WO2019114423A1

WO2019114423A1 - Method and apparatus for merging model prediction values, and device

Info

Publication number: WO2019114423A1
Application number: PCT/CN2018/111824
Authority: WO
Inventors: 方文静; 周俊
Original assignee: 阿里巴巴集团控股有限公司
Priority date: 2017-12-15
Filing date: 2018-10-25
Publication date: 2019-06-20
Also published as: CN108052979A; TW201928709A; TWI718422B

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 device for fusing model prediction values

Technical field

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

Background technique

Machine learning algorithms are a class of algorithms that can automatically analyze and obtain rules from data and use rules to predict unknown data. They are widely used in many fields.

In practical applications, including online prediction models and offline prediction models, offline prediction models are usually implemented with timing tasks. The advantage is that they can incorporate higher-dimensional features and use more complex algorithms to achieve more accurate predictions. The effect; however, due to the large number of features and the complexity of the algorithm, the prediction process is usually time consuming. Compared with offline prediction models, online prediction models can use more dimensional features and simpler algorithms to achieve more efficient predictions. The disadvantage is that the features are not rich enough and the accuracy is not high. It can be seen that the online prediction model and the offline prediction model each have their own advantages, and how to properly integrate the two is an urgent problem to be solved in the industry.

Summary of the invention

For the above technical problem, the embodiment of the present specification provides a method, device and device for fusing a model prediction value, and the technical solution is as follows:

In one aspect, a method for fusing a model predictor includes:

Dividing the predicted value of the online prediction model and the predicted value of the offline prediction model according to a set binning method according to a given number of samples, wherein each of the plurality of samples includes: a first predicted value a second predicted value and a label of the sample, the first predicted value being predicted by an online prediction model, and the second predicted value being predicted by an offline prediction model;

And converting, according to the result of the binning, the first predicted value of each sample into a first interval feature corresponding to the interval in which the first predicted value is located, and converting the second predicted value of each sample into the second predicted value a second interval feature corresponding to the interval;

The transformed first sample feature, the second interval feature, and the sample tag of each sample constitute transformed sample data, and the transformed sample data is used to train the model, and the trained completed model is used for online The predicted value of the prediction model is combined with the predicted value of the offline prediction model to obtain a final predicted value.

In one aspect, a method for fusing a model predictor includes:

Obtaining the service data generated by the target user in the first time period, determining the input feature according to the service data, inputting the online prediction model, and outputting the first predicted value;

Obtaining, by using an offline prediction model, a second predicted value corresponding to the target user, where an input feature of the offline prediction model is determined according to a service feature generated by the target user in a second time period;

Obtaining a result of binning the first predicted value of the online prediction model and the second predicted value of the offline prediction model, respectively determining a first interval in which the first predicted value is located and a second predicted value Second interval

And merging the first predicted value and the second predicted value by using a pre-trained model according to the first interval and the second interval to obtain a final fused predicted value, where the fused predicted value is used To determine the label of the target user.

In one aspect, an apparatus for fusing a model predictor includes:

a binning unit, based on a given number of samples, binning the predicted value of the online predictive model and the predicted value of the offline predictive model according to a set binning method, wherein each of the plurality of samples comprises: a first predicted value, a second predicted value, and a label of the sample, the first predicted value being predicted by an online prediction model, and the second predicted value being predicted by an offline prediction model;

The feature conversion unit converts the first predicted value of each sample into a first interval feature corresponding to the interval in which the first predicted value is located according to the result of the binning, and converts the second predicted value of each sample into the first predicted value a second interval feature corresponding to the interval in which the predicted value is located;

a training unit that constructs the transformed sample data by using the first interval feature, the second interval feature, and the sample tag corresponding to each sample, and training the model by using the transformed sample data, and the trained model is used The predicted value of the online prediction model and the predicted value of the offline prediction model are combined to obtain a final predicted value.

In one aspect, an apparatus for fusing a model predictor includes:

The online score prediction unit acquires service data generated by the target user in a first time period before the trigger time, determines an input feature according to the service data, inputs the online input prediction model, and outputs a first predicted value, the online prediction model a label used to predict the user;

An offline score obtaining unit obtains a second predicted value corresponding to the target user obtained by using an offline prediction model, wherein an input feature of the offline prediction model is generated according to the target user in a past second time period Determined by a service feature, the offline prediction model is used to predict a user's tag;

The interval determining unit determines, according to a result of binning the predicted value of the online prediction model and the predicted value of the offline prediction model in advance, respectively determining the first interval and the second predicted value at which the first predicted value is located Second interval;

a score fusion unit that fuses 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 value is used to determine the label of the target user.

In one aspect, a computer apparatus is provided, comprising:

processor;

a memory for storing processor executable instructions;

The processor is configured to:

In one aspect, a computer apparatus is provided, comprising:

processor;

a memory for storing processor executable instructions;

The processor is configured to:

The effects of the technical solutions provided by the embodiments of the present specification include:

The machine-learned model is used to fuse the predicted value of the line prediction model with the predicted value of the offline prediction model, and finally the score obtained by the fusion is used to predict the user's label, thereby improving the user's The accuracy of the predictions of the tags also meets the requirements of the business for low latency.

The above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the embodiments.

Moreover, any of the embodiments of the present specification does not need to achieve all of the above effects.

DRAWINGS

In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings to be used in the embodiments or the description of the prior art will be briefly described below. Obviously, the drawings in the following description are only It is a few embodiments described in the embodiments of the present specification, and other drawings can be obtained from those skilled in the art based on these drawings.

1 is a schematic flow chart of a method for fusing a model prediction value according to an embodiment of the present specification;

2 is a process for determining a fusion weight provided by an embodiment of the present specification;

3 is a schematic structural diagram of a device (weight training phase) for integrating model prediction values according to an embodiment of the present disclosure;

4 is a schematic structural diagram of a device (a score fusion phase) for fusing a model prediction value according to an embodiment of the present specification;

Figure 5 is a block diagram showing the structure of an apparatus for configuring an apparatus of an embodiment of the present specification.

Detailed ways

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 accompanying drawings in the embodiments of the present specification. The examples are only a part of the embodiments of the specification, and not all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this specification should fall within the scope of protection.

Referring to FIG. 1 , in an embodiment of the present specification, a method for fusing a model prediction value is used to fuse a score obtained by an online prediction model with a score obtained by an offline prediction model. The method can include the following steps 101-104, wherein:

Step 101: Acquire service data generated by the target user in the first time period, determine an input feature according to the service data, input the online prediction model, and output a first predicted value.

Step 102: Acquire a second prediction value corresponding to the target user obtained by using an offline prediction model, where an input feature of the offline prediction model is determined according to a service feature generated by the target user in a second time period. of.

Herein, the online prediction model and the offline prediction model are models constructed by using a machine learning algorithm to predict a user's tags. The user tags that the two models need to predict may be related to specific services. For example, for a network payment service, the user tags required for prediction can be classified into: “high-risk users”, “medium risk users”, “ Low-risk users", and so on. For an information recommendation service, the user tags required for prediction can be classified into: "sports class", "education class", "financial class", and the like. Both the online prediction model and the offline prediction model are trained by using a certain number of training samples, and each of the training samples may include: a sample generated by the sample user in participating in a specific service (such as a network payment service). Or multiple behavioral data, as well as the label that the sample user is identified. The same batch of samples may be used to train the online prediction model and the offline prediction model, or two different samples may be used to train the online prediction model and the offline prediction model, which are not limited herein.

In the embodiment of the present specification, the offline prediction model may be implemented by a timing task, such as: performing offline score prediction every day at a specified time or a specified time period, the prediction process may be for a full amount of users; and online prediction The model can be triggered by the operation of a specific user. For example, the behavior of a user clicking a web page can trigger a score calculation process of the online prediction model.

Because the offline prediction model is generally more advanced than the online prediction model, the time span of the feature data can be longer, and more complex algorithms can be used. As shown in FIG. 1 , in a specific example, on the T day, the offline prediction model can obtain the service data (feature A) generated by each user in the process of participating in a specific service on the T-1 day, according to the obtained service. The 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 of each user (ie, the second predicted values in the text) are obtained and written into the database X. For the online prediction model, the online feature data (feature B) of the user can be continuously collected and written into the database Y, wherein the online feature data can be a quasi-real-time service generated by the user in the process of participating in a specific service. The data, for example, the triggering time of the online prediction is t1, and the online feature data may be the business data generated during the period from t0 to t1 (eg, 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 of which is to read from the database X the second predicted value corresponding to the target user obtained by the offline prediction model calculation; The second is to read the online feature data of the target user from the database Y to perform the score prediction process of the next online prediction model.

At this point, for any target user, a predictive score can be obtained through the online predictive model, and a predicted score can be obtained through the offline predictive model.

Step 103: Determine, according to a result of binning the predicted value of the online prediction model and the predicted value of the offline prediction model, respectively, the first interval in which the first predicted value is located and the second predicted value Second interval.

Step 104: merging the first predicted value and the second predicted value by using a pre-trained model according to the first interval and the second interval, to obtain a final fusion predicted value, where The fusion prediction value is used to determine the 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 second weight corresponding to the second interval, based on a predetermined weight corresponding to each interval obtained by the binning. The parameters to be trained of the model include weights corresponding to the intervals obtained by the binning.

Step 1042: Determine, by using the first weight and the second weight, a fusion prediction value, where the fusion prediction value is used to determine a label of the target user.

Since the above steps 103 to 104 need to be implemented based on the binning result and the weight corresponding to each segment obtained by the binning, before the steps 103 to 104 are described 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: Bind the predicted value of the online prediction model and the predicted value of the offline prediction model according to a set binning method according to a given number of samples, wherein each sample of the plurality of samples includes: a predicted value, a second predicted value, and a label of the sample, the first predicted value being predicted by an online prediction model, and the second predicted value being predicted by an offline prediction model.

The sample mentioned in the step 201 may be the same as the sample used to train the above-mentioned offline prediction model and/or the online prediction model. Of course, it may be a different sample, which is not limited thereto.

In an 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 achieved after binning. The benefit of the entropy-based binning method is the ability to show better discrimination in high score areas. Of course, the setting binning method may also be a Gini-based binning method, an equal-frequency binning method, or the like.

Step 202: Convert, according to the result of the binning, the first predicted value of each sample into a first interval feature corresponding to the interval in which the first predicted value is located, and convert the second predicted value of each sample into the second predicted value. The second interval feature corresponding to the interval in which the predicted value is located.

In an example, if the first predicted value and the second predicted value are both between 0 and 1, after the predicted value of the online predictive model is binned, the obtained split points include: 0, 0.1, 0.13. , 0.15, 0.2, 0.3, 0.5, 1; after binning the predicted values of the offline prediction model, the obtained segmentation points include: 0, 0.03, 0.05, 0.08, 0.09, 0.11, 0.13, 1; The output values of the online prediction model and the offline prediction model are respectively obtained in 7 intervals after binning.

In an embodiment, the one-hot rule may be employed to implement the feature conversion of step 202. Suppose a sample has a first predicted value of 0.17 and a second predicted value of 0.12. Since 0.17 is in the 4th interval (0.15, 0.2) and 0.12 is in the 6th interval (0.11, 0.13), one-hot is used. The rule may convert the first predicted value: 0.17 into the first interval feature: on-bin-0001000 ("on-bin" is the identifier of the online prediction model), and convert the second predicted value: 0.12 into the second interval feature: off -bin-0000010 ("off-bin" is the identifier 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 one pair of other samples.

Step 203: constituting the transformed sample data by using the first interval feature, the second interval feature, and the sample tag corresponding to each sample, and training the model by using the transformed sample data, and the trained model is used. The predicted value of the online prediction model and the predicted value of the offline prediction model are combined to obtain a final predicted value.

The converted sample data may include other data in addition to the first interval feature, the second interval feature, and the 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 users"};

After the feature is transformed, the new sample data obtained is, for example:

{0001000,0000010, "medium risk users"}

The model to be trained in this paper may be a linear model or a nonlinear model. In an embodiment adopting a linear model, the parameters to be trained of the model may include weights corresponding to the intervals obtained by binning, and the weights may be used. The predicted values of the line prediction model and the predicted values of the offline prediction model are combined to obtain a final predicted value. The model to be trained may be a Logistic Regression (LR) model, in which each interval obtained by binning may be assigned a weight, and the weight is trained as a parameter of the LR model, and finally the weight values can be solved. . The above weights can be a score for the corresponding interval, which is not only between different model features (online and offline models), but also a global importance trade-off and learning between the score segments.

Following the example mentioned above, you can finally get the following weights:

The weight of the interval (0, 0.1) is on-bin-1=1.054,

......

The weight of the interval (0.5, 1) is on-bin-7=4.439;

The weight of the interval (0, 0.03) off-bin-1=0.604,

......

The weight of the interval (0.13, 1) off-bin-7 = 3.237.

Next, the above steps 103 to 104 will be described in conjunction with the above specific examples. Suppose that for a certain target user, the first prediction value obtained by the online prediction model is 0.66, and the second prediction value obtained by the offline prediction model is 0.25, then in combination with the above example, first in step 103, the first prediction is determined. The first interval in which the value 0.4 is located is: (0.5, 1), and the second interval in which the second predicted value is 0.25 is: (0.13, 1). Then, in step 1041, based on the predetermined weight corresponding to each interval obtained by the binning, the first weight corresponding to the first interval: (0.5, 1) can be obtained: 4.439, and the second interval. The second weight corresponding to :(0.13,1) is: 3.237.

Finally, in step 1042, a final fusion prediction value may be determined according to the first weight and the second weight, and in an optional embodiment, the first weight and the second weight may be summed. The summation result is taken as the fusion prediction value, that is, the fusion prediction value = 4.439 + 3.237 = 7.766. Of course, the specific way of integration is not limited to summation, such as: averaging. Finally, it is possible to decide how to apply the fusion prediction value according to the specific business.

The weights obtained by machine learning are used to fuse the predicted value of the line prediction model with the predicted value of the offline prediction model, and finally the score obtained by the fusion is used to predict the user's label, thereby improving the user's The accuracy of the predictions of the tags also meets the requirements of the business for low latency. In addition, the entropy-based binning and logistic regression models are used to effectively integrate online model scores and offline model scores, so that the comparability between online offline scores is adaptively adjusted in the machine learning process.

Corresponding to the above method embodiment, the embodiment of the present specification further provides an apparatus for fusing a model prediction value.

Referring to FIG. 3, in an embodiment, in the training phase of the fusion weight, a device 300 for determining the fusion weight may include:

The binning unit 301 is configured to bin the predicted value of the online predictive model and the predicted value of the offline predictive model according to a set binning method, respectively, according to a given number of samples, wherein the plurality of samples Each sample includes: a first predicted value, a second predicted value, and a label of the sample, the first predicted value being predicted by an online prediction model, and the second predicted value being predicted by an offline prediction model;

The feature conversion unit 302 is configured to: convert the first predicted value of each sample into a first interval feature corresponding to the interval in which the first predicted value is located according to the result of the binning, and set a second predicted value of each sample Converting into a 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 corresponding to each sample, the second interval feature, and the label of the sample constitute the converted sample data, and use the converted sample data to train the model, The trained model is used to fuse the predicted value of the online predictive model with the predicted value of the offline predictive model to obtain the final predicted value.

Referring to FIG. 4, in an embodiment, in the score fusion phase, a device 400 for fusing a model prediction value may include:

The online score prediction unit 401 is configured to: acquire service data generated by the target user in a first time period before the trigger time, determine an input feature according to the service data, input the online prediction model, and output a first predicted value, The online prediction model is used to predict a user's tag;

The offline score obtaining unit 402 is configured to: acquire a second predicted value corresponding to the target user obtained by using an offline prediction model, wherein an input feature of the offline prediction model is according to the target user in the past Determined by a service feature generated within two time periods, the offline prediction model is used to predict a user's tag;

The section determining unit 403 is configured to determine, according to a result of binning the predicted value of the online prediction model and the predicted value of the offline prediction model in advance, respectively determining the first interval and the second portion where the first predicted value is located The second interval in which the predicted value is located;

The weight determining unit 404 is configured to: fuse the first predicted value and the second predicted value by using a pre-trained model according to the first interval and the second interval to obtain a final fusion a predicted value, the blended predicted value used to determine a label of the target user.

In an optional embodiment, the score fusion unit 404 can include:

The weight determining subunit obtains a first weight corresponding to the first interval and a second weight corresponding to the second interval, based on a predetermined weight corresponding to each interval obtained by the binning;

a fusion subunit, using the first weight and the second weight to determine a fusion prediction value, the fusion prediction value used to determine a label of the target user.

In an embodiment, the fusion subunit may be configured to:

The first weight and the second weight are summed, and the summation result is used as a fusion prediction value.

For details of the implementation process of the functions and functions of the modules in the foregoing devices, refer to the implementation process of the corresponding steps in the foregoing methods, and details are not described herein again.

The embodiment of the present specification further provides a computer device (such as a server), comprising at least a memory, a processor, and a computer program stored on the memory and operable on the processor, wherein the processor implements the foregoing method when the program is executed .

FIG. 5 is a schematic diagram showing a hardware structure of a more specific computing device provided by an embodiment of the present 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 communication connections within the device with each other through the bus 1050.

The processor 1010 can be implemented by using a general-purpose CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits for performing correlation. The program is implemented to implement the technical solutions provided by the embodiments of the present specification.

The memory 1020 can be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, or the like. The memory 1020 can 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 related program code is saved in the memory 1020 and is called and executed by the processor 1010.

The input/output interface 1030 is used to connect an input/output module to implement information input and output. The input/output/module can be configured as a component in the device (not shown) or 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 types of sensors, and the like, and the output device may include a display, a speaker, a vibrator, an indicator light, and the like.

The communication interface 1040 is for connecting a communication module (not shown) to implement communication interaction between the device and other devices. The communication module can communicate by wired means (such as USB, network cable, etc.), or can communicate by wireless means (such as mobile network, WIFI, Bluetooth, etc.).

Bus 1050 includes a path for communicating information between various components of the device, such as processor 1010, memory 1020, input/output interface 1030, and 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 a specific implementation, the device may also include necessary for normal operation. Other components. In addition, it will be understood by those skilled in the art that the above-mentioned devices may also include only the components necessary for implementing the embodiments of the present specification, and do not necessarily include all the components shown in the drawings.

It can be clearly understood by those skilled in the art that the embodiments of the present specification can be implemented by means of software plus a necessary general hardware platform. Based on such understanding, the technical solution of the embodiments of the present specification may be embodied in the form of a software product in essence or in the form of a software product, which may be stored in a storage medium such as a ROM/RAM. Disks, optical disks, and the like, including instructions for causing a computer device (which may be a personal computer, server, or network device, etc.) to perform the methods described in various embodiments of the embodiments of the present specification or embodiments.

The system, device, module or unit illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product having a certain function. A typical implementation device is a computer, and the specific form of the computer may be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email transceiver, and a game control. A combination of a tablet, a tablet, a wearable device, or any of these devices.

The various embodiments in the specification are described in a progressive manner, and the same or similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment. The device embodiments described above are merely illustrative, and the modules described as separate components may or may not be physically separated, and the functions of the modules may be the same in the implementation of the embodiments of the present specification. Or implemented in multiple 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 the embodiment. Those of ordinary skill in the art can understand and implement without any creative effort.

The above is only a specific embodiment of the embodiments of the present specification, and it should be noted that those skilled in the art can make some improvements and refinements without departing from the principles of the embodiments of the present specification. Improvements and retouching should also be considered as protection of embodiments of the present specification.

Claims

A method of fusing model predictions, including:

Dividing the predicted value of the online prediction model and the predicted value of the offline prediction model according to a set binning method according to a given number of samples, wherein each of the plurality of samples includes: a first predicted value a second predicted value and a label of the sample, the first predicted value being predicted by an online prediction model, and the second predicted value being predicted by an offline prediction model;

And converting, according to the result of the binning, the first predicted value of each sample into a first interval feature corresponding to the interval in which the first predicted value is located, and converting the second predicted value of each sample into the second predicted value a second interval feature corresponding to the interval;

The transformed first sample feature, the second interval feature, and the sample tag of each sample constitute transformed sample data, and the transformed sample data is used to train the model, and the trained completed model is used for online The predicted value of the prediction model is combined with the predicted value of the offline prediction model to obtain a final predicted value.
The method according to claim 1, wherein the setting binning method comprises an entropy-based binning method, or a Gini-based binning method, or an equal-frequency binning method.
The method according to claim 1, wherein the parameter to be trained of the model includes a weight corresponding to each interval obtained by binning, and the weight is used to fuse the predicted value of the line prediction model with the predicted value of the offline prediction model. The final predicted value.
A method of fusing model predictions, including:

Obtaining the service data generated by the target user in the first time period, determining the input feature according to the service data, inputting the online prediction model, and outputting the first predicted value;

Obtaining, by using an offline prediction model, a second predicted value corresponding to the target user, where an input feature of the offline prediction model is determined according to a service feature generated by the target user in a second time period;

Obtaining a result of binning the first predicted value of the online prediction model and the second predicted value of the offline prediction model, respectively determining a first interval in which the first predicted value is located and a second predicted value Second interval

And merging the first predicted value and the second predicted value by using a pre-trained model according to the first interval and the second interval to obtain a final fused predicted value, where the fused predicted value is used To determine the label of the target user.
The method according to claim 3, wherein the merging the first predicted value and the second predicted value by using a pre-trained model to obtain a final fused predicted value comprises:

Obtaining, according to a predetermined weight corresponding to each interval obtained by the binning, a first weight corresponding to the first interval and a second weight corresponding to the second interval, where the parameters to be trained of the model include and The weight corresponding to each interval obtained by the box;

The fusion prediction value is determined using the first weight and the second weight.
The method according to claim 5, wherein the determining the fusion prediction value by using the first weight and the second weight comprises:

The first weight and the second weight are summed, and the summation result is used as a fusion prediction value.
A device for fusing model prediction values, comprising:

a binning unit, based on a given number of samples, binning the predicted value of the online predictive model and the predicted value of the offline predictive model according to a set binning method, wherein each of the plurality of samples comprises: a first predicted value, a second predicted value, and a label of the sample, the first predicted value being predicted by an online prediction model, and the second predicted value being predicted by an offline prediction model;

The feature conversion unit converts the first predicted value of each sample into a first interval feature corresponding to the interval in which the first predicted value is located according to the result of the binning, and converts the second predicted value of each sample into the first predicted value a second interval feature corresponding to the interval in which the predicted value is located;

a training unit that constructs the transformed sample data by using the first interval feature, the second interval feature, and the sample tag corresponding to each sample, and training the model by using the transformed sample data, and the trained model is used The predicted value of the online prediction model and the predicted value of the offline prediction model are combined to obtain a final predicted value.
The apparatus according to claim 7, wherein the setting binning method comprises an entropy-based binning method, or a Gini-based binning method, or an equal-frequency binning method.
The apparatus according to claim 7, wherein the parameter to be trained of the model includes a weight corresponding to each section obtained by binning, and the weight is used to fuse the predicted value of the line prediction model with the predicted value of the offline prediction model. The final predicted value.
A device for fusing model prediction values, comprising:

The online score prediction unit acquires service data generated by the target user in a first time period before the trigger time, determines an input feature according to the service data, inputs the online input prediction model, and outputs a first predicted value, the online prediction model a label used to predict the user;

An offline score obtaining unit obtains a second predicted value corresponding to the target user obtained by using an offline prediction model, wherein an input feature of the offline prediction model is generated according to the target user in a past second time period Determined by a service feature, the offline prediction model is used to predict a user's tag;

The interval determining unit determines, according to a result of binning the predicted value of the online prediction model and the predicted value of the offline prediction model in advance, respectively determining the first interval and the second predicted value at which the first predicted value is located Second interval;

a score fusion unit that fuses 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 value is used to determine the label of the target user.
The apparatus according to claim 10, wherein the score fusion unit comprises:

The weight determining subunit obtains a first weight corresponding to the first interval and a second weight corresponding to the second interval, based on a predetermined weight corresponding to each interval obtained by the binning;

a fusion subunit, using the first weight and the second weight to determine a fusion prediction value, the fusion prediction value used to determine a label of the target user.
The apparatus of claim 11 wherein said fusion subunit is configured to:

The first weight and the second weight are summed, and the summation result is taken as a fusion prediction value.
A computer device comprising:

processor;

a memory for storing processor executable instructions;

The processor is configured to:

Dividing the predicted value of the online prediction model and the predicted value of the offline prediction model according to a set binning method according to a given number of samples, wherein each of the plurality of samples includes: a first predicted value a second predicted value and a label of the sample, the first predicted value being predicted by an online prediction model, and the second predicted value being predicted by an offline prediction model;

And converting, according to the result of the binning, the first predicted value of each sample into a first interval feature corresponding to the interval in which the first predicted value is located, and converting the second predicted value of each sample into the second predicted value a second interval feature corresponding to the interval;

The transformed first sample feature, the second interval feature, and the sample tag of each sample constitute transformed sample data, and the transformed sample data is used to train the model, and the trained completed model is used for online The predicted value of the prediction model is combined with the predicted value of the offline prediction model to obtain a final predicted value.
A computer device comprising:

processor;

a memory for storing processor executable instructions;

The processor is configured to:

Obtaining the service data generated by the target user in the first time period, determining the input feature according to the service data, inputting the online prediction model, and outputting the first predicted value;

Obtaining, by using an offline prediction model, a second predicted value corresponding to the target user, where an input feature of the offline prediction model is determined according to a service feature generated by the target user in a second time period;

Obtaining a result of binning the first predicted value of the online prediction model and the second predicted value of the offline prediction model, respectively determining a first interval in which the first predicted value is located and a second predicted value Second interval

And merging the first predicted value and the second predicted value by using a pre-trained model according to the first interval and the second interval to obtain a final fused predicted value, where the fused predicted value is used To determine the label of the target user.