TW201928709A - Method and apparatus for merging model prediction values, and device - Google Patents

Abstract

Disclosed are a method and apparatus for merging model prediction values, and a device. The method for merging model prediction values comprises: on the basis of a given number of samples, binning a prediction value of an online prediction model and a prediction value of an offline prediction model separately according to a set binning method; according to the binning result, converting the first prediction value of each sample into a first interval feature corresponding to the interval in which the first prediction value is located, and converting the second prediction value of each sample into a second interval feature corresponding to the interval in which the second prediction value is located; and using the first interval feature and the second interval feature corresponding to each sample, and a sample tag to form sample data after the conversion, and using said sample data to train a model, the trained model being used for merging the prediction value of the online prediction model and the prediction value of the offline prediction model to obtain a final prediction value.

Description

Method, device and equipment for fusing model prediction values

This specification relates to the field of machine learning technology, and in particular, to a method, a device, and a device for fusing model prediction values.

Machine learning algorithms are a type of algorithms that can automatically analyze patterns from data and use the patterns to predict unknown data. They are widely used in many fields. In practical applications, it includes online prediction models and offline prediction models. Among them, offline prediction models are usually implemented by timing tasks. Its advantages are that it can incorporate higher-dimensional features and use more complex algorithms to achieve more accurate. Prediction effect; however, the prediction process is usually time consuming due to its many features and complex algorithms. Compared with offline prediction models, online prediction models can use lower-dimensional features and simpler algorithms to achieve more efficient predictions. The disadvantages are insufficient features and low accuracy. It can be seen that the online prediction model and the offline prediction model each have their own advantages. How to reasonably integrate the two is an issue to be solved in the industry.

In view of the above technical problems, the embodiments of the present specification provide a method, an apparatus, and a device for fusing model predictions. The technical solution is as follows: In one aspect, a method for fusing model predictions is provided, including: Based on a given For several samples, the predicted values of the online prediction model and the predicted values of the offline prediction model are binned respectively according to a set binning method, wherein each of the plurality of samples includes a first predicted value and a second predicted value. And the label of the sample, the first predicted value is predicted by the online prediction model, and the second predicted value is predicted by the offline prediction model; According to the results of the binning, the first predicted value of each sample is converted to the first prediction. A first interval feature corresponding to the interval in which the value is located, converting the second prediction value of each sample into a second interval feature corresponding to the interval in which the second prediction value is located; the first interval corresponding to each sample Feature, the second interval feature, and the label of the sample constitute the transformed sample data, and The trained sample data is used to train the model. The trained model is used to fuse the prediction value of the online prediction model and the prediction value of the offline prediction model to obtain the final prediction value. In one aspect, a method for fusing model predictions is provided, including: Obtaining business data generated by a target user within a first period of time, determining input features based on the business data and inputting to an online prediction model, and outputting the first A prediction value; obtaining a second prediction value corresponding to the target user obtained by using an offline prediction model, wherein the input characteristics of the offline prediction model are based on the business generated by the target user in the second time period; Determined by characteristics; obtaining a binning result of the first prediction value of the online prediction model and the second prediction value of the offline prediction model, and determining the first interval and the second prediction respectively where the first prediction value is located; The second interval in which the value is located; according to the first interval and the second interval, using a model trained in advance to fuse the first prediction value and the second prediction value to obtain a final fusion prediction Value, the fusion prediction value is used to determine a label of the target user. In one aspect, a device for fusing model predictions is provided, including: a binning unit, based on a given number of samples, according to a set binning method to separately predict the predictions of an online prediction model and the predictions of an offline prediction model Binning is performed, wherein each of the plurality of samples includes: a first prediction value, a second prediction value, and a label of the sample, the first prediction value is predicted by an online prediction model, and the second prediction value is obtained by offline The prediction model is predicted; the feature conversion unit converts the first prediction value of each sample into a first interval feature corresponding to the interval where the first prediction value is located according to the binning result, and converts the second prediction value of each sample Transformed into a second interval feature corresponding to the interval in which the second predicted value is located; a training unit, forming the transformed sample with the first interval feature, the second interval feature, and a label of the sample corresponding to each sample Data, and use the transformed sample data to train the model. The trained model is used for the online prediction model. The predicted value and the measured value of the offline prediction models obtained are fused final prediction value. In one aspect, a device for fusing model predictions is provided, including: Online score prediction unit, which acquires business data generated by a target user within a first time period before a trigger time, and determines an input based on the business data The features are input to an online prediction model, and a first prediction value is output, the online prediction model is used to predict a user's label; an offline score obtaining unit, which obtains a second corresponding to the target user obtained by using the offline prediction model; Prediction value, wherein the input characteristics of the offline prediction model are determined according to the business characteristics of the target user in the past second time period, and the offline prediction model is used to predict the label of the user; The determining unit determines the first interval in which the first prediction value is located and the third position in which the second prediction value is located according to a result of binning the prediction value of the online prediction model and the prediction value of the offline prediction model in advance. Two intervals; score fusion unit, according to the first interval and the second interval Interval, using a model trained in advance to fuse the first prediction value and the second prediction value to obtain a final fusion prediction value, where the fusion prediction value is used to determine a label of the target user. In one aspect, a proposed computer device includes: (i) a processor; (ii) a memory for storing processor-executable instructions; (ii) the processor is configured to: (ii) based on a given number of samples, respectively, according to a set binning method; The prediction value of the online prediction model and the prediction value of the offline prediction model are binned, wherein each of the samples includes: a first prediction value, a second prediction value, and a label of the sample, and the first prediction value Obtained from the online prediction model and the second predicted value from the offline prediction model; According to the results of the binning, convert the first prediction value of each sample into the first interval feature corresponding to the interval where the first prediction value is located. , Converting the second prediction value of each sample into a second interval feature corresponding to the interval in which the second prediction value is located; using the first interval feature, the second interval feature, and the sample's The labels constitute the transformed sample data and the transformed sample data is used to train the model. The training As a model for a prediction value predicted values and off-line prediction model predictive models are fused to give a final prediction value. In one aspect, a proposed computer device includes: a processor; a memory for storing processor-executable instructions; a processor configured to: an online score prediction unit that obtains a target user's Business data generated in the first time period, determining input characteristics according to the business data and inputting it to an online prediction model, and outputting a first prediction value, the online prediction model is used to predict a user's label; an offline score obtaining unit, A second prediction value corresponding to the target user obtained by using an offline prediction model is obtained, wherein the input characteristics of the offline prediction model are based on business characteristics generated by the target user in the past second time period. It is determined that the offline prediction model is used to predict a user's tag; an interval determination unit determines the first prediction value according to a result of binning the prediction value of the online prediction model and the prediction value of the offline prediction model in advance; The first interval and the second prediction A second interval in which it is located; a score fusion unit, based on the first interval and the second interval, using a model trained in advance to fuse the first prediction value and the second prediction value to obtain The final fusion prediction value is used to determine a label of the target user. The effects produced by the technical solutions provided in the embodiments of the present specification include: 融合 Fusion of the prediction value of the line prediction model and the prediction value of the offline prediction model through a model obtained by machine learning, and finally using the fused score To predict the tags of users, while improving the accuracy of predicting the tags of users, it also meets the requirements of the service for low latency. It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and should not limit the embodiments of the present specification. In addition, any one of the embodiments in this specification does not need to achieve all the effects described above.

In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present specification, the technical solutions in the embodiments of the present specification will be described in detail below with reference to the drawings in the embodiments of the present specification. Obviously, the described implementations The examples are only a part of the embodiments of this specification, but not all the examples. Based on the embodiments in this specification, all other embodiments obtained by those with ordinary knowledge in the art should fall within the scope of protection. Referring to FIG. 1, in an embodiment of the present specification, a method for fusing model predictions is used to fuse scores obtained from online prediction models and scores obtained from offline prediction models. The method It may include the following steps 101 to 104, where: Step 101: Obtain business data generated by a target user within a first period of time, determine input characteristics according to the business data and input them to an online prediction model, and output a first prediction value. Step 102: Obtain a second prediction value corresponding to the target user obtained by using an offline prediction model, wherein the input characteristics of the offline prediction model are based on business characteristics generated by the target user within a second time period. To determine. In this paper, both the online prediction model and the offline prediction model are models constructed by using machine learning algorithms to predict user labels. The user tags predicted by these two models can be related to specific services. For example, for an online payment service, the user tags predicted need to be divided into: "high-risk users", "medium-risk use" "," Low-risk users ", and so on. For an information recommendation business, the user labels that need to be predicted can be divided into: "sports", "educational", "financial", and so on. Both the online prediction model and the offline prediction model are trained using a certain number of training samples. Each of these training samples may include: sample users generated in the process of participating in a specific business (such as online payment business) One or more behavioral data, and tags identified by sample users. Among them, the same batch of samples can be used to train the above online prediction model and offline prediction model, or two different batches of samples can be used to train the online prediction model and offline prediction model, which is not limited in this article. In the embodiment of the present specification, the offline prediction model may be implemented by a timing task, such as: performing offline score prediction once a day at a specified time or a specified time period, and the prediction process may be directed to all users; while online The prediction model can be triggered by the operation of a specific user. For example, the behavior of a user clicking a webpage can trigger the score calculation process of the online prediction model. Because offline prediction models generally use higher-dimensional feature data than online prediction models, the time range of feature data can also be longer, and more sophisticated algorithms can be used. As shown in Figure 1, for a specific example, on day T, the offline prediction model can obtain business data (characteristic A) generated by each user during the process of participating in a specific business on day T-1. According to the obtained The business data (feature A) is processed accordingly, and the input features can be obtained and input into the offline prediction model, and the offline prediction scores (that is, the second prediction values in the text) of each user are obtained and written into the database X. As for the online prediction model, the user's online feature data (feature B) can be continuously collected and written into the database Y, where the online feature data can be a standard generated by the user in the process of participating in a specific business Real-time business data, for example: the trigger time of online prediction is t1, then the online characteristic data can be business data generated during the time period from t0 to t1 (such as 3 minutes). It can be seen that after the user request for initiating the prediction process arrives, the scheduler needs to perform two tasks. One is to read from the database X the last time corresponding to the target user calculated by the offline prediction model. The second prediction value; the second is to read the online characteristic data of the target user from the database Y to perform the subsequent value prediction process of the online prediction model. So far, for any target user, a prediction score can be obtained through the online prediction model, and a prediction score can be obtained through the offline prediction model. Step 103: According to the results of binning the prediction values of the online prediction model and the prediction value of the offline prediction model in advance, determine the first interval in which the first prediction value is located and the second interval in which the second prediction value is located. Second interval. Step 104: According to the first interval and the second interval, using a pre-trained model to fuse the first prediction value and the second prediction value to obtain a final fusion prediction value. The fusion prediction value is used to determine a label of the target user. In an optional embodiment, step 104 may specifically include: Step 1041: Obtain a first weight corresponding to the first interval and a first weight corresponding to the first interval based on a predetermined weight corresponding to each interval obtained by binning. The second weight corresponding to the two intervals. The parameters to be trained of the model include weights corresponding to the intervals obtained by binning. Step 1042: Use the first weight and the second weight to determine a fusion prediction value, where the fusion prediction value is used to determine a label of the target user. Because the above steps 103 to 104 need to be implemented based on the binning results and the weights corresponding to the intervals obtained by the binning, before introducing the steps 103 to 104 in detail, a method for determining the fusion weight needs to be introduced. As shown in FIG. 2, in an embodiment, the method includes steps 201 to 203, where: Step 201: Based on a given number of samples, the prediction value of the online prediction model and the offline prediction are respectively set according to a set binning method. The prediction value of the model is binned, wherein each of the plurality of samples includes: a first prediction value, a second prediction value, and a label of the sample, the first prediction value being obtained from the online prediction model prediction, and the second The predicted value is obtained by the offline prediction model.的 The samples mentioned in step 201 may be the same as the samples used to train the offline prediction model and / or the online prediction model. Of course, the samples may also be different samples, which is not limited. In one embodiment, the set binning method may be an entropy-based binning method. The entropy-based binning method considers the value of the dependent variable when binning, so that the minimum entropy (minimumentropy) is reached after binning. The advantage of the entropy-based binning method is that it can show better discrimination in high score areas. Of course, the set binning method may also be a Gini-based binning method, an iso-frequency binning method, or the like. Step 202: According to the binning result, the first prediction value of each sample is converted into a first interval feature corresponding to the interval in which the first prediction value is located, and the second prediction value of each sample is converted into the second prediction value. The second interval feature corresponding to the interval in which the predicted value is located. In an example, assuming that the first prediction value and the second prediction value are both between 0 and 1, after dividing the prediction value of the online prediction model, the obtained segmentation points include: 0, 0.1, 0.13 , 0.15, 0.2, 0.3, 0.5, 1; After binning the prediction values of the offline prediction model, the resulting segmentation points include: 0, 0.03, 0.05, 0.08, 0.09, 0.11, 0.13, 1; that is, The output values of the online prediction model and the offline prediction model are divided into 7 intervals after binning. In one embodiment, the one-hot rule can be used to implement the feature conversion in step 202. Assume that the first prediction value of a sample is 0.17 and the second prediction value is 0.12. Since 0.17 is in the fourth interval (0.15, 0.2) and 0.12 is in the sixth interval (0.11, 0.13), one-hot is used. The rule can convert the first prediction value: 0.17 to the first interval feature: on-bin-0001000 ("on-bin" is the identifier of the online prediction model), and the second prediction value: 0.12 to the second interval feature: off -bin-0000010 ("off-bin" is an identification of the offline prediction model). In the same way, feature conversion can be performed on the first predicted value and the second predicted value in other samples one by one. Step 203: Use the first interval feature, the second interval feature, and the label of the sample to form transformed sample data, and use the transformed sample data to train a model. The trained model is used The final prediction value is obtained by fusing the prediction value of the online prediction model and the prediction value of the offline prediction model. Wherein, the transformed sample data may include other data in addition to the first interval feature, the second interval feature, and a label of the sample. That is, the "composition" is not closed. In the above example, before the feature conversion, a piece of sample data is, for example: {0.17, 0.12, "medium risk user"}; After the feature conversion, a new piece of sample data is, for example: {0001000, 0000010, "Medium-risk users"} 的 The model to be trained in this article can be a linear model or a non-linear model. In an embodiment using a linear model, the parameters to be trained of the model can include weights corresponding to the intervals obtained by binning. The weight may be used to fuse the prediction value of the line prediction model and the prediction value of the offline prediction model to obtain a final prediction value. The model to be trained can be a Logistic Regression (LR) model, in which each interval obtained by binning can be assigned a weight, and the weight is used as a parameter of the LR model for training, and finally each weight value can be solved . The above weight may be a score of the corresponding interval. The score is not only a global importance trade-off and learning between different model features (online or offline models), but also between each score interval. Following the example mentioned above, the following weights can finally be obtained: the weight of the interval (0,0.1) on-bin-1 = 1.054, ... the weight of the interval 0.5 (0.5,1) on-bin-7 = 4.439; the interval (( 0,0.03) weight off-bin-1 = 0.604, ... interval (0.13,1) weight off-bin-7 = 3.237. Next, the above steps 103 to 104 will be described in combination with the above specific examples. Assume that for a target user, the first prediction value obtained through the online prediction model is 0.66, and the second prediction value obtained through the offline prediction model is 0.25. In combination with the above example, first in step 103, determine the first The first interval where the predicted value 0.4 is located is: (0.5, 1), and the second interval where the second predicted value 0.25 is located is: (0.13, 1). Then in step 1041, based on a predetermined weight corresponding to each interval obtained by binning, a first weight corresponding to the first interval: (0.5, 1) is: 4.439, and the second interval : (0.13,1) The corresponding second weight is: 3.237. Finally, in step 1042, the final fusion prediction value may be determined according to the first weight and the second weight. In an optional embodiment, the first weight and the second weight may be summed, The summed result is used as the fusion prediction value, that is, the fusion prediction value = 4.439 + 3.237 = 7.676. Of course, the specific way of fusion is not limited to summation, such as averaging. Finally, how to use the fusion prediction value may be decided according to a specific service. The effects produced by the technical solutions provided in the embodiments of the present specification include: 融合 Fusion of the prediction value of the line prediction model and the prediction value of the offline prediction model through weights obtained by machine learning, and finally using the score obtained by the fusion To predict the tags of users, while improving the accuracy of predicting the tags of users, it also meets the requirements of the service for low latency. In addition, using entropy-based binning and logistic regression models, the online model scores and offline model scores are effectively integrated, so that the comparability between online and offline scores is adaptively adjusted during the machine learning process. Corresponding to the above method embodiment, the embodiment of the present specification also provides a device for fusing model prediction values. Referring to FIG. 3, in an embodiment, in the training phase of fusion weights, a device 300 for determining fusion weights may include: a binning unit 301 configured to: based on a given number of samples, according to a set binning method To bin the prediction value of the online prediction model and the prediction value of the offline prediction model, wherein each of the samples includes: a first prediction value, a second prediction value, and a label of the sample; A prediction value is obtained from the online prediction model, and a second prediction value is obtained from the offline prediction model. The feature conversion unit 302 is configured to convert the first prediction value of each sample into the first prediction value according to the binning result. The first interval feature corresponding to the interval in which the predicted value is located converts the second predicted value of each sample into the second interval feature corresponding to the interval in which the second predicted value is located. The training unit 303 is configured to: The first interval feature, the second interval feature, and the label of the sample corresponding to a sample constitute the transformed sample data, and use the conversion The trained sample data is used to train the model. The trained model is used to fuse the prediction value of the online prediction model and the prediction value of the offline prediction model to obtain the final prediction value. Referring to FIG. 4, in an embodiment, in a score fusion stage, a device 400 for fusing model prediction values may include: an online score prediction unit 401 configured to: obtain a target user before a trigger time Business data generated in the first period of time, determine input characteristics according to the business data and input it to an online prediction model, and output a first prediction value, the online prediction model is used to predict a user's label; an offline score obtaining unit 402, configured to: obtain a second prediction value corresponding to the target user obtained by using an offline prediction model, wherein an input feature of the offline prediction model is based on a second time period of the target user in the past The offline prediction model is used to predict the user's label based on the internally generated business characteristics. The interval determination unit 403 is configured to perform binning according to the prediction value of the online prediction model and the prediction value of the offline prediction model in advance. Results of determining the first interval and the second prediction where the first prediction value is located, respectively A second interval in which the value is located; a weight determining unit 404 is configured to: according to the first interval and the second interval, use a model trained in advance to perform a first prediction on the second prediction and the second prediction Value fusion to obtain a final fusion prediction value, and the fusion prediction value is used to determine a label of the target user. In an optional embodiment, the score fusion unit 404 may include: a weight determination sub-unit that obtains a first weight corresponding to the first interval based on a predetermined weight corresponding to each interval obtained by binning; And a second weight corresponding to the second interval; a fusion subunit, using the first weight and the second weight to determine a fusion prediction value, where the fusion prediction value is used to determine a label of the target user . In an embodiment, the fusion subunit may be configured to: sum the first weight and the second weight, and use the summed result as a fusion prediction value.的 For the implementation process of the functions and functions of each module in the above device, see the implementation process of the corresponding steps in the above method for details. An embodiment of the present specification also provides a computer device (such as a server), which includes at least a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor implements the program when the program is executed. The aforementioned method. FIG. 5 shows a more specific schematic diagram of the hardware structure of a computing device provided by an embodiment of this specification. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. . The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 implement a communication connection within the device through a bus 1050. The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit, central processing unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits. Relevant programs are executed to implement the technical solutions provided by the embodiments of this specification. Memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage devices, dynamic storage devices, and the like. The memory 1020 may store an operating system and other applications. When the technical solution provided by the embodiment of the present specification is implemented by software or firmware, the relevant code is stored in the memory 1020 and is called and executed by the processor 1010. . The input / output interface 1030 is used to connect input / output modules to realize information input and output. The input / output / module can be configured as a component in the device (not shown in the figure), or it can be externally connected to the device to provide the corresponding function. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, and the like, and the output device may include a display, a speaker, a vibrator, and an indicator light. The communication interface 1040 is used to connect a communication module (not shown in the figure), so as to realize communication interaction between the device and other devices. The communication module can realize communication through a wired method (such as USB, network cable, etc.), and can also realize communication through a wireless method (such as mobile network, WIFI, Bluetooth, etc.). The bus 1050 includes a path for transmitting information between various components of the device (such as the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040). It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in the specific implementation process, the device may further include a device that implements normal execution. Required additional components. In addition, it can be understood by those having ordinary knowledge in the art that the above-mentioned device may also include only components necessary to implement the solutions of the embodiments of the present specification, and not necessarily all the components shown in the drawings. From the description of the above embodiments, it can be known that those with ordinary knowledge in the art can clearly understand that the embodiments of this specification can be implemented by means of software plus necessary universal hardware platforms. Based on this understanding, the technical solutions of the embodiments of the present specification can be reflected in the form of software products that are essentially or contribute to the existing technology. The computer software products can be stored in storage media, such as ROM / RAM, magnetic Disks, optical discs, etc. include several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the methods described in various embodiments or portions of the embodiments of this specification.的 The system, device, module, or unit described in the above embodiments may be implemented by a computer chip or entity, or a product with a certain function. A typical implementation device is a computer. The specific form of the computer can be a personal computer, a notebook computer, a mobile phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an e-mail receiving and sending device, and a game console. , Tablet, wearable, or a combination of any of these devices.实施 Each embodiment in this specification is described in a gradual manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, as for the device embodiment, since it is basically similar to the method embodiment, it is described relatively simply. For the relevant part, refer to the description of the method embodiment. The device embodiments described above are only schematic, and the modules described as separate components may or may not be physically separated. When implementing the solutions of the embodiments of this specification, the functions of each module may be In one or more software and / or hardware. It is also possible to select some or all of the modules according to actual needs to achieve the purpose of the solution of this embodiment. Those with ordinary knowledge in the field can understand and implement it without paying progressive labor. The above are only specific implementations of the embodiments of the present specification. It should be noted that for those with ordinary knowledge in the art, without departing from the principles of the embodiments of the present specification, several improvements and retouches can be made. These Improvement and retouching should also be regarded as the protection scope of the embodiments of the present specification.

201‧‧‧ steps

202‧‧‧step

203‧‧‧step

300‧‧‧ device

301‧‧‧dividing unit

302‧‧‧Feature Conversion Unit

303‧‧‧ training unit

400‧‧‧ device

401‧‧‧online score prediction unit

402‧‧‧Offline score obtaining unit

403‧‧‧ interval determination unit

404‧‧‧point fusion unit

1010‧‧‧ processor

1020‧‧‧Memory

1030‧‧‧ input / output interface

1040‧‧‧ communication interface

1050‧‧‧Bus

In order to more clearly explain the embodiments of the present specification or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only the present invention. Some of the embodiments described in the description of the specification may be obtained by those with ordinary knowledge in the art based on these drawings. FIG. 1 is a schematic flowchart of a method for fusing model prediction values according to an embodiment of the present specification; FIG. 2 is a process for determining a fusion weight provided by an embodiment of the present specification; FIG. 3 is an example of a model provided by the present specification. Schematic diagram of a device (weight training phase) for fusing predicted values; FIG. 4 is a schematic diagram of a device (score value fusion phase) for fusing model prediction values provided by an embodiment of the present specification; FIG. 5 is used to configure the device. Schematic diagram of the structure of a device in the embodiment of the specification.

Claims (14)

A method for fusing model prediction values, including: based on a given number of samples, and binning the prediction values of an online prediction model and the prediction values of an offline prediction model according to a set binning method, wherein the plurality of samples Each of the samples includes a first prediction value, a second prediction value, and a label of the sample. The first prediction value is obtained from an online prediction model, and the second prediction value is obtained from an offline prediction model. According to the results of binning , Converting the first prediction value of each sample into a first interval feature corresponding to the interval in which the first prediction value is located, and converting the second prediction value of each sample into a corresponding interval in which the second prediction value is located Second interval feature; The first interval feature, the second interval feature, and the label of the sample corresponding to each sample constitute the transformed sample data, and the transformed sample data is used to train the model. The model is used to fuse the prediction value of the online prediction model and the prediction value of the offline prediction model to obtain the final prediction value. According to the method of claim 1, the set binning method includes: an entropy-based binning method, a Gini-based binning method, or an iso-frequency binning method. According to the method described in claim 1, the parameters to be trained of the model include weights corresponding to each interval obtained by binning, and the weights are used to obtain the fusion of the prediction value of the line prediction model and the prediction value of the offline prediction model. The final predicted value. A method for fusing model predictions, including: acquiring business data generated by a target user within a first period of time, determining input characteristics based on the business data and inputting them to an online prediction model, and outputting a first prediction value; obtaining utilization A second prediction value corresponding to the target user obtained by the offline prediction model, wherein the input characteristics of the offline prediction model are determined according to the business characteristics generated by the target user in the second time period; Results of binning the first prediction value of the online prediction model and the second prediction value of the offline prediction model, and respectively determining a first interval in which the first prediction value is located and a second interval in which the second prediction value is located. Interval; using a model trained in advance to fuse the first predicted value and the second predicted value according to the first interval and the second interval to obtain a final fusion predicted value, the fusion prediction The value is used to determine the label of the target user. According to the method of claim 3, the use of a pre-trained model to fuse the first prediction value and the second prediction value to obtain a final fusion prediction value includes: based on a predetermined determination and binning The weights corresponding to the intervals are obtained, a first weight corresponding to the first interval and a second weight corresponding to the second interval are obtained, and the parameters to be trained of the model include those corresponding to each interval obtained by binning. Weights: (1) Use the first weight and the second weight to determine a fusion prediction value. According to the method of claim 5, the using the first weight and the second weight to determine a fusion prediction value includes: 求 summing the first weight and the second weight, and And the result is used as the fusion prediction value. A device for fusing model prediction values, including: (ii) a binning unit, based on a given number of samples, according to a set binning method to bin the prediction values of an online prediction model and the prediction values of an offline prediction model, respectively, Each of the plurality of samples includes: a first prediction value, a second prediction value, and a label of the sample, the first prediction value is obtained by an online prediction model prediction, and the second prediction value is obtained by an offline prediction model prediction; The conversion unit converts the first prediction value of each sample into a first interval feature corresponding to the interval in which the first prediction value is located, and converts the second prediction value of each sample into the second prediction value according to the binning result. The second interval feature corresponding to the interval in which the predicted value is located; The training unit uses the first interval feature, the second interval feature, and the label of the sample to form transformed sample data, and uses the transformed sample data. To train the model using the sample data of the model. The trained model is used to predict the prediction value of the online prediction model and the offline prediction model. Fusion to obtain a final measurement value for the predicted value. According to the device of claim 7, the set binning method includes: an entropy-based binning method, a Gini-based binning method, or an iso-frequency binning method. According to the apparatus of claim 7, the parameters to be trained of the model include weights corresponding to each interval obtained by binning, and the weights are used to obtain the fusion of the prediction value of the line prediction model and the prediction value of the offline prediction model. The final predicted value. A device for fusing model prediction values, comprising: an online score prediction unit, which acquires business data generated by a target user within a first time period before a trigger time, determines input characteristics based on the business data, and inputs the data to online prediction; Model, outputting a first prediction value, the online prediction model is used to predict a user's label; an offline score obtaining unit, obtaining a second prediction value corresponding to the target user obtained by using the offline prediction model, wherein The input characteristics of the offline prediction model are determined according to the business characteristics of the target user in the past second time period, and the offline prediction model is used to predict the user's label; Result of binning the prediction value of the online prediction model and the prediction value of the offline prediction model to determine the first interval in which the first prediction value is located and the second interval in which the second prediction value is located; Score fusion A unit, obtained by using pre-training according to the first interval and the second interval Model to fuse the first prediction value and the second prediction value to obtain a final fusion prediction value, and the fusion prediction value is used to determine a label of the target user. According to the apparatus of claim 10, the score fusion unit includes: a weight determining subunit, which obtains a first weight corresponding to the first interval based on a predetermined weight corresponding to each interval obtained by the binning and A second weight corresponding to the second interval; a fusion subunit, using the first weight and the second weight to determine a fusion prediction value, where the fusion prediction value is used to determine a label of the target user. According to the apparatus of claim 11, the fusion subunit is configured to: sum the first weight and the second weight, and use the summed result as a fusion prediction value. A computer device, comprising: a processor; a memory for storing processor-executable instructions; a processor configured to: a predicted value of an online prediction model according to a set of binning methods based on a given number of samples Binning with the prediction value of the offline prediction model, wherein each of the several samples includes a first prediction value, a second prediction value, and a label of the sample, and the first prediction value is obtained by prediction by the online prediction model. , The second prediction value is obtained by the offline prediction model; According to the binning result, the first prediction value of each sample is converted into a first interval feature corresponding to the interval in which the first prediction value is located, and the first The two prediction values are converted into a second interval feature corresponding to the interval in which the second prediction value is located; The first interval feature, the second interval feature, and the label of the sample corresponding to each sample constitute a transformed sample. Data, and use the transformed sample data to train the model. The trained model is used to predict online Prediction value of the prediction models and the offline models are fused to give a final prediction value. A computer device, comprising: a processor; a memory for storing processor-executable instructions; a processor configured to: obtain a business data generated by a target user within a first time period, and determine based on the business data Input features and input to the online prediction model, and output the first prediction value; obtain a second prediction value corresponding to the target user obtained by using an offline prediction model, wherein the input features of the offline prediction model are based on the target Determined by the user's business characteristics generated in the second time period; Obtaining the results of binning the first prediction value of the online prediction model and the second prediction value of the offline prediction model, and determining the first prediction value. The first interval at the second interval and the second interval at which the second predicted value is located; according to the first interval and the second interval, using a model trained in advance to compare the first predicted value and the first predicted value The two prediction values are fused to obtain a final fusion prediction value, and the fusion prediction value is used to Determining a label of the target user.