WO2022218068A1 - 素材投放方法、装置、设备和介质 - Google Patents

素材投放方法、装置、设备和介质 Download PDF

Info

Publication number
WO2022218068A1
WO2022218068A1 PCT/CN2022/080125 CN2022080125W WO2022218068A1 WO 2022218068 A1 WO2022218068 A1 WO 2022218068A1 CN 2022080125 W CN2022080125 W CN 2022080125W WO 2022218068 A1 WO2022218068 A1 WO 2022218068A1
Authority
WO
WIPO (PCT)
Prior art keywords
vector
historical
target
target material
indicator data
Prior art date
Application number
PCT/CN2022/080125
Other languages
English (en)
French (fr)
Inventor
陈维识
Original Assignee
北京字节跳动网络技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京字节跳动网络技术有限公司 filed Critical 北京字节跳动网络技术有限公司
Publication of WO2022218068A1 publication Critical patent/WO2022218068A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
    • G06F16/43—Querying
    • G06F16/435—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • Embodiments of the present disclosure relate to the field of computer technologies, and in particular, to a material delivery method, apparatus, electronic device, and computer-readable medium.
  • the creative may also not perform as well as the number of servings increases.
  • the usual method is to determine whether the material is delivered again by manually analyzing the results of the last delivery of the material.
  • some embodiments of the present disclosure provide a material delivery method, the method includes: inputting a historical index data sequence related to a target material into a pre-trained time series coding network to obtain a first vector; wherein the above historical index
  • the historical index data in the data sequence includes the index data of the above-mentioned target material on the delivery channel; the above-mentioned first vector is input into the pre-trained first decoding network to obtain the prediction index data of the above-mentioned target material at the future target time point;
  • the prediction indicator data is the prediction result of the delivery effect of the target material on the delivery channel at the target time point in the future; the first vector is input into the pre-trained second decoding network to obtain at least one of the historical indicator data sequences.
  • Abnormal information of a historical indicator data according to the abnormal information and the prediction indicator data of the target material, it is determined whether the target material is released again.
  • some embodiments of the present disclosure provide a material delivery device, the device includes: a first input unit configured to input a historical index data sequence related to a target material into a pre-trained time series coding network to obtain a first input unit.
  • the historical indicator data in the above-mentioned historical indicator data sequence includes the indicator data of the above-mentioned target material on the delivery channel
  • the second input unit is configured to input the above-mentioned first vector into the pre-trained first decoding network, and obtain The prediction index data of the target material at the future target time point
  • the prediction index data is the prediction result of the delivery effect of the target material on the delivery channel at the future target time point
  • the third input unit is configured to Inputting the above-mentioned first vector into a pre-trained second decoding network to obtain abnormality information of at least one historical indicator data in the above-mentioned historical indicator data sequence; Determines whether the above target creative will serve again.
  • some embodiments of the present disclosure provide an electronic device, comprising: one or more processors; a storage device on which one or more programs are stored, when one or more programs are stored by one or more The processors execute such that the one or more processors implement a method as in any of the first aspects.
  • some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method according to any one of the first aspects.
  • an embodiment of the present disclosure provides a computer program, comprising: instructions, when executed by a processor, the instructions cause the processor to perform any one of the foregoing methods.
  • embodiments of the present disclosure provide a computer program product, comprising instructions, which when executed by a processor cause the processor to perform any one of the foregoing methods.
  • FIG. 1 is a schematic diagram of an application scenario diagram of a material delivery method according to some embodiments of the present disclosure
  • FIG. 2 is a flowchart of some embodiments of a material delivery method according to the present disclosure
  • FIG. 3 is a schematic diagram of another application scenario diagram of the material delivery method according to some embodiments of the present disclosure.
  • FIG. 4 is a flowchart of other embodiments of the material delivery method according to the present disclosure.
  • FIG. 5 is a schematic diagram of a network model structure corresponding to the material delivery method according to some embodiments of the present disclosure
  • FIG. 6 is a schematic structural diagram of some embodiments of an apparatus for generating placement information according to the present disclosure
  • FIG. 7 is a schematic structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure.
  • the result of the last delivery of the material may be abnormal, and the effect of the next delivery may be unsatisfactory. Therefore, the last delivery result of the material cannot be effectively used as a judgment basis to determine whether the material is worthy of being delivered again. In addition, it is impossible to comprehensively consider various factors that lead to the unsatisfactory effect of material delivery.
  • Some embodiments of the present disclosure propose a material delivery method, apparatus, device, and computer-readable medium to solve the technical problems mentioned in the background section above.
  • FIG. 1 is a schematic diagram of an application scenario diagram of a material delivery method according to some embodiments of the present disclosure.
  • the electronic device 101 can obtain the effect 104 of the last delivery of the target material 102 on the delivery channel of the target application 103 and the last delivery of the target material 102 on the delivery channel of the target application 103 .
  • the indicator data 105 is used to determine whether to place the target material 102 on the target application 103 again.
  • the placement effect may characterize the popularity information of the target material 102 .
  • the delivery effect of the target material 102 may be the ranking of the target material 102 in each material. The higher the ranking of the target material 102 in each material, the more the target material is favored by the user.
  • the indicator data may be data indicators of various aspects generated by the target material 102 during the delivery process.
  • the indicator data may include, but is not limited to, at least one of the following: the conversion rate during the delivery of the target material 102 , or the download amount during the delivery of the target material 102 . It should be noted that for the indicator data and delivery effect, there may be situations where the indicator data of the material is better in all aspects, but the delivery effect is not good. There are also cases where the indicator data is medium in all aspects, but the delivery effect is good. Therefore, we should comprehensively consider whether the material is worthy of being delivered next time from the aspects of delivery effect and indicator data.
  • the target material 102 may be: an article; the above-mentioned indicator data 105 may be: "conversion rate: 0.45, download amount: 875".
  • the last delivery effect 104 may be that the target material 102 ranks 21st in each material.
  • the above executive body may determine whether to deliver the material again according to whether the last ranking of the target material 102 in each material is less than 100, whether the conversion rate in the indicator data is greater than 0.3, and whether the download amount in the indicator data is greater than 500.
  • the above executive body may consider that the target material 102 can be placed in the The target application 103 is delivered on the delivery channel.
  • the current material delivery method only considers the delivery effect of the last target material 102 and the indicator data generated in the delivery process. It may be an anomaly that the delivery performance and metrics data of the last target creative are not taken into account.
  • the abnormal situation of the target material 102 can include the following two situations:
  • the delivery effect and delivery data of the target material 102 in the last delivery process were relatively poor, and the actual delivery effect and delivery data were far from the expected results. Such conditions may be exceptional.
  • the occurrence of the last abnormal situation of the target material 102 may affect the next delivery of the target material with a high probability.
  • the last delivery time of the target material was the Lantern Festival, and the delivery effect of the target material 102 for saving water may not be very good in delivery effect and indicator data.
  • This situation often occurs because people pay more attention to materials related to the Lantern Festival, resulting in poor delivery effect and index data of the target material 102 .
  • This abnormal situation often cannot be used as a reference for whether the target material 102 is to be delivered next time.
  • the delivery effect and delivery data of the target material 102 in the last delivery process are relatively good, and the actual delivery effect and delivery data far exceed the expected results. Such conditions may be exceptional.
  • the occurrence of the abnormal situation of the target material 102 in the last time may lead to an unsatisfactory delivery of the target material next time with a high probability.
  • the last delivery time of the target material was World Water Day
  • the delivery effect of the target material 102 for saving water may far exceed the expected results.
  • This situation often occurs because people paid more attention to the material related to the World Water Day at that time, resulting in the result that the delivery effect and index data of the target material 102 far exceeded expectations.
  • This abnormal situation often cannot be used as a reference for whether the target material 102 is to be delivered next time.
  • the existing material delivery methods only consider the delivery effect of the last target material 102 and the indicator data generated in the delivery process.
  • the impact of delivery effects and index data of other materials similar to the target material 102 on the target material 102 is not considered, wherein the above-mentioned materials similar to the target material 102 may be the same type of material as the target material 102 .
  • target material 102 may be an article with a prompt to conserve water.
  • a material similar to the target material 102 may be an article of water waste cause analysis.
  • encoder-decoder structure and multi-objective prediction model are two popular branches of deep learning currently gaining more and more attention.
  • an encoder-decoder combined with a multi-objective prediction model can be considered to determine whether to serve the material again.
  • MT-Learning Model For the method of using the encoder-decoder combined with the multi-target prediction model (MT-Learning Model), it is required that the above-mentioned encoder-decoder combined with the multi-target prediction model can pay more attention to the historical index data sequence related to the target material.
  • the sequence of indicator data serves as the input to the encoder in the encoder-decoder structure.
  • the above historical indicator data sequence can represent the historical delivery effect of the target material and the material related to the target material.
  • the game material of a small game has been continuously delivered on App A for more than 30 days. Therefore, when predicting whether the next game material will be delivered, it is hoped that the 30-day historical delivery effect of the above game material can be considered.
  • the above encoder-decoder combined with the multi-objective prediction model can detect abnormal information of at least one historical indicator data in the historical indicator data sequence.
  • the above-mentioned abnormal information can be used as a determining factor for determining whether the target material is delivered again.
  • the encoder-decoder combined with the multi-objective prediction model can be considered to determine whether to serve the material again.
  • the material delivery method may be performed by the electronic device 101 .
  • the above electronic device 101 may be hardware or software.
  • the electronic device When the electronic device is hardware, it can be implemented as a distributed cluster composed of multiple servers or terminal devices, or can be implemented as a single server or a single terminal device.
  • the electronic device 101 When the electronic device 101 is embodied as software, it can be implemented as a plurality of software or software modules for providing distributed services, for example, or as a single software or software module. There is no specific limitation here.
  • FIG. 1 the number of electronic devices in FIG. 1 is merely illustrative. There may be any number of electronic devices depending on implementation needs.
  • the material delivery method includes the following steps:
  • Step 201 Input the historical index data sequence related to the target material into a pre-trained time series coding network to obtain a first vector.
  • the execution body of the material delivery method may input the historical index data sequence related to the target material into the pre-trained time series coding network to obtain the first vector.
  • the historical indicator data in the above-mentioned historical indicator data sequence includes indicator data of the above-mentioned target material on the delivery channel.
  • the above target material may include, but is not limited to, at least one of the following: target video, target music, or target article.
  • the above-mentioned time series encoding network may be an encoding network that processes time series data.
  • the above-mentioned time series encoding network may include, but is not limited to, at least one of the following: Recurrent Neural Network (RNN)-autoencoder, or Long Short-Term Memory Network (LSTM, Long Short-Term Memory)-autoencoder network.
  • the above-mentioned delivery channel may be the channel information of material delivery.
  • the above-mentioned delivery channels may include, but are not limited to, at least one of the following: delivery on the mobile terminal of the target brand, delivery on the mobile terminal using the target operating system, or delivery on the target application of the mobile terminal.
  • the material can be placed on the target application of the mobile terminal of the target operating system of the target brand.
  • the above historical indicator data may be various indicator parameters in the historical delivery process of the target material.
  • the above-mentioned historical indicator data includes information on the target material's predetermined time indicator parameters on the target delivery channels in the delivery channel set, information on the above-mentioned target material's predetermined time indicator parameters on each delivery channel in the delivery channel set, and the above-mentioned target material Information about the pre-determined time indicator parameters of the relevant material on each of the above distribution channels.
  • Indicator parameters may include, but are not limited to, at least one of the following: click-through rate (CTR, Click-Through-Rate), conversion rate (CVR, Conversion Rate), rate of return (Return on Investment, ROI), cost per action (Cost Per Action, CPA), or installs.
  • CTR click-through rate
  • CVR Click-Through-Rate
  • conversion rate CVR, Conversion Rate
  • rate of return Return on Investment
  • ROI cost per action
  • CPA Cost Per Action
  • the above historical indicator data sequence may include historical indicator data of the material associated with the target material.
  • the purpose of using the historical index data of the material related to the target material as input is to consider the impact of the material of the same topic type as the target material on the delivery revenue of the target material. As a result, the network model can determine whether the information on whether to deliver the above target material again is more accurate.
  • inputting the historical index data sequence related to the target material into the pre-trained time series coding network to obtain the first vector may include the following steps.
  • each historical indicator data in the historical indicator data sequence is input into the corresponding convolutional neural network in the pre-trained convolutional neural network set to output a fourth vector to obtain a fourth vector sequence.
  • each fourth vector in the above-mentioned fourth vector sequence is input to the corresponding data flattening layer in the data flattening layer set to output a fifth vector, and a fifth vector sequence is obtained.
  • the above-mentioned data flattening layer can transform a matrix with a matrix dimension of (n, m) into a matrix of (n*m, 1), that is, flatten multi-dimensional data to one-dimensional data.
  • the vector dimension of the target matrix may be (4, 5).
  • the above target matrix is input to the data flattening layer, and the matrix of (20, 1) is obtained. As a result, the elements of the matrix input to the data flattening layer do not change, and the corresponding data dimension changes.
  • the above-mentioned fifth vector sequence is input into the pre-trained time series coding network to obtain the first vector.
  • the training of the time-series coding network may be combined with the subsequent first decoding network and the second decoding network, and the specific training steps are as follows.
  • the first step is to determine the network structure of the initial network model and initialize the network parameters of the initial network model, wherein the initial network model includes: a time sequence encoding network, a first decoding network and a second decoding network.
  • a training sample set is obtained, wherein the training sample set includes a sample set and a label information set corresponding to the above-mentioned sample set.
  • the third step is to use the sample set in the training sample set and the labeling information set as the input and expected output of the initial network model, respectively, and train the initial network model by using a deep learning method.
  • the above-mentioned initial network model obtained by training is determined as the trained network model.
  • Step 202 inputting the first vector into a pre-trained first decoding network to obtain prediction index data of the target material at a future target time point.
  • the execution body may input the first vector into a pre-trained first decoding network to obtain prediction index data of the target material at a future target time point, wherein the prediction index data is that the target material is in the above On the delivery channel, the forecast result of the above-mentioned future target time point.
  • the above-mentioned future target time point may be preset.
  • the above-mentioned first decoding network corresponds to the above-mentioned time-series encoding network.
  • the above-mentioned first decoding network may be a network for performing regression tasks.
  • the above-mentioned network for performing the regression task may include, but is not limited to, at least one of the following: multi-layer fully connected layers, linear regression, or Support Vector Machine (SVM).
  • SVM Support Vector Machine
  • the above-mentioned first vector may be firstly input into a pre-trained convolutional neural network with a predetermined number of layers to obtain an output result. Then, the above-mentioned output result is input to the pre-trained regression network, and the prediction index data of the above-mentioned target material at the future target time point is obtained.
  • the historical indicator data includes indicator data of the target material in the target delivery channel; and the first vector is input into the pre-trained first decoding network to obtain the future target time Clicking on the forecast indicator data of the above target material can include the following steps:
  • the above-mentioned first vector is input into the pre-trained first decoding network to obtain the prediction index data of the above-mentioned target material on the above-mentioned target delivery channel at a future target time point.
  • the historical indicator data may include indicator data of the target material in the target delivery channel. It can be more targeted to determine whether the material can be re-delivered in a certain delivery channel.
  • the above-mentioned target delivery channel may be an application A.
  • the above application A is the main application run by the party delivering the target material. Therefore, it is necessary to know whether the target material is delivered in application A. Therefore, it is necessary to input the above-mentioned first vector into the pre-trained first decoding network to obtain the prediction index data of the above-mentioned target material in the above-mentioned A application at a future target time point.
  • Step 203 Input the above-mentioned first vector into a pre-trained second decoding network to obtain abnormal information of at least one historical indicator data in the above-mentioned historical indicator data sequence.
  • the above-mentioned execution body may input the above-mentioned first vector into a pre-trained second decoding network to obtain abnormal information of at least one historical indicator data in the above-mentioned historical indicator data sequence.
  • the above-mentioned second decoding network may include, but is not limited to, at least one of the following: a fully connected network, or a recurrent neural network.
  • the above abnormal information may be whether there is abnormal historical indicator data in the historical indicator data sequence, and which indicator data is abnormal.
  • the above network model includes: a time sequence encoding network, a first decoding network and a second decoding network.
  • Step 204 according to the abnormal information and the prediction index data of the target material, determine whether the target material is released again.
  • the execution subject may determine whether to deliver the target material again according to the abnormality information and the prediction index data of the target material.
  • the above-mentioned execution subject may determine that the above-mentioned target material is no longer delivered.
  • the execution entity may determine whether the target material is re-delivered according to the prediction result of the target material on the delivery channel and the future target time point.
  • the material delivery methods of some embodiments of the present disclosure can more accurately and effectively determine whether the target material is to be delivered again through the generated prediction indicator data and abnormal information.
  • the last delivery result of the material may be abnormal, which may cause the next delivery effect to be unsatisfactory. Therefore, the last delivery result of the material cannot be effectively used as a judgment basis for determining whether the material is worthy of being delivered again.
  • the material delivery methods of some embodiments of the present disclosure firstly use the historical indicator data sequence related to the target material as the input of the time series coding network, which can more comprehensively consider various factors that may lead to unsatisfactory delivery of the target material. Then, the above-mentioned first vector is input to the pre-trained first decoding network, so that the prediction index data of the above-mentioned target material at the future target time point can be generated more accurately and effectively. Further, the above-mentioned first vector is input into the pre-trained second decoding network to obtain abnormal information of at least one historical indicator data in the above-mentioned historical indicator data sequence.
  • the contingency of the historical delivery result of the target material can be more effectively excluded.
  • the above abnormal information and the prediction index data of the above target material it can be more efficiently and conveniently determined whether the above target material is released again.
  • FIG. 3 is a schematic diagram of another application scenario diagram of the material delivery method according to some embodiments of the present disclosure.
  • the electronic device 301 may first input the historical index data sequence 302 related to the target material into the pre-trained time series coding network 303 to obtain the first vector.
  • the above-mentioned time series encoding network 303 may be a long short-term memory network.
  • the historical indicator data sequence 302 may include historical indicator data 3021 , historical indicator data 3022 , and historical indicator data 3023 .
  • the time corresponding to the historical indicator data 3021 in the historical indicator data sequence 302 is earlier than the time corresponding to the historical indicator data 3022 .
  • the time corresponding to the historical indicator data 3022 is earlier than the time corresponding to the historical indicator data 3023 .
  • Each historical indicator data can include 3 layers of data. Each layer of historical indicator data is an indicator of multiple dimensions.
  • the first layer of historical indicator data may include: various indicator data of the target material on the target delivery channel.
  • the second-level data of the historical indicator data may include: various indicator data of the target material in each delivery channel.
  • the third-level data of the historical indicator data may include: various indicator data of each material associated with the target material in each delivery channel.
  • the above-mentioned first vector is input to the pre-trained first decoding network 304 to obtain prediction index data 306 of the above-mentioned target material at a future target time point.
  • the above-mentioned prediction index data 306 is the prediction result of the above-mentioned target material on the above-mentioned delivery channel and the above-mentioned future target time point. Further, the above-mentioned first vector is input to the pre-trained second decoding network 305 to obtain abnormal information 307 of at least one historical indicator data in the above-mentioned historical indicator data sequence 302 .
  • the above-mentioned second decoding network may be a network composed of a set of fully connected networks and a set of deconvolutional networks.
  • the above-mentioned execution subject may determine that the above-mentioned target material is no longer delivered.
  • the abnormal information 307 is represented as normal in the historical index data sequence, it may be determined whether the target material is re-delivered according to the prediction result of the target material at the future target time point on the delivery channel.
  • the material delivery method may be performed by the electronic device 301 .
  • the above electronic device 301 may be hardware or software.
  • the electronic device When the electronic device is hardware, it can be implemented as a distributed cluster composed of multiple servers or terminal devices, or can be implemented as a single server or a single terminal device.
  • the electronic device 301 When the electronic device 301 is embodied as software, it can be implemented as a plurality of software or software modules for providing distributed services, for example, or as a single software or software module. There is no specific limitation here.
  • FIG. 3 is merely illustrative. There may be any number of electronic devices depending on implementation needs.
  • the material delivery method includes the following steps.
  • Step 401 Input the historical index data sequence related to the target material into a pre-trained time series coding network to obtain a first vector.
  • Step 402 Input the above-mentioned first vector into a pre-trained first decoding network to obtain the prediction index data of the above-mentioned target material at a future target time point.
  • Step 403 Input the above-mentioned first vector into the target fully-connected network in the above-mentioned pre-trained fully-connected network set to obtain a second vector.
  • the execution body of the material delivery method may input the above-mentioned first vector into the target fully-connected network in the above-mentioned pre-trained set of fully-connected networks, to obtain the first vector Two vectors, wherein each historical indicator data in the above historical indicator data sequence corresponds to a historical time point.
  • the number of fully-connected layers in the above-mentioned target fully-connected network may vary with historical time points. The longer the historical time point is, the more fully connected layers are included in the corresponding target fully connected network.
  • the number of fully connected layers corresponding to the first historical time point is less than the number of fully connected layers corresponding to the second historical time point.
  • the number of fully connected layers corresponding to the second historical time point is smaller than the number of fully connected layers corresponding to the third historical time point.
  • Step 404 Input the second vector into the deconvolution network corresponding to the target historical time point in the deconvolution network set to obtain a third vector.
  • the execution body may input the second vector into a deconvolution network corresponding to the target historical time point in the deconvolution network set to obtain a third vector, wherein the data of the third vector
  • the dimension is the same as the data dimension of the historical indicator data corresponding to the above target historical time point.
  • the third vector and the historical indicator data corresponding to the target historical time point may be presented in the form of a matrix.
  • the fully connected network set and the deconvolution network set can play a role in stabilizing the network structure.
  • the above-mentioned third vector is generated by assigning different weights to different indicators through the above-mentioned deconvolution network.
  • Step 405 determine the abnormal information of the historical indicator data corresponding to the target historical time point.
  • the above-mentioned execution body may determine abnormal information of the historical indicator data corresponding to the above-mentioned target historical time point according to the above-mentioned third vector.
  • the above-mentioned execution body may first determine the cosine value between the above-mentioned third vector and the historical indicator data corresponding to the above-mentioned target historical time point. Then, in response to the cosine value being smaller than the preset threshold, no abnormality occurs in the historical indicator data corresponding to the target historical time point.
  • the above-mentioned determining the abnormal information of the historical indicator data corresponding to the above-mentioned target historical time point according to the above-mentioned third vector may include the following steps.
  • the first step is to determine the difference value between the third vector and the historical indicator data corresponding to the target historical time point.
  • the execution body may first determine the difference between the values of the matrix elements at the positions corresponding to the third vector and the historical indicator data corresponding to the target historical time point, and obtain a difference set as the difference value.
  • the second step in response to determining that the difference value is greater than or equal to a preset threshold, determine that the historical indicator data corresponding to the target historical time point is abnormal.
  • Step 406 according to the abnormality information and the prediction index data of the target material, determine whether the target material is released again.
  • the above-mentioned determining whether the above-mentioned target material is re-launched according to the above-mentioned abnormal information and the above-mentioned prediction index data of the above-mentioned target material may include the following steps.
  • the above-mentioned execution body may, according to the above-mentioned abnormal information, determine by means of a query whether there is a time point in which the historical indicator data is abnormal in at least one of the above-mentioned historical time points.
  • the second step in response to a time point when the historical indicator data is abnormal, receive the instruction sent by the terminal whether to execute the instruction to determine whether the above-mentioned target material is released again.
  • Step 3 In response to executing the above instruction, determine whether the prediction index data of the target material satisfies a preset condition, wherein the prediction index data of the target material may include a plurality of index data predicted by the network model. Each indicator data corresponds to its own normal data value range. Therefore, the preset condition may be that the predicted value corresponding to each index data is within the normal data value range.
  • the fourth step in response to the prediction index data of the target material satisfying a preset condition, determine that the target material is to be released again.
  • steps 401-402 and 406 for the specific implementation of steps 401-402 and 406 and the technical effects brought about by them, reference may be made to steps 201-202 and 204 in those embodiments corresponding to FIG. 2 , which will not be repeated here.
  • the process 400 of the material delivery method in some embodiments corresponding to FIG. 4 highlights the specific steps for generating abnormal information. Therefore, the solutions described in these embodiments can generate abnormal information more accurately and effectively by using the fully connected network set and the deconvolution network set. In this way, it is further determined whether the target material is to be released again according to the abnormality information and the prediction index data of the target material.
  • FIG. 5 is a schematic diagram of a network model structure corresponding to material delivery methods according to some embodiments of the present disclosure.
  • the historical indicator data sequence is obtained. Because the above historical indicator data may be various indicator data related to the target material in the historical delivery process. Regarding whether the previous material is re-delivered, the above-mentioned historical indicator data may also include information on the target material's predetermined time indicator data on the target delivery channel in the delivery channel set, and the target material's predetermined time indicator data on each delivery channel in the delivery channel set. information, and information about the predetermined time indicator data of the above-mentioned target material-related material on each of the above-mentioned delivery channels.
  • the above-mentioned indicator parameters may include, but are not limited to, at least one of the following: click-through rate, conversion rate, return rate, cost per action, or install volume.
  • the input of multi-category data enables the subsequent network model to learn more useful feature information.
  • each historical indicator data in the historical indicator data sequence is input into the corresponding convolutional neural network in the convolutional neural network set for preliminary extraction of characteristic information of the above historical indicator data.
  • the first historical indicator data 501 in the historical indicator data sequence is input to the first convolutional neural network 504 in the set of convolutional neural networks.
  • the second historical indicator data 502 in the historical indicator data sequence is input to the second convolutional neural network 505 in the set of convolutional neural networks.
  • the third historical indicator data 503 in the historical indicator data sequence is input to the third convolutional neural network 506 in the set of convolutional neural networks.
  • the output vector of the first convolutional neural network 504 is input to the first data flattening layer 507 to adjust the vector dimension of the output vector to the input vector dimension of the above-mentioned long short-term memory network 510 .
  • the output vector of the second convolutional neural network 505 is input to the second data flattening layer 508 to adjust the vector dimension of the output vector to the input vector dimension of the long short-term memory network 510 described above.
  • the output vector of the third convolutional neural network 506 is input to the third data flattening layer 509 to adjust the vector dimension of the output vector to the input vector dimension of the above-mentioned long short-term memory network 510 .
  • the output results of the first data leveling layer 507 , the second data leveling layer 508 and the third data leveling layer 509 are input to the corresponding units in the long short-term memory network 510 . That is, the output result of the above-mentioned first data leveling layer 507 is used as the input of the first unit 5101 of the long short-term memory network.
  • the output result of the above-mentioned second data leveling layer 508 is used as the input of the second unit 5102 of the long short-term memory network.
  • the output result of the above-mentioned third data flattening layer 509 is used as the input of the first unit 5013 of the long short-term memory network.
  • the above-mentioned long short-term memory network can learn the time series information between the historical indicator data in the historical indicator data sequence.
  • the output result of the long short-term memory network 510 is input to the first fully connected layer 511 to obtain the output result of the first fully connected layer 511 .
  • the output result of the first fully connected layer 511 is input to the first fully connected network 512 to obtain the output result of the first fully connected network 512 .
  • the output result of the first fully connected layer 512 is input to the third fully connected layer 513 to obtain the output result of the third fully connected layer 513 .
  • the first fully connected layer 511 , the second fully connected layer 512 and the third fully connected layer 513 can learn nonlinear feature information due to the existence of the activation function.
  • a recurrent neural network-autoencoder is often used for the previous temporal encoding network.
  • the time series coding network in the material delivery method of the present disclosure may adopt a long short-term memory network-autoencoder network.
  • Long short-term memory network-autoencoder network can effectively solve the long-term dependency problem. It avoids the information transfer problem of the long input sequence of the recurrent neural network-autoencoder.
  • regression prediction is performed on the output result of the third fully connected layer 513 to obtain the prediction index data 514 of the target material on the delivery channel at the target time point in the future.
  • the output result of the long short-term memory network 510 is input to the second fully connected layer 515 to obtain the output result of the second fully connected layer 515 .
  • the output of the second fully connected layer 515 is input to the deconvolution network 517 to obtain a third vector 519.
  • the output result of the second fully connected layer 515 is input to the second fully connected layer 516 to obtain the output result of the second fully connected layer 516 .
  • the output of the second fully connected layer 516 is input to the deconvolution network 518 to obtain a third vector 520 .
  • each third vector with the same data dimension as the historical indicator data is generated by using a deconvolution network, and at least one historical indicator in the above historical indicator data sequence can be obtained by comparing each third vector with the corresponding historical indicator data. Exception information for data.
  • the above deconvolutional network can be used as a decoding network to decode the long short-term memory network.
  • each second fully connected layer and the long short-term memory network the data form and indicator data category of the historical indicator data input by the model can be expanded.
  • the present disclosure provides some embodiments of an apparatus for generating advertisement information. These apparatus embodiments correspond to the above-mentioned method embodiments in FIG. 2 . Specifically, the apparatus may Used in various electronic devices.
  • the material delivery apparatus 600 in some embodiments includes: a first input unit 601 , a second input unit 602 , a third input unit 603 and a determination unit 604 .
  • the first input unit 601 is configured to input the historical indicator data sequence related to the target material into the pre-trained time series coding network to obtain a first vector; the historical indicator data in the historical indicator data sequence includes the target material on the delivery channel. indicator data.
  • the second input unit 602 is configured to input the first vector into a pre-trained first decoding network to obtain prediction index data of the target material at a future target time point, wherein the prediction index data is the future The prediction result of the delivery effect of the target material on the delivery channel at the target time point.
  • the third input unit 603 is configured to input the above-mentioned first vector into the pre-trained second decoding network to obtain abnormal information of at least one historical indicator data in the above-mentioned historical indicator data sequence.
  • the determining unit 604 is configured to determine whether the target material is released again according to the abnormal information and the prediction index data of the target material.
  • the historical indicator data includes indicator data of the target material in the target delivery channel.
  • the second input unit 602 may be further configured to: input the first vector into the pre-trained first decoding network to obtain the prediction index data of the target material on the target delivery channel at the future target time point.
  • the above-mentioned second decoding network includes: a set of fully connected networks and a set of deconvolutional networks.
  • the third input unit 603 may be further configured to: input the above-mentioned first vector into the target fully-connected network in the pre-trained set of fully-connected networks to obtain a second vector; wherein, the above-mentioned target fully-connected network is the same as the target historical time The network associated with the point; the above-mentioned second vector is input into the above-mentioned deconvolution network set in the deconvolution network corresponding to the target historical time point, to obtain a third vector; wherein, the data dimension of the above-mentioned third vector and the above-mentioned target The data dimensions of the historical indicator data corresponding to the historical time point are the same; according to the above-mentioned third vector, the abnormal information of the historical indicator data corresponding to the above-mentioned target historical time point is determined.
  • the third input unit 603 may be further configured to: determine a difference value between the above-mentioned third vector and the historical indicator data corresponding to the above-mentioned target historical time point; in response to determining the above-mentioned difference If the value is greater than or equal to the preset threshold, it is determined that the historical indicator data corresponding to the above target historical time point is abnormal.
  • the determining unit 604 may be further configured to: determine, according to the above-mentioned abnormal information, whether there is a time point in which the historical indicator data is abnormal in at least one of the above-mentioned historical time points; in response to the existence of the historical indicator data At the abnormal time point, whether to execute the instruction sent by the receiving terminal to determine whether the above-mentioned target material is re-launched; in response to executing the above-mentioned instruction, determine whether the prediction index data of the above-mentioned target material satisfies the preset conditions; If the forecast indicator data satisfies the pre-set conditions, it is determined that the above-mentioned target material is delivered again.
  • the above-mentioned third vector is generated by assigning different weights to different indicators through the above-mentioned deconvolution network.
  • the first input unit 601 may be further configured to: input each historical indicator data in the historical indicator data sequence to the corresponding convolutional neural network in the pre-trained convolutional neural network set The network outputs a fourth vector to obtain a fourth vector sequence; each fourth vector in the fourth vector sequence is input to the corresponding data flattening layer in the data flattening layer set to output a fifth vector to obtain a fifth vector sequence; The fifth vector sequence is input into a pre-trained time-series coding network to obtain a first vector.
  • each historical indicator data in the above historical indicator data sequence corresponds to a historical time point.
  • the units recorded in the apparatus 600 correspond to the respective steps in the method described with reference to FIG. 2 . Therefore, the operations, features and beneficial effects described above with respect to the method are also applicable to the apparatus 600 and the units included therein, and details are not described herein again.
  • FIG. 7 a schematic structural diagram of an electronic device (eg, the electronic device of FIG. 1 ) 700 suitable for implementing some embodiments of the present disclosure is shown.
  • the electronic device shown in FIG. 7 is only an example, and should not impose any limitation on the function and scope of use of the embodiments of the present disclosure.
  • an electronic device 700 may include a processing device (eg, a central processing unit, a graphics processor, etc.) 701 that may be loaded into random access according to a program stored in a read only memory (ROM) 702 or from a storage device 708 Various appropriate actions and processes are executed by the programs in the memory (RAM) 703 . In the RAM 703, various programs and data required for the operation of the electronic device 700 are also stored.
  • the processing device 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704.
  • An input/output (I/O) interface 705 is also connected to bus 704 .
  • the following devices can be connected to the I/O interface 705: input devices 706 including, for example, a touch screen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; including, for example, a liquid crystal display (LCD), speakers, vibration An output device 707 of a computer, etc.; a storage device 708 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 709. Communication means 709 may allow electronic device 700 to communicate wirelessly or by wire with other devices to exchange data.
  • FIG. 7 shows an electronic device 700 having various means, it should be understood that not all of the illustrated means are required to be implemented or provided. More or fewer devices may alternatively be implemented or provided. Each block shown in FIG. 7 can represent one device, and can also represent multiple devices as required.
  • the processes described above with reference to the flowcharts may be implemented as computer software programs.
  • some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the method illustrated in the flowchart.
  • the computer program may be downloaded and installed from a network via communication device 709, or from storage device 708, or from ROM 702.
  • the processing device 701 the above-mentioned functions defined in the methods of some embodiments of the present disclosure are performed.
  • the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the foregoing two.
  • the computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or a combination of any of the above.
  • a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
  • a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code therein.
  • Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • a computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device .
  • Program code embodied on a computer readable medium may be transmitted using any suitable medium including, but not limited to, electrical wire, optical fiber cable, RF (radio frequency), etc., or any suitable combination of the foregoing.
  • the client and server can use any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol) to communicate, and can communicate with digital data in any form or medium Communication (eg, a communication network) interconnects.
  • HTTP HyperText Transfer Protocol
  • Examples of communication networks include local area networks (“LAN”), wide area networks (“WAN”), the Internet (eg, the Internet), and peer-to-peer networks (eg, ad hoc peer-to-peer networks), as well as any currently known or future development network of.
  • the above-mentioned computer-readable medium may be included in the above-mentioned apparatus; or may exist alone without being assembled into the electronic device.
  • the computer-readable medium carries one or more programs, and when the one or more programs are executed by the electronic device, the electronic device: input the historical index data sequence related to the target material into the pre-trained time series coding network, Obtain a first vector; wherein, the historical indicator data in the historical indicator data sequence includes the indicator data of the target material on the delivery channel; input the first vector into the pre-trained first decoding network to obtain the future target time point The prediction index data of the target material; wherein, the prediction index data is the prediction result of the delivery effect of the target material on the delivery channel at the future target time point; input the above-mentioned first vector into the pre-trained No.
  • the decoding network obtains the abnormal information of at least one historical indicator data in the above historical indicator data sequence; according to the above abnormal information and the prediction indicator data of the above target material, it is determined whether the above target material is
  • Computer program code for carrying out operations of some embodiments of the present disclosure may be written in one or more programming languages, including object-oriented programming languages—such as Java, Smalltalk, C++, or a combination thereof, Also included are conventional procedural programming languages - such as the "C" language or similar programming languages.
  • the program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server.
  • the remote computer may be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (eg, using an Internet service provider to via Internet connection).
  • LAN local area network
  • WAN wide area network
  • each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more logical functions for implementing the specified functions executable instructions.
  • the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.
  • each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations can be implemented in dedicated hardware-based systems that perform the specified functions or operations , or can be implemented in a combination of dedicated hardware and computer instructions.
  • the units described in some embodiments of the present disclosure may be implemented by means of software, and may also be implemented by means of hardware.
  • the described unit may also be provided in the processor, for example, it may be described as: a processor includes a first input unit, a second input unit, a third input unit and a determination unit. Wherein, the names of these units do not constitute a limitation of the unit itself under certain circumstances.
  • the first input unit can also be described as "inputting the historical index data sequence related to the target material into the pre-trained time series coding network. , get the unit of the first vector".
  • exemplary types of hardware logic components include: Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), Systems on Chips (SOCs), Complex Programmable Logical Devices (CPLDs) and more.
  • FPGAs Field Programmable Gate Arrays
  • ASICs Application Specific Integrated Circuits
  • ASSPs Application Specific Standard Products
  • SOCs Systems on Chips
  • CPLDs Complex Programmable Logical Devices
  • a material delivery method comprising: inputting a sequence of historical indicator data related to a target material into a pre-trained time series coding network to obtain a first vector; wherein the above historical indicator data
  • the historical index data in the sequence includes the index data of the above-mentioned target material on the delivery channel; the first vector is input into the pre-trained first decoding network to obtain the prediction index data of the target material at the future target time point; wherein,
  • the prediction indicator data is the prediction result of the delivery effect of the target material on the delivery channel at the target time point in the future; the first vector is input into the pre-trained second decoding network, and the sequence of the historical indicator data is obtained.
  • Abnormal information of at least one historical index data according to the above abnormal information and the predicted index data of the above target material, determine whether the above target material is released again.
  • the historical indicator data includes indicator data of the target material in the target delivery channel; and the first vector is input into the pre-trained first decoding network to obtain the target time point in the future.
  • the prediction index data of the target material includes: inputting the first vector into a pre-trained first decoding network to obtain the prediction index data of the target material on the target delivery channel at a future target time point.
  • each historical indicator data in the above historical indicator data sequence corresponds to a historical time point.
  • the second decoding network includes: a set of fully connected networks and a set of deconvolutional networks; and the above-mentioned first vector is input into the pre-trained second decoding network to obtain the above-mentioned historical indicator
  • the abnormal information of at least one historical indicator data in the data sequence includes: inputting the above-mentioned first vector into the target fully-connected network in the above-mentioned pre-trained fully-connected network set to obtain the second vector; wherein, the above-mentioned target fully-connected network is the same as the The network associated with the target historical time point; input the above-mentioned second vector into the deconvolution network corresponding to the above-mentioned target historical time point in the above-mentioned deconvolution network set to obtain a third vector; wherein, the data dimension of the above-mentioned third vector The data dimension of the historical indicator data corresponding to the above target historical time point is the same; according to the above third vector, the abnormal information of
  • determining the abnormal information of the historical indicator data corresponding to the target historical time point according to the third vector includes: determining the historical indicator corresponding to the third vector and the target historical time point The difference value between the data; in response to determining that the above-mentioned difference value is greater than or equal to a preset threshold, it is determined that the historical indicator data corresponding to the above-mentioned target historical time point is abnormal.
  • the above-mentioned determining whether the above-mentioned target material is re-launched according to the above-mentioned abnormal information and the prediction index data of the above-mentioned target material includes: according to the above-mentioned abnormal information, determining whether at least one of the above-mentioned historical time points exists The time point when the historical indicator data is abnormal; in response to the time point when the historical indicator data is abnormal, whether to execute the instruction sent by the receiving terminal to determine whether the above-mentioned target material is re-launched; Whether the preset conditions are met; in response to the prediction index data of the above-mentioned target material satisfying the preset conditions, it is determined that the above-mentioned target material is released again.
  • the above-mentioned third vector is generated by assigning different weights to different indicators through the above-mentioned deconvolution network.
  • inputting the historical indicator data sequence related to the target material into the pre-trained time series coding network, and obtaining the first vector includes: inputting each historical indicator data in the historical indicator data sequence into the pre-training The corresponding convolutional neural network in the set of convolutional neural networks to output a fourth vector to obtain a fourth vector sequence; input each fourth vector in the fourth vector sequence to the corresponding data leveling layer in the data leveling layer set to output a fifth vector to obtain a fifth vector sequence; input the fifth vector sequence to a pre-trained time series coding network to obtain a first vector.
  • a material delivery device comprising: a first input unit, configured to input a historical indicator data sequence related to a target material into a pre-trained time series coding network to obtain a first input unit.
  • the historical indicator data in the above-mentioned historical indicator data sequence includes indicator data of the above-mentioned target material on the delivery channel;
  • the second input unit is configured to input the first vector into the pre-trained first decoding network, Obtain the prediction index data of the target material at the future target time point; wherein, the prediction index data is the prediction result of the delivery effect of the target material on the delivery channel at the future target time point;
  • the third input unit which is configured to input the above-mentioned first vector into a pre-trained second decoding network to obtain abnormality information of at least one historical indicator data in the above-mentioned historical indicator data sequence;
  • the determining unit is configured to be configured according to the above-mentioned abnormal information and the above-mentioned prediction index of the target material data to determine whether the above target material is served again.
  • the historical indicator data includes indicator data of the target material in the target delivery channel.
  • the second input unit may be further configured to: input the first vector into a pre-trained first decoding network to obtain the prediction index data of the target material on the target delivery channel at a future target time point.
  • the above-mentioned second decoding network includes: a set of fully connected networks and a set of deconvolutional networks.
  • the third input unit may be further configured to: input the above-mentioned first vector into the target fully-connected network in the pre-trained set of fully-connected networks to obtain a second vector; wherein, the above-mentioned target fully-connected network is the same as the target historical time point Associated network; input the above-mentioned second vector into the deconvolution network corresponding to the above-mentioned target historical time point in the above-mentioned deconvolution network set to obtain a third vector; wherein, the data dimension of the above-mentioned third vector and the above-mentioned target history The data dimensions of the historical indicator data corresponding to the time point are the same; according to the third vector, the abnormal information of the historical indicator data corresponding to the target historical time point is determined.
  • the third input unit may be further configured to: determine a difference value between the third vector and the historical indicator data corresponding to the target historical time point; in response to determining that the difference value is greater than or equal to the preset threshold, it is determined that the historical indicator data corresponding to the above target historical time point is abnormal.
  • the determining unit may be further configured to: determine, according to the abnormality information, whether there is a time point in which the historical indicator data is abnormal in at least one of the above-mentioned historical time points; At a time point, whether to execute the instruction sent by the terminal to determine whether the target material is re-launched; in response to executing the above instruction, determine whether the prediction index data of the target material satisfies a preset condition; in response to the prediction index of the target material If the data satisfies the pre-set conditions, it is determined that the above-mentioned target material is delivered again.
  • the above-mentioned third vector is generated by assigning different weights to different indicators through the above-mentioned deconvolution network.
  • the first input unit is further configured to: input each historical indicator data in the historical indicator data sequence into a corresponding convolutional neural network in the set of pre-trained convolutional neural networks to output a fourth vector to obtain a fourth vector sequence; input each fourth vector in the fourth vector sequence into the corresponding data flattening layer in the data flattening layer set to output a fifth vector to obtain a fifth vector sequence;
  • the fifth vector sequence is input to the pre-trained time-series coding network to obtain the first vector.
  • each historical indicator data in the above historical indicator data sequence corresponds to a historical time point.
  • an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, when the one or more programs are stored by one or more
  • the execution of the processor causes one or more processors to implement a method as described in any of the above embodiments.
  • a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the foregoing embodiments.
  • a computer program comprising instructions that, when executed by a processor, cause the processor to perform any one of the aforementioned methods.
  • a computer program product comprising instructions that, when executed by a processor, cause the processor to perform any of the foregoing methods.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Multimedia (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

一种素材投放方法、装置、电子设备和计算机可读介质。该方法包括：将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量；其中，历史指标数据序列中的历史指标数据包括目标素材在投放渠道上的指标数据；将第一向量输入至预先训练的第一解码网络，得到未来目标时间点该目标素材的预测指标数据；将第一向量输入至预先训练的第二解码网络，得到历史指标数据序列中至少一个历史指标数据的异常信息；根据该异常信息和该目标素材的预测指标数据，确定该目标素材是否再次投放。

Description

素材投放方法、装置、设备和介质
相关申请的交叉引用
本申请是以中国申请号为202110399719.7，申请日为2021年4月14日的申请为基础，并主张其优先权，该中国申请的公开内容在此作为整体引入本申请中。
技术领域
本公开的实施例涉及计算机技术领域，具体涉及素材投放方法、装置、电子设备和计算机可读介质。
背景技术
目前，现实生活中存在大量热门或关注度较大的素材。素材也可能随着投放次数的增多效果没有那么理想。对于确定素材是否值得再次被投放，通常采用的方式为：通过对素材的上一次投放结果进行人工分析来确定素材是否再次被投放。
发明内容
本公开的内容部分用于以简要的形式介绍构思，这些构思将在后面的具体实施方式部分被详细描述。本公开的内容部分并不旨在标识要求保护的技术方案的关键特征或必要特征，也不旨在用于限制所要求的保护的技术方案的范围。
第一方面，本公开的一些实施例提供了一种素材投放方法，该方法包括：将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量；其中，上述历史指标数据序列中的历史指标数据包括上述目标素材在投放渠道上的指标数据；将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材的预测指标数据；其中，所述预测指标数据为所述未来目标时间点所述目标素材在所述投放渠道上投放效果的预测结果；将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息；根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。
第二方面，本公开的一些实施例提供了一种素材投放装置，装置包括：第一输入单元，被配置成将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量；其中，上述历史指标数据序列中的历史指标数据包括上述目标素材在 投放渠道上的指标数据；第二输入单元，被配置成将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材的预测指标数据；其中，所述预测指标数据为所述未来目标时间点所述目标素材在所述投放渠道上投放效果的预测结果；第三输入单元，被配置成将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息；确定单元，被配置成根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。
第三方面，本公开的一些实施例提供了一种电子设备，包括：一个或多个处理器；存储装置，其上存储有一个或多个程序，当一个或多个程序被一个或多个处理器执行，使得一个或多个处理器实现如第一方面中任一的方法。
第四方面，本公开的一些实施例提供了一种计算机可读介质，其上存储有计算机程序，其中，程序被处理器执行时实现如第一方面中任一的方法。
第五方面，本公开实施例提供一种计算机程序，包括：指令，所述指令当由处理器执行时使所述处理器执行前述任意一种方法。
第六方面，本公开实施例提供一种计算机程序产品，包括指令，所述指令当由处理器执行时使所述处理器执行前述任意一种方法。
附图说明
结合附图并参考以下具体实施方式，本公开各实施例的上述和其他特征、优点及方面将变得更加明显。贯穿附图中，相同或相似的附图标记表示相同或相似的元素。应当理解附图是示意性的，原件和元素不一定按照比例绘制。
图1是本公开的一些实施例的素材投放方法的一个应用场景图的示意图；
图2是根据本公开的素材投放方法一些实施例的流程图；
图3是本公开的一些实施例的素材投放方法的另一个应用场景图的示意图；
图4是根据本公开的素材投放方法的另一些实施例的流程图；
图5是本公开的一些实施例的素材投放方法对应的一个网络模型结构的示意图；
图6是根据本公开的投放信息生成装置的一些实施例的结构示意图；
图7是适于用来实现本公开的一些实施例的电子设备的结构示意图。
具体实施方式
下面将参照附图更详细地描述本公开的实施例。虽然附图中显示了本公开的某些 实施例，然而应当理解的是，本公开可以通过各种形式来实现，而且不应该被解释为限于这里阐述的实施例。相反，提供这些实施例是为了更加透彻和完整地理解本公开。应当理解的是，本公开的附图及实施例仅用于示例性作用，并非用于限制本公开的保护范围。
另外还需要说明的是，为了便于描述，附图中仅示出了与有关发明相关的部分。在不冲突的情况下，本公开中的实施例及实施例中的特征可以相互组合。
需要注意，本公开中提及的“第一”、“第二”等概念仅用于对不同的装置、模块或单元进行区分，并非用于限定这些装置、模块或单元所执行的功能的顺序或者相互依存关系。
需要注意，本公开中提及的“一个”、“多个”的修饰是示意性而非限制性的，本领域技术人员应当理解，除非在上下文另有明确指出，否则应该理解为“一个或多个”。
本公开实施方式中的多个装置之间所交互的消息或者信息的名称仅用于说明性的目的，而并不是用于对这些消息或信息的范围进行限制。
通过对素材的上一次投放结果进行人工分析来确定素材是否再次被投放，经常会存在如下技术问题：素材的上一次投放结果可能属于异常情况，以此造成下次投放效果可能并不理想。因此，素材的上一次的投放结果不能有效的作为判断依据来确定素材是否值得再次被投放。除此之外，不能全面的考虑到各种导致素材投放效果不理想的因素。
本公开的一些实施例提出了素材投放方法、装置、设备和计算机可读介质，来解决以上背景技术部分提到的技术问题。
下面将参考附图并结合实施例来详细说明本公开。
图1是本公开的一些实施例的素材投放方法的一个应用场景图的示意图。
如图1所示，对于目前素材投放所存在的方法，电子设备101可以获取目标素材102在目标应用103的投放渠道的上一次投放效果104以及目标素材102在目标应用103的投放渠道的上一次指标数据105来确定是否再次在目标应用103上投放目标素材102。投放效果可以表征目标素材102的受欢迎程度信息。例如，本次素材投放过程中，目标素材102的投放效果可以是目标素材102在各个素材中的排名。目标素材102在各个素材中的排名越高，表征目标素材越受用户喜爱。反之，目标素材102在各个素材中的排名越低，表征目标素材越不受用户喜爱。指标数据可以是目标素材102 在投放过程中所产生的各方面的数据指标。指标数据可以包括但不限于以下至少一项：目标素材102投放过程中的转化率，或目标素材102投放过程中的下载量。需要说明的是，对于指标数据和投放效果，由于可能存在素材的指标数据各方面较优，但投放效果不佳的情况。同样存在指标数据各方面中等，但投放效果较好的情况。所以从投放效果和指标数据两方面来综合考量素材是否值得下次被投放。
在本应用场景中，目标素材102可以是：文章；上述指标数据105可以是：“转换率：0.45，下载量：875”。上一次的投放效果104可以是目标素材102在各个素材中的排名为第21位。上述执行主体可以根据上一次目标素材102在各个素材中的排名是否小于100、指标数据中的转换率是否大于0.3和指标数据中的下载量是否大于500来确定素材是否再次投放。在本应用场景中，由于上一次目标素材102在各个素材中的排名在100名之内、并且它的转换率大于0.3以及下载量大于500，以此，上述执行主体可以认为目标素材102可以在目标应用103的投放渠道上投放。
对于上述目前素材投放所存在的方法，存在以下问题。
第一，目前素材投放所存在的方法仅仅考量了上一次目标素材102的投放效果和投放过程中所产生的指标数据。未考虑到上一次目标素材的投放效果和指标数据可能是异常情况。目标素材102异常情况可以包括以下两种情况：
1、目标素材102在上一次投放过程中的投放效果和投放数据都较为差，实际投放效果和投放数据远没有达到所预期的结果。此类情况可能为异常情况。上一次目标素材102异常情况的发生可能会较大概率的会影响目标素材的下次投放。
作为示例，上一次目标素材的投放时间为元宵节，对于节约用水的目标素材102投放效果可能投放效果和指标数据都不太好。此种情况的发生往往是人们更为关注与元宵节相关的素材，导致目标素材102的投放效果和指标数据较差。这种异常情况往往不能作为目标素材102下次是否投放的参考依据。
2、目标素材102在上一次投放过程中的投放效果和投放数据都较为优异，实际投放效果和投放数据远超过所预期的结果。此类情况可能为异常情况。上一次目标素材102异常情况的发生可能会较大概率的导致下次目标素材投放不理想。
作为示例，上一次目标素材的投放时间为世界水日，对于节约用水的目标素材102投放效果可能投放效果和指标数据远超过预期的结果。此种情况的发生往往是人们当时更为关注与世界水日相关的素材，导致目标素材102的投放效果和指标数据远超预期的结果。这种异常情况往往也不能作为目标素材102下次是否投放的参考依据。
第二，目前素材投放方法所存在的方法，仅仅考量了上一次目标素材102的投放效果和投放过程中所产生的指标数据。未考量其他与目标素材102相类似的素材的投放效果和指标数据对目标素材102的影响，其中，上述与目标素材102相类似的素材可以是与目标素材102相同类型的素材。
作为示例，目标素材102可以是与提示节约用水的文章。与目标素材102相类似的素材可以是水浪费原因分析的文章。
除此之外，目前，编码器-解码器结构和多目标预测模型(MT-Learning Model)是当前获得越来越多关注的深度学习的两大热门分支。
由此，可以考虑采用编码器-解码器结合多目标预测模型的方式来确定素材是否再次投放。对于采用编码器-解码器结合多目标预测模型(MT-Learning Model)的方式，需要上述编码器-解码器结合多目标预测模型可以更为关注目标素材相关的历史指标数据序列，即，将历史指标数据序列作为编码器-解码器结构中编码器的输入。上述历史指标数据序列可以表征着目标素材和目标素材相关素材的历史投放效果。
作为示例，某小游戏的游戏类素材在应用程序A上连续投放了超过30天，所以预测下一次游戏类素材是否投放时，希望可以考虑上述游戏类素材30天的历史投放效果。
除此之外，还需要上述编码器-解码器结合多目标预测模型可以检测历史指标数据序列中至少一个历史指标数据的异常信息。在这里，上述异常信息可以用于作为确定目标素材是否再次投放的决定因素。
由此可以得到，可以考量采用编码器-解码器结合多目标预测模型的方式来确定素材是否再次投放。
需要说明的是，素材投放方法可以是由电子设备101来执行。上述电子设备101可以是硬件，也可以是软件。当电子设备为硬件时，可以实现成多个服务器或终端设备组成的分布式集群，也可以实现成单个服务器或单个终端设备。当电子设备101体现为软件时，可以实现成例如用来提供分布式服务的多个软件或软件模块，也可以实现成单个软件或软件模块。在此不做具体限定。
应该理解，图1中的电子设备的数目仅仅是示意性的。根据实现需要，可以具有任意数目的电子设备。
继续参考图2，示出了根据本公开的素材投放方法的一些实施例的流程200。该 素材投放方法，包括以下步骤：
步骤201，将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量。
在一些实施例中，素材投放方法的执行主体(例如图1所示的电子设备101)可以将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量。上述历史指标数据序列中的历史指标数据包括上述目标素材在投放渠道上的指标数据。上述目标素材可以包括但不限于以下至少一项：目标视频，目标音乐，或目标文章。上述时序编码网络可以是处理时序数据的编码网络。上述时序编码网络可以包括但不限于以下至少一项：循环神经网络(Recurrent Neural Network,RNN)-自编码器，或长短期记忆网络(LSTM，Long Short-Term Memory)-自编码器网络。上述投放渠道可以是素材投放的途径信息。上述投放渠道可以包括但不限于以下至少一项：在目标品牌的移动终端上投放，在使用目标操作系统的移动终端上投放，或在移动终端的目标应用上投放。例如，素材可以投放在目标品牌的目标操作系统的移动终端的目标应用上。上述历史指标数据可以是目标素材历史投放过程中的各个指标参数。作为示例，上述历史指标数据包括上述目标素材在投放渠道集合中目标投放渠道上预定时间指标参数的信息、上述目标素材在投放渠道集合中各个投放渠道上预定时间指标参数的信息、与上述目标素材相关素材在上述各个投放渠道上预定时间指标参数的信息。指标参数可以包括但不限于以下至少一项：点击通过率(CTR，Click-Through-Rate)、转化率(CVR，Conversion Rate)、回报率(Return on Investment，ROI)、每次行动成本(Cost Per Action，CPA)、或安装量。上述历史指标数据序列中各个历史指标数据是依照对应历史时间点的先后顺序排列的。
需要指出的是，上述历史指标数据序列中可以包括与目标素材相关联的素材的历史指标数据。采用与目标素材相关的素材的历史指标数据作为输入的目的在于将与目标素材相同题型的素材对目标素材的投放收益影响进行考量。由此，网络模型可以确定是否再次投放上述目标素材的信息更为准确。
作为示例，将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量，可以包括以下步骤。
第一步，将历史指标数据序列中每个历史指标数据输入至预先训练的卷积神经网络集合中对应的卷积神经网络以输出第四向量，得到第四向量序列。
第二步，将上述第四向量序列中每个第四向量输入至数据拉平层集合中对应的数 据拉平层以输出第五向量，得到第五向量序列。上述数据拉平层可以将矩阵维度为(n，m)的矩阵变化为(n*m，1)的矩阵，即，将多维数据拉平至一维数据。作为示例，目标矩阵的向量维度可以是(4，5)。将上述目标矩阵输入至数据拉平层，得到(20，1)的矩阵。由此，输入至数据拉平层的矩阵的元素不发生变化，对应的数据维度发生变化。
第三步，将上述第五向量序列输入至预先训练的时序编码网络，得到第一向量。
时序编码网络的训练可以是结合后续的第一解码网络和第二解码网络一起训练的，具体的训练步骤如下。
第一步，确定上述初始网络模型的网络结构以及初始化上述初始网络模型的网络参数，其中，上述初始网络模型包括：时序编码网络、第一解码网络和第二解码网络。
第二步，获取训练样本集，其中，训练样本集包括样本集合和与上述样本集合对应的标注信息集合。
第三步，将上述训练样本集中的样本集合和与上述标注信息集合分别作为上述初始网络模型的输入和期望输出，利用深度学习方法训练上述初始网络模型。
第四步，将训练得到的上述初始网络模型确定为训练后的网络模型。
步骤202，将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材的预测指标数据。
在一些实施例中，上述执行主体可以将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材的预测指标数据，其中，上述预测指标数据为上述目标素材在上述投放渠道上，上述未来目标时间点的预测结果。上述未来目标时间点可以预先设置的。上述第一解码网络与上述时序编码网络相对应。作为示例，上述第一解码网络可以是用于做回归任务的网络。上述用于做回归任务的网络可以包括但不限以下至少一项：多层全连接层，线性回归，或支持向量机(Support Vector Machine，SVM)。
作为示例，可以首先将上述第一向量输入至预先训练的预定数目层卷积神经网络，得到输出结果。然后，将上述输出结果输入至预先训练的回归网络，得到未来目标时间点上述目标素材的预测指标数据。
在一些实施例的一些可选的实现方式中，上述历史指标数据包括上述目标素材在目标投放渠道的指标数据；以及上述将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材的预测指标数据，可以包括以下步骤：
将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材在上述目标投放渠道上的预测指标数据。
在这里，在步骤202的情况下，进一步限定历史指标数据可以包括目标素材在目标投放渠道的指标数据。可以更有针对性地确定在某一确定的投放渠道可以进行素材是否再次投放。例如，上述目标投放渠道可以是A应用。上述A应用为投放目标素材方所运行的主要应用。所以需要了解目标素材在A应用是否进行投放。由此，需要将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材在上述A应用上的预测指标数据。
步骤203，将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息。
在一些实施例中，上述执行主体可以将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息。上述第二解码网络可以包括但不限于以下至少一项：全连接网络，或循环神经网络。上述异常信息可以是历史指标数据序列中是否有历史指标数据存在异常，以及哪些指标数据出现了异常。
需要说明的是，考虑上述历史指标数据序列中至少一个历史指标数据的异常信息的原因在于，网络模型的特征学习比较依赖于历史指标数据。如果网络模型的输入历史指标数据中存有异常情况，可能网络模型会学习到上述异常情况的特征信息，使得预测未来目标时间点上述目标素材的预测指标数据不够精准。上述网络模型包括：时序编码网络、第一解码网络和第二解码网络。
步骤204，根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。
在一些实施例中，上述执行主体可以根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。作为示例，响应于确定异常信息表征为历史指标数据序列中某一历史指标数据出现异常，上述执行主体可以确定上述目标素材不再投放。响应于确定异常信息表征为历史指标数据序列中正常，上述执行主体可以根据目标素材在上述投放渠道上，上述未来目标时间点的预测结果，确定上述目标素材是否再次投放。
本公开的上述各个实施例中具有如下有益效果：本公开的一些实施例的素材投放方法通过生成的预测指标数据和异常信息来更加准确、有效的确定出目标素材是否再 次投放。具体来说，素材的上一次投放结果可能属于异常情况，以此造成下次投放效果可能并不理想。因此，素材的上一次的投放结果不能有效的作为确定素材是否值得再次被投放的判断依据。
除此之外，相关技术不能全面的考虑到各种导致素材投放效果不理想的因素。基于此，本公开的一些实施例的素材投放方法首先将目标素材相关的历史指标数据序列作为时序编码网络的输入，可以更为全面的考虑到各方面可能导致目标素材投放不理想的因素。然后，将上述第一向量输入至预先训练的第一解码网络，可以更加准确、有效的生成未来目标时间点上述目标素材的预测指标数据。进而，将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息。在这里，通过对历史指标数据序列中异常的至少一个历史指标数据的确定，可以更加有效地排除目标素材的历史投放结果的偶然性。最后，根据上述异常信息和上述目标素材的预测指标数据，能够更高效、便捷的确定出上述目标素材是否再次投放。
图3是本公开的一些实施例的素材投放方法的另一个应用场景图的示意图。
如图3所示，电子设备301可以首先将目标素材相关的历史指标数据序列302输入至预先训练的时序编码网络303，得到第一向量。可选的，上述时序编码网络303可以是长短期记忆网络。在本应用场景中，历史指标数据序列302可以包括历史指标数据3021，历史指标数据3022，历史指标数据3023。历史指标数据序列302中历史指标数据3021对应的时间要早于历史指标数据3022对应的时间。历史指标数据3022对应的时间要早于历史指标数据3023对应的时间。上述历史指标数据3021对应的时间与历史指标数据3021对应的时间之间的时间间隔可以与历史指标数据3021对应的时间与历史指标数据3023对应的时间之间的时间间隔相同。每个历史指标数据可以包括3层数据。历史指标数据的每层数据为多个维度的指标。历史指标数据的第一层数据可以包括：目标素材在目标投放渠道上的各个指标数据。历史指标数据的第二层数据可以包括：目标素材在各个投放渠道的各个指标数据。历史指标数据的第三层数据可以包括：与目标素材相关联的各个素材在各个投放渠道的各个指标数据。
然后，将上述第一向量输入至预先训练的第一解码网络304，得到未来目标时间点上述目标素材的预测指标数据306。
上述预测指标数据306为上述目标素材在上述投放渠道上、上述未来目标时间点 的预测结果。进而，将上述第一向量输入至预先训练的第二解码网络305，得到上述历史指标数据序列302中至少一个历史指标数据的异常信息307。可选的，上述第二解码网络可以是由全连接网络集合和反卷积网络集合组成的网络。最后，根据上述异常信息307和上述目标素材的预测指标数据306，确定上述目标素材是否再次投放。可选的，响应于确定异常信息307表征为历史指标数据序列中某一历史指标数据出现异常，上述执行主体可以确定上述目标素材不再投放。响应于确定异常信息307表征为历史指标数据序列中正常，可以根据目标素材在上述投放渠道上上述未来目标时间点的预测结果，确定上述目标素材是否再次投放。
需要说明的是，素材投放方法可以是由电子设备301来执行。上述电子设备301可以是硬件，也可以是软件。当电子设备为硬件时，可以实现成多个服务器或终端设备组成的分布式集群，也可以实现成单个服务器或单个终端设备。当电子设备301体现为软件时，可以实现成例如用来提供分布式服务的多个软件或软件模块，也可以实现成单个软件或软件模块。在此不做具体限定。
应该理解，图3中的电子设备的数目仅仅是示意性的。根据实现需要，可以具有任意数目的电子设备。
继续参考图4，示出了根据本公开的素材投放方法的另一些实施例的流程400。该素材投放方法，包括以下步骤。
步骤401，将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量。
步骤402，将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材的预测指标数据。
步骤403，将上述第一向量输入至预先训练的上述全连接网络集合中的目标全连接网络，得到第二向量。
在一些实施例中，素材投放方法的执行主体(例如图1或图3所示的电子设备)可以将上述第一向量输入至预先训练的上述全连接网络集合中的目标全连接网络，得到第二向量，其中，上述历史指标数据序列中每个历史指标数据对应一个历史时间点。
需要说明的是，上述目标全连接网络中全连接层的个数可以是与历史时间点的不同而变化的。历史时间点越久，对应目标全连接网络中包括的全连接层数目越多。
作为示例，存在3个历史时间点，依据时间先后的顺序排序为：第一历史时间点、 第二历史时间点、第三历史时间点。则第一历史时间点对应的全连接层的数目小于第二历史时间点对应的全连接层的数目。第二历史时间点对应的全连接层的数目小于第三历史时间点对应的全连接层的数目。
步骤404，将上述第二向量输入至上述反卷积网络集合中与上述目标历史时间点对应的反卷积网络，得到第三向量。
在一些实施例中，上述执行主体可以将上述第二向量输入至上述反卷积网络集合中与上述目标历史时间点对应的反卷积网络，得到第三向量，其中，上述第三向量的数据维度与上述目标历史时间点对应的历史指标数据的数据维度相同。上述第三向量和目标历史时间点对应的历史指标数据可以是以矩阵的形式展现的。
需要强调的是，在网络模型训练中，全连接网络集合和反卷积网络集合可以起到稳定网络结构的作用。
在一些实施例的一些可选的实现方式中，上述第三向量是通过上述反卷积网络赋予不同指标不同权重生成的。
步骤405，根据上述第三向量，确定上述目标历史时间点对应的历史指标数据的异常信息。
在一些实施例中，上述执行主体可以根据上述第三向量，确定上述目标历史时间点对应的历史指标数据的异常信息。
作为示例，上述执行主体可以首先确定上述第三向量与上述目标历史时间点对应的历史指标数据之间的余弦值。然后，响应于上述余弦值小于预先设定的阈值，则上述目标历史时间点对应的历史指标数据没有发生异常。
在一些实施例的一些可选的实现方式中，上述根据上述第三向量，确定上述目标历史时间点对应的历史指标数据的异常信息，可以包括以下步骤。
第一步，确定上述第三向量与上述目标历史时间点对应的历史指标数据之间的差异值。作为示例，上述执行主体可以首先确定上述第三向量与上述目标历史时间点对应的历史指标数据对应位置的矩阵元素数值的差值，得到差值集合作为差异值。
第二步，响应于确定上述差异值大于或等于预先设定的阈值，确定上述目标历史时间点对应的历史指标数据异常。
步骤406，根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。
在一些实施例的一些可选的实现方式中，上述根据上述异常信息和上述目标素材 的预测指标数据，确定上述目标素材是否再次投放，可以包括以下步骤。
第一步，根据上述异常信息，确定至少一个上述历史时间点中是否存在历史指标数据异常的时间点。作为示例，上述执行主体可以依据上述异常信息，通过查询的方式来确定至少一个上述历史时间点中是否存在历史指标数据异常的时间点。
第二步，响应于存在历史指标数据异常的时间点，接收终端发送的是否执行确定上述目标素材是否再次投放的指令。
第三步，响应于执行上述指令，确定上述目标素材的预测指标数据的是否满足预先设定的条件，其中，上述目标素材的预测指标数据可以包括多个由网络模型预测出来的指标数据。各个指标数据分别对应着独自的正常数据取值范围。由此，预先设定的条件可以是预测出来每个指标数据对应的数值处于正常数据取值范围。
第四步，响应于上述目标素材的预测指标数据满足预先设定的条件，确定上述目标素材再次投放。
在一些实施例中，步骤401-402、406的具体实现及所带来的技术效果可以参考图2对应的那些实施例中的步骤201-202、204，在此不再赘述。
从图4中可以看出，与图2对应的一些实施例的描述相比，图4对应的一些实施例中的素材投放方法的流程400更加突出了异常信息生成的具体步骤。由此，这些实施例描述的方案通过全连接网络集合和反卷积网络集合来更为精准、有效的生成异常信息。以此，进一步通过上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。
图5是本公开的一些实施例的素材投放方法对应的一个网络模型结构的示意图。
如图5所示，首先，获取历史指标数据序列。由于上述历史指标数据可以是历史投放过程中与目标素材相关的各个指标数据。针对之前素材是否再次投放，上述历史指标数据还可以包括上述目标素材在投放渠道集合中目标投放渠道上预定时间指标数据的信息、上述目标素材在投放渠道集合中各个投放渠道上预定时间指标数据的信息、与上述目标素材相关素材在上述各个投放渠道上预定时间指标数据的信息。上述指标参数可以包括但不限于以下至少一项：点击通过率、转化率、回报率、每次行动成本、或安装量。
在这里，多类别数据的输入使得后续网络模型学习到更多有用的特征信息。通过学习更多的特征信息以考虑各种指标数据、与目标素材相关素材对目标素材是否再次 投放的影响。使得确定目标素材是否再次投放更为精准。
然后，将历史指标数据序列中的每个历史指标数据输入至卷积神经网络集合中对应的卷积神经网络用于初步提取上述历史指标数据的特征信息。历史指标数据序列中的第一历史指标数据501输入至卷积神经网络集合中的第一卷积神经网络504。历史指标数据序列中的第二历史指标数据502输入至卷积神经网络集合中的第二卷积神经网络505。历史指标数据序列中的第三历史指标数据503输入至卷积神经网络集合中的第三卷积神经网络506。
进而，将第一卷积神经网络504的输出向量输入至第一数据拉平层507用以将输出向量的向量维度调整为上述长短期记忆网络510的可输入向量维度。将第二卷积神经网络505的输出向量输入至第二数据拉平层508用以将输出向量的向量维度调整为上述长短期记忆网络510的可输入向量维度。将第三卷积神经网络506的输出向量输入至第三数据拉平层509用以将输出向量的向量维度调整为上述长短期记忆网络510的可输入向量维度。将第一数据拉平层507、第二数据拉平层508和第三数据拉平层509的输出结果输入至长短期记忆网络510中对应的单元。即，将上述第一数据拉平层507的输出结果作为长短期记忆网络第一单元5101的输入。将上述第二数据拉平层508的输出结果作为长短期记忆网络第二单元5102的输入。将上述第三数据拉平层509的输出结果作为长短期记忆网络第一单元5013的输入。在这里，上述长短期记忆网络可以学习到历史指标数据序列中历史指标数据之间的时序信息。
接着，将长短期记忆网络510的输出结果输入至第一全连接层511，得到第一全连接层511的输出结果。将第一全连接层511的输出结果输入至第一全连接网络512，得到第一全连接网络512的输出结果。将第一全连接层512的输出结果输入至第三全连接层513，得到第三全连接层513的输出结果。其中，上述第一全连接层511、第二全连接层512和第三全连接层513由于激活函数的存在可以学习到非线性特征信息。
在这里，针对以往时序编码网络，常常采用循环神经网络-自编码器。本公开的素材投放方法中的时序编码网络可以采用长短期记忆网络-自编码器网络。长短期记忆网络-自编码器网络可以有效解决长期依赖问题。避免了循环神经网路-自编码器出现的长输入序列的信息传递问题。
最后，将第三全连接层513的输出结果进行回归预测，得到未来目标时间点投放渠道上的上述目标素材的预测指标数据514。除此之外，将长短期记忆网络510的输出结果输入至第二全连接层515，得到第二全连接层515的输出结果。将第二全连接 层515的输出结果输入至反卷积网络517，得到第三向量519。另外，第二全连接层515的输出结果输入至第二全连接层516，得到第二全连接层516的输出结果。将第二全连接层516的输出结果输入至反卷积网络518，得到第三向量520。在这里，利用反卷积网络来生成与历史指标数据的数据维度相同的各个第三向量，可以通过各个第三向量与对应的历史指标数据相对比，得到上述历史指标数据序列中至少一个历史指标数据的异常信息。
在这里，上述反卷积网络可以作为解码网络来针对长短期记忆网络进行解码。通过各个反卷积网络、各个第二全连接层与长短期记忆网络的结合可以来扩大模型输入的历史指标数据的数据形态和指标数据类别。
继续参考图6，作为对上述各图上述方法的实现，本公开提供了一种投放信息生成装置的一些实施例，这些装置实施例与图2上述的那些方法实施例相对应，该装置具体可以应用于各种电子设备中。
如图6所示，一些实施例的素材投放装置600包括：第一输入单元601、第二输入单元602、第三输入单元603和确定单元604。第一输入单元601，被配置成将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量；上述历史指标数据序列中的历史指标数据包括上述目标素材在投放渠道上的指标数据。第二输入单元602，被配置成将所述第一向量输入至预先训练的第一解码网络，得到未来目标时间点所述目标素材的预测指标数据，其中，所述预测指标数据为所述未来目标时间点所述目标素材在所述投放渠道上投放效果的预测结果。第三输入单元603，被配置成将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息。确定单元604，被配置成根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。
在一些实施例的一些可选实现方式中，上述历史指标数据包括上述目标素材在目标投放渠道的指标数据。第二输入单元602可以进一步被配置成：将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材在上述目标投放渠道上的预测指标数据。
在一些实施例的一些可选实现方式中，上述第二解码网络包括：全连接网络集合和反卷积网络集合。第三输入单元603可以进一步被配置成：将上述第一向量输入至预先训练的上述全连接网络集合中的目标全连接网络，得到第二向量；其中，上述目 标全连接网络是与目标历史时间点相关联的网络；将上述第二向量输入至上述反卷积网络集合中与上述目标历史时间点对应的反卷积网络，得到第三向量；其中，上述第三向量的数据维度与上述目标历史时间点对应的历史指标数据的数据维度相同；根据上述第三向量，确定上述目标历史时间点对应的历史指标数据的异常信息。
在一些实施例的一些可选实现方式中，第三输入单元603可以进一步被配置成：确定上述第三向量与上述目标历史时间点对应的历史指标数据之间的差异值；响应于确定上述差异值大于或等于预先设定的阈值，确定上述目标历史时间点对应的历史指标数据异常。
在一些实施例的一些可选实现方式中，确定单元604可以进一步被配置成：根据上述异常信息，确定至少一个上述历史时间点中是否存在历史指标数据异常的时间点；响应于存在历史指标数据异常的时间点，接收终端发送的是否执行确定上述目标素材是否再次投放的指令；响应于执行上述指令，确定上述目标素材的预测指标数据的是否满足预先设定的条件；响应于上述目标素材的预测指标数据满足预先设定的条件，确定上述目标素材再次投放。
在一些实施例的一些可选实现方式中，上述第三向量是通过上述反卷积网络赋予不同指标不同权重生成的。
在一些实施例的一些可选实现方式中，第一输入单元601可以进一步被配置成：将历史指标数据序列中每个历史指标数据输入至预先训练的卷积神经网络集合中对应的卷积神经网络以输出第四向量，得到第四向量序列；将所述第四向量序列中每个第四向量输入至数据拉平层集合中对应的数据拉平层以输出第五向量，得到第五向量序列；将所述第五向量序列输入至预先训练的时序编码网络，得到第一向量。
在一些实施例的一些可选实现方式中，上述历史指标数据序列中每个历史指标数据对应一个历史时间点。
可以理解的是，该装置600中记载的诸单元与参考图2描述的方法中的各个步骤相对应。由此，上文针对方法描述的操作、特征以及产生的有益效果同样适用于装置600及其中包含的单元，在此不再赘述。
下面参考图7，其示出了适于用来实现本公开的一些实施例的电子设备(例如图1的电子设备)700的结构示意图。图7示出的电子设备仅仅是一个示例，不应对本公开的实施例的功能和使用范围带来任何限制。
如图7所示，电子设备700可以包括处理装置(例如中央处理器、图形处理器等)701，其可以根据存储在只读存储器(ROM)702中的程序或者从存储装置708加载到随机访问存储器(RAM)703中的程序而执行各种适当的动作和处理。在RAM 703中，还存储有电子设备700操作所需的各种程序和数据。处理装置701、ROM 702以及RAM 703通过总线704彼此相连。输入/输出(I/O)接口705也连接至总线704。
通常，以下装置可以连接至I/O接口705：包括例如触摸屏、触摸板、键盘、鼠标、摄像头、麦克风、加速度计、陀螺仪等的输入装置706；包括例如液晶显示器(LCD)、扬声器、振动器等的输出装置707；包括例如磁带、硬盘等的存储装置708；以及通信装置709。通信装置709可以允许电子设备700与其他设备进行无线或有线通信以交换数据。虽然图7示出了具有各种装置的电子设备700，但是应理解的是，并不要求实施或具备所有示出的装置。可以替代地实施或具备更多或更少的装置。图7中示出的每个方框可以代表一个装置，也可以根据需要代表多个装置。
特别地，根据本公开的一些实施例，上文参考流程图描述的过程可以被实现为计算机软件程序。例如，本公开的一些实施例包括一种计算机程序产品，其包括承载在计算机可读介质上的计算机程序，该计算机程序包含用于执行流程图所示的方法的程序代码。在这样的一些实施例中，该计算机程序可以通过通信装置709从网络上被下载和安装，或者从存储装置708被安装，或者从ROM 702被安装。在该计算机程序被处理装置701执行时，执行本公开的一些实施例的方法中限定的上述功能。
需要说明的是，本公开的一些实施例上述的计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质或者是上述两者的任意组合。计算机可读存储介质例如可以是——但不限于——电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子可以包括但不限于：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机访问存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本公开的一些实施例中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。而在本公开的一些实施例中，计算机可读信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。计算机可读 信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读信号介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。计算机可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于：电线、光缆、RF(射频)等等，或者上述的任意合适的组合。
在一些实施方式中，客户端、服务器可以利用诸如HTTP(HyperText Transfer Protocol，超文本传输协议)之类的任何当前已知或未来研发的网络协议进行通信，并且可以与任意形式或介质的数字数据通信(例如，通信网络)互连。通信网络的示例包括局域网(“LAN”)，广域网(“WAN”)，网际网(例如，互联网)以及端对端网络(例如，ad hoc端对端网络)，以及任何当前已知或未来研发的网络。
上述计算机可读介质可以是上述装置中所包含的；也可以是单独存在，而未装配入该电子设备中。上述计算机可读介质承载有一个或者多个程序，当上述一个或者多个程序被该电子设备执行时，使得该电子设备：将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量；其中，上述历史指标数据序列中的历史指标数据包括上述目标素材在投放渠道上的指标数据；将所述第一向量输入至预先训练的第一解码网络，得到未来目标时间点所述目标素材的预测指标数据；其中，所述预测指标数据为所述未来目标时间点所述目标素材在所述投放渠道上投放效果的预测结果；将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息；根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。
可以以一种或多种程序设计语言或其组合来编写用于执行本公开的一些实施例的操作的计算机程序代码，上述程序设计语言包括面向对象的程序设计语言—诸如Java、Smalltalk、C++，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络——包括局域网(LAN)或广域网(WAN)——连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。
附图中的流程图和框图，图示了按照本公开各种实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段、或代码的一部分，该模块、程序段、或代码的一部分包 含一个或多个用于实现规定的逻辑功能的可执行指令。也应当注意，在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个接连地表示的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或操作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
描述于本公开的一些实施例中的单元可以通过软件的方式实现，也可以通过硬件的方式来实现。所描述的单元也可以设置在处理器中，例如，可以描述为：一种处理器包括第一输入单元、第二输入单元、第三输入单元和确定单元。其中，这些单元的名称在某种情况下并不构成对该单元本身的限定，例如，第一输入单元还可以被描述为“将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量的单元”。
本文中以上描述的功能可以至少部分地由一个或多个硬件逻辑部件来执行。例如，非限制性地，可以使用的示范类型的硬件逻辑部件包括：现场可编程门阵列(FPGA)、专用集成电路(ASIC)、专用标准产品(ASSP)、片上系统(SOC)、复杂可编程逻辑设备(CPLD)等等。
根据本公开的一个或多个实施例，提供了一种素材投放方法，包括：将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量；其中，上述历史指标数据序列中的历史指标数据包括上述目标素材在投放渠道上的指标数据；将所述第一向量输入至预先训练的第一解码网络，得到未来目标时间点所述目标素材的预测指标数据；其中，所述预测指标数据为所述未来目标时间点所述目标素材在所述投放渠道上投放效果的预测结果；将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息；根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。
根据本公开的一个或多个实施例，上述历史指标数据包括上述目标素材在目标投放渠道的指标数据；以及上述将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材的预测指标数据，包括：将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材在上述目标投放渠道上的预测指标数据。
根据本公开的一个或多个实施例，上述历史指标数据序列中每个历史指标数据对 应一个历史时间点。
根据本公开的一个或多个实施例，上述第二解码网络包括：全连接网络集合和反卷积网络集合；以及上述将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息，包括：将上述第一向量输入至预先训练的上述全连接网络集合中的目标全连接网络，得到第二向量；其中，上述目标全连接网络是与目标历史时间点相关联的网络；将上述第二向量输入至上述反卷积网络集合中与上述目标历史时间点对应的反卷积网络，得到第三向量；其中，上述第三向量的数据维度与上述目标历史时间点对应的历史指标数据的数据维度相同；根据上述第三向量，确定上述目标历史时间点对应的历史指标数据的异常信息。
根据本公开的一个或多个实施例，上述根据上述第三向量，确定上述目标历史时间点对应的历史指标数据的异常信息，包括：确定上述第三向量与上述目标历史时间点对应的历史指标数据之间的差异值；响应于确定上述差异值大于或等于预先设定的阈值，确定上述目标历史时间点对应的历史指标数据异常。
根据本公开的一个或多个实施例，上述根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放，包括：根据上述异常信息，确定至少一个上述历史时间点中是否存在历史指标数据异常的时间点；响应于存在历史指标数据异常的时间点，接收终端发送的是否执行确定上述目标素材是否再次投放的指令；响应于执行上述指令，确定上述目标素材的预测指标数据的是否满足预先设定的条件；响应于上述目标素材的预测指标数据满足预先设定的条件，确定上述目标素材再次投放。
根据本公开的一个或多个实施例，上述第三向量是通过上述反卷积网络赋予不同指标不同权重生成的。
根据本公开的一个或多个实施例，将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量包括：将历史指标数据序列中每个历史指标数据输入至预先训练的卷积神经网络集合中对应的卷积神经网络以输出第四向量，得到第四向量序列；将所述第四向量序列中每个第四向量输入至数据拉平层集合中对应的数据拉平层以输出第五向量，得到第五向量序列；将所述第五向量序列输入至预先训练的时序编码网络，得到第一向量。
根据本公开的一个或多个实施例，提供了一种素材投放装置，包括：第一输入单元，被配置成将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络， 得到第一向量；其中，上述历史指标数据序列中的历史指标数据包括上述目标素材在投放渠道上的指标数据；第二输入单元，被配置成将所述第一向量输入至预先训练的第一解码网络，得到未来目标时间点所述目标素材的预测指标数据；其中，所述预测指标数据为所述未来目标时间点所述目标素材在所述投放渠道上投放效果的预测结果；第三输入单元，被配置成将上述第一向量输入至预先训练的第二解码网络，得到上述历史指标数据序列中至少一个历史指标数据的异常信息；确定单元，被配置成根据上述异常信息和上述目标素材的预测指标数据，确定上述目标素材是否再次投放。
根据本公开的一个或多个实施例，上述历史指标数据包括上述目标素材在目标投放渠道的指标数据。第二输入单元可以进一步被配置成：将上述第一向量输入至预先训练的第一解码网络，得到未来目标时间点上述目标素材在上述目标投放渠道上的预测指标数据。
根据本公开的一个或多个实施例，上述第二解码网络包括：全连接网络集合和反卷积网络集合。第三输入单元可以进一步被配置成：将上述第一向量输入至预先训练的上述全连接网络集合中的目标全连接网络，得到第二向量；其中，上述目标全连接网络是与目标历史时间点相关联的网络；将上述第二向量输入至上述反卷积网络集合中与上述目标历史时间点对应的反卷积网络，得到第三向量；其中，上述第三向量的数据维度与上述目标历史时间点对应的历史指标数据的数据维度相同；根据上述第三向量，确定上述目标历史时间点对应的历史指标数据的异常信息。
根据本公开的一个或多个实施例，第三输入单元可以进一步被配置成：确定上述第三向量与上述目标历史时间点对应的历史指标数据之间的差异值；响应于确定上述差异值大于或等于预先设定的阈值，确定上述目标历史时间点对应的历史指标数据异常。
根据本公开的一个或多个实施例，确定单元可以进一步被配置成：根据上述异常信息，确定至少一个上述历史时间点中是否存在历史指标数据异常的时间点；响应于存在历史指标数据异常的时间点，接收终端发送的是否执行确定上述目标素材是否再次投放的指令；响应于执行上述指令，确定上述目标素材的预测指标数据的是否满足预先设定的条件；响应于上述目标素材的预测指标数据满足预先设定的条件，确定上述目标素材再次投放。
根据本公开的一个或多个实施例，上述第三向量是通过上述反卷积网络赋予不同指标不同权重生成的。
根据本公开的一个或多个实施例，第一输入单元进一步被配置为：将历史指标数据序列中每个历史指标数据输入至预先训练的卷积神经网络集合中对应的卷积神经网络以输出第四向量，得到第四向量序列；将所述第四向量序列中每个第四向量输入至数据拉平层集合中对应的数据拉平层以输出第五向量，得到第五向量序列；将所述第五向量序列输入至预先训练的时序编码网络，得到第一向量。
根据本公开的一个或多个实施例，上述历史指标数据序列中每个历史指标数据对应一个历史时间点。
根据本公开的一个或多个实施例，提供了一种电子设备，包括：一个或多个处理器；存储装置，其上存储有一个或多个程序，当一个或多个程序被一个或多个处理器执行，使得一个或多个处理器实现如上述任一实施例描述的方法。
根据本公开的一个或多个实施例，提供了一种计算机可读介质，其上存储有计算机程序，其中，程序被处理器执行时实现如上述任一实施例描述的方法。
根据本公开的一个或多个实施例，提供了一种计算机程序，包括：指令，所述指令当由处理器执行时使所述处理器执行前述任意一种方法。
根据本公开的一个或多个实施例，提供了一种计算机程序产品，包括指令，所述指令当由处理器执行时使所述处理器执行前述任意一种方法。
以上描述仅为本公开的一些较佳实施例以及对所运用技术原理的说明。本领域技术人员应当理解，本公开的实施例中所涉及的发明范围，并不限于上述技术特征的特定组合而成的技术方案，同时也应涵盖在不脱离上述发明构思的情况下，由上述技术特征或其等同特征进行任意组合而形成的其它技术方案。例如上述特征与本公开的实施例中公开的(但不限于)具有类似功能的技术特征进行互相替换而形成的技术方案。

Claims (13)

  1. 一种素材投放方法，包括：
    将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量；其中，所述历史指标数据序列中的历史指标数据包括所述目标素材在投放渠道上的指标数据；
    将所述第一向量输入至预先训练的第一解码网络，得到未来目标时间点所述目标素材的预测指标数据；其中，所述预测指标数据为所述未来目标时间点所述目标素材在所述投放渠道上投放效果的预测结果；
    将所述第一向量输入至预先训练的第二解码网络，得到所述历史指标数据序列中至少一个历史指标数据的异常信息；
    根据所述异常信息和所述目标素材的预测指标数据，确定所述目标素材是否再次投放。
  2. 根据权利要求1所述的方法，其中，所述历史指标数据包括所述目标素材在目标投放渠道的指标数据；以及
    所述将所述第一向量输入至预先训练的第一解码网络，得到未来目标时间点所述目标素材的预测指标数据，包括：
    将所述第一向量输入至预先训练的第一解码网络，得到未来目标时间点所述目标素材在所述目标投放渠道上的预测指标数据。
  3. 根据权利要求1所述的方法，其中，所述历史指标数据序列中每个历史指标数据对应一个历史时间点。
  4. 根据权利要求3所述的方法，其中，所述第二解码网络包括：全连接网络集合和反卷积网络集合；以及
    所述将所述第一向量输入至预先训练的第二解码网络，得到所述历史指标数据序列中至少一个历史指标数据的异常信息，包括：
    将所述第一向量输入至预先训练的所述全连接网络集合中的目标全连接网络，得到第二向量；其中，所述目标全连接网络是与目标历史时间点相关联的网络；
    将所述第二向量输入至所述反卷积网络集合中与所述目标历史时间点对应的反卷积网络，得到第三向量；其中，所述第三向量的数据维度与所述目标历史时间点对应的历史指标数据的数据维度相同；
    根据所述第三向量，确定所述目标历史时间点对应的历史指标数据的异常信息。
  5. 根据权利要求4所述的方法，其中，所述根据所述第三向量，确定所述目标历史时间点对应的历史指标数据的异常信息，包括：
    确定所述第三向量与所述目标历史时间点对应的历史指标数据之间的差异值；
    响应于确定所述差异值大于或等于预先设定的阈值，确定所述目标历史时间点对应的历史指标数据异常。
  6. 根据权利要求3所述的方法，其中，所述根据所述异常信息和所述目标素材的预测指标数据，确定所述目标素材是否再次投放，包括：
    根据所述异常信息，确定至少一个所述历史时间点中是否存在历史指标数据异常的时间点；
    响应于存在历史指标数据异常的时间点，接收终端发送的是否执行确定所述目标素材是否再次投放的指令；
    响应于执行所述指令，确定所述目标素材的预测指标数据的是否满足预先设定的条件；
    响应于所述目标素材的预测指标数据满足预先设定的条件，确定所述目标素材再次投放。
  7. 根据权利要求4所述的方法，其中，所述第三向量是通过所述反卷积网络赋予不同指标不同权重生成的。
  8. 根据权利要求1所述的方法，其中，所述将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量包括：
    将历史指标数据序列中每个历史指标数据输入至预先训练的卷积神经网络集合中对应的卷积神经网络以输出第四向量，得到第四向量序列；
    将所述第四向量序列中每个第四向量输入至数据拉平层集合中对应的数据拉平 层以输出第五向量，得到第五向量序列；
    将所述第五向量序列输入至预先训练的时序编码网络，得到第一向量。
  9. 一种素材投放装置，包括：
    第一输入单元，被配置成将目标素材相关的历史指标数据序列输入至预先训练的时序编码网络，得到第一向量；其中，所述预测指标数据为所述未来目标时间点所述目标素材在所述投放渠道上投放效果的预测结果；
    第二输入单元，被配置成将所述第一向量输入至预先训练的第一解码网络，得到未来目标时间点所述目标素材的预测指标数据；其中，所述预测指标数据为所述目标素材在所述投放渠道上所述未来目标时间点的预测结果；
    第三输入单元，被配置成将所述第一向量输入至预先训练的第二解码网络，得到所述历史指标数据序列中至少一个历史指标数据的异常信息；
    确定单元，被配置成根据所述异常信息和所述目标素材的预测指标数据，确定所述目标素材是否再次投放。
  10. 一种电子设备，包括：
    一个或多个处理器；
    存储装置，用于存储一个或多个程序；
    当所述一个或多个程序被所述一个或多个处理器执行时，使得所述一个或多个处理器实现如权利要求1-7中任一所述的方法。
  11. 一种计算机可读介质，其上存储有计算机程序，其中，所述程序被处理器执行时实现如权利要求1-7中任一所述的方法。
  12. 一种计算机程序，包括：指令，所述指令当由处理器执行时使所述处理器执行根据权利要求1-8中任一所述的方法。
  13. 一种计算机程序产品，包括指令，所述指令当由处理器执行时使所述处理器执行根据权利要求1-8中任一所述的方法。
PCT/CN2022/080125 2021-04-14 2022-03-10 素材投放方法、装置、设备和介质 WO2022218068A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110399719.7A CN112989203B (zh) 2021-04-14 2021-04-14 素材投放方法、装置、设备和介质
CN202110399719.7 2021-04-14

Publications (1)

Publication Number Publication Date
WO2022218068A1 true WO2022218068A1 (zh) 2022-10-20

Family

ID=76338413

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/080125 WO2022218068A1 (zh) 2021-04-14 2022-03-10 素材投放方法、装置、设备和介质

Country Status (2)

Country Link
CN (1) CN112989203B (zh)
WO (1) WO2022218068A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118410189A (zh) * 2024-07-02 2024-07-30 上海大汉三通通信股份有限公司 一种多媒体素材管理方法、装置、设备及存储介质

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112989203B (zh) * 2021-04-14 2023-11-24 抖音视界有限公司 素材投放方法、装置、设备和介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180181611A1 (en) * 2016-12-28 2018-06-28 Intel Corporation Methods and apparatus for detecting anomalies in electronic data
CN111144937A (zh) * 2019-12-20 2020-05-12 北京达佳互联信息技术有限公司 广告素材确定方法、装置、设备及存储介质
CN111966915A (zh) * 2019-05-20 2020-11-20 腾讯科技（深圳）有限公司 信息巡检方法、计算机设备及存储介质
CN112989203A (zh) * 2021-04-14 2021-06-18 北京字节跳动网络技术有限公司 素材投放方法、装置、设备和介质

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11748596B2 (en) * 2019-05-23 2023-09-05 International Business Machines Corporation Context based vehicular traffic prediction

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180181611A1 (en) * 2016-12-28 2018-06-28 Intel Corporation Methods and apparatus for detecting anomalies in electronic data
CN111966915A (zh) * 2019-05-20 2020-11-20 腾讯科技（深圳）有限公司 信息巡检方法、计算机设备及存储介质
CN111144937A (zh) * 2019-12-20 2020-05-12 北京达佳互联信息技术有限公司 广告素材确定方法、装置、设备及存储介质
CN112989203A (zh) * 2021-04-14 2021-06-18 北京字节跳动网络技术有限公司 素材投放方法、装置、设备和介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118410189A (zh) * 2024-07-02 2024-07-30 上海大汉三通通信股份有限公司 一种多媒体素材管理方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN112989203B (zh) 2023-11-24
CN112989203A (zh) 2021-06-18

Similar Documents

Publication Publication Date Title
CN109299348B (zh) 一种数据查询方法、装置、电子设备及存储介质
WO2022218068A1 (zh) 素材投放方法、装置、设备和介质
JP2020517004A (ja) パイプ漏れを予測する新規な自律的人工知能システム
US11176508B2 (en) Minimizing compliance risk using machine learning techniques
CN110929799B (zh) 用于检测异常用户的方法、电子设备和计算机可读介质
CN109471783B (zh) 预测任务运行参数的方法和装置
WO2022121801A1 (zh) 信息处理方法、装置和电子设备
CN107392259B (zh) 构建不均衡样本分类模型的方法和装置
CN117131281B (zh) 舆情事件处理方法、装置、电子设备和计算机可读介质
CN110766184A (zh) 订单量预测方法和装置
CN112927050A (zh) 待推荐金融产品确定方法、装置、电子设备及存储介质
CN110866625A (zh) 促销指标信息生成方法和装置
CN111949678A (zh) 跨时间窗口的非累加指标的处理方法和装置
CN117236653A (zh) 基于业务量预测的车辆调度方法、装置、电子设备
CN117035842A (zh) 模型训练方法、业务量预测方法、装置、设备和介质
CN113723712B (zh) 风电功率预测方法、系统、设备及介质
CN111915115A (zh) 执行策略设置方法和装置
CN112860999B (zh) 信息推荐方法、装置、设备和存储介质
CN110837907A (zh) 一种预测波次订单量的方法和装置
CN113362097B (zh) 一种用户确定方法和装置
CN113823368B (zh) 资源配置方法和装置
CN117235535B (zh) 异常供应端断电方法、装置、电子设备和介质
CN118245341B (zh) 业务模型的切换方法、装置、电子设备和计算机可读介质
CN117391763B (zh) 申请信息趋势确定方法、装置、电子设备和存储介质
CN111738536B (zh) 设备操作方法、装置、电子设备和计算机可读介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22787298

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 130224)

122 Ep: pct application non-entry in european phase

Ref document number: 22787298

Country of ref document: EP

Kind code of ref document: A1