Method and apparatus for training a prediction label click model, predicting clicked labels

Abstract

The application provides a method and device for training a prediction label click model and predicting a clicked label. The method comprises determining a plurality of time differences between each of a plurality of buried point labels and a corresponding next clicked buried point label of each of the plurality of buried point labels; performing quantitative splitting processing on each of the plurality of time differences to obtain a plurality of time sequences corresponding to each of the plurality of time differences; combining the plurality of time sequences corresponding to each of the plurality of time differences, the plurality of buried point labels, and a plurality of next clicked buried point labels corresponding to the plurality of buried point labels into training samples; and training a basic prediction model by using the training samples to obtain the prediction label click model. The method can accurately predict the next buried point label.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of predicting a clicked label, and in particular, to a method and device for training a prediction label clicking model and predicting a clicked label. BACKGROUND

[0002] At present, in order to let customers know about relevant products, most users are through the serialization of data of a clicked label to predict a next clicked label, and then recommend the next clicked label to the customers.

[0003] The above method has great limitations. After clicking a label, the data is analyzed. Due to different standards for analyzing the data of each relevant label, the accuracy of predicting the next label is also low.

[0004] Therefore, how to accurately predict the next label is a technical problem to be solved. SUMMARY

[0005] The embodiments of the present application provide a method for training a prediction label clicking model and predicting a clicked label. The technical scheme of the embodiments of the present application can achieve the effect of accurately predicting the next label.

[0006] In a first aspect, the embodiments of the present application provide a method for training a prediction label clicking model, comprising: determining a plurality of time differences between each label and a corresponding next clicked label in a plurality of labels; quantitatively splitting each time difference in the plurality of time differences to obtain a plurality of time sequences corresponding to each time difference; combining the plurality of time sequences corresponding to each time difference, the plurality of labels and the plurality of next clicked labels corresponding to the plurality of labels into a training sample; training a basic prediction model through the training sample to obtain a prediction label clicking model.

[0007] In the above embodiments, the present application quantitatively splits the time difference between two labels in the training data, and uses the obtained plurality of time sequences and the labels clicked before and after as feature samples to train a prediction label clicking model. The model can accurately predict the next label to be clicked through the plurality of time sequence features corresponding to the time difference between different labels. The method can achieve the effect of accurately predicting the next label.

[0008] In some embodiments, the plurality of labels and the plurality of next clicked labels are obtained by collecting and accumulating historical label data in a target field.

[0009] In the above embodiment, the historical data can be collected and accumulated multiple times to obtain sufficient historical data for training the model in real time, so that the final predicted label click model can accurately predict the next buried point label.

[0010] In some embodiments, each of the plurality of time differences is quantitatively split to obtain a plurality of time sequences corresponding to each of the plurality of time differences, including:

[0011] Each of the plurality of time differences is quantitatively split to obtain a plurality of processed time differences;

[0012] The plurality of processed time differences are digitized to obtain a plurality of time sequences corresponding to each of the plurality of time differences.

[0013] In the above embodiment, the time difference after quantitative splitting is digitized to sequence, so that the model can accurately predict the next buried point label according to the characteristics of the time sequence.

[0014] In some embodiments, each of the plurality of time differences is quantitatively split to obtain a plurality of time sequences corresponding to each of the plurality of time differences, including:

[0015] Each of the plurality of time differences is quantitatively split to obtain a plurality of time sequences corresponding to each of the plurality of time differences;

[0016] Each of the plurality of time sequences is quantitatively split again to obtain a plurality of time sequences corresponding to each of the plurality of time differences.

[0017] In the above embodiment, the time difference is quantitatively split into smaller time sequences, so that the model can accurately predict the next buried point label according to detailed characteristics when extracting time characteristics.

[0018] In a second aspect, the embodiments of the present application provide a method for predicting a clicked label, including: obtaining a current buried point label; predicting the next clicked buried point label of the current buried point label by a predicted label click model to obtain a prediction result, wherein the predicted label click model is obtained by training a basic prediction model by a training sample, the training sample is obtained by quantitatively splitting a plurality of time differences between each of a plurality of buried point labels and a corresponding next clicked buried point label of each of the plurality of buried point labels, and the plurality of training samples include a plurality of time sequences corresponding to each of the plurality of time differences, the plurality of buried point labels, and a plurality of next clicked buried point labels corresponding to the plurality of buried point labels.

[0019] In the above embodiment, the time difference between two buried point labels in the training data is quantitatively split, and the obtained multi-segment time sequence and the buried point labels before and after the click are used as characteristic samples to train a prediction label click model. The model can accurately predict the next buried point label corresponding to the current buried point label through the multi-segment time sequence corresponding to the time difference between different buried points. Through the method, the next buried point label can be accurately predicted.

[0020] In some embodiments, the prediction label click model is used to predict the next clicked buried point label of the current buried point label to obtain a prediction result, including:

[0021] The prediction label click model is used to calculate a plurality of time segments of the current buried point label and each buried point label in the set of buried point labels.

[0022] The plurality of time segments are used to determine the next clicked buried point label corresponding to the current buried point label.

[0023] In the above embodiment, through smaller time segments, more detailed time characteristics can be extracted, and the next buried point label can be accurately predicted according to the time characteristics.

[0024] In a third aspect, an embodiment of the present application provides a device for training a prediction label click model, including:

[0025] A determination module is configured to determine a plurality of time differences between each buried point label in a plurality of buried point labels and the next clicked buried point label corresponding to each buried point label.

[0026] A quantitative splitting module is configured to quantitatively split each time difference in the plurality of time differences to obtain a plurality of time sequences corresponding to each time difference.

[0027] A combination module is configured to combine the plurality of time sequences corresponding to each time difference, the plurality of buried point labels, and the plurality of next clicked buried point labels corresponding to the plurality of buried point labels into training samples.

[0028] A training module is configured to train a basic prediction model through the training samples to obtain a prediction label click model.

[0029] Optionally, the plurality of buried point labels and the plurality of next clicked buried point labels are obtained through multiple collection and accumulation of historical buried point label data in a target field.

[0030] Optionally, the quantitative splitting module is specifically configured to:

[0031] Quantitatively split each time difference in the plurality of time differences to obtain a plurality of processed time differences.

[0032] The processed multiple time differences are digitized to obtain multiple time sequences corresponding to each time difference.

[0033] Optionally, the quantitative splitting module is specifically configured to:

[0034] The multiple time differences are quantitatively split to obtain multiple time segments corresponding to each time difference.

[0035] Each time segment is quantitatively split again to obtain multiple time sequences corresponding to each time difference.

[0036] In a fourth aspect, an embodiment of the present application provides a device for predicting a clicked label, comprising:

[0037] An acquisition module configured to acquire a current buried point label.

[0038] A prediction module configured to predict a next clicked buried point label of the current buried point label by using a prediction label clicking model, to obtain a prediction result, wherein the prediction label clicking model is obtained by training a basic prediction model by using training samples, the training samples are obtained by quantitatively splitting multiple time differences between each buried point label and a next clicked buried point label corresponding to each buried point label in multiple buried point labels, and the multiple training samples comprise multiple time sequences corresponding to each time difference, the multiple buried point labels, and multiple next clicked buried point labels corresponding to the multiple buried point labels.

[0039] Optionally, the prediction module is specifically configured to:

[0040] The prediction module is specifically configured to:

[0041] The prediction module is specifically configured to:

[0042] In a fifth aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores computer readable instructions, and when the computer readable instructions are executed by the processor, the steps in the method provided in the first aspect or the second aspect are executed.

[0043] In a sixth aspect, an embodiment of the present application provides a readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the steps in the method provided in the first aspect or the second aspect are executed.

[0044] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art upon examination of the following or can be learned by practice of the application. The objects and other advantages of the application can be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments of the present application. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be considered as a limitation to the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0046] Figure 1 A flow chart of a method for training a prediction label click model provided by the embodiments of the present application;

[0047] Figure 2 A schematic diagram of a method for quantitative splitting processing provided by the embodiments of the present application;

[0048] Figure 3 A flow chart of a method for predicting a clicked label provided by the embodiments of the present application;

[0049] Figure 4 A schematic block diagram of an apparatus for training a prediction label click model provided by the embodiments of the present application;

[0050] Figure 5 A schematic block diagram of an apparatus for predicting a clicked label provided by the embodiments of the present application;

[0051] Figure 6 A structural schematic block diagram of an apparatus for training a prediction label click model provided by the embodiments of the present application;

[0052] Figure 7 A structural schematic block diagram of an apparatus for predicting a clicked label provided by the embodiments of the present application. DETAILED DESCRIPTION

[0053] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0054] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0055] First, some of the terms used in the embodiments of this application will be explained to facilitate understanding by those skilled in the art.

[0056] Data tracking (or event tracking) is a term in the field of data acquisition (especially user behavior data acquisition). It refers to the related technologies and implementation processes for capturing, processing, and transmitting specific user behaviors or events. The technical essence of data tracking is to first listen for events during the operation of the software application, and then judge and capture them when the events of interest occur.

[0057] Time series (or dynamic data series) refers to a data series that arranges the values ​​of the same statistical indicator in chronological order of their occurrence. The main purpose of time series analysis is to predict the future based on existing historical data. Most economic data is presented in time series form. Depending on the observation period, the time in a time series can be a year, quarter, month, or any other time format.

[0058] RNN (Recurrent Neural Network) is a basic multilayer feedback neural network where nodes are oriented and connected in loops. Compared to feedforward neural networks, RNNs have strong internal memory, allowing them to process input sequences of arbitrary temporal order.

[0059] This application is applied to the scenario of predicting click tracking tags, specifically the scenario of predicting the next tracking tag based on the currently clicked tracking tag.

[0060] However, at present, in order to enable the customer to understand the related product, most users are to serialize the data of the click tag after clicking the different click tags, to predict the next click tag, and then recommend the next click tag to the customer. The above method has great limitations. After clicking the click tag, the data is analyzed. Due to the different standards for analyzing the data of each related click tag, the accuracy of predicting the next click tag will also be low.

[0061] Therefore, the present application determines a plurality of time differences between each click tag in a plurality of click tags and the next clicked click tag corresponding to each click tag; quantitatively splits each time difference in the plurality of time differences to obtain a plurality of time sequences corresponding to each time difference; combines the plurality of time sequences corresponding to each time difference, the plurality of click tags and the plurality of next clicked click tags corresponding to the plurality of click tags into training samples; trains a basic prediction model through the training samples to obtain a prediction tag click model. By quantitatively splitting the time difference between two click tags in the training data, the plurality of time sequences obtained and the click tags before and after are used as feature samples to train the prediction tag click model. The model can accurately predict the next click tag to be clicked through the plurality of time sequence features corresponding to the time difference between different click tags. Through the method, the next click tag can be accurately predicted.

[0062] In the embodiment of the present application, the execution subject can be a training prediction tag click model device in the training prediction tag click model system. In actual application, the training prediction tag click model device can be a terminal device and a server, etc. electronic device, which is not limited here.

[0063] The following will be described in detail Figure 1 The method for training the prediction tag click model of the embodiment of the present application is described in detail.

[0064] Please refer to Figure 1 , Figure 1 The flow chart of the method for training the prediction tag click model provided by the embodiment of the present application is shown in Figure 1 The method for training the prediction tag click model includes:

[0065] Step 110: determining a plurality of time differences between each click tag in a plurality of click tags and the next clicked click tag corresponding to each click tag.

[0066] Among them, in the historical data, the click tag can be a dialogue node, can be a product introduction, can be any event, for example, in the insurance field, each insurance type can be a click tag, and there is a time difference between the clicks of different insurance types.

[0067] In some embodiments of the present application, the plurality of buried point tags and the plurality of next clicked buried point tags are obtained by collecting and accumulating historical buried point tag data multiple times in a target field.

[0068] In the above process, the present application can obtain sufficient historical data in real time by collecting and accumulating historical buried point tag data multiple times to train the model, so that the final predicted tag click model can accurately predict the next buried point tag.

[0069] The target field can be the insurance field, the service field, the catering field, the news field, etc. The present application is not limited thereto. When the target field is the insurance field, the buried point tag can be different types of insurance, or a product or event under one type of insurance. When the target field is the service field, the buried point tag can be different service items. When the target field is the catering field, the buried point tag can be different foods. When the target field is the news field, the buried point tag can be different news events. By collecting and accumulating historical buried point tag data multiple times, data can be collected in real time, and the result of the final predicted tag click model predicting the next buried point tag can be more accurate.

[0070] Step 120: quantitatively splitting each of the plurality of time differences to obtain a plurality of time sequences corresponding to each of the plurality of time differences.

[0071] The quantitative splitting can be quantitative splitting of each time difference according to a preset rule, or can be quantitative splitting of each time difference according to a user setting.

[0072] In some embodiments of the present application, the plurality of time differences are quantitatively split to obtain a plurality of time sequences corresponding to each of the plurality of time differences, including: quantitatively splitting each of the plurality of time differences to obtain a plurality of processed time differences; and digitizing the plurality of processed time differences to obtain a plurality of time sequences corresponding to each of the plurality of time differences.

[0073] In the above process, the present application can accurately predict the next buried point tag according to the characteristics of the time sequence by digitizing the time difference after quantitative splitting.

[0074] The time sequence obtained after digitizing the processed multiple time differences can be fitted into the prediction label click model as a time feature, so that the prediction label click model can extract the time feature between two buried points according to time, and predict the next buried point label through the time feature. The quantitative splitting processing is to divide the time difference to obtain different interval marks. In addition, the click sequence of the buried point label and the features of the above-mentioned multiple time sequences can be embedded into the prediction label click model at the same time, so that the prediction label click model can predict the next buried point label according to the click sequence features and time sequence features of the buried point label.

[0075] In some embodiments of the present application, the quantitative splitting processing is performed on each of the multiple time differences to obtain a multiple time sequence corresponding to each time difference, including: performing quantitative splitting processing on each of the multiple time differences to obtain multiple time corresponding to each time difference; and performing quantitative splitting processing on each of the multiple time again to obtain a multiple time sequence corresponding to each time difference.

[0076] In the above process, by quantitatively splitting the time difference into a smaller time sequence, the model can more accurately predict the next buried point label according to detailed features when extracting time features.

[0077] Through the above two quantitative splitting processes or more quantitative splitting processes, the time difference between the click buried point labels can be split in more detail, and the obtained time features are more detailed, so that the final prediction label click model can extract more detailed features to accurately predict the next buried point label to be clicked by the user.

[0078] In addition, the specific quantitative splitting processing method can be described in detail by Figure 2 .

[0079] Please refer to Figure 2 , Figure 2 for a schematic diagram of a quantitative splitting processing method provided by an embodiment of the present application, as shown in the quantitative splitting processing method includes: Figure 2

[0080] ​There are n common buried point labels (the buried point labels are label1, label2, label3…labeln), and n is a positive integer greater than or equal to 2. There are n-1 corresponding time differences (the time differences are Time1, Time2…Time(n-1)). Each time difference is quantitatively divided. Take Time1 as an example. Time1 is quantified into smaller time periods (the time periods are Time11, Time12, Time13…Time1m), and m is a positive integer greater than or equal to 2. For example, Time1 is 30s, and Time11, Time12, Time13…Time1m can be 2s, 2s, 2s…2s. The buried point label and the quantified buried point time period combination feature are embedded into the prediction label click model. The buried point label and the quantified buried point time period combination feature can be in the form of a two-dimensional array, or in the form of a one-dimensional array obtained by inserting each other.

[0081] Step 130: combining the multiple time sequences corresponding to each time difference, the multiple buried point labels, and the multiple next clicked buried point labels corresponding to the multiple buried point labels into training samples.

[0082] The multiple time sequences can be digitized time periods, which are used to embed the prediction label click model. The multiple buried point labels are input samples. The multiple output results obtained according to the input samples can be compared with the multiple next clicked buried point labels. The difference between the multiple output results and the multiple next clicked buried point labels is determined by similarity calculation method. Then, the parameters of the model are repeatedly adjusted according to the difference, and the final accurate prediction label click model can be obtained.

[0083] Step 140: training the basic prediction model by using the training samples to obtain the prediction label click model.

[0084] The basic prediction model can be a basic neural network with an RNN structure, or an existing prediction model.

[0085] In the above Figure 1In the process shown, the application determines the time differences between each of the multiple buried point tags and the next clicked buried point tag corresponding to each of the multiple buried point tags; quantitatively splits each of the multiple time differences to obtain multiple time sequences corresponding to each of the time differences; combines the multiple time sequences corresponding to each of the time differences, the multiple buried point tags, and the multiple next clicked buried point tags corresponding to the multiple buried point tags into training samples; trains a basic prediction model using the training samples to obtain a prediction tag click model. By quantitatively splitting the time difference between two buried point tags in the training data, the multiple time sequences obtained and the buried point tags clicked before and after are used as feature samples to train the prediction tag click model. The model can accurately predict the next buried point tag to be clicked by using the multiple time sequence features corresponding to the time differences between different buried points. This method can accurately predict the next buried point tag.

[0086] The foregoing Figures 1-2 describes a method of training a prediction tag click model. The following Figure 3 describes a method of predicting a clicked tag.

[0087] The following Figure 3 describes a method of predicting a clicked tag according to an embodiment of the application.

[0088] Please refer to Figure 3 , Figure 3 A flowchart of a method of predicting a clicked tag according to an embodiment of the application is shown in Figure 3 The method of training a prediction tag click model includes:

[0089] Step 310: Obtain a current buried point tag.

[0090] Step 320: Use the prediction tag click model to predict the next clicked buried point tag of the current buried point tag to obtain a prediction result.

[0091] The prediction tag click model is obtained by training a basic prediction model using training samples. The training samples are obtained by quantitatively splitting the time differences between each of the multiple buried point tags and the next clicked buried point tag corresponding to each of the multiple buried point tags. The multiple training samples include multiple time sequences corresponding to each of the time differences, the multiple buried point tags, and the multiple next clicked buried point tags corresponding to the multiple buried point tags.

[0092] In the above embodiments, this application quantitatively splits the time difference between two tracking labels in the training data, and uses the resulting multiple time series and the tracking labels of the previous and subsequent clicks as feature samples to train a tag click prediction model. This model can accurately predict the next tracking label corresponding to the current tracking label through multiple time series features corresponding to the time difference between different tracking points. This method can achieve the effect of accurately predicting the next tracking label.

[0093] In some embodiments of this application, the prediction of the next clicked tracking tag is performed on the current tracking tag by a tag click prediction model, and the prediction result is obtained. This includes: calculating multiple time periods for the current tracking tag and each tracking tag in the tracking tag set by the tag click prediction model; and determining the next clicked tracking tag corresponding to the current tracking tag by using multiple time periods.

[0094] In the above process, this application can extract more detailed time features through a shorter time period, and accurately predict the next embedding label based on the time features.

[0095] Among them, determining the next clicked tracking tag corresponding to the current tracking tag through multiple time periods includes extracting features from multiple time periods using a tag click prediction model, and predicting the next clicked tracking tag based on the time features stored in the system. This can achieve the effect of predicting the next tracking tag in advance and providing personalized product or service recommendations to users based on the predicted tracking tags.

[0096] In addition, after determining the next clicked tracking tag corresponding to the current tracking tag, the model is further trained and debugged again by manual annotation using the current tracking tag and the next clicked tracking tag corresponding to the current tracking tag. This can optimize the predictive tag click model and obtain an optimized predictive tag click model.

[0097] The previous text passed Figures 1-3 The methods for training a label-predicting click model and predicting the clicked label are described below. Figures 4-7 Describe the apparatus for training a model to predict label clicks and the apparatus for predicting the clicked label.

[0098] Please refer to Figure 4 This is a schematic block diagram of a device 400 for training and predicting label clicks according to an embodiment of this application. The device 400 can be a module, program segment, or code on an electronic device. This device 400 is related to the above... Figure 1 The method implementation corresponds to this and can be executed. Figure 1 The various steps involved in the method embodiment, and the specific functions of the device 400, can be found in the following description. To avoid repetition, detailed descriptions are omitted here.

[0099] Optionally, the device 400 includes:

[0100] The determination module 410 is used to determine multiple time differences between each tracking tag and the next clicked tracking tag corresponding to each tracking tag;

[0101] The quantitative splitting module 420 is used to perform quantitative splitting on each time difference among multiple time differences to obtain multiple time series corresponding to each time difference;

[0102] The combination module 430 is used to combine multiple time series corresponding to each time difference, multiple tracking labels, and multiple next clicked tracking labels corresponding to the multiple tracking labels into training samples.

[0103] Training module 440 is used to train the basic prediction model using training samples to obtain the prediction label click model.

[0104] Optionally, multiple event tracking labels and multiple next clicked event tracking labels are obtained by collecting and accumulating historical event tracking label data multiple times within the target domain.

[0105] Optionally, the quantitative splitting module is specifically used for:

[0106] Each time difference among multiple time differences is quantitatively split to obtain multiple processed time differences; the processed multiple time differences are then digitized to obtain multiple time series corresponding to each time difference.

[0107] Optionally, the quantitative splitting module is specifically used for:

[0108] Each time difference among multiple time differences is quantitatively divided to obtain multiple time segments corresponding to each time difference; each time segment among multiple time segments is quantitatively divided again to obtain multiple time series corresponding to each time difference.

[0109] Please refer to Figure 5 This is a schematic block diagram of a device 500 for predicting clicked tags provided in an embodiment of this application. The device 500 can be a module, program segment, or code on an electronic device. This device 500 is related to the above... Figure 3 The method implementation corresponds to this and can be executed. Figure 3 The various steps involved in the method embodiment, and the specific functions of the device 500, can be found in the description below. To avoid repetition, detailed descriptions are appropriately omitted here.

[0110] Optionally, the device 500 includes:

[0111] Module 510 is used to obtain the current tracking tag;

[0112] The prediction module 520 is configured to perform prediction on the next clicked burying point label of the current burying point label by using a prediction label click model to obtain a prediction result, wherein the prediction label click model is obtained by training a basic prediction model by using training samples, the training samples are obtained by quantitatively splitting a plurality of time differences between each burying point label and the next clicked burying point label corresponding to the burying point label, and the plurality of training samples include a plurality of time sequences corresponding to each time difference, a plurality of burying point labels and a plurality of next clicked burying point labels corresponding to the plurality of burying point labels.

[0113] Optionally, the prediction module is specifically configured to:

[0114] calculate a plurality of time periods of the current burying point label and each burying point label in the burying point label set by using the prediction label click model; and determine the next clicked burying point label corresponding to the current burying point label by using the plurality of time periods.

[0115] Please refer to Figure 6 The device for training a prediction label click model provided in the embodiments of the present application has a structure schematic block diagram, which can include a memory 610 and a processor 620. Optionally, the device can further include a communication interface 630 and a communication bus 640. The device corresponds to the method embodiments described above, and can perform each step involved in the method embodiments. The specific functions of the device can be referred to the description below. Figure 1 The device corresponds to the method embodiments described above, and can perform each step involved in the method embodiments. The specific functions of the device can be referred to the description below. Figure 1 The device corresponds to the method embodiments described above, and can perform each step involved in the method embodiments. The specific functions of the device can be referred to the description below.

[0116] Specifically, the memory 610 is configured to store computer readable instructions.

[0117] The processor 620 is configured to process the readable instructions stored in the memory, and can perform each step in the method. Figure 1 The processor 620 is configured to process the readable instructions stored in the memory, and can perform each step in the method.

[0118] The communication interface 630 is configured to perform signaling or data communication with other node devices. For example, it is configured to perform communication with a server or a terminal, or communication with other device nodes, and the embodiments of the present application are not limited to this.

[0119] The communication bus 640 is configured to realize direct connection communication of the above-mentioned components.

[0120] In this embodiment, the communication interface 630 of the device is used for signaling or data communication with other node devices. The memory 610 can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Optionally, the memory 610 can also be at least one storage device located remotely from the aforementioned processor. The memory 610 stores computer-readable instructions, which, when executed by the processor 620, enable the electronic device to perform the aforementioned... Figure 1 The method process is shown. Processor 620 can be used on device 400 and is used to perform the functions described in this application. Exemplarily, the processor 620 described above can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components; the embodiments of this application are not limited thereto.

[0121] Please refer to Figure 7 This is a schematic block diagram of a device for predicting clicked tags provided in an embodiment of this application. The device may include a memory 710 and a processor 720. Optionally, the device may further include a communication interface 730 and a communication bus 740. This device is similar to the one described above. Figure 3 The method implementation corresponds to this and can be executed. Figure 3 The specific functions of the device involved in the method embodiments can be found in the following description.

[0122] Specifically, memory 710 is used to store computer-readable instructions.

[0123] Processor 720 is used to process readable instructions stored in memory and is capable of executing... Figure 3 Each step in the method.

[0124] The communication interface 730 is used for signaling or data communication with other node devices. For example, it is used for communication with a server or terminal, or for communication with other device nodes, but the embodiments of this application are not limited thereto.

[0125] Communication bus 740 is used to enable direct communication between the above components.

[0126] The communication interface 730 of the device in the embodiments of the present application is configured to communicate signaling or data with other node devices. The memory 710 can be a high-speed RAM memory or a non-volatile memory such as at least one disk memory. The memory 710 can alternatively be at least one storage device located away from the aforementioned processor. The memory 710 stores computer-readable instructions which, when executed by the processor 720, cause the electronic device to perform the method processes described above. Figure 3 The processor 720 can be used in the device 500 and configured to perform the functions in the embodiments of the present application. By way of example, the processor 720 described above can be a general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and the embodiments of the present application are not limited thereto.

[0127] The embodiments of the present application further provide a readable storage medium. When the computer program is executed by the processor, the method processes performed by the electronic device in the method embodiments described above are performed. Figure 1 or Figure 3 The method processes performed by the electronic device in the method embodiments described above are performed.

[0128] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the device described above can refer to the corresponding processes in the foregoing methods, which will not be described in detail herein.

[0129] In summary, the embodiments of the present application provide a method and device for training a prediction label click model and predicting a clicked label. The method includes determining a plurality of time differences between each of a plurality of buried point labels and a corresponding next clicked buried point label of each of the plurality of buried point labels; performing quantitative splitting processing on each of the plurality of time differences to obtain a plurality of time sequences corresponding to each of the plurality of time differences; combining the plurality of time sequences corresponding to each of the plurality of time differences, the plurality of buried point labels, and a plurality of next clicked buried point labels corresponding to the plurality of buried point labels into training samples; and training a basic prediction model by using the training samples to obtain the prediction label click model. The method can accurately predict the next buried point label.

[0130] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can also be implemented by other means. The apparatus embodiments described above are only illustrative, for example, the flowcharts and block diagrams in the drawings show the possible implementation architecture, function and operation of the apparatus, method and computer program product according to the embodiments of the present application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order from that shown in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0131] In addition, the functional modules in the various embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0132] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0133] The above merely provides an example of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, and thus, once an item is defined in one drawing, it need not be further defined and explained in subsequent drawings.

[0134] The above merely provides an example of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, and thus, once an item is defined in one drawing, it need not be further defined and explained in subsequent drawings.

[0135] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one from another entity or action without necessarily requiring or implying any actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

Claims

1. A method for training a predictive label click model, characterized in that, include: Determine multiple time differences between each of the multiple tracking tags and the next clicked tracking tag corresponding to each tracking tag; Each of the multiple time differences is quantitatively split to obtain multiple time series corresponding to each time difference; The multiple time series corresponding to each time difference, the multiple tracking labels, and the multiple next clicked tracking labels corresponding to the multiple tracking labels are combined into training samples; The basic prediction model is trained using the training samples to obtain a predicted label click model; The quantitative decomposition of each of the multiple time differences to obtain multiple time series corresponding to each time difference includes: Each of the multiple time differences is quantitatively divided to obtain the processed multiple time differences; The processed time differences are digitized to obtain multiple time series corresponding to each time difference; Among them, multiple time series obtained by digitizing the processed multiple time differences are used as time features and fitted into the prediction tag click model, so that the prediction tag click model can extract the time features between two tracking points based on time, and predict the tag of the next tracking point based on the time features. Quantitative splitting involves dividing the time difference to obtain different interval markers; The click sequence of the event tracking tag and the features of multiple time series are simultaneously embedded into the tag prediction model, so that the tag prediction model can predict the next event tracking tag based on the click sequence features and time series features of the event tracking tag.

2. The method according to claim 1, characterized in that, The multiple tracking tags and the multiple next clicked tracking tags are obtained by collecting and accumulating historical tracking tag data multiple times within the target domain.

3. A method for predicting clicked tags, characterized in that, The method includes: Get the current tracking tag; The prediction tag click model obtained by training the prediction tag click model based on the method of claim 1 or 2 predicts the next clicked tag for the current tracking tag, and obtains the prediction result. The prediction tag click model is obtained by training the basic prediction model with training samples. The training samples are obtained by quantitatively splitting multiple time differences between each tracking tag and the next clicked tag corresponding to each tracking tag. The multiple training samples include multiple time series corresponding to each time difference, the multiple tracking tags, and multiple next clicked tags corresponding to the multiple tracking tags.

4. The method according to claim 3, characterized in that, The predicted label click model obtained by training the predicted label click model based on the method described in claim 1 or 2 predicts the next clicked label for the current embedded label, and obtains the prediction result, including: The predicted tag click model is used to calculate multiple time periods for the current tracking tag and each tracking tag in the tracking tag set; By using the multiple time periods, the next clicked tracking tag corresponding to the current tracking tag is determined.

5. An apparatus for training a predictive label click model, implemented based on the method of claim 1, characterized in that, include: The determination module is used to determine multiple time differences between each of the multiple tracking tags and the next clicked tracking tag corresponding to each tracking tag; The quantitative splitting module is used to perform quantitative splitting processing on each of the multiple time differences to obtain multiple time series corresponding to each time difference; The combination module is used to combine multiple time series corresponding to each time difference, multiple tracking labels, and multiple next clicked tracking labels corresponding to the multiple tracking labels into training samples. The training module is used to train the basic prediction model using the training samples to obtain the prediction tag click model; The quantitative splitting module is specifically used to quantitatively split each of the multiple time differences to obtain multiple processed time differences; digitize the multiple processed time differences to obtain multiple time series corresponding to each time difference; wherein, the multiple time series obtained after digitizing the multiple processed time differences are used as time features and fitted into the prediction tag click model, so that the prediction tag click model extracts the time features between two tracking points based on time, and predicts the tag for the next tracking point based on the time features; Quantitative splitting involves dividing the time difference to obtain different interval markers; the click sequence of the embedded tag and the features of multiple time series are simultaneously embedded into the prediction tag click model, so that the prediction tag click model can predict the next embedded tag based on the click sequence features and time series features of the embedded tag.

6. An electronic device, characterized in that, include: A memory and a processor, the memory storing computer-readable instructions that, when executed by the processor, perform the steps of the method as claimed in any one of claims 1-2 or the steps of the method as claimed in any one of claims 3-4.

7. A computer-readable storage medium, characterized in that, include: A computer program that, when run on a computer, causes the computer to perform the method as described in any one of claims 1-2 or the method as described in any one of claims 3-4.

Images