseline model enhancement training is performed, specifically by then training data through a CatBoost model, using logits values as an initial prediction baseline (Baseline), learning the bias between the initial prediction baseline and the original prediction baseline, and finally performing performance evaluation by cross-validation and mean square error (MSE) to evaluate the performance of the trained CatBoost model.

[0085] The following describes each step in detail. The application first recommends relevant associated data from a public or business data set, which usually includes input features and label data (output). The feature data is used to describe the attributes or characteristics of the classification object, such as user behavior data, device features, environmental information, etc. The label data is used to represent the target variable of the classification result, such as whether to pass the audit, whether to purchase goods, disease category, etc.

[0086] Then, the data is split into features (input) and labels (output), and the obtained associated data is divided into a sample training set (used for subsequent weak model and CatBoost model training) and a sample test set (used for CatBoost model evaluation), usually in a ratio of 80:20, as shown in Figure 8 .

[0087] It can be understood that the application will also normalize or standardize the data in the process of dividing into a sample training set and a sample test set to reduce the influence of different characteristic value ranges on the subsequent model.

[0088] The application determines an initial prediction baseline through out-of-fold prediction, where the out-of-fold prediction is a prediction result generated through cross-validation for maximizing data utilization. Specifically, the sample training set is divided into several folds (usually 5 folds), and each time 4 folds of data are used for training and the remaining 1 fold of data is used for prediction, and finally all prediction values are combined to obtain the initial prediction baseline. In the application, a ridge regression model is used as a weak model to generate OOF prediction values of the training data through 5-fold cross-validation and to predict the test data. The out-of-fold prediction values reflect the understanding of the weak model for data distribution and can be used as the initial prediction baseline of the gradient boosting model.

[0089] It can be understood that 5-fold cross-validation is performed on the sample training set, and each time the model is trained using only 4 folds of data. Prediction values are generated on the remaining 1 fold of data, and after the prediction of all fold data is completed in turn, complete prediction values (OOF values) are obtained. The sample test set is used to generate prediction values for the trained ridge regression model, the above steps are repeated multiple times, and the average prediction value of the sample test set in multiple predictions is taken as the first prediction value of the test set.

[0090] As shown in Figure 9 The application also converts the prediction value P into logits value as the initial prediction baseline, and the formula is as follows:

[0091]

[0092] The logits value represents the log-odds, which retains more information than the probability value and is suitable for the baseline input of CatBoost.

[0093] It can be understood that the CatBoost model uses logits value as the initial prediction baseline (Baseline) during training, rather than simply using the original target value. CatBoost optimizes the performance of the model itself by learning the deviation (Residual) between the original target value and the initial prediction baseline.

[0094] Continuing the above example, the target prediction baseline of the CatBoost model is the original target value minus the logits baseline value, that is, the target prediction baseline = original target value - initial prediction baseline.

[0095] As shown in Figure 10As shown, after the CatBoost model training is completed, in the prediction stage, the result generated by the CatBoost model represents the deviation value, which needs to be added to the logits baseline value of the sample test set to obtain the final prediction value. The sample test set is used to test the trained weak model, and the obtained prediction value is recorded as the first prediction value. The first prediction value is converted into a first logical value. The sample test set is used to test the trained initial recommendation model, and the obtained prediction value is recorded as the second prediction value. The second prediction value is converted into a second logical value, then the target prediction value = CatBoost prediction deviation value (second prediction value) + first prediction value.

[0096] It is understood that in order to ensure that the CatBoost model can be fully trained, this application sets the number of iterations of the CatBoost model to 200.

[0097] like Figure 11 As shown, this application compares the predicted values ​​of the test set with the true values ​​and calculates the final error. Specifically, the model performance is evaluated through 5-fold cross-validation and the mean square error (MSE) is calculated:

[0098] This application uses the hyperparameter tuning of the Ridge base model and adopts the method of multiple iterations × 5-fold cross-validation to generate multiple prediction results and take the average of them. Compared with the results of single model training, this multiple integration strategy can effectively smooth the prediction error, reduce the impact of randomness on model performance, and further improve stability and generalization ability. Using the out-of-fold prediction (OOF) value as the initial prediction baseline of CatBoost can reflect the global characteristics of the data while reducing the risk of CatBoost overfitting. This method makes full use of the prediction information of the weak model (Ridge), provides a more optimized initial learning starting point for the gradient boosting algorithm (CatBoost), and significantly improves the performance and training efficiency of the model.

[0099] This application uses the predicted probabilities generated by the Ridge model as input to the CatBoost model after logits conversion, rather than directly using classification labels (such as 1 / 0). In the feature construction of the CatBoost model, the multi-dimensional information of the original data features and baseline features is retained, enabling the model to learn multi-level data representations simultaneously. This feature enhancement strategy minimizes the compression or information loss of the predicted values, ensures the effective combination of baseline predictions and original features in training, and thus improves the learning ability and prediction accuracy of the model.

[0100] In some embodiments, after the target recommendation model outputs recommendation data based on the association data and performs target recommendation based on the recommendation data, the method further includes: displaying the target recommendation result through the interactive interface, so that the relevant personnel can quickly learn the target recommendation result.

[0101] All the optional technical solutions described above can be combined to form optional embodiments of the present application, which will not be described one by one here.

[0102] The following is an apparatus embodiment of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the apparatus embodiments of the present application, please refer to the method embodiments of the present application.

[0103] The present embodiment also provides a target recommendation device, as shown in the figure, the device comprises: Figure 12

[0104] The acquisition module 1201 is configured to acquire association data of an object.

[0105] The processing module 1202 is configured to input the association data into a target recommendation model, the target recommendation model being generated based on a sample training set and an initial prediction baseline, the initial prediction baseline being determined by performing out-of-sample prediction on the sample training set.

[0106] The recommendation module 1203 is configured to acquire recommendation data output by the target recommendation model based on the association data, and perform target recommendation based on the recommendation data.

[0107] In some examples, the processing module 1202 is further configured to acquire the sample training set; perform multiple sample divisions on the sample training set to obtain multiple sample combinations, each sample combination including multiple groups of training samples and one group of test samples; train a weak model according to the multiple groups of training samples in each sample combination, and test the trained weak model based on the test samples to obtain a prediction value corresponding to each sample combination; and determine the initial prediction baseline according to the prediction value corresponding to each sample combination.

[0108] In some examples, the processing module 1202 is further configured to determine an average prediction value according to the prediction values corresponding to the multiple sample combinations; and determine the initial prediction baseline according to the average prediction value.

[0109] ​In some examples, the processing module 1202 is further configured to determine a target prediction baseline of the initial recommendation model according to the initial prediction baseline; train the initial recommendation model according to the target prediction baseline and the sample training set; determine a target prediction value of the trained initial recommendation model according to the sample test set, and determine an error value of the trained initial recommendation model based on a standard value corresponding to the sample test set and the target prediction value; and if the error value of the trained initial recommendation model is lower than a pre-set error threshold, take the trained initial recommendation model as the target recommendation model.

[0110] In some examples, the processing module 1202 is further configured to obtain a sample test set; test the trained weak model based on the sample test set to obtain a first prediction value corresponding to the sample test set; test the trained initial recommendation model based on the sample test set to obtain a second prediction value corresponding to the sample test set; and determine a target prediction value according to the first prediction value and the second prediction value.

[0111] In some examples, the processing module 1202 is further configured to convert the first prediction value into a first logic value, and convert the second prediction value into a second logic value; and add the first logic value and the second logic value to obtain the target prediction value.

[0112] In some examples, the recommendation module 1203 is further configured to display the target recommendation result through an interactive interface.

[0113] According to the technical scheme provided in the embodiments of the present application, the target recommendation device obtains the associated data of the object; inputs the associated data into the target recommendation model, the target recommendation model is generated based on the sample training set and the initial prediction baseline, and the initial prediction baseline is determined by out-of-sample prediction on the sample training set; obtains the recommendation data output by the target recommendation model based on the associated data, and performs target recommendation based on the recommendation data, wherein the initial prediction baseline determined by out-of-sample prediction on the sample training set retains most of the information of the original data features and the baseline features, so that the target recommendation model in the training stage can learn multiple levels of data identification at the same time, the features of the sample training set are enhanced, the learning ability and prediction accuracy of the target recommendation model are improved, and then the target recommendation model can accurately predict based on the associated data to obtain accurate recommendation data, thereby avoiding the problem that in related technologies, the user cannot be accurately recommended the target of interest, resulting in low recommendation accuracy.

[0114] Figure 13 FIG. 13 is a schematic diagram of an electronic device 13 provided in an embodiment of the present application. As shown in FIG. 13, the electronic device 13 includes a processor 1301, a memory 1302, a communication interface 1303, and a power supply 1304. Figure 13As shown, the electronic device 13 of this embodiment includes a processor 1301, a memory 1302, and a computer program 1303 stored in the memory 1302 and executable on the processor 1301. The processor 1301 implements the steps in each of the above method embodiments when executing the computer program 1303. Alternatively, the processor 1301 implements the functions of each module / unit in each of the above apparatus embodiments when executing the computer program 1303.

[0115] The electronic device 13 can be a desktop computer, a notebook computer, a palm computer, a cloud server, or the like. The electronic device 13 can include but is not limited to the processor 1301 and the memory 1302. Those skilled in the art can understand that the electronic device 13 can include more or less components, or different components than those shown. Figure 13 The electronic device 13 is merely an example and does not constitute a limitation on the electronic device 13, which can include more or less components, or different components than those shown.

[0116] The processor 1301 can be a central processing unit (CPU), or other general purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or the like.

[0117] The memory 1302 can be an internal storage unit of the electronic device 13, such as a hard disk or a memory of the electronic device 13. The memory 1302 can also be an external storage device of the electronic device 13, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, or the like. The memory 1302 can also include both an internal storage unit and an external storage device of the electronic device 13. The memory 1302 is used to store computer programs and other programs and data required by the electronic device.

[0118] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of software functional unit.

[0119] If the integrated module / unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be instructed by a computer program to related hardware to complete. The computer program can be stored in a computer readable storage medium, and the computer program can implement the steps of each method embodiment described above when executed by a processor. The computer program can include computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium can include any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electric carrier signal and telecommunication signal.

[0120] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A target recommendation method, characterized in that: The method comprises: Get the associated data of the object; Inputting the associated data into a target recommendation model, wherein the target recommendation model is generated by training a sample training set and an initial prediction baseline, wherein the initial prediction baseline is determined by performing out-of-fold prediction on the sample training set; The recommended data output by the target recommendation model based on the associated data is obtained, and a target recommendation is performed based on the recommended data.

2. The method according to claim 1, characterized in that Before inputting the associated data into the target recommendation model, the method further includes: Obtaining the sample training set; Performing multiple sample divisions on the sample training set to obtain multiple sample combinations, each of the sample combinations including multiple groups of training samples and one group of test samples; Training the weak model based on the multiple groups of training samples in each of the sample combinations, and testing the trained weak model based on the test samples to obtain a prediction value corresponding to each of the sample combinations; The initial prediction baseline is determined according to the prediction value corresponding to each sample combination.

3. The method according to claim 2, characterized in that Determining the initial prediction baseline according to the prediction value corresponding to each sample combination includes: Determining an average prediction value based on the prediction values ​​corresponding to the plurality of sample combinations; The initial prediction baseline is determined according to the average prediction value.

4. The method according to claim 2, characterized in that After determining the initial prediction baseline according to the prediction value corresponding to each sample combination, the method further includes: Determining a target prediction baseline for an initial recommendation model based on the initial prediction baseline; Training an initial recommendation model based on the target prediction baseline and the sample training set; Determining a target prediction value of the trained initial recommendation model based on a sample test set, and determining an error value of the trained initial recommendation model based on a standard value corresponding to the sample test set and the target prediction value; If the error value of the trained initial recommendation model is lower than a preset error threshold, the trained initial recommendation model is used as the target recommendation model.

5. The method according to claim 4, characterized in that Determining a target prediction value of the trained initial recommendation model based on the sample test set includes: Obtaining the sample test set; Testing the trained weak model based on the sample test set to obtain a first prediction value corresponding to the sample test set; Testing the trained initial recommendation model according to the sample test set to obtain a second prediction value corresponding to the sample test set; The target prediction value is determined according to the first prediction value and the second prediction value.

6. The method according to claim 5, characterized in that Determining the target predicted value according to the first predicted value and the second predicted value includes: Converting the first predicted value into a first logical value, and converting the second predicted value into a second logical value; The first logic value and the second logic value are added to obtain the target prediction value.

7. The method according to claim 3, characterized in that After obtaining the recommendation data output by the target recommendation model based on the associated data and performing target recommendation based on the recommendation data, the method further includes: displaying the target recommendation result through an interactive interface.

8. A target recommendation device, characterized in that: The device comprises: The acquisition module is used to obtain the associated data of the object; a processing module, configured to input the associated data into a target recommendation model, wherein the target recommendation model is generated by training a sample training set and an initial prediction baseline, wherein the initial prediction baseline is determined by performing out-of-fold prediction on the sample training set; The recommendation module is used to obtain the recommendation data output by the target recommendation model based on the associated data, and perform target recommendation based on the recommendation data.

9. An electronic device comprising 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 of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images