WO2024251023A1 - 投放结果的确定方法、装置、可读介质和电子设备 - Google Patents

投放结果的确定方法、装置、可读介质和电子设备 Download PDF

Info

Publication number
WO2024251023A1
WO2024251023A1 PCT/CN2024/096290 CN2024096290W WO2024251023A1 WO 2024251023 A1 WO2024251023 A1 WO 2024251023A1 CN 2024096290 W CN2024096290 W CN 2024096290W WO 2024251023 A1 WO2024251023 A1 WO 2024251023A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature
features
target
object feature
candidate object
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2024/096290
Other languages
English (en)
French (fr)
Inventor
解至煊
孙斌
郭勤振
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Youzhuju Network Technology Co Ltd
Original Assignee
Beijing Youzhuju Network Technology Co Ltd
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 Beijing Youzhuju Network Technology Co Ltd filed Critical Beijing Youzhuju Network Technology Co Ltd
Publication of WO2024251023A1 publication Critical patent/WO2024251023A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0269—Targeted advertisements based on user profile or attribute
    • G06Q30/0271—Personalized advertisement
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0242—Determining effectiveness of advertisements
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0277—Online advertisement

Definitions

  • the present disclosure relates to the field of electronic information technology, and in particular, to a method, device, readable medium and electronic device for determining a delivery result.
  • advertisements can be placed in different positions on the user's browsing interface. Whether an advertisement can achieve a good delivery effect affects whether the advertisement conversion rate can be improved and the brand reputation of the advertiser. Therefore, the results of advertising delivery are of great significance to the precise delivery of advertisements.
  • the present disclosure provides a method for determining a delivery result, the method comprising:
  • the target object features are input into a prediction model to obtain a target delivery result corresponding to the target object output by the prediction model; the target delivery result is used to characterize user acceptance of the target object.
  • the present disclosure provides a device for determining a delivery result, the device comprising:
  • a first acquisition module is configured to acquire a plurality of object features corresponding to the target object
  • a first determining module is configured to determine a target object feature from the plurality of object features according to feature types corresponding to the plurality of object features when a total number of first features of the plurality of object features is greater than or equal to a preset number threshold;
  • the prediction module is configured to input the target object features into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model; the target delivery result is used to characterize the user acceptance of the target object.
  • the present disclosure provides a computer-readable medium having a computer program and/or instructions stored thereon, which, when executed by a processing device, implements the steps of the method for determining the delivery result described in the first aspect above.
  • an electronic device including:
  • the processing device is configured to execute the computer program and/or instructions in the storage device to implement the steps of the method for determining the delivery result described in the first aspect above.
  • the present disclosure provides a computer program product, comprising instructions, which, when executed by a processor, enable the processor to implement the steps of the method for determining the delivery result described in the first aspect above.
  • the present disclosure provides a computer program, including program code, which, when executed by a processor, leads to the implementation of the steps of the method for determining the delivery result described in the first aspect above.
  • FIG1 is a flow chart of a method for determining a delivery result according to an exemplary embodiment
  • FIG2 is a flow chart of another method for determining delivery results according to an exemplary embodiment
  • FIG3 is a flow chart of another method for determining delivery results according to an exemplary embodiment
  • FIG4 is a flow chart of another method for determining delivery results according to an exemplary embodiment
  • FIG5 is a flow chart of another method for determining delivery results according to an exemplary embodiment
  • FIG6 is a flow chart of another method for determining delivery results according to an exemplary embodiment
  • FIG7 is a flow chart of a method for determining a delivery result according to an exemplary embodiment
  • FIG8 is a schematic diagram of a process of predicting a target delivery result according to an exemplary embodiment
  • FIG9 is a block diagram of a device for determining a delivery result according to an exemplary embodiment
  • FIG10 is a block diagram of another device for determining delivery results according to an exemplary embodiment
  • FIG11 is a block diagram of another device for determining delivery results according to an exemplary embodiment
  • FIG12 is a block diagram of another device for determining delivery results according to an exemplary embodiment
  • Fig. 13 is a block diagram of an electronic device according to an exemplary embodiment.
  • a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information.
  • the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.
  • the prompt information in response to receiving the active request from the user, may be sent to the user in the form of a pop-up window, in which the prompt information may be in the form of text.
  • the pop-up window may also carry a selection control for the user to choose "agree” or “disagree” to provide personal information to the electronic device.
  • the results of advertising are mainly determined by sampling user data after the advertisements are delivered in batches. This method requires the advertisements to be delivered first. When the delivery volume is small, the error is large, the evaluation accuracy is low, and there is a certain lag, which cannot accurately reflect the delivery effect of the advertisements.
  • the present disclosure proposes improved data delivery related processing, and in particular proposes improved delivery result determination, which will be described in detail with reference to the accompanying drawings.
  • the following description is mainly for the delivery of advertisements, commodities, etc., it should be noted that this is only an example and not a limitation.
  • the solution of the embodiment of the present disclosure can be equally/equivalently applied to data delivery/presentation of other applications, etc., which is essentially an improved data analysis and processing for more accurate or appropriate screening/providing of data.
  • FIG. 1 is a flow chart of a method for determining a delivery result according to an exemplary embodiment. As shown in FIG. 1 , the method may include the following steps:
  • step S101 a plurality of object features corresponding to a target object are obtained.
  • the target object may be, for example, an object to be promoted or released, such as an advertisement or a product.
  • an advertisement such as an advertisement or a product.
  • the effect of the advertisement can be predicted, that is, the user's preference and acceptance can be predicted, so as to facilitate more accurate advertisement delivery (for example, where in the browsing interface the advertisement is to be delivered, at what time period, in what area, etc.).
  • the user's preference and acceptance rate for the advertisement can be increased, and Can maximize advertising conversion rate.
  • the object characteristics may include, but are not limited to, one or more of advertiser characteristics (such as advertiser name, advertiser category, advertiser source), advertising plan characteristics (such as the number of planned creatives, advertisement creation duration, comment permission switch, and the industry to which the advertisement belongs), advertising material characteristics (such as title, copy, picture, video, and landing page), advertiser qualification characteristics (such as qualification authorization information and advertisement validity period), advertising delivery characteristics (such as advertising delivery budget, advertising delivery area, delivery time period, user acceptance (such as the number of likes, shares, dislikes, reports, and comments)), etc.
  • advertiser characteristics such as advertiser name, advertiser category, advertiser source
  • advertising plan characteristics such as the number of planned creatives, advertisement creation duration, comment permission switch, and the industry to which the advertisement belongs
  • advertising material characteristics such as title, copy, picture, video, and landing page
  • advertiser qualification characteristics such as qualification authorization information and advertisement validity period
  • advertising delivery characteristics such as advertising delivery budget, advertising delivery area, delivery time period, user acceptance (such as the number of likes, shares, dislikes
  • step S102 when the total number of first features of the plurality of object features is greater than or equal to a preset quantity threshold, a target object feature is determined from the plurality of object features according to feature types corresponding to the plurality of object features.
  • a target object often corresponds to more object features, and the more object features there are, the longer the calculation time of the subsequent model will be.
  • a maximum feature quantity limit that the model can handle that is, a preset quantity threshold, can be set.
  • the object features can be deleted to reduce the number of object features and obtain the target object features.
  • targeted screening is performed according to the feature types corresponding to the multiple object features, and different screening methods are used for different types of object features to obtain the target object features.
  • the feature types of object features may include discrete features (such as advertiser category, comment permission switch, industry to which the advertisement belongs, etc.), numerical features (such as advertising budget, number of likes, etc.), image features (such as pictures), and text features (such as titles, copy, etc.).
  • discrete features and numerical features since such features can often directly reflect the properties of the target object itself, the relevance of the object features to the target object can be analyzed in advance for such features (for example, based on Experience may be obtained, or it may be obtained through input model analysis, which is not specifically limited in the present disclosure), and the priority between the features is set according to the correlation obtained by the analysis, and the higher the priority, the higher the correlation between the object feature and the target object.
  • the features with lower priorities may be deleted first.
  • features with a priority lower than a specific priority threshold may be deleted.
  • features may be deleted from low to high priority until a specific number of features are retained, and the specific number may be less than or equal to the total number of first features, or any other pre-fixed number.
  • the features may be deduplicated and/or sampled to reduce the number of features.
  • the multiple object features can be directly used as the target object features.
  • step S103 the target object feature is input into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model.
  • the target delivery result is used to characterize the user acceptance of the target object, for example, the number of likes, shares, dislikes, reports, comments, etc.
  • the prediction model may be, for example but not limited to, a Transformer structure, a residual structure, or a graph network or other structures.
  • this embodiment can be used to predict the effect of advertising before the advertisement is officially released, and can also be used to predict the subsequent advertising effect after the advertisement is officially released.
  • the user acceptance in the above object features can be a preset value (for example, all 0).
  • the user acceptance in the above object features can be the real user acceptance in the historical time period obtained after the actual release.
  • the prediction model generally outputs an output vector, so it can also be mapped into a single value through the network layer, that is, to obtain the target delivery result.
  • the target delivery result of the target object can be predicted according to the target object features corresponding to the target object.
  • the delivery target of the target object is determined in advance according to the target delivery result, so as to facilitate more accurate delivery of the target object.
  • the object features can be screened according to the feature type of the object features to obtain the target object features, and the target object features can be input into the prediction model, which effectively improves the processing efficiency of the model.
  • the prediction model may be pre-generated in the following manner:
  • Step A obtaining a training sample set, wherein the training sample set includes sample object features corresponding to the sample object and actual delivery results corresponding to the sample object.
  • the actual delivery result is used to characterize the user acceptance of the sample object after delivery. That is, the actual delivery result is the user acceptance obtained after the sample object is actually delivered, which can be used to further modify the model and improve the accuracy of model prediction.
  • Step B iteratively updating the preset model according to the preset loss function, the sample object characteristics and the actual delivery result to obtain the prediction model.
  • the sample object features can be input into the preset model to obtain the sample delivery result corresponding to the sample object output by the prediction model, the sample delivery result is compared with the actual delivery result, the difference is calculated through the preset loss function, and the model parameters are updated using gradient feedback.
  • the model iteration step can be executed cyclically until it is determined that the updated preset model meets the preset stop iteration condition according to the preset loss function, and the updated preset model is used as the prediction model.
  • the model iteration steps include:
  • the sample delivery result is used to characterize the user acceptance of the sample object.
  • the preset model can be, for example but not limited to, a Transformer structure, a residual structure Or other structures such as graph networks.
  • the target loss value is used to characterize the degree of difference between the sample delivery result and the actual delivery result.
  • the stop iteration condition can be at least one of various appropriate conditions, such as the loss value has become stable, such as the loss is less than a specific loss threshold, such as the loss value change in a specific number of consecutive iterations is less than a specific change threshold, and a specific number of iterations has been performed, etc. These are only examples and not limitations and will not be described in detail.
  • the iteration can be stopped when any iteration condition is met.
  • step S104 the target object feature is dimensionally transformed to transform vector dimensions of different target object features into the same vector dimension.
  • the target object features can be first dimensionally converted to map all target object features to features of the same dimension. For example, all target object features can be mapped to 768 dimensions.
  • the dimensionality conversion methods are different. For example, for discrete features and numerical features, dimensionality conversion can be performed through methods such as AutoDis (Automatic Discretization) encoding, bucket mapping, and numerical padding to map the features to a fixed dimension.
  • dimension conversion can be performed by methods such as image vision models (such as Resnet, Vit (Vision Transformer) models) to map features to fixed dimensions. For example, taking the image features including video as an example, the video can be split into N image frames first. Then, the image vision model is used to map to N ⁇ 768 dimensions.
  • image vision models such as Resnet, Vit (Vision Transformer) models
  • sentence processing can be performed so that the text length of the text after sentence processing is less than the preset length threshold.
  • the text after sentence processing can be dimensionally converted by methods such as natural language models (such as Bert, MLXR, etc. models) to map features to fixed dimensions. For example, taking texts with longer text lengths such as copy and landing page text as an example, it can be segmented into K short texts with a length less than 128 by sentence processing. Then, a natural language model can be used to map it to K ⁇ L ⁇ 768 dimensions.
  • the above method divides the long text into sentences and inputs it into the natural language model, that is, it is input sentence by sentence, and the corresponding output is also output sentence by sentence. At this time, they can be compressed into features of 1*768 dimensions, that is, they can be pooled to K ⁇ 768 dimensions.
  • dimensionality conversion can be performed directly through methods such as natural language models to map the features to fixed dimensions. For example, texts with a text length of less than 128, such as titles and advertiser names, can be mapped to L ⁇ 768 dimensions using a natural language model, and then pooled to 768 dimensions.
  • the target object feature is input into the prediction model, and the target delivery result corresponding to the target object output by the prediction model includes:
  • the target object features after dimension conversion are input into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model.
  • FIG3 is a flow chart of another method for determining a delivery result according to an exemplary embodiment. As shown in FIG3 , the method may further include the following steps:
  • step S105 according to the feature type of the target object feature, the target feature type code corresponding to the target object feature is determined.
  • feature type codes corresponding to different feature types may be pre-set to obtain a coding correspondence relationship, and the feature type code corresponding to the feature type may be obtained through the coding correspondence relationship according to the feature type of the target object feature as the target feature type code.
  • the target object feature is input into the prediction model, and the target delivery result corresponding to the target object output by the prediction model includes:
  • the target object feature is added to the target feature type code and then input into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model.
  • the embodiments of FIG. 2 and FIG. 3 can also be combined for execution. That is, the target object feature can be first dimensionally converted to convert the vector dimensions of different target object features into the same vector dimension. And according to the feature type of the target object feature, the target feature type code corresponding to the target object feature is determined. Then, the target object feature after dimension conversion is added to the target feature type code and input into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model.
  • step S102 in which, when the total number of first features of the plurality of object features is greater than or equal to a preset quantity threshold, determining a target object feature from the plurality of object features according to feature types corresponding to the plurality of object features.
  • step S102 may include the following steps:
  • step S1021 the object feature whose feature type is the first specified type is used as the first candidate object feature.
  • the first specified type may be, for example, an image feature and a text feature.
  • step S1022 second candidate object features are determined from the first candidate object features.
  • determining the second candidate object feature from the first candidate object feature may include: performing a deduplication process on the first candidate object feature, and performing the deduplication process on the first candidate object feature.
  • the object feature is selected as the second candidate object feature.
  • the advertiser's profile picture, the first frame of the video, or the consecutive frames in the video may be the same. Therefore, for image features, all image features can be deduplicated through hash coding.
  • the purpose of all text expressions is to increase the user's willingness to convert. Therefore, for text features, the text can be filtered first, that is, fixed-format fields and invalid symbols can be removed.
  • determining the second candidate object feature from the first candidate object feature may include: first, determining the similarity between every two first candidate object features. Then, when the similarity is greater than or equal to a preset similarity threshold, any one of the two first candidate object features corresponding to the similarity is used as a fourth candidate object feature. And the fourth candidate object feature and the first candidate object feature whose similarity is less than the preset similarity threshold are used as the second candidate object feature.
  • the similarity between each two first candidate object features can be calculated by cosine similarity. If the similarity is greater than or equal to a preset similarity threshold, it indicates that the similarity between the two first candidate object features is high. In this case, one of them can be arbitrarily retained as the fourth candidate object feature.
  • the fourth candidate object feature and the first candidate object feature whose similarity is less than the preset similarity threshold are used as the second candidate object feature.
  • determining the second candidate object feature from the first candidate object feature may include: first, taking the first candidate object feature whose feature type is the second specified type as the fifth candidate object feature. Then, performing sampling processing on the fifth candidate object feature. And taking the fifth candidate object feature and the sixth candidate object feature after the sampling processing as the second candidate object feature.
  • the sixth candidate object feature is an object feature in the first candidate object feature other than the fifth candidate object feature.
  • the second specified type may be, for example, a video frame image feature and a long text feature (the image feature may include a single picture feature and a video frame image feature, and the text feature may include a long text feature).
  • the sampling process of the fifth candidate object feature includes: determining the total number of second features of the first candidate object feature and the total number of third features of the fifth candidate object feature. And according to the preset quantity threshold, the total number of second features and the total number of third features, the target feature quantity is determined.
  • the third candidate object feature (wherein the third candidate object feature is an object feature other than the first candidate object feature among multiple object features) can be, for example, a discrete feature and a numerical feature.
  • the number of the third candidate object features can be calculated according to the total number of first features and the total number of second features, that is, the number of discrete features and numerical features is determined.
  • the number of features other than the fifth candidate object feature in the first candidate object feature is determined according to the total number of second features and the total number of third features, that is, the number of single picture features and short text features is determined.
  • L is the number of target features, and represents the number of samples required for each fifth candidate object feature.
  • the fifth candidate object feature is sampled according to the number of target features.
  • the fifth candidate object feature can be uniformly sampled according to the number of target features.
  • the corresponding sampling method can also be set according to different target feature numbers. For example, if L is less than 1, it means that each video or long text does not need to be retained. At this time, the video frame image and long text corresponding to the video can be deleted.
  • the first sampling method can be used.
  • the first sampling method can be used for the video frame image feature.
  • the image at the first position in the video frame can be sampled.
  • the first paragraph of the long text can be sampled.
  • L is greater than or equal to 2 and less than 3
  • the first and last position sampling method can be used for sampling.
  • L is greater than or equal to 3
  • the linear uniform sampling method can be used for sampling.
  • deduplication processing can be performed first by the method in the first exemplary embodiment. If the total number of remaining object features after deduplication processing is less than the preset number threshold, the object features after deduplication processing can be used as the second candidate object features. If the total number of remaining object features after deduplication processing is still greater than or equal to the preset number threshold, the similarity can be calculated by the second exemplary embodiment, and one of the object features with higher similarity can be deleted.
  • the deleted object features can be used as the second candidate object features. If the total number of remaining object features after deletion is still greater than or equal to the preset number threshold, the sampling processing can be continued by the third exemplary embodiment, and the object features after sampling processing can be used as the second candidate object features. It can also be any other combination, such as combining the first exemplary embodiment with the third exemplary embodiment, combining the first exemplary embodiment with the second exemplary embodiment, combining the second exemplary embodiment with the third exemplary embodiment, and so on, which will not be listed one by one here.
  • step S1023 the target object feature is determined according to the second candidate object feature and the third candidate object feature.
  • the third candidate object feature is an object feature among the multiple object features except the first candidate object feature.
  • the second candidate object features and the third candidate object features can be directly used as the target object features.
  • the total number of fourth features of the second candidate object features and the third candidate object features can also be determined first, and when the total number of fourth features is less than a preset number threshold, the second candidate object features can be used as the target object features.
  • the feature and the third candidate object feature are used as the target object feature.
  • step S102 may include the following steps:
  • step S1024 the object feature whose feature type is the third specified type is used as the seventh candidate object feature.
  • the third specified type may be, for example but not limited to, a discrete feature and a numerical feature.
  • step S1025 the preset priority corresponding to the seventh candidate object feature is obtained.
  • the preset priority may be set after analyzing the relevance between the feature and the target object in advance (for example, it may be obtained based on experience or through input model analysis).
  • step S1026 an eighth candidate object feature is determined from the seventh candidate object feature according to the preset priority.
  • the eighth candidate object feature can be determined from the seventh candidate object feature according to the preset priority and the total number of first features, so that the total number of features after screening does not exceed a preset number threshold.
  • step S1027 the target object feature is determined according to the eighth candidate object feature and the ninth candidate object feature.
  • the ninth candidate object feature is an object feature other than the seventh candidate object feature among the plurality of object features.
  • the eighth candidate object feature and the ninth candidate object feature can be directly used as the target object feature.
  • the total number of the fifth features of the eighth candidate object feature and the ninth candidate object feature can be first determined, and when the total number of the fifth features is less than a preset number threshold, the eighth candidate object feature and the ninth candidate object feature can be used as the target object feature.
  • the embodiments of FIG. 4 and FIG. 5 can be implemented separately or in combination, and the implementation order can also be freely combined, and the present disclosure does not specifically limit this.
  • the fourth feature of FIG. 5 can also be used. The method in the embodiment further reduces the number of features so that the total number of features is less than a preset number threshold.
  • FIG6 is a flow chart of another method for determining a delivery result according to an exemplary embodiment. As shown in FIG6 , the method may further include the following steps:
  • step S106 the historical delivery results corresponding to the target object are obtained.
  • the historical delivery result is used to characterize the user acceptance of the target object at the historical moment;
  • the historical delivery result is the delivery result corresponding to the target object output by the prediction model obtained by inputting the historical object features corresponding to the target object at the historical moment into the prediction model.
  • the embodiment of FIG1 can also be used in the target object delivery process. After the target object is delivered, the values of user acceptance and the like will increase from 0. At this time, in order to more accurately determine the target delivery result of the target object, the target delivery result can be corrected.
  • the triggering moment of the correction can also be customized.
  • the target delivery result of the target object can be re-predicted every preset time after delivery, that is, the method in the embodiment of Figure 1 is executed.
  • the target delivery result of the target object can be re-predicted.
  • the target delivery result of the target object is re-predicted.
  • the latest object features of the target object may be obtained to more accurately reflect the current features of the target object.
  • step S107 the target delivery result is corrected according to the historical delivery result.
  • the Momentum algorithm can be used to make corrections based on historical delivery results and target delivery results.
  • Corresponding weights can also be set for the historical delivery results and the target delivery results, and the historical delivery results and the target delivery results are multiplied by the corresponding weights and then added together to obtain the corrected results.
  • step S108 the corrected target delivery result is used as the new target delivery result of the target object. Post the results.
  • the predicted target delivery results can be corrected in real time during the delivery of the target object, thereby improving the accuracy of the delivery result prediction after the delivery.
  • the target delivery result of the target object can be predicted according to the target object features corresponding to the target object.
  • the delivery target of the target object is determined in advance according to the target delivery result, so as to facilitate more accurate delivery of the target object.
  • the object features can be screened according to the feature type of the object features to obtain the target object features, and the target object features can be input into the prediction model, which effectively improves the processing efficiency of the model.
  • FIG. 7 is a flow chart of a method for determining a delivery result according to an exemplary embodiment. As shown in FIG. 7 , the method may include the following steps:
  • step S201 a plurality of object features corresponding to the target object are obtained.
  • step S202 when the total number of first features of the plurality of object features is less than a preset number threshold, the plurality of object features are used as target object features.
  • step S203 when the total number of first features of the plurality of object features is greater than or equal to a preset number threshold, the object feature whose feature type is the first specified type is used as a first candidate object feature.
  • the first specified type may be, for example, image features and text features. Since image features and text features are prone to have repeated features or features with the same meaning, image features and text features may be screened preferentially in this embodiment.
  • step S204 the first candidate object feature is deduplicated, and the total number of sixth features of the first candidate object feature and the third candidate object feature after deduplication is determined.
  • the third candidate object feature is an object feature among the multiple object features except the first candidate object feature.
  • step S205 when the total number of sixth features is less than the preset number threshold, the duplicate features are removed.
  • the processed first candidate object features and the third candidate object features are used as target object features.
  • step S206 when the total number of sixth features is greater than or equal to a preset number threshold, the similarity between every two first candidate object features after deduplication processing is determined.
  • step S207 when the similarity is greater than or equal to the preset similarity threshold, any one of the two deduplicated first candidate object features corresponding to the similarity is used as the fourth candidate object feature, and the fourth candidate object feature and the deduplicated first candidate object feature whose similarity is less than the preset similarity threshold are used as the tenth candidate object feature.
  • the first candidate object can be preferentially deduplicated to retain the object features to the greatest extent. If the remaining features after deduplication still exceed the preset number threshold, the similarity can be calculated to further reduce the number of features.
  • step S208 the total number of the tenth candidate object feature and the seventh feature of the third candidate object feature is determined.
  • step S209 when the total number of the seventh features is less than the preset number threshold, the tenth candidate object feature and the third candidate object feature are used as target object features.
  • step S210 when the total number of the seventh features is greater than or equal to the preset number threshold, the tenth candidate object feature whose feature type is the second specified type is used as the eleventh candidate object feature.
  • the second specified type may be, for example, a video frame image feature and a long text feature.
  • step S211 the eleventh candidate object feature is sampled.
  • the sampling process for the eleventh candidate object feature may refer to the method for sampling the fifth candidate object feature in the above embodiment, which will not be described in detail here.
  • step S212 the total number of the eighth features of the eleventh candidate object feature and the third candidate object feature after the sampling process is determined.
  • step S213 when the total number of the eighth features is less than the preset number threshold, the tenth candidate object feature and the third candidate object feature after sampling are used as target object features.
  • step S214 when the total number of the eighth features is greater than or equal to the preset number threshold, Obtain a preset priority corresponding to the third candidate object feature.
  • the preset priority may be set after analyzing the relevance between the feature and the target object in advance (for example, it may be obtained based on experience or through input model analysis).
  • step S215 a twelfth candidate object feature is determined from the third candidate object feature according to the preset priority.
  • the number of remaining features can be reduced to less than the preset number threshold according to the preset priority and the preset number threshold.
  • step S216 the tenth candidate object feature and the twelfth candidate object feature after the sampling process are used as target object features.
  • step S217 the target object feature is dimensionally converted to convert vector dimensions of different target object features into the same vector dimension.
  • step S2128 the target feature type code corresponding to the target object feature is determined according to the feature type of the target object feature.
  • step S219 the target object feature after dimension conversion is added to the target feature type code and then input into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model.
  • the object features can be preprocessed, such as splitting the video into video frame images and splitting the long text into short text. Then perform feature screening, and after screening, map each feature to the same dimension. After that, add each feature to the corresponding feature type code. Finally, input the added features into the prediction model to obtain the prediction result output by the prediction model, that is, the target delivery result.
  • the order in which the features are input into the prediction model can be arbitrary.
  • multiple output vectors may be output. In this embodiment, we only need to pay attention to the output vectors that reflect the delivery results, such as the number of likes and the number of comments. In this way, users can input multiple object feature types of different businesses into the prediction model according to their needs, which improves the diversity of application scenarios. Easy to expand and migrate.
  • the target delivery result of the target object can be predicted according to the target object features corresponding to the target object.
  • the delivery target of the target object is determined in advance according to the target delivery result, so as to facilitate more accurate delivery of the target object.
  • the object features can be screened according to the feature type of the object features to obtain the target object features, and the target object features can be input into the prediction model, which effectively improves the processing efficiency of the model.
  • the determination device provided by the embodiment of the present disclosure can execute the method for determining the delivery results provided by any embodiment of the present disclosure.
  • the embodiment of the present disclosure can divide the device into functional units according to the above method example. For example, each functional module/unit can be divided corresponding to each function, or two or more functions can be integrated into one processing module. It is worth noting that the various modules/units included in the above-mentioned device are only divided according to functional logic, but are not limited to the division described in the text.
  • each module/unit can be implemented in various appropriate ways, such as hardware, firmware, or any appropriate combination.
  • FIG. 9 is a block diagram of a device for determining a delivery result according to an exemplary embodiment. As shown in FIG. 9 , the device 300 includes:
  • a first acquisition module 301 is configured to acquire a plurality of object features corresponding to a target object
  • the first determining module 302 is configured to, when the total number of first features of the plurality of object features is greater than or equal to a preset number threshold, select a first feature from the plurality of object features according to the feature types corresponding to the plurality of object features. Determine the target object characteristics from the characteristics;
  • the prediction module 303 is configured to input the target object feature into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model; the target delivery result is used to characterize the user acceptance of the target object.
  • the first determination module 302 is configured to use the object feature whose feature type is the first specified type as the first candidate object feature; determine the second candidate object feature from the first candidate object feature; determine the target object feature based on the second candidate object feature and the third candidate object feature; wherein the third candidate object feature is an object feature among multiple object features other than the first candidate object feature.
  • the first determining module 302 is configured to perform deduplication processing on the first candidate object feature; and use the deduplication-processed first candidate object feature as the second candidate object feature.
  • the first determination module 302 is configured to determine the similarity between every two first candidate object features; when the similarity is greater than or equal to a preset similarity threshold, any one of the two first candidate object features corresponding to the similarity is used as a fourth candidate object feature; the fourth candidate object feature and the first candidate object feature whose similarity is less than the preset similarity threshold are used as the second candidate object feature.
  • the first determination module 302 is configured to use the first candidate object feature whose feature type is the second specified type as the fifth candidate object feature; perform sampling processing on the fifth candidate object feature; use the fifth candidate object feature and the sixth candidate object feature after the sampling processing as the second candidate object feature; wherein the sixth candidate object feature is the object feature in the first candidate object feature except the fifth candidate object feature.
  • the first determination module 302 is configured to determine the total number of second features of the first candidate object feature and the total number of third features of the fifth candidate object feature; determine the target feature quantity based on the preset quantity threshold, the total number of second features and the total number of third features; and perform sampling processing on the fifth candidate object feature based on the target feature quantity.
  • the first determination module 302 is configured to use the object feature whose feature type is the third specified type as the seventh candidate object feature; obtain a preset priority corresponding to the seventh candidate object feature; determine an eighth candidate object feature from the seventh candidate object feature according to the preset priority; determine the target object feature according to the eighth candidate object feature and the ninth candidate object feature; wherein the ninth candidate object feature is an object feature among multiple object features other than the seventh candidate object feature.
  • the prediction model is pre-generated by:
  • the training sample set including sample object features corresponding to the sample object and actual delivery results corresponding to the sample object; the actual delivery results are used to characterize user acceptance of the sample object after delivery;
  • the preset model is iteratively updated according to the preset loss function, the sample object characteristics and the actual delivery result to obtain the prediction model.
  • the iterative updating of the preset model according to the preset loss function, the sample object characteristics and the actual delivery result to obtain the prediction model includes:
  • the model iteration step is executed cyclically until it is determined that the updated preset model satisfies the preset stop iteration condition according to the preset loss function, and the updated preset model is used as the prediction model;
  • the model iteration steps include:
  • a target loss value between the sample delivery result and the actual delivery result corresponding to the sample object is determined; the target loss value is used to characterize the degree of difference between the sample delivery result and the actual delivery result;
  • the parameters of the preset model are updated according to the target loss value to obtain a new preset model.
  • the device 300 further includes:
  • the second acquisition module 304 is configured to acquire the historical delivery result corresponding to the target object, the historical delivery result is used to characterize the user acceptance of the target object at a historical moment; the historical delivery result is the delivery result corresponding to the target object output by the prediction model obtained by inputting the historical object features corresponding to the target object into the prediction model at a historical moment;
  • the correction module 305 is configured to correct the target delivery result according to the historical delivery result
  • the second determination module 306 is configured to use the corrected target delivery result as a new target delivery result for the target object.
  • the first determination module 302 is further configured to use multiple object features as the target object features when the total number of the first features is less than the preset number threshold.
  • the device 300 further includes:
  • the conversion module 307 is configured to perform dimension conversion on the target object feature to convert vector dimensions of different target object features into the same vector dimension;
  • the prediction module 303 is configured to input the target object features after dimension conversion into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model.
  • the device 300 further includes:
  • the third determination module 308 is configured to determine the target feature type code corresponding to the target object feature according to the feature type of the target object feature;
  • the prediction module 303 is configured to add the target object feature and the target feature type code and input the result into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model.
  • the target delivery result of the target object can be predicted according to the target object features corresponding to the target object.
  • the delivery target of the target object is determined in advance according to the target delivery result, so as to facilitate more accurate delivery to the target object.
  • the object features can be screened according to the feature type of the object features to obtain the target object features, and the target object features can be used as the input model.
  • the target object features are input into the prediction model, which effectively improves the processing efficiency of the model.
  • the electronic device may include hardware structures and/or software modules corresponding to the execution of each function.
  • the present disclosure can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present disclosure.
  • FIG. 13 shows a schematic diagram of the structure of an electronic device (e.g., a terminal device or a server) 400 suitable for implementing the embodiments of the present disclosure.
  • the terminal device in the embodiments of the present disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (e.g., vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc.
  • the electronic device shown in FIG. 13 is only an example and should not impose any limitations on the functions and scope of use of the embodiments of the present disclosure.
  • the electronic device 400 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage device 408 into a random access memory (RAM) 403.
  • a processing device e.g., a central processing unit, a graphics processing unit, etc.
  • RAM random access memory
  • various programs and data required for the operation of the electronic device 400 are also stored.
  • the processing device 401, the ROM 402, and the RAM 403 are connected to each other via a bus 404.
  • An input/output (I/O) interface 405 is also connected to the bus 404.
  • the following devices may be connected to the I/O interface 405: input devices 406 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 407 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 408 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 409.
  • the communication device 409 may allow the electronic device 400 to communicate wirelessly or wired with other devices to exchange data.
  • FIG. 13 shows an electronic device 400 with various devices, it should be understood that it is not required to implement or have all the devices shown. More or fewer devices may be implemented or have alternatively.
  • an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart.
  • the computer program can be downloaded and installed from a network through a communication device 409, or installed from a storage device 408, or installed from a ROM 402.
  • the processing device 401 the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.
  • the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above.
  • Computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
  • a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or device.
  • a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein the carrier wave The computer-readable program code is loaded.
  • This propagated data signal can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above.
  • the computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or device.
  • the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.
  • the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network).
  • HTTP HyperText Transfer Protocol
  • Examples of communication networks include a local area network ("LAN”), a wide area network ("WAN”), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.
  • the computer-readable medium may be included in the electronic device, or may exist independently without being installed in 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, any step or combination of the method provided in any embodiment of the present disclosure is executed.
  • Computer program code for performing operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including, but not limited to, object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages.
  • the program code may execute entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially 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 type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to the user's computer through a remote computer or server.
  • LAN local area network
  • WAN wide area network
  • an Internet service provider for example, AT&T, MCI, Sprint, EarthLink, MSN, GTE, etc.
  • each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function.
  • the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.
  • each square box in the block diagram and/or flow chart, and the combination of the square boxes in the block diagram and/or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
  • modules involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a module does not, in some cases, limit the module itself.
  • 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 logic devices (CPLDs), and the like.
  • FPGAs field programmable gate arrays
  • ASICs application specific integrated circuits
  • ASSPs application specific standard products
  • SOCs systems on chips
  • CPLDs complex programmable logic devices
  • a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device.
  • a machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium.
  • a machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include electrical connections based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable and removable hard disk, or a computer program product. Programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
  • Example 1 provides a method for determining a delivery result, the method comprising: obtaining multiple object features corresponding to a target object; determining a target object feature from the multiple object features according to the feature types corresponding to the multiple object features when the total number of first features of the multiple object features is greater than or equal to a preset quantity threshold; inputting the target object feature into a prediction model to obtain a target delivery result corresponding to the target object output by the prediction model; the target delivery result is used to characterize user acceptance of the target object.
  • Example 2 provides the method of Example 1, wherein determining a target object feature from the multiple object features according to the feature types corresponding to the multiple object features comprises: taking an object feature whose feature type is a first specified type as a first candidate object feature; determining a second candidate object feature from the first candidate object feature; determining the target object feature based on the second candidate object feature and a third candidate object feature; wherein the third candidate object feature is an object feature among the multiple object features other than the first candidate object feature.
  • Example 3 provides the method of Example 2, wherein determining the second candidate object feature from the first candidate object feature includes: performing deduplication processing on the first candidate object feature; and using the deduplicated first candidate object feature as the second candidate object feature.
  • Example 4 provides the method of Example 2, wherein determining the second candidate object feature from the first candidate object feature includes: determining the similarity between every two first candidate object features; when the similarity is greater than or equal to a preset similarity threshold, taking any one of the two first candidate object features corresponding to the similarity as a fourth candidate object feature; and taking the fourth candidate object feature and the first candidate object feature whose similarity is less than the preset similarity threshold as the second candidate object feature.
  • Example 5 provides the method of Example 2, wherein Determining the second candidate object feature from the first candidate object feature includes: using the first candidate object feature whose feature type is the second specified type as the fifth candidate object feature; performing sampling processing on the fifth candidate object feature; using the fifth candidate object feature and the sixth candidate object feature after the sampling processing as the second candidate object feature; wherein the sixth candidate object feature is the object feature of the first candidate object feature other than the fifth candidate object feature.
  • Example 6 provides the method of Example 5, wherein the sampling processing of the fifth candidate object feature includes: determining the total number of second features of the first candidate object feature and the total number of third features of the fifth candidate object feature; determining the target feature quantity based on the preset quantity threshold, the total number of the second features and the total number of the third features; and sampling processing of the fifth candidate object feature based on the target feature quantity.
  • Example 7 provides the method of Example 1, wherein determining the target object feature from the multiple object features according to the feature types corresponding to the multiple object features includes: taking the object feature whose feature type is a third specified type as the seventh candidate object feature; obtaining a preset priority corresponding to the seventh candidate object feature; determining an eighth candidate object feature from the seventh candidate object feature according to the preset priority; determining the target object feature according to the eighth candidate object feature and the ninth candidate object feature; wherein the ninth candidate object feature is an object feature among the multiple object features other than the seventh candidate object feature.
  • Example 8 provides the method of Example 1, wherein the prediction model is pre-generated in the following manner: obtaining a training sample set, wherein the training sample set includes sample object features corresponding to sample objects and actual delivery results corresponding to the sample objects; the actual delivery results are used to characterize user acceptance after the sample objects are delivered; and the preset model is iteratively updated according to a preset loss function, the sample object features, and the actual delivery results to obtain the prediction model.
  • Example 9 provides the method of Example 8, wherein the preset model is analyzed based on the preset loss function, the sample object characteristics, and the actual delivery result. Iterative update to obtain the prediction model includes: looping the model iteration step until it is determined according to the preset loss function that the updated preset model meets the preset stop iteration condition, and using the updated preset model as the prediction model; the model iteration step includes: inputting the sample object feature into the preset model to obtain the sample delivery result corresponding to the sample object output by the preset model, and the sample delivery result is used to characterize the user acceptance of the sample object; according to the preset loss function, determining the target loss value between the sample delivery result corresponding to the sample object and the actual delivery result; the target loss value is used to characterize the degree of difference between the sample delivery result and the actual delivery result; when it is determined according to the target loss value that the preset model does not meet the preset stop iteration condition, updating the parameters of the preset model according to
  • Example 10 provides the method of Example 1, which further includes: obtaining historical delivery results corresponding to the target object, the historical delivery results being used to characterize user acceptance of the target object at a historical moment; the historical delivery results being inputting historical object features corresponding to the target object into the prediction model at a historical moment, and obtaining delivery results corresponding to the target object output by the prediction model; correcting the target delivery results based on the historical delivery results; and using the corrected target delivery results as the new target delivery results for the target object.
  • Example 11 provides the method of Example 1, which further includes: when the total number of the first features is less than the preset number threshold, using the multiple object features as the target object features.
  • Example 12 provides a method of any one of Examples 1 to 11, wherein the method further includes: performing dimension conversion on the target object feature to convert vector dimensions of different target object features into the same vector dimension; the inputting the target object feature into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model includes: inputting the target object feature after dimension conversion into the prediction model to obtain the target delivery result corresponding to the target object output by the prediction model The target delivery result corresponding to the target object is obtained.
  • Example 13 provides the method of any one of Examples 1 to 11, the method further comprising: determining a target feature type code corresponding to the target object feature according to a feature type of the target object feature; inputting the target object feature into a prediction model to obtain a target delivery result corresponding to the target object output by the prediction model comprises: adding the target object feature to the target feature type code and inputting the result into the prediction model to obtain a target delivery result corresponding to the target object output by the prediction model.
  • Example 14 provides a device for determining a delivery result, the device comprising: a first acquisition module, used to acquire multiple object features corresponding to a target object; a first determination module, used to determine a target object feature from the multiple object features according to the feature types corresponding to the multiple object features when the total number of first features of the multiple object features is greater than or equal to a preset quantity threshold; a prediction module, used to input the target object feature into a prediction model to obtain a target delivery result corresponding to the target object output by the prediction model; the target delivery result is used to characterize user acceptance of the target object.
  • Example 15 provides a computer-readable medium having a computer program stored thereon, which implements the steps of the method described in any one of Examples 1 to 13 when executed by a processing device.
  • Example 16 provides an electronic device, comprising: a storage device on which a computer program is stored; and a processing device for executing the computer program in the storage device to implement the steps of the method described in any one of Examples 1 to 13.
  • Example 17 provides a computer program product, comprising instructions, which, when executed by a processor, enable the processor to implement the steps of the method described in any one of Examples 1 to 13.
  • Example 18 provides a computer program, including a program code, which, when executed by a processor, causes implementation of any one of Examples 1 to 13. The steps of the method are illustrated.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本公开涉及一种投放结果的确定方法、装置、可读介质和电子设备，该方法包括：获取目标对象对应的多个对象特征；在多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据多个对象特征对应的特征类型，从多个对象特征中确定目标对象特征；将目标对象特征输入预测模型中，得到预测模型输出的目标对象对应的目标投放结果；目标投放结果用于表征目标对象的用户接受度。

Description

投放结果的确定方法、装置、可读介质和电子设备
相关申请的交叉引用
本申请是以申请号为202310678334.3、申请日为2023年6月8日的中国申请为基础，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
本公开涉及电子信息技术领域，具体地，涉及一种投放结果的确定方法、装置、可读介质和电子设备。
背景技术
随着互联网的不断发展，给广告提供了更多样的投放形式。对于终端设备来说，广告可以投放在用户浏览界面的不同位置中。广告能否取得较好的投放效果，影响着是否能够提高广告转换率，以及广告商的品牌口碑。由此可见，广告的投放结果对于广告的精准投放有着十分重要的意义。
发明内容
提供该发明内容部分以便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。该发明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，本公开提供一种投放结果的确定方法，所述方法包括：
获取目标对象对应的多个对象特征；
在所述多个对象特征的第一特征总数大于或等于预设数量阈值的情况 下，根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征；
将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果；所述目标投放结果用于表征所述目标对象的用户接受度。
第二方面，本公开提供一种投放结果的确定装置，所述装置包括：
第一获取模块，被配置为获取目标对象对应的多个对象特征；
第一确定模块，被配置为在所述多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征；
预测模块，被配置为将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果；所述目标投放结果用于表征所述目标对象的用户接受度。
第三方面，本公开提供一种计算机可读介质，其上存储有计算机程序和/或指令，该计算机程序和/或指令被处理装置执行时实现上述第一方面所述投放结果的确定方法的步骤。
第四方面，本公开提供一种电子设备，包括：
存储装置，其上存储有计算机程序和/或指令；
处理装置，被配置为执行所述存储装置中的所述计算机程序和/或指令，以实现上述第一方面所述投放结果的确定方法的步骤。
第五方面，本公开提供一种计算机程序产品，包含指令，该指令在由处理器执行时使得处理器实现上述第一方面所述投放结果的确定方法的步骤。
第六方面，本公开提供一种计算机程序，包括程序代码，该程序代码在由处理器执行时导致实现上述第一方面所述投放结果的确定方法的步骤。
本公开的其他特征和优点将在随后的具体实施方式部分予以详细说明。
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。在附图中：
图1是根据一示例性实施例示出的一种投放结果的确定方法的流程图；
图2是根据一示例性实施例示出的另一种投放结果的确定方法的流程图；
图3是根据一示例性实施例示出的另一种投放结果的确定方法的流程图；
图4是根据一示例性实施例示出的另一种投放结果的确定方法的流程图；
图5是根据一示例性实施例示出的另一种投放结果的确定方法的流程图；
图6是根据一示例性实施例示出的另一种投放结果的确定方法的流程图；
图7是根据一示例性实施例示出的一种投放结果的确定方法的流程图；
图8是根据一示例性实施例示出的一种目标投放结果预测的流程示意图；
图9是根据一示例性实施例示出的一种投放结果的确定装置的框图；
图10是根据一示例性实施例示出的另一种投放结果的确定装置的框图；
图11是根据一示例性实施例示出的另一种投放结果的确定装置的框图；
图12是根据一示例性实施例示出的另一种投放结果的确定装置的框图；
图13是根据一示例性实施例示出的一种电子设备的框图。
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。
应当理解，本公开的方法实施方式中记载的各个步骤可以按照不同的顺序执行，和/或并行执行。此外，方法实施方式可以包括附加的步骤和/或省略执行示出的步骤。本公开的范围在此方面不受限制。
本文使用的术语“包括”及其变形是开放性包括，即“包括但不限于”。术语“基于”是“至少部分地基于”。术语“一个实施例”表示“至少一个实施例”；术语“另一实施例”表示“至少一个另外的实施例”；术语“一些实施例”表示“至少一些实施例”。其他术语的相关定义将在下文描述中给出。
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”。
本公开实施例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。
可以理解的是，在使用本公开各实施例公开的技术方案之前，均应当依据相关法律法规通过恰当的方式对本公开所涉及个人信息的类型、使用范围、使用场景等告知用户并获得用户的授权。
例如，在响应于接收到用户的主动请求时，向用户发送提示信息，以明确地提示用户，其请求执行的操作将需要获取和使用到用户的个人信息。从而，使得用户可以根据提示信息来自主地选择是否向执行本公开技术方案的操作的电子设备、应用程序、服务器或存储介质等软件或硬件提供个人信息。
作为一种可选的但非限定性的实现方式，响应于接收到用户的主动请求，向用户发送提示信息的方式例如可以是弹窗的方式，弹窗中可以以文字的方 式呈现提示信息。此外，弹窗中还可以承载供用户选择“同意”或者“不同意”向电子设备提供个人信息的选择控件。
可以理解的是，上述通知和获取用户授权过程仅是示意性的，不对本公开的实现方式构成限定，其它满足相关法律法规的方式也可应用于本公开的实现方式中。
此外，可以理解的是，本技术方案所涉及的数据(包括但不限于数据本身、数据的获取或使用)应当遵循相应法律法规及相关规定的要求。
目前，确定广告投放结果主要是在广告进行批量投放之后，通过采样用户数据来进行评估。这种方法需要先将广告进行投放，当投放量级较少时误差较大，评估准确性较低，且存在一定的滞后性，无法准确反映广告的投放效果。
鉴于此，本公开提出了改进的数据投放相关处理，尤其提出了改进的投放结果确定，以下将参照附图进行详细描述。应指出，虽然以下主要针对广告、商品等的投放进行描述，但是应指出，这仅是示例而非限制，本公开实施例的方案可同样/等同地适用于其他应用的数据投放/呈现等等，实质上属于一种改进的数据分析和处理，以用于更加准确或适当地筛选/提供数据。
下面结合附图对本公开的具体实施方式进行详细说明。
图1是根据一示例性实施例示出的一种投放结果的确定方法的流程图，如图1所示，该方法可以包括以下步骤：
在步骤S101中，获取目标对象对应的多个对象特征。
其中，该目标对象例如可以是待宣传或发布的对象，如广告或商品等。举例来说，在广告进行投放之前，可以对广告的投放效果进行预测，也即对用户的喜爱度以及接受度等进行预测，从而便于更加精准的对广告进行投放(例如是将广告具体投放至浏览界面中的什么位置、什么时间段投放广告、投放至什么区域等等)。这样，以提高用户对该广告的喜爱度和接受率，并 能够最大化提高广告转换率。
以目标对象为广告为例，该对象特征例如但不限于可以包括广告主特征(如广告主名称、广告主类别、广告主来源)、广告计划特征(如计划创意数、广告创建时长、评论权限开关、广告所属行业)、广告素材特征(如标题、文案、图片、视频、落地页)、广告主资质特征(如资质授权信息、广告有效期)、广告投放特征(如广告投放预算、广告投放区域、投放时间段、用户接受度(如点赞数、分享数、不喜欢数、举报数、评论数))等中的一个或多个。
需要说明的是，上述示例仅为举例说明，本公开并不局限于此。
在步骤S102中，在多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据多个对象特征对应的特征类型，从多个对象特征中确定目标对象特征。
可以理解的是，一个目标对象往往对应较多的对象特征，而当对象特征的数量越多，后续模型的计算时间就越长。考虑到这一问题，为了保证模型的处理效率，在本实施例中，可以设置模型可处理的最大特征数量限值，也即预设数量阈值。在多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，可以对对象特征进行删减，以减少对象特征的数量，得到目标对象特征。具体地，在多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据多个对象特征对应的特征类型进行有针对性的筛选，对于不同类型的对象特征采用不同的筛选方式，进而得到目标对象特征。
示例地，对象特征的特征类型例如可以包括离散特征(如广告主类别、评论权限开关、广告所属行业等)、数值特征(如广告投放预算、点赞数等)、图像特征(如图片)和文本特征(如标题、文案等)。对于离散特征和数值特征来说，由于该类特征往往能够直接反映目标对象的自身属性，因此针对这类特征可以预先对对象特征与目标对象的相关度进行分析(例如可以根据 经验得到，或者可以通过输入模型分析得到，本公开对此不作具体限定)，并根据分析得到的相关度设定特征之间的优先级，优先级越高表明对象特征与目标对象的相关度越高。在多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，对于优先级较低的特征可以优先删除。作为一个示例，可以删除优先级低于特定优先级阈值的特征。还作为示例，可以按照优先级从低到高来删除特征，直至保留特定数量的特征，该特定数量可以小于或等于第一特征总数，或者任何其他预先固定的数量。对于图像特征和文本特征来说，由于该类特征中易出现重复或相似的特征，因而可以对特征进行特征去重和/或采样，以减少特征的数量。
另外，在多个对象特征的第一特征总数小于预设数量阈值的情况下，表明对象特征的数量没有超过模型可处理的最大特征数量限值，此时可以直接将多个对象特征作为该目标对象特征。
在步骤S103中，将该目标对象特征输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果。
其中，该目标投放结果用于表征该目标对象的用户接受度。例如，点赞数、分享数、不喜欢数、举报数、评论数等。该预测模型例如但不限于可以是Transformer结构、残差结构或图网络等其他结构。
以目标对象为广告为例，需要说明的是，该本实施例可用于广告未正式投放之前，对广告投放效果进行预测，还可以用于广告正式投放之后，对后续的广告投放效果进行预测。在本实施例用于广告未正式投放之前的场景时，上述对象特征中的用户接受度可以为预设值(例如均为0)。在本实施例用于广告正式投放之后的场景时，上述对象特征中的用户接受度可以为实际投放后获取的历史时间段内用户真实的接受度。
在本实施例中，预测模型一般输出的为输出向量，因此还可以通过网络层将其映射为单个数值，也即得到目标投放结果。
采用上述方法，可以根据目标对象对应的目标对象特征来预测目标对象的目标投放结果。在目标对象未投放时，预先根据目标投放结果来确定目标对象的投放目标，从而便于更加精准的对目标对象进行投放。此外，附加地或者可替代地，在输入预测模型前，若对象特征的第一特征总数过大，则可以根据对象特征的特征类型对对象特征进行筛选，得到目标对象特征，并将目标对象特征输入预测模型中，有效的提高了模型的处理效率。
在一些实施例中，该预测模型可以通过以下方式预先生成：
步骤A，获取训练样本集，该训练样本集中包含样本对象对应的样本对象特征以及该样本对象对应的真实投放结果。
其中，该真实投放结果用于表征该样本对象投放后的用户接受度。也即该真实投放结果为该样本对象在实际进行投放后获取到的用户接受度，可用于对模型进行进一步修正，提高模型预测的准确性。
步骤B，根据预设损失函数、该样本对象特征以及该真实投放结果对预设模型进行迭代更新，得到该预测模型。
示例地，可以将样本对象特征输入预设模型中，得到预测模型输出的该样本对象对应的样本投放结果，将该样本投放结果与真实投放结果进行比较，通过预设损失函数计算差值，使用梯度回传来更新模型参数。
具体地，可以循环执行模型迭代步骤，直至根据该预设损失函数确定更新后的预设模型满足预设停止迭代条件，将更新后的预设模型作为该预测模型。
其中，该模型迭代步骤包括：
S1，将该样本对象特征输入该预设模型中，得到该预设模型输出的该样本对象对应的样本投放结果。
其中，该样本投放结果用于表征该样本对象的用户接受度。
同样地，该预设模型例如但不限于可以是Transformer结构、残差结构 或图网络等其他结构。
S2，根据该预设损失函数，确定该样本对象对应的该样本投放结果与该真实投放结果的目标损失值。
其中，该目标损失值用于表征该样本投放结果和该真实投放结果之间的差异程度。
S3，在根据该目标损失值确定该预设模型不满足该预设停止迭代条件的情况下，根据该目标损失值更新该预设模型的参数，得到新的预设模型。然后，可基于新的预设模型来执行下一次迭代处理，直到基于目标损失值判断满足了预设停止迭代条件为止。在本公开的实施例中，停止迭代条件可以为各种适当的条件中的至少一者，例如损失值已经变得稳定，诸如损失小于特定损失阈值，诸如连续特定次迭代中的损失值变化小于特定变化阈值，还例如已经执行了特定迭代次数，等等，这些仅作为示例而非限制，将不再详细描述。在一些实施例中，可以在满足任一迭代条件时停止迭代。
图2是根据一示例性实施例示出的另一种投放结果的确定方法的流程图，如图2所示，该方法还可以包括以下步骤：
在步骤S104中，对该目标对象特征进行维度转换，以将不同的目标对象特征的向量维度转换为相同的向量维度。
在一些实施例中，对于预测模型来说，需要对多个目标对象特征进行分析处理，不同的目标对象特征之间需要相互运算。因此，为了便于模型对目标对象特征进行处理，可以先对目标对象特征进行维度转换，以将所有的目标对象特征映射为相同维度的特征。其中，例如可以将所有的目标对象特征映射至768维度。
对于不同类型的特征，维度转换的方法有所不同。示例地，对于离散特征和数值特征来说，可以通过AutoDis(Automatic Discretization)编码、桶映射、数值填充等方法进行维度转换，以将特征映射至固定维度。对于图像 特征(其中包括单个图片和/或视频拆分后得到的N个图像帧)来说，可以通过图像视觉模型(如Resnet、Vit(Vision Transformer)等模型)等方法进行维度转换，以将特征映射至固定维度。例如，以图像特征包括视频为例，可以先视频拆分为N个图像帧。然后，使用图像视觉模型映射至N×768维度。对于文本特征来说，由于各个文本中包含的文本长度是不同的，例如标题、广告主名称等往往文本长度较小，而文案、落地页文本往往文本长度较大。因此，针对文本长度大于或等于预设长度阈值(例如但不限于可以为128)的文本特征，可以进行分句处理，以使得分句处理后文本的文本长度小于预设长度阈值。进一步地，可以将分句处理后的文本通过自然语言模型(如Bert、MLXR等模型)等方法进行维度转换，以将特征映射至固定维度。例如，以文案、落地页文本等文本长度较大的文本为例，可以通过分句处理分割为K个长度小于128的短文本。然后，可以使用自然语言模型，映射至K×L×768维度。可以理解的是，由于上述方法中是将长文本进行分句后输入自然语言模型中的，也即是一句句输入的，得到的也是对应的一句句输出的，此时可将它们压缩成1*768维度的特征，也即可以池化至K×768维度。另外，针对文本长度小于预设长度阈值的文本特征，可以直接通过自然语言模型等方法进行维度转换，以将特征映射至固定维度。例如，以标题、广告主名称等文本长度小于128的文本，可以使用自然语言模型，映射至L×768维度，然后池化至768维度。
相应地，上述步骤S103中将该目标对象特征输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果包括：
将维度转换后的目标对象特征输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果。
图3是根据一示例性实施例示出的另一种投放结果的确定方法的流程图，如图3所示，该方法还可以包括以下步骤：
在步骤S105中，根据该目标对象特征的特征类型，确定该目标对象特征对应的目标特征类型编码。
在本步骤中，可以预先设置不同的特征类型对应的特征类型编码，进而得到编码对应关系，可以根据目标对象特征的特征类型，通过该编码对应关系得到该特征类型对应的特征类型编码，作为目标特征类型编码。
相应地，上述步骤S103中将该目标对象特征输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果包括：
将该目标对象特征与该目标特征类型编码相加后输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果。
需要说明的是，在本实施例中，图2和图3的实施例还可以结合起来执行。也即可以先对该目标对象特征进行维度转换，以将不同的目标对象特征的向量维度转换为相同的向量维度。并根据该目标对象特征的特征类型，确定该目标对象特征对应的目标特征类型编码。然后，将维度转换后的目标对象特征与该目标特征类型编码相加后输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果。
下面针对上述步骤S102，在多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据多个对象特征对应的特征类型，从多个对象特征中确定目标对象特征进行详细说明。
在一种实施例中，如图4所示，上述步骤S102可以包括以下步骤：
在步骤S1021中，将该特征类型为第一指定类型的对象特征，作为第一候选对象特征。
其中，该第一指定类型例如可以是图像特征和文本特征。
在步骤S1022中，从第一候选对象特征中确定第二候选对象特征。
在第一示例性实施例中，从第一候选对象特征中确定第二候选对象特征可以包括：对该第一候选对象特征进行去重处理，并将去重处理后的第一候 选对象特征作为该第二候选对象特征。
示例地，以目标对象对广告为例，在一个广告计划中，广告主头像、视频首帧图像或视频中的连续帧之间都可能是相同的。因此，针对图像特征来说，可以通过哈希编码对所有图像特征进行去重处理。一个广告计划中，所有文本的表述目的都是为了提高用户的转化意愿。因此，针对文本特征来说，可以先对文本进行过滤，也即可以将固定格式的字段、无效符号去除。
在第二示例性实施例中，从第一候选对象特征中确定第二候选对象特征可以包括：首先，确定每两个第一候选对象特征之间的相似度。然后，在该相似度大于或等于预设相似度阈值的情况下，将该相似度对应的两个第一候选对象特征中的任意一个作为第四候选对象特征。并将该第四候选对象特征和该相似度小于该预设相似度阈值的第一候选对象特征，作为该第二候选对象特征。
示例地，可以通过余弦相似度来计算每两个第一候选对象特征之间的相似度，若相似度大于或等于预设相似度阈值，则表明这两个第一候选对象特征之间相似度较高，此时，可以任意保留其中一个作为第四候选对象特征。并将该第四候选对象特征和该相似度小于该预设相似度阈值的第一候选对象特征，作为该第二候选对象特征。
在第三示例性实施例中，从第一候选对象特征中确定第二候选对象特征可以包括：首先，可以将特征类型为第二指定类型的第一候选对象特征，作为第五候选对象特征。然后，对该第五候选对象特征进行采样处理。并将采样处理后的第五候选对象特征和第六候选对象特征作为该第二候选对象特征。其中，该第六候选对象特征为该第一候选对象特征中除该第五候选对象特征以外的对象特征。
示例地，该第二指定类型例如可以是视频帧图像特征和长文本特征(图像特征可以包括单个图片特征和视频帧图像特征，文本特征可以包括长文本 特征和短文本特征)。首先，可以将特征类型为第二指定类型的第一候选对象特征，作为第五候选对象特征。然后，对该第五候选对象特征进行采样处理包括：确定该第一候选对象特征的第二特征总数和该第五候选对象特征的第三特征总数。并根据该预设数量阈值、该第二特征总数和该第三特征总数，确定目标特征数量。举例来说，在第一候选对象特征包括图像特征和文本特征的情况下，第三候选对象特征(其中，该第三候选对象特征为多个对象特征中除该第一候选对象特征以外的对象特征)例如可以为离散特征和数值特征，可以先根据第一特征总数和第二特征总数，计算出第三候选对象特征的数量，也即确定离散特征和数值特征的数量。然后，在第一候选对象特征包括图像特征和文本特征，且第五候选对象特征包括视频帧图像特征和长文本特征的情况下，根据第二特征总数和第三特征总数，确定第一候选对象特征中除第五候选对象特征以外的特征数量，也即确定单个图片特征和短文本特征的数量。之后，可以根据预设数量阈值、离散特征和数值特征的数量、单个图片特征和短文本特征的数量，确定第五候选对象特征可用的最大剩余长度，也即视频帧图像特征和长文本特征可用的最大剩余长度，记为R＝预设数量阈值-离散特征和数值特征的数量-单个图片特征和短文本特征的数量。然后，再确定视频个数N和长文本个数M，既可以得到每个视频或长文本需采样至L＝R/(M+N)个特征，L即为目标特征数量，表征每个第五候选对象特征需采样的数量。之后，根据该目标特征数量，对该第五候选对象特征进行采样处理。其中，可以根据目标特征数量，对第五候选对象特征进行均匀采样。还可以根据不同的目标特征数量，设定相应的采样方式，举例来说，若L小于1，则表明每个视频或长文本可以都不需要保留，此时可以将视频对应的视频帧图像和长文本都删除掉。并且还可以给予提示，建议修改预设数量阈值，以尽可能使得目标对象特征中包含各种类型的对象特征。若L大于等于1且小于2，则可以采用首位采样的方式，例如对于视频帧图像特征 来说，可以将视频帧图像中位于首位的图像进行采样，对于长文本特征来说，可以将该长文本中首段的内容进行采样。若L大于等于2且小于3，则可以采用首、尾位采样的方式进行采样。若L大于或等于3，则可以采样线性均匀采样的方式进行采样。
需要说明的是，上述三个示例性实施例可以单独实施，还可以选择其中任意两个或三个组合实施，且实施顺序也可以自由组合，本公开对此不作具体限定。举例来说，可以先通过第一示例性实施例中的方法进行去重处理，若进行去重处理后剩余对象特征的总数小于预设数量阈值，则可以将去重处理后的对象特征作为第二候选对象特征。若进行去重处理后剩余对象特征的总数仍大于或等于预设数量阈值，则可以继续通过第二示例性实施例进行相似度的计算，并删除相似度较高的其中一个对象特征。若删除后剩余对象特征的总数小于预设数量阈值，则可以将删除后的对象特征作为第二候选对象特征。若删除后剩余对象特征的总数仍大于或等于预设数量阈值，则可以继续通过第三示例性实施例进行采样处理，将采样处理后的对象特征作为第二候选对象特征。还可以是其他的任意组合方式，例如将第一示例性实施例与第三示例性实施例进行组合，第一示例性实施例与第二示例性实施例进行组合，第二示例性实施例与第三示例性实施例进行组合等等，此处不再一一列举说明。
在步骤S1023中，根据该第二候选对象特征和第三候选对象特征，确定该目标对象特征。
其中，该第三候选对象特征为多个对象特征中除该第一候选对象特征以外的对象特征。
在本步骤中，可以直接将第二候选对象特征和第三候选对象特征作为目标对象特征。还可以先确定第二候选对象特征和第三候选对象特征的第四特征总数，并在第四特征总数小于预设数量阈值的情况下，将第二候选对象特 征和第三候选对象特征作为目标对象特征。
在另一种实施例中，如图5所示，上述步骤S102可以包括以下步骤：
在步骤S1024中，将该特征类型为第三指定类型的对象特征，作为第七候选对象特征。
其中，该第三指定类型例如但不限于可以是离散特征和数值特征。
在步骤S1025中，获取该第七候选对象特征对应的预设优先级。
其中，该预设优先级可以是预先对特征与目标对象的相关度进行分析(例如可以根据经验得到，或者可以通过输入模型分析得到)后设定的。
在步骤S1026中，根据该预设优先级，从该第七候选对象特征中确定第八候选对象特征。
在本步骤中，可以根据预设优先级以及第一特征总数，从该第七候选对象特征中确定第八候选对象特征，以使得筛选后的特征总数不超过预设数量阈值。
在步骤S1027中，根据该第八候选对象特征和第九候选对象特征，确定该目标对象特征。
其中，该第九候选对象特征为多个对象特征中除该第七候选对象特征以外的对象特征。
在本步骤中，可以直接将第八候选对象特征和第九候选对象特征作为目标对象特征。还可以先确定第八候选对象特征和第九候选对象特征的第五特征总数，并在第五特征总数小于预设数量阈值的情况下，将第八候选对象特征和第九候选对象特征作为目标对象特征。
需要说明的是，在本实施例中，图4和图5的实施例可以单独实施，还可以组合实施，并且实施顺序也可以自由组合，本公开对此不作具体限定。举例来说，若通过图4实施例的方法得到的第二候选对象特征和第三候选对象特征的第四特征总数大于或等于预设数量阈值的情况下，还可以通过图5 实施例中的方法对特征数量进行进一步的缩减，以使得总的特征数量小于预设数量阈值。
图6是根据一示例性实施例示出的另一种投放结果的确定方法的流程图，如图6所示，该方法还可以包括以下步骤：
在步骤S106中，获取该目标对象对应的历史投放结果。
其中，该历史投放结果用于表征历史时刻该目标对象的用户接受度；该历史投放结果为历史时刻将该目标对象对应的历史对象特征输入该预测模型中，得到的该预测模型输出的该目标对象对应的投放结果。
在本实施例中，图1的实施例还可以用于目标对象投放过程中，当目标对象投放之后，用户接受度等数值会从0开始增加。此时，为了更加准确的确定目标对象的目标投放结果，可以对目标投放结果进行校正。
在一些实施例中，还可以自定义校正的触发时刻，例如可以根据业务需求，在投放后每经过预设时长，对目标对象的目标投放结果进行重新预测，也即执行图1实施例中的方法。又例如，还可以在目标识别结果对应的用户接受度达到预设接受度的情况下，对目标对象的目标投放结果进行重新预测。例如点赞数、分享数、评论数、不喜欢数、举报数中一个或多个达到预设阈值时，对目标对象的目标投放结果进行重新预测。
需要说明的是，在本实施例中，当重新对目标对象的目标投放结果进行预测时，为了保证准确度，可以获取目标对象最新的对象特征，以更加准确的反映目标对象当前的特征。
在步骤S107中，根据该历史投放结果，对该目标投放结果进行校正。
示例地，可以通过Momentum算法，基于历史投放结果和目标投放结果，进行校正。还可以给历史投放结果和目标投放结果设定相应的权重，并分别将历史投放结果和目标投放结果乘以相应的权重后相加，得到校正后的结果。
在步骤S108中，将校正后的目标投放结果作为该目标对象新的目标投 放结果。
这样，可以在目标对象投放的过程中，对预测的目标投放结果进行实时校正，提高了投放之后，投放结果预测的准确性。
采用上述方法，可以根据目标对象对应的目标对象特征来预测目标对象的目标投放结果。在目标对象未投放时，预先根据目标投放结果来确定目标对象的投放目标，从而便于更加精准的对目标对象进行投放。此外，附加地或者可替代地，在输入预测模型前，若对象特征的第一特征总数过大，则可以根据对象特征的特征类型对对象特征进行筛选，得到目标对象特征，并将目标对象特征输入预测模型中，有效的提高了模型的处理效率。
图7是根据一示例性实施例示出的一种投放结果的确定方法的流程图，如图7所示，该方法可以包括以下步骤：
在步骤S201中，获取目标对象对应的多个对象特征。
在步骤S202中，在多个对象特征的第一特征总数小于预设数量阈值的情况下，将多个对象特征作为目标对象特征。
在步骤S203中，在多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，将该特征类型为第一指定类型的对象特征，作为第一候选对象特征。
其中，该第一指定类型例如可以是图像特征和文本特征。由于图像特征和文本特征中易存在重复或表达含义相同的特征，因此，在本实施例中可以优先对图像特征和文本特征进行筛选。
在步骤S204中，对该第一候选对象特征进行去重处理，并确定去重处理后的第一候选对象特征和第三候选对象特征的第六特征总数。
其中，该第三候选对象特征为多个对象特征中除该第一候选对象特征以外的对象特征。
在步骤S205中，在第六特征总数小于预设数量阈值的情况下，将去重 处理后的第一候选对象特征和第三候选对象特征作为目标对象特征。
在步骤S206中，在第六特征总数大于或等于预设数量阈值的情况下，确定每两个去重处理后的第一候选对象特征之间的相似度。
在步骤S207中，在该相似度大于或等于预设相似度阈值的情况下，将该相似度对应的两个去重处理后的第一候选对象特征中的任意一个作为第四候选对象特征，并将该第四候选对象特征和该相似度小于该预设相似度阈值的去重处理后的第一候选对象特征，作为第十候选对象特征。
这样，可以优先对第一候选对象进行去重处理，以最大程度的保留对象特征，若去重处理后的剩余特征仍然超过预设数量阈值，则可以计算相似度来进一步减少特征数量。
在步骤S208中，确定第十候选对象特征和第三候选对象特征的第七特征总数。
在步骤S209中，在第七特征总数小于预设数量阈值的情况下，将第十候选对象特征和第三候选对象特征作为目标对象特征。
在步骤S210中，在第七特征总数大于或等于预设数量阈值的情况下，将特征类型为第二指定类型的第十候选对象特征，作为第十一候选对象特征。
其中，第二指定类型例如可以是视频帧图像特征和长文本特征。
在步骤S211中，对该第十一候选对象特征进行采样处理。
其中，对第十一候选对象特征进行采样处理可参照上述实施例中对第五候选对象特征进行采样处理的方法，此处不再赘述。
在步骤S212中，确定采样处理后的第十一候选对象特征和第三候选对象特征的第八特征总数。
在步骤S213中，在第八特征总数小于预设数量阈值的情况下，将采样处理后的第十候选对象特征和第三候选对象特征作为目标对象特征。
在步骤S214中，在第八特征总数大于或等于预设数量阈值的情况下， 获取第三候选对象特征对应的预设优先级。
其中，该预设优先级可以是预先对特征与目标对象的相关度进行分析(例如可以根据经验得到，或者可以通过输入模型分析得到)后设定的。
在步骤S215中，根据该预设优先级，从该第三候选对象特征中确定第十二候选对象特征。
其中，可以根据预设优先级和预设数量阈值，将剩余特征的数量缩减至小于预设数量阈值。
在步骤S216中，将采样处理后的第十候选对象特征和第十二候选对象特征作为目标对象特征。
在步骤S217中，对该目标对象特征进行维度转换，以将不同的目标对象特征的向量维度转换为相同的向量维度。
在步骤S218中，根据该目标对象特征的特征类型，确定该目标对象特征对应的目标特征类型编码。
在步骤S219中，将维度转换后的目标对象特征与该目标特征类型编码相加后输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果。
为了更加直观的理解，如图8所示，示出了目标对象的目标投放结果预测的流程示意图，首先可以对对象特征进行预处理，如将视频拆分为视频帧图像，将长文本拆分为短文本。然后进行特征筛选，并在筛选之后将各个特征映射至相同维度。之后，将各个特征与对应的特征类型编码进行相加。最后，将相加后的特征输入预测模型中，得到预测模型输出的预测结果，也即目标投放结果。需要说明的是，特征输入预测模型的顺序可以是任意的。对于预测模型来说，可能会输出多个输出向量，在本实施例中，只需关注例如点赞数、评论数等反映投放结果的输出向量即可。这样，用户可以根据需求，将不同业务的多种对象特征类型输入预测模型中，提高了应用场景的多样性， 易扩展易迁移。
采用上述方法，可以根据目标对象对应的目标对象特征来预测目标对象的目标投放结果。在目标对象未投放时，预先根据目标投放结果来确定目标对象的投放目标，从而便于更加精准的对目标对象进行投放。此外，附加地或者可替代地，在输入预测模型前，若对象特征的第一特征总数过大，则可以根据对象特征的特征类型对对象特征进行筛选，得到目标对象特征，并将目标对象特征输入预测模型中，有效的提高了模型的处理效率。
需要说明的是，关于上述实施例中的方法，其中各个步骤执行操作的具体方式已经在有关图1至图6的实施例中进行了详细描述，此处将不做详细阐述说明。
以下将描述根据本公开的实施例的投放结果的确定装置。本公开实施例所提供的确定装置可执行本公开任意实施例所提供的投放结果的确定方法，本公开实施例可以根据上述方法示例对装置进行功能单元的划分，例如，可以对应各个功能划分各个功能模块/单元，也可以将两个或两个以上的功能集成在一个处理模块中。值得注意的是，上述装置所包括的各个模块/单元只是按照功能逻辑进行划分的，但并不局限于文中所述的划分，实际实现时可以有另外的划分方式，只要能够实现相应的功能即可；另外，各模块/单元的具体名称也只是为了便于相互区分，并不用于限制本公开实施例的保护范围。此外，各模块/单元可采用各种适当方式来实现，例如硬件、固件、或任何适当组合来实现的。
图9是根据一示例性实施例示出的一种投放结果的确定装置的框图，如图9所示，该装置300包括：
第一获取模块301，被配置为获取目标对象对应的多个对象特征；
第一确定模块302，被配置为在多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据多个对象特征对应的特征类型，从多个对象 特征中确定目标对象特征；
预测模块303，被配置为将该目标对象特征输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果；该目标投放结果用于表征该目标对象的用户接受度。
可选地，第一确定模块302，被配置为将该特征类型为第一指定类型的对象特征，作为第一候选对象特征；从第一候选对象特征中确定第二候选对象特征；根据该第二候选对象特征和第三候选对象特征，确定该目标对象特征；其中，该第三候选对象特征为多个对象特征中除该第一候选对象特征以外的对象特征。
可选地，第一确定模块302，被配置为对该第一候选对象特征进行去重处理；将去重处理后的第一候选对象特征作为该第二候选对象特征。
可选地，第一确定模块302，被配置为确定每两个第一候选对象特征之间的相似度；在该相似度大于或等于预设相似度阈值的情况下，将该相似度对应的两个第一候选对象特征中的任意一个作为第四候选对象特征；将该第四候选对象特征和该相似度小于该预设相似度阈值的第一候选对象特征，作为该第二候选对象特征。
可选地，第一确定模块302，被配置为将特征类型为第二指定类型的第一候选对象特征，作为第五候选对象特征；对该第五候选对象特征进行采样处理；将采样处理后的第五候选对象特征和第六候选对象特征作为该第二候选对象特征；其中，该第六候选对象特征为该第一候选对象特征中除该第五候选对象特征以外的对象特征。
可选地，第一确定模块302，被配置为确定该第一候选对象特征的第二特征总数和该第五候选对象特征的第三特征总数；根据该预设数量阈值、该第二特征总数和该第三特征总数，确定目标特征数量；根据该目标特征数量，对该第五候选对象特征进行采样处理。
可选地，第一确定模块302，被配置为将该特征类型为第三指定类型的对象特征，作为第七候选对象特征；获取该第七候选对象特征对应的预设优先级；根据该预设优先级，从该第七候选对象特征中确定第八候选对象特征；根据该第八候选对象特征和第九候选对象特征，确定该目标对象特征；其中，该第九候选对象特征为多个对象特征中除该第七候选对象特征以外的对象特征。
可选地，该预测模型是通过以下方式预先生成的：
获取训练样本集，该训练样本集中包含样本对象对应的样本对象特征以及该样本对象对应的真实投放结果；该真实投放结果用于表征该样本对象投放后的用户接受度；
根据预设损失函数、该样本对象特征以及该真实投放结果对预设模型进行迭代更新，得到该预测模型。
可选地，该根据预设损失函数、该样本对象特征以及该真实投放结果对预设模型进行迭代更新，得到该预测模型包括：
循环执行模型迭代步骤，直至根据该预设损失函数确定更新后的预设模型满足预设停止迭代条件，将更新后的预设模型作为该预测模型；
该模型迭代步骤包括：
将该样本对象特征输入该预设模型中，得到该预设模型输出的该样本对象对应的样本投放结果，该样本投放结果用于表征该样本对象的用户接受度；
根据该预设损失函数，确定该样本对象对应的该样本投放结果与该真实投放结果的目标损失值；该目标损失值用于表征该样本投放结果和该真实投放结果之间的差异程度；
在根据该目标损失值确定该预设模型不满足该预设停止迭代条件的情况下，根据该目标损失值更新该预设模型的参数，得到新的预设模型。
可选地，如图10所示，该装置300还包括：
第二获取模块304，被配置为获取该目标对象对应的历史投放结果，该历史投放结果用于表征历史时刻该目标对象的用户接受度；该历史投放结果为历史时刻将该目标对象对应的历史对象特征输入该预测模型中，得到的该预测模型输出的该目标对象对应的投放结果；
校正模块305，被配置为根据该历史投放结果，对该目标投放结果进行校正；
第二确定模块306，被配置为将校正后的目标投放结果作为该目标对象新的目标投放结果。
可选地，第一确定模块302，还被配置为在该第一特征总数小于该预设数量阈值的情况下，将多个对象特征作为该目标对象特征。
可选地，如图11所示，该装置300还包括：
转换模块307，被配置为对该目标对象特征进行维度转换，以将不同的目标对象特征的向量维度转换为相同的向量维度；
预测模块303，被配置为将维度转换后的目标对象特征输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果。
可选地，如图12所示，该装置300还包括：
第三确定模块308，被配置为根据该目标对象特征的特征类型，确定该目标对象特征对应的目标特征类型编码；
预测模块303，被配置为将该目标对象特征与该目标特征类型编码相加后输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果。
采用上述装置，可以根据目标对象对应的目标对象特征来预测目标对象的目标投放结果。在目标对象未投放时，预先根据目标投放结果来确定目标对象的投放目标，从而便于更加精准的对目标对象进行投放。此外，附加地或者可替代地，在输入预测模型前，若对象特征的第一特征总数过大，则可以根据对象特征的特征类型对对象特征进行筛选，得到目标对象特征，并将 目标对象特征输入预测模型中，有效的提高了模型的处理效率。
需要说明的是，关于上述实施例中的装置，其中各个模块执行操作的具体方式已经在有关图1至图8的实施例中进行了详细描述，此处将不做详细阐述说明。
上述主要对本公开实施例提供的方案进行了介绍。可以理解的是，为了实现上述功能，电子设备可以包含了执行各个功能相应的硬件结构和/或软件模块。本领域技术人员应该很容易意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，本公开能够以硬件或硬件和计算机软件的结合形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本公开的范围。
下面参考图13，其示出了适于用来实现本公开实施例的电子设备(例如终端设备或服务器)400的结构示意图。本公开实施例中的终端设备可以包括但不限于诸如移动电话、笔记本电脑、数字广播接收器、PDA(个人数字助理)、PAD(平板电脑)、PMP(便携式多媒体播放器)、车载终端(例如车载导航终端)等等的移动终端以及诸如数字TV、台式计算机等等的固定终端。图13示出的电子设备仅仅是一个示例，不应对本公开实施例的功能和使用范围带来任何限制。
如图13所示，电子设备400可以包括处理装置(例如中央处理器、图形处理器等)401，其可以根据存储在只读存储器(ROM)402中的程序或者从存储装置408加载到随机访问存储器(RAM)403中的程序而执行各种适当的动作和处理。在RAM 403中，还存储有电子设备400操作所需的各种程序和数据。处理装置401、ROM 402以及RAM 403通过总线404彼此相连。输入/输出(I/O)接口405也连接至总线404。
通常，以下装置可以连接至I/O接口405：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406；包括例如液晶显示器(LCD)、扬声器、振动器等的输出装置407；包括例如磁带、硬盘等的存储装置408；以及通信装置409。通信装置409可以允许电子设备400与其他设备进行无线或有线通信以交换数据。虽然图13示出了具有各种装置的电子设备400，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。
特别地，根据本公开的实施例，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，本公开的实施例包括一种计算机程序产品，其包括承载在非暂态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。在这样的实施例中，该计算机程序可以通过通信装置409从网络上被下载和安装，或者从存储装置408被安装，或者从ROM 402被安装。在该计算机程序被处理装置401执行时，执行本公开实施例的方法中限定的上述功能。
需要说明的是，本公开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。
在一些实施例中，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电子设备中所包含的；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执行本公开任意实施例所提供的方法的任一步骤或组合。
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操作的计算机程序代码，上述程序设计语言包括但不限于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。
描述于本公开实施例中所涉及到的模块可以通过软件的方式实现，也可以通过硬件的方式来实现。其中，模块的名称在某种情况下并不构成对该模块本身的限定。
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。
在本公开的上下文中，机器可读介质可以是有形的介质，其可以包含或存储以供指令执行系统、装置或设备使用或与指令执行系统、装置或设备结合地使用的程序。机器可读介质可以是机器可读信号介质或机器可读储存介质。机器可读介质可以包括但不限于电子的、磁性的、光学的、电磁的、红外的、或半导体系统、装置或设备，或者上述内容的任何合适组合。机器可读存储介质的更具体示例会包括基于一个或多个线的电气连接、便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦除可 编程只读存储器(EPROM或快闪存储器)、光纤、便捷式紧凑盘只读存储器(CD-ROM)、光学储存设备、磁储存设备、或上述内容的任何合适组合。
根据本公开的一个或多个实施例，示例1提供了一种投放结果的确定方法，所述方法包括：获取目标对象对应的多个对象特征；在所述多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征；将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果；所述目标投放结果用于表征所述目标对象的用户接受度。
根据本公开的一个或多个实施例，示例2提供了示例1的方法，所述根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征包括：将所述特征类型为第一指定类型的对象特征，作为第一候选对象特征；从第一候选对象特征中确定第二候选对象特征；根据所述第二候选对象特征和第三候选对象特征，确定所述目标对象特征；其中，所述第三候选对象特征为所述多个对象特征中除所述第一候选对象特征以外的对象特征。
根据本公开的一个或多个实施例，示例3提供了示例2的方法，所述从第一候选对象特征中确定第二候选对象特征包括：对所述第一候选对象特征进行去重处理；将去重处理后的第一候选对象特征作为所述第二候选对象特征。
根据本公开的一个或多个实施例，示例4提供了示例2的方法，所述从第一候选对象特征中确定第二候选对象特征包括：确定每两个第一候选对象特征之间的相似度；在所述相似度大于或等于预设相似度阈值的情况下，将所述相似度对应的两个第一候选对象特征中的任意一个作为第四候选对象特征；将所述第四候选对象特征和所述相似度小于所述预设相似度阈值的第一候选对象特征，作为所述第二候选对象特征。
根据本公开的一个或多个实施例，示例5提供了示例2的方法，所述从 第一候选对象特征中确定第二候选对象特征包括：将特征类型为第二指定类型的第一候选对象特征，作为第五候选对象特征；对所述第五候选对象特征进行采样处理；将采样处理后的第五候选对象特征和第六候选对象特征作为所述第二候选对象特征；其中，所述第六候选对象特征为所述第一候选对象特征中除所述第五候选对象特征以外的对象特征。
根据本公开的一个或多个实施例，示例6提供了示例5的方法，所述对所述第五候选对象特征进行采样处理包括：确定所述第一候选对象特征的第二特征总数和所述第五候选对象特征的第三特征总数；根据所述预设数量阈值、所述第二特征总数和所述第三特征总数，确定目标特征数量；根据所述目标特征数量，对所述第五候选对象特征进行采样处理。
根据本公开的一个或多个实施例，示例7提供了示例1的方法，所述根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征包括：将所述特征类型为第三指定类型的对象特征，作为第七候选对象特征；获取所述第七候选对象特征对应的预设优先级；根据所述预设优先级，从所述第七候选对象特征中确定第八候选对象特征；根据所述第八候选对象特征和第九候选对象特征，确定所述目标对象特征；其中，所述第九候选对象特征为所述多个对象特征中除所述第七候选对象特征以外的对象特征。
根据本公开的一个或多个实施例，示例8提供了示例1的方法，所述预测模型是通过以下方式预先生成的：获取训练样本集，所述训练样本集中包含样本对象对应的样本对象特征以及所述样本对象对应的真实投放结果；所述真实投放结果用于表征所述样本对象投放后的用户接受度；根据预设损失函数、所述样本对象特征以及所述真实投放结果对预设模型进行迭代更新，得到所述预测模型。
根据本公开的一个或多个实施例，示例9提供了示例8的方法，所述根据预设损失函数、所述样本对象特征以及所述真实投放结果对预设模型进行 迭代更新，得到所述预测模型包括：循环执行模型迭代步骤，直至根据所述预设损失函数确定更新后的预设模型满足预设停止迭代条件，将更新后的预设模型作为所述预测模型；所述模型迭代步骤包括：将所述样本对象特征输入所述预设模型中，得到所述预设模型输出的所述样本对象对应的样本投放结果，所述样本投放结果用于表征所述样本对象的用户接受度；根据所述预设损失函数，确定所述样本对象对应的所述样本投放结果与所述真实投放结果的目标损失值；所述目标损失值用于表征所述样本投放结果和所述真实投放结果之间的差异程度；在根据所述目标损失值确定所述预设模型不满足所述预设停止迭代条件的情况下，根据所述目标损失值更新所述预设模型的参数，得到新的预设模型。
根据本公开的一个或多个实施例，示例10提供了示例1的方法，所述方法还包括：获取所述目标对象对应的历史投放结果，所述历史投放结果用于表征历史时刻所述目标对象的用户接受度；所述历史投放结果为历史时刻将所述目标对象对应的历史对象特征输入所述预测模型中，得到的所述预测模型输出的所述目标对象对应的投放结果；根据所述历史投放结果，对所述目标投放结果进行校正；将校正后的目标投放结果作为所述目标对象新的目标投放结果。
根据本公开的一个或多个实施例，示例11提供了示例1的方法，所述方法还包括：在所述第一特征总数小于所述预设数量阈值的情况下，将所述多个对象特征作为所述目标对象特征。
根据本公开的一个或多个实施例，示例12提供了示例1至11任一项的方法，所述方法还包括：对所述目标对象特征进行维度转换，以将不同的目标对象特征的向量维度转换为相同的向量维度；所述将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果包括：将维度转换后的目标对象特征输入预测模型中，得到所述预测模型输 出的所述目标对象对应的目标投放结果。
根据本公开的一个或多个实施例，示例13提供了示例1至11任一项的方法，所述方法还包括：根据所述目标对象特征的特征类型，确定所述目标对象特征对应的目标特征类型编码；所述将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果包括：将所述目标对象特征与所述目标特征类型编码相加后输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果。
根据本公开的一个或多个实施例，示例14提供了一种投放结果的确定装置，所述装置包括：第一获取模块，用于获取目标对象对应的多个对象特征；第一确定模块，用于在所述多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征；预测模块，用于将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果；所述目标投放结果用于表征所述目标对象的用户接受度。
根据本公开的一个或多个实施例，示例15提供了一种计算机可读介质，其上存储有计算机程序，该程序被处理装置执行时实现示例1至13中任一项示例所述方法的步骤。
根据本公开的一个或多个实施例，示例16提供了一种电子设备，包括：存储装置，其上存储有计算机程序；处理装置，用于执行所述存储装置中的所述计算机程序，以实现示例1至13中任一项示例所述方法的步骤。
根据本公开的一个或多个实施例，示例17提供了一种计算机程序产品，包含指令，该指令在由处理器执行时使得处理器实现示例1至13中任一项示例所述方法的步骤。
根据本公开的一个或多个实施例，示例18提供了一种计算机程序，包括程序代码，该程序代码在由处理器执行时导致实现示例1至13中任一项 示例所述方法的步骤。
以上描述仅为本公开的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，本公开中所涉及的公开范围，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公开构思的情况下，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。
此外，虽然采用特定次序描绘了各操作，但是这不应当理解为要求这些操作以所示出的特定次序或以顺序次序执行来执行。在一定环境下，多任务和并行处理可能是有利的。同样地，虽然在上面论述中包含了若干具体实现细节，但是这些不应当被解释为对本公开的范围的限制。在单独的实施例的上下文中描述的某些特征还可以组合地实现在单个实施例中。相反地，在单个实施例的上下文中描述的各种特征也可以单独地或以任何合适的子组合的方式实现在多个实施例中。
尽管已经采用特定于结构特征和/或方法逻辑动作的语言描述了本主题，但是应当理解所附权利要求书中所限定的主题未必局限于上面描述的特定特征或动作。相反，上面所描述的特定特征和动作仅仅是实现权利要求书的示例形式。关于上述实施例中的装置，其中各个模块执行操作的具体方式已经在有关该方法的实施例中进行了详细描述，此处将不做详细阐述说明。

Claims (28)

  1. 一种投放结果的确定方法，所述方法包括：
    获取目标对象对应的多个对象特征；
    在所述多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征；
    将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果；所述目标投放结果用于表征所述目标对象的用户接受度。
  2. 根据权利要求1所述的方法，其中，所述根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征包括：
    将所述特征类型为第一指定类型的对象特征，作为第一候选对象特征；
    从第一候选对象特征中确定第二候选对象特征；
    根据所述第二候选对象特征和第三候选对象特征，确定所述目标对象特征；
    其中，所述第三候选对象特征为所述多个对象特征中除所述第一候选对象特征以外的对象特征。
  3. 根据权利要求2所述的方法，其中，所述从第一候选对象特征中确定第二候选对象特征包括：
    对所述第一候选对象特征进行去重处理；
    将去重处理后的第一候选对象特征作为所述第二候选对象特征。
  4. 根据权利要求2所述的方法，其中，所述从第一候选对象特征中确 定第二候选对象特征包括：
    确定每两个第一候选对象特征之间的相似度；
    在所述相似度大于或等于预设相似度阈值的情况下，将所述相似度对应的两个第一候选对象特征中的任意一个作为第四候选对象特征；
    将所述第四候选对象特征和所述相似度小于所述预设相似度阈值的第一候选对象特征，作为所述第二候选对象特征。
  5. 根据权利要求2所述的方法，其中，所述从第一候选对象特征中确定第二候选对象特征包括：
    将特征类型为第二指定类型的第一候选对象特征，作为第五候选对象特征；
    对所述第五候选对象特征进行采样处理；
    将采样处理后的第五候选对象特征和第六候选对象特征作为所述第二候选对象特征；
    其中，所述第六候选对象特征为所述第一候选对象特征中除所述第五候选对象特征以外的对象特征。
  6. 根据权利要求5所述的方法，其中，所述对所述第五候选对象特征进行采样处理包括：
    确定所述第一候选对象特征的第二特征总数和所述第五候选对象特征的第三特征总数；
    根据所述预设数量阈值、所述第二特征总数和所述第三特征总数，确定目标特征数量；
    根据所述目标特征数量，对所述第五候选对象特征进行采样处理。
  7. 根据权利要求1所述的方法，其中，所述根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征包括：
    将所述特征类型为第三指定类型的对象特征，作为第七候选对象特征；
    获取所述第七候选对象特征对应的预设优先级；
    根据所述预设优先级，从所述第七候选对象特征中确定第八候选对象特征；
    根据所述第八候选对象特征和第九候选对象特征，确定所述目标对象特征；
    其中，所述第九候选对象特征为所述多个对象特征中除所述第七候选对象特征以外的对象特征。
  8. 根据权利要求1所述的方法，其中，所述预测模型是通过以下方式预先生成的：
    获取训练样本集，所述训练样本集中包含样本对象对应的样本对象特征以及所述样本对象对应的真实投放结果；所述真实投放结果用于表征所述样本对象投放后的用户接受度；
    根据预设损失函数、所述样本对象特征以及所述真实投放结果对预设模型进行迭代更新，得到所述预测模型。
  9. 根据权利要求8所述的方法，其中，所述根据预设损失函数、所述样本对象特征以及所述真实投放结果对预设模型进行迭代更新，得到所述预测模型包括：
    循环执行模型迭代步骤，直至根据所述预设损失函数确定更新后的预设模型满足预设停止迭代条件，将更新后的预设模型作为所述预测模型；
    所述模型迭代步骤包括：
    将所述样本对象特征输入所述预设模型中，得到所述预设模型输出的所述样本对象对应的样本投放结果，所述样本投放结果用于表征所述样本对象的用户接受度；
    根据所述预设损失函数，确定所述样本对象对应的所述样本投放结果与所述真实投放结果的目标损失值；所述目标损失值用于表征所述样本投放结果和所述真实投放结果之间的差异程度；
    在根据所述目标损失值确定所述预设模型不满足所述预设停止迭代条件的情况下，根据所述目标损失值更新所述预设模型的参数，得到新的预设模型。
  10. 根据权利要求1所述的方法，其中，所述方法还包括：
    获取所述目标对象对应的历史投放结果，所述历史投放结果用于表征历史时刻所述目标对象的用户接受度；所述历史投放结果为历史时刻将所述目标对象对应的历史对象特征输入所述预测模型中，得到的所述预测模型输出的所述目标对象对应的投放结果；
    根据所述历史投放结果，对所述目标投放结果进行校正；
    将校正后的目标投放结果作为所述目标对象新的目标投放结果。
  11. 根据权利要求1所述的方法，其中，所述方法还包括：
    在所述第一特征总数小于所述预设数量阈值的情况下，将所述多个对象特征作为所述目标对象特征。
  12. 根据权利要求1至11中任一项所述的方法，其中，所述方法还包括：
    对所述目标对象特征进行维度转换，以将不同的目标对象特征的向量维 度转换为相同的向量维度；
    所述将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果包括：
    将维度转换后的目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果。
  13. 根据权利要求1至11中任一项所述的方法，其中，所述方法还包括：
    根据所述目标对象特征的特征类型，确定所述目标对象特征对应的目标特征类型编码；
    所述将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果包括：
    将所述目标对象特征与所述目标特征类型编码相加后输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果。
  14. 一种投放结果的确定装置，其中，所述装置包括：
    第一获取模块，被配置为获取目标对象对应的多个对象特征；
    第一确定模块，被配置为在所述多个对象特征的第一特征总数大于或等于预设数量阈值的情况下，根据所述多个对象特征对应的特征类型，从所述多个对象特征中确定目标对象特征；
    预测模块，被配置为将所述目标对象特征输入预测模型中，得到所述预测模型输出的所述目标对象对应的目标投放结果；所述目标投放结果用于表征所述目标对象的用户接受度。
  15. 根据权利要求14所述的装置，其中，所述第一确定模块进一步被 配置为：
    将该特征类型为第一指定类型的对象特征，作为第一候选对象特征；从第一候选对象特征中确定第二候选对象特征；根据该第二候选对象特征和第三候选对象特征，确定该目标对象特征；其中，该第三候选对象特征为多个对象特征中除该第一候选对象特征以外的对象特征。
  16. 根据权利要求15所述的装置，其中，所述第一确定模块进一步被配置为：对该第一候选对象特征进行去重处理；将去重处理后的第一候选对象特征作为该第二候选对象特征。
  17. 根据权利要求14或15所述的装置，其中，所述第一确定模块进一步被配置为：确定每两个第一候选对象特征之间的相似度；在该相似度大于或等于预设相似度阈值的情况下，将该相似度对应的两个第一候选对象特征中的任意一个作为第四候选对象特征；将该第四候选对象特征和该相似度小于该预设相似度阈值的第一候选对象特征，作为该第二候选对象特征。
  18. 根据权利要求14所述的装置，其中，所述第一确定模块进一步被配置为：将特征类型为第二指定类型的第一候选对象特征，作为第五候选对象特征；对该第五候选对象特征进行采样处理；将采样处理后的第五候选对象特征和第六候选对象特征作为该第二候选对象特征；其中，该第六候选对象特征为该第一候选对象特征中除该第五候选对象特征以外的对象特征。
  19. 根据权利要求18所述的装置，其中，所述第一确定模块进一步被配置为：确定该第一候选对象特征的第二特征总数和该第五候选对象特征的第三特征总数；根据该预设数量阈值、该第二特征总数和该第三特征总数， 确定目标特征数量；根据该目标特征数量，对该第五候选对象特征进行采样处理。
  20. 根据权利要求14所述的装置，其中，所述第一确定模块进一步被配置为：将该特征类型为第三指定类型的对象特征，作为第七候选对象特征；获取该第七候选对象特征对应的预设优先级；根据该预设优先级，从该第七候选对象特征中确定第八候选对象特征；根据该第八候选对象特征和第九候选对象特征，确定该目标对象特征；其中，该第九候选对象特征为多个对象特征中除该第七候选对象特征以外的对象特征。
  21. 根据权利要求14所述的装置，其中，所述装置还包括：
    第二获取模块，被配置为获取该目标对象对应的历史投放结果，该历史投放结果用于表征历史时刻该目标对象的用户接受度；该历史投放结果为历史时刻将该目标对象对应的历史对象特征输入该预测模型中，得到的该预测模型输出的该目标对象对应的投放结果；
    校正模块，被配置为根据该历史投放结果，对该目标投放结果进行校正；
    第二确定模块，被配置为将校正后的目标投放结果作为该目标对象新的目标投放结果。
  22. 根据权利要求14所述的装置，其中，第一确定模块还被配置为：
    在该第一特征总数小于该预设数量阈值的情况下，将多个对象特征作为该目标对象特征。
  23. 根据权利要求14所述的装置，其中，所述装置还包括：
    转换模块，被配置为对该目标对象特征进行维度转换，以将不同的目标 对象特征的向量维度转换为相同的向量维度；
    预测模块，被配置为将维度转换后的目标对象特征输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果。
  24. 根据权利要求14所述的装置，其中，所述装置还包括：
    第三确定模块，被配置为根据该目标对象特征的特征类型，确定该目标对象特征对应的目标特征类型编码；
    预测模块，被配置为将该目标对象特征与该目标特征类型编码相加后输入预测模型中，得到该预测模型输出的该目标对象对应的目标投放结果。
  25. 一种计算机可读介质，其上存储有计算机程序和/或指令，其中，该计算机程序和/或指令被处理装置执行时使得所述处理装置实现根据权利要求1至13中任一项所述方法。
  26. 一种电子设备，其中，包括：
    存储装置，其上存储有计算机程序和/或指令；
    处理装置，被配置为执行所述存储装置中的所述计算机程序和/或指令，以实现根据权利要求1至13中任一项所述方法。
  27. 一种计算机程序产品，包含指令，该指令在由处理器执行时使得处理器实现根据权利要求1到13中任一项所述的方法。
  28. 一种计算机程序，包括程序代码，该程序代码在由处理器执行时导致实现根据权利要求1到13中任一项所述的方法。
PCT/CN2024/096290 2023-06-08 2024-05-30 投放结果的确定方法、装置、可读介质和电子设备 Ceased WO2024251023A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202310678334.3 2023-06-08
CN202310678334.3A CN116757752A (zh) 2023-06-08 2023-06-08 投放结果的确定方法、装置、可读介质和电子设备

Publications (1)

Publication Number Publication Date
WO2024251023A1 true WO2024251023A1 (zh) 2024-12-12

Family

ID=87952611

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/096290 Ceased WO2024251023A1 (zh) 2023-06-08 2024-05-30 投放结果的确定方法、装置、可读介质和电子设备

Country Status (2)

Country Link
CN (1) CN116757752A (zh)
WO (1) WO2024251023A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120144586A (zh) * 2025-03-03 2025-06-13 每日互动股份有限公司 一种确定目标区域异常用电状态的数据处理系统

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116757752A (zh) * 2023-06-08 2023-09-15 北京有竹居网络技术有限公司 投放结果的确定方法、装置、可读介质和电子设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104536983A (zh) * 2014-12-08 2015-04-22 北京掌阔技术有限公司 一种预测广告点击率的方法和装置
CN113256335A (zh) * 2021-05-27 2021-08-13 腾讯科技（深圳）有限公司 数据筛选方法、多媒体数据的投放效果预测方法及装置
CN114330578A (zh) * 2021-12-31 2022-04-12 北京字跳网络技术有限公司 对象分类方法、装置、可读介质及电子设备
CN116757752A (zh) * 2023-06-08 2023-09-15 北京有竹居网络技术有限公司 投放结果的确定方法、装置、可读介质和电子设备

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118043802B (zh) * 2021-09-29 2025-11-21 华为技术有限公司 一种推荐模型训练方法及装置
CN115601088A (zh) * 2022-10-25 2023-01-13 杭州网易云音乐科技有限公司（Cn） 对象投放方法、装置、存储介质及电子设备
CN116089873A (zh) * 2023-02-10 2023-05-09 北京百度网讯科技有限公司 模型训练方法、数据分类分级方法、装置、设备及介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104536983A (zh) * 2014-12-08 2015-04-22 北京掌阔技术有限公司 一种预测广告点击率的方法和装置
CN113256335A (zh) * 2021-05-27 2021-08-13 腾讯科技（深圳）有限公司 数据筛选方法、多媒体数据的投放效果预测方法及装置
CN114330578A (zh) * 2021-12-31 2022-04-12 北京字跳网络技术有限公司 对象分类方法、装置、可读介质及电子设备
CN116757752A (zh) * 2023-06-08 2023-09-15 北京有竹居网络技术有限公司 投放结果的确定方法、装置、可读介质和电子设备

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120144586A (zh) * 2025-03-03 2025-06-13 每日互动股份有限公司 一种确定目标区域异常用电状态的数据处理系统
CN120144586B (zh) * 2025-03-03 2025-11-11 每日互动股份有限公司 一种确定目标区域异常用电状态的数据处理系统

Also Published As

Publication number Publication date
CN116757752A (zh) 2023-09-15

Similar Documents

Publication Publication Date Title
US11068658B2 (en) Dynamic word embeddings
CN111061956B (zh) 用于生成信息的方法和装置
WO2024251023A1 (zh) 投放结果的确定方法、装置、可读介质和电子设备
CN111782968A (zh) 内容推荐方法、装置、可读介质及电子设备
WO2022188534A1 (zh) 信息推送的方法和装置
CN110688528A (zh) 生成视频的分类信息的方法、装置、电子设备和介质
CN111783810A (zh) 用于确定用户的属性信息的方法和装置
WO2024099171A1 (zh) 视频生成方法和装置
CN113159453B (zh) 资源数据预测方法、装置、设备及存储介质
CN113033680B (zh) 视频分类方法、装置、可读介质及电子设备
CN118228200B (zh) 基于多模态模型的电力设备异常识别方法、装置与设备
CN110866040B (zh) 用户画像生成方法、装置和系统
WO2024198920A1 (zh) 内容推送模型的训练方法、装置、设备及存储介质
WO2024109860A1 (zh) 一种交互方法、装置、电子设备、计算机可读介质
CN115062119B (zh) 政务事件办理推荐方法、装置
CN116992115A (zh) 推荐方法、装置、存储介质及电子设备
CN111767290B (zh) 用于更新用户画像的方法和装置
WO2024235211A1 (zh) 转场位置确定方法、装置、介质和电子设备
CN109857838B (zh) 用于生成信息的方法和装置
CN116800834B (zh) 虚拟礼物合并方法、装置、电子设备和计算机可读介质
CN118293945A (zh) 一种导航方法、装置、设备、介质、产品
CN113742593B (zh) 用于推送信息的方法和装置
CN116737917A (zh) 文本推荐信息生成方法、电子设备和计算机可读介质
CN112101399B (zh) 一种数据处理方法和装置
CN113177174B (zh) 特征构建方法、内容显示方法及相关装置

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: 24818555

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE