Machine learning for digital image selection across object variations
Through machine learning technology, the features and user preferences are extracted from multiple digital images, the prediction scores are generated, and the most suitable digital image is selected, which solves the problem of changes in the object depiction method in the digital image, and improves the accuracy of recommendations and resource utilization efficiency.
Patent Information
- Application Number
- CN202011260870.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-01-28
- Filing Date
- 2020-11-12
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2040-11-12
AI Technical Summary
The prior art is difficult to deal with changes in the way objects are depicted in digital images, resulting in difficult identification of user preference differences, conventional systems cannot accurately recommend digital images of interest, and computational and hardware resource usage efficiency is inefficient.
Using machine learning technology, extract features from multiple digital images, generate predictive scores based on user preferences, select digital images that are most likely to produce the desired result, process image metadata and user profiles through convolutional neural networks, and train individual machine learning Model to process multiple digital images of objects.
It improves the accuracy of digital image selection, can handle multiple object presentation methods, avoids cold start problems and waste of computing resources in conventional technologies, and achieves a more efficient user experience.
Smart Images

Figure CN113255713B_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present disclosure relate to depicting objects in digital images, and more particularly to machine learning for digital image selection across object variations. Background Art
[0002] The way an object is depicted in a digital image is one of the primary ways used to drive user interest in that object. For example, digital images may be configured to follow popular style trends, themes of popular television shows, etc. In such instances, the characteristics of the object itself may remain the same (e.g., color, shape), but the way the object is depicted in different digital images changes. Therefore, one challenge for the service provider system in determining possible digital images of interest involves the way the object is depicted in the digital image.
[0003] This challenge is even more serious because user preferences can change significantly between preferences for the way objects are depicted. In fact, it has been shown that each user has its own choices and affinities for the way objects are depicted. For example, a first user may have preferences related to a favorite television program (e.g., objects are shown in a mid-century modern setting), while a second user may prefer to view objects in a neutral setting, such as on a white background for the clarity of the color of the object. Therefore, it is difficult, if not impossible, for humans to determine which preferences are associated with each user, especially in the face of potentially millions of users who can access digital content with digital images (e.g., web pages that provide services in real time). Conventional service provider systems also cannot handle how to depict objects in digital images. This is because conventional techniques rely on identifiers of digital images as a whole, and therefore, these techniques cannot address the actual visual characteristics of the depicted objects, nor the way those characteristics are related to other digital images. Therefore, conventional service provider systems may be inaccurate and may result in inefficient use of computing and hardware resources in order to recommend digital images of interest. Summary of the invention
[0004] Digital image selection techniques are described that employ machine learning to select a digital image of an object from a plurality of digital images that relate to variations in the way the object is depicted. For example, the plurality of digital images may capture an object but differ in the way the object is depicted in the digital images, such as different models wearing the same clothing, backgrounds, etc. The likelihood of a result (e.g., conversion of a good or service depicted by the object) is then increased by processing user preferences relating to these variations to select the digital images for inclusion in digital content.
[0005] In one example, a digital image selection technique is described that uses machine learning to select a digital image of an object from a plurality of digital images involving changes to the object. First, a user ID is received by a service provider system as part of a request to obtain digital content (e.g., a web page). Then, a user profile is obtained by the service provider system based on the user ID. The service provider system also selects a digital image from a plurality of digital images having changes to the object to include it as part of the digital content.
[0006] To this end, image metadata including features extracted from digital images (e.g., using a convolutional neural network) is used along with a user profile to generate a prediction score for each of a plurality of digital images having variations. The digital image indicated as being most likely to produce a desired result (e.g., a conversion) is then selected by the system for inclusion as part of digital content (e.g., a web page). As a result, the service provider system can select a digital image of an object from a plurality of digital images of the object that has an increased likelihood of obtaining a desired result, and can handle many different ways in which the object can be presented to the user.
[0007] This Summary introduces a selection of concepts in a simplified form that are further described below in the Detailed Description. As such, this Summary is neither intended to identify key features of the claimed subject matter nor to serve as an aid in determining the scope of the claimed subject matter. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] The detailed description is described with reference to the accompanying drawings. Entities shown in the drawings may refer to one or more entities, and thus singular or plural entities may be interchangeably referenced in the discussion.
[0009] Figure 1 is an illustration of an environment in an example implementation that is operable to employ the machine learning digital image selection techniques described herein.
[0010] Figure 2 Depicted are systems in an example implementation that generate training data to train a machine learning model.
[0011] Figure 3 is a flowchart depicting a process in an example implementation that uses exploration / exploitation techniques to generate training data and trains a machine learning model to select digital images from multiple digital images of an object to be used to generate digital content.
[0012] Figure 4 Describes the use Figure 2 The system in the example implementation uses training data to train a machine learning model.
[0013] Figure 5 is a flowchart depicting a process in an example implementation using training data having interaction events, the training data for interaction events comprising: a user profile; image metadata comprising image features extracted from a digital image; and result data for training a machine learning model.
[0014] Figure 6 Describes the use Figure 4 A system in an example implementation in which a machine learning model trained in selects a digital image from multiple digital images of an object.
[0015] Figure 7 is a flow diagram depicting a process in an example implementation for generating digital content having digital images selected by a machine learning model based on a user profile and image metadata.
[0016] Figure 8 An example system including various components of an example device is illustrated, which can be implemented as described in reference Figures 1 to 7 Any type of computing device described and / or used to implement embodiments of the techniques described herein. DETAILED DESCRIPTION
[0017] Overview
[0018] It has been observed that in real-world scenarios, even in instances where the visual characteristics of the object itself remain unchanged, each user has different affinities for different aspects of changes in the way an object is depicted in a digital image. However, conventional techniques and systems for selecting digital images so that they are included as part of digital content do not handle these changes, such as a specific view of an object to be included in a web page, a thumbnail to be used to represent a digital video, a model for wearing a piece of clothing, etc. On the contrary, conventional techniques rely on a general approach when dealing with specific objects. In addition, conventional techniques typically train a dedicated machine learning model for each digital image, and therefore cannot exploit visual similarities with other digital images, and encounter a "cold start" problem as described below. Therefore, this can lead to inefficient use of computing and network resources for providing and receiving digital content (such as web pages, network-enabled applications, etc.) including these digital images.
[0019] Therefore, a digital image selection technique is described that adopts machine learning to select a digital image of an object from a plurality of digital images of an object. A plurality of digital images capture an object separately so as to include it as a part of generating digital content, such as a web page, a thumbnail for representing a digital video, etc. However, a plurality of digital images include changes in the way of depicting an object, and the preference of the way of depicting an object may vary from user to user. In one example, a user ID is received by a service provider system as a part of a request to obtain digital content (such as a web page). Then, a user profile is obtained by the service provider system based on the user ID. For example, a user profile can describe user interactions with a digital content item, a digital image, user demographic information, a place where a digital content request is initiated, etc.
[0020] The service provider system then selects a digital image from the plurality of digital images of the object to include as part of the digital content. For example, the plurality of digital images may be located based on an object ID associated with the requested digital content. In this example, the plurality of digital images each captures an object of interest to be represented in the digital content, but has differences in at least one visual characteristic that support variations in the manner in which the object is depicted. For example, the object may be a piece of clothing in a particular color, but worn by different human models. Other examples of variations are also contemplated, including background characteristics of the scene in which the object is captured, different angles, arrangements, orientations, lighting, etc.
[0021] In one example of digital image selection, a service provider system determines whether to explore or exploit a user behavior associated with a user ID in response to a request as part of selecting a digital image. A determination to explore user behavior involves selecting a digital image in order to learn more about user behavior with respect to an object depicted in the digital image, i.e., user preferences with respect to different depictions of the object. On the other hand, a determination to exploit user behavior is made in order to maximize the likelihood of obtaining a desired result when exposed to the object via the digital image, such as recommending items of interest, conversions, etc.
[0022] Thus, when a determination is made to explore user behavior, the service provider system randomly selects a digital image from a plurality of digital images having an object. When a determination is made to exploit user behavior, the service provider system selects a digital image from a plurality of digital images based on a user profile using a machine learning model, such as a neural network. However, in either instance, training data is generated based on user interactions with the selected digital images to train and / or update the training of the machine learning model, such as to capture current trends.
[0023] For example, the training data may be formed as a plurality of interaction events, at least a portion of which correspond to requests made for digital content. Each interaction event may include a user ID that initiated the request, a user profile associated with the user ID, an image ID of a digital image selected in response to the request, image metadata associated with the digital image, and output data describing a result of exposure of the digital image as part of the digital content. For example, the result may describe a conversion, such as whether a digital image was selected (e.g., as a thumbnail that initiated a digital video), whether it resulted in a purchase of a corresponding good or service corresponding to an object depicted in the digital image, etc.
[0024] Image metadata used as part of the training data and / or for selecting digital images for subsequent requests can support increased functionality over conventional techniques. In conventional systems, because a single model is trained for each image, the image ID is only used to identify the correspondence between the digital image, the user ID, and the obtained results. Therefore, the image ID does not describe the visual characteristics extracted directly from the digital image and the changes in those visual characteristics, and thus cannot support the determination of the similarity between one digital image and another digital image. Therefore, conventional techniques suffer from the "cold start" problem, and predictions about digital images are not accurate enough until a sufficient amount of training data is received (usually within a few weeks). In order to collect this data, it is resource-intensive and computationally intensive, and this leads to user frustration caused by inaccurate predictions.
[0025] However, in the techniques described herein, image metadata used to train machine learning models utilizes image features extracted from corresponding digital images, such as vectors generated by feature extraction using a neural network. In this way, image metadata describes what is visually contained within a digital image, and can describe that content with improved accuracy compared to other techniques, such as image tagging that relies on the user's ability to express and manually tag what is contained in a digital image. By mapping features extracted from a digital image to a feature space using a machine learning model, the visual similarity of a digital image to other digital images can be determined, and the visual similarity can be used to avoid the cold start problem of conventional techniques and to handle changes in object depiction. This is used to improve the operation of a computing device that implements these techniques and to improve the accuracy of the computing device.
[0026] In order to map features and user profiles to feature space, the machine learning model is trained and updated using user profiles and image metadata (e.g., image features extracted from digital images using a neural network) included in corresponding interaction events in the training data. For example, the service provider system can process the user profile and image metadata as part of the machine learning together with the loss function using the corresponding result data. In this way, a single machine learning model is trained to generate a prediction score for each combination of a subsequent user profile and image data extracted from the corresponding digital image of the object. This overcomes the limitations of conventional techniques that only generate a dedicated machine learning model for each digital image, and therefore do not support similarities between digital images, and therefore suffer from the cold start problem as previously described.
[0027] Continuing with the above example, to select a digital image of an object from a plurality of digital images of the object in response to an exploitation determination, the service provider system generates a prediction score for each digital image (e.g., retrieved based on the object ID) using the associated image metadata and user profile corresponding to the user ID associated with the request. The digital image indicated as being most likely to produce a desired result (e.g., a conversion) is then selected by the system for inclusion as part of the digital content (e.g., a web page).
[0028] The digital image is then used by the service provider system (whether randomly selected as part of the exploration or based on a machine learning model as part of the development) to generate digital content (e.g., a web page) that is transmitted back to the originator of the request. As a result, the service provider system can select a digital image of an object from multiple digital images of the object that has an increased likelihood of obtaining the desired result, and can handle many different ways that the object can be presented to the user. Due to the many differences in user affinity for different visual characteristics that cannot be detected by humans, this is impossible to perform by humans alone. Further discussion of these and other examples is included in the following sections and is shown in the corresponding figures.
[0029] Terminology Examples
[0030] "Digital content" includes any type of data that can be presented by a computing device. Examples of digital content include web pages, digital video, digital media, digital audio, digital images, user interfaces, etc.
[0031] A "neural network" generally includes a series of layers modeled as having nodes (ie, neurons) and connections between neurons that process data to obtain an output (such as classifying an input as exhibiting or not exhibiting a particular characteristic). An example of a neural network is a convolutional neural network.
[0032] A "loss function" is a function that maps the value of one or more explanatory variables (e.g., features) to a real number representing the cost associated with an event, and in optimization design, the loss function is minimized in order to train a machine learning model. In classification, for example, the loss function is a penalty for incorrect classification, such as whether the outcome described in the output data did or did not occur.
[0033] "Explore / Exploit" is used to determine whether to explore or exploit user behavior. A determination to explore user behavior involves selecting digital images in order to learn more about user behavior regarding objects depicted in digital images, such as user preferences for how objects are depicted. On the other hand, a determination to exploit user behavior is made to maximize the likelihood of obtaining a desired result when exposed to an object via a digital image, such as recommending items of interest, conversions, etc.
[0034] A "conversion" may correspond to a variety of actions. Examples of such actions include whether an interaction with a digital image occurs (e.g., hovering or "clicking"), whether a corresponding product (e.g., an object) or service is added to a shopping cart, whether a corresponding product or service is purchased, selection of a thumbnail that launches a digital video or digital audio, etc.
[0035] In the following discussion, an example environment is described in which the techniques described herein may be employed. Example procedures are also described that may be performed in the example environment as well as other environments. Therefore, the execution of the example procedures is not limited to the example environment, and the example environment is not limited to the execution of the example procedures.
[0036] Example Environment
[0037] Figure 1 1 is an illustration of a digital media environment 100 in an example implementation operable to employ the machine learning and digital image selection techniques described herein that support changes in the manner in which objects are depicted within digital images. The illustrated environment 100 includes a service provider system 102 and a client device 104 communicatively coupled via a network 106 (e.g., the Internet). The computing devices implementing the service provider system 102 and the client device 104 can be configured in a variety of ways.
[0038] For example, the computing device may be configured as a desktop computer, a laptop computer, a mobile device (e.g., assuming a handheld configuration, such as a tablet or mobile phone as described for client device 104), etc. Thus, computing devices may range from full-resource devices with large amounts of memory and processor resources (e.g., personal computers, game consoles) to low-resource devices with limited memory and / or processing resources (e.g., mobile devices). Additionally, although a single computing device is shown and described in certain examples, the computing device may represent a plurality of different devices, such as those used by an enterprise to perform tasks such as in Figure 8 Multiple servers operating "on the cloud" as described in .
[0039] The client device 104 includes a communication module 108 (e.g., a web browser, a network-enabled application, etc.) configured to form a request 110 for transmission to a digital content generation system 112 via a network 106. In response, the digital content generation system 112 generates digital content 114 for transmission to and presentation in a user interface 116 via a display device 118 of the client device 104. The digital content 114 may be configured in a variety of ways, such as a web page, a user interface screen, a digital video, etc.
[0040] As part of generating the digital content 114, the digital content generation system employs a digital image selection module 120 to select digital images for inclusion as part of the digital content 114. For example, a digital image 122(n) may be selected from a plurality of digital images 122(1), 122(2), ..., 122(N) stored by a storage device 124, each of the plurality of digital images depicting an object but having a visual difference in the manner in which the object is depicted.
[0041] In the illustrated example, the digital images 122(1)-122(N) each include a pair of shoes, but are captured from different perspectives, have different arrangements, etc. Thus, in this example, the visual characteristics of the object itself remain the same (e.g., color), but the manner in which the object is depicted is different. Other examples are also contemplated, such as differences in the background of the digital images, differences in the objects themselves, a human model displayed with the objects, etc. Although the digital image selection module 120 is illustrated as being implemented at the service provider system 102, the functionality of the digital image selection module 120 may also be implemented in whole or in part at the client device 104.
[0042] To select the digital image, the digital image selection module 120 employs a machine learning module 126 implementing a machine learning model 128 to select the digital image that is most likely to achieve a desired outcome. For example, the outcome may include a conversion, such as whether a corresponding good or service associated with an object depicted in the digital image is purchased, whether the digital image is selected (i.e., a "click" to initiate output of corresponding digital content in a digital video scene), etc.
[0043] For example, the digital image selection module 120 may receive a request 110 for digital content 114 and obtain digital images 122(1)-122(N) associated with an object ID 130 associated with the digital content 114. The machine learning model 128 then calculates a probability score. The probability score is calculated for each of the plurality of digital images 122(1)-122(N) based on the visual characteristics of the digital image and a user profile associated with the user ID received as part of the request. The probability score indicates the probability of obtaining a desired result.
[0044] In order to determine which visual characteristics are included in the corresponding digital image, image features are extracted from the corresponding digital image. For example, the image features can be extracted using an embedding layer of a neural network (e.g., a convolutional neural network) to map the digital image to a lower dimensional embedding space. In this way, a single machine learning model 128 can be used for multiple different digital images, and thus handle visual similarities between these digital images. This is not possible in conventional techniques that use a single dedicated model for each digital image.
[0045] As a result, the techniques described herein overcome challenges of conventional techniques and improve the operation of computing devices that implement these techniques. The first such example is known as a counterfactual problem. Assume that data indicating a user selection (e.g., a transformation) of a digital image is received, which does not indicate how the user will interact with other digital images. To avoid this problem, conventional techniques use a separate dedicated machine learning model for each digital image. The problem with this approach is that since each digital image is associated with a single machine learning model, it is impossible to learn patterns in the digital images. This adversely affects the accuracy of the machine learning model and, therefore, the operation of the corresponding computing device.
[0046] However, in the technology described herein, a system is described that is configured to process multiple digital images of an object using a single machine learning model. As part of this, image features are extracted from the corresponding digital images using an embedding layer of a neural network to map the images as vectors to a lower dimensional space. In this way, the proximity of the vectors within the embedding space is a measure of the visual similarity between the digital images, and therefore, the machine learning model 128 can implement a decision strategy on the digital images of the object, which results in improved accuracy in making predictions.
[0047] Furthermore, because conventional techniques train a separate machine learning model for each digital image, these conventional techniques are not inherently based on image content, but rather are based only on image IDs. In contrast, the techniques described herein may be used to train a single machine learning model 128 that learns image features extracted from multiple digital images 122(1) to 122(N). This may be performed, for example, by using a pre-trained convolutional neural network to extract image features as embeddings learned by the last layer of the network. This enables the machine learning model 128 to learn patterns in digital images that are not possible with conventional techniques.
[0048] In addition, conventional techniques suffer from a cold start problem. In conventional techniques, when a new digital image is added to support personalization of digital content, it may take several weeks until the digital image is ready for accurate personalization. This is because a separate machine learning model is trained for each digital image, and therefore, when a new digital image is added, it takes about two weeks in real-world implementations to collect sufficient training data (e.g., "click" data) through exploration techniques for digital images. This is a significant challenge even when the new digital image is a minor variation of an existing digital image.
[0049] However, in the technology described herein, this problem is solved in a number of ways. First, by using an embedding layer, the machine learning model 128 maps image identifiers to a low-dimensional vector space. Therefore, when a new digital image is added, a few training examples are sufficient to map the digital image to a vector in the embedding space of the machine learning model 128. Therefore, the patterns learned for digital images mapped to vectors close to the vectors of the new digital image can now be used by the machine learning model 128 to control the distribution, for example as part of digital content, without having to wait for weeks as would be required using conventional techniques. In addition, by using image metadata, the machine learning model 128 can utilize patterns learned from other digital images with similar image metadata (e.g., shape, color, etc.). Further discussion of these examples and other examples is included in the following sections and is shown in the corresponding figures.
[0050] In general, the functionality, features, and concepts described with respect to the above examples and the following examples can be adopted in the context of the example processes described in this chapter. In addition, the functionality, features, and concepts described with respect to the different figures and examples in this document can be interchangeable with each other, and are not limited to the implementation in the context of a specific figure or process. Moreover, the frames associated with the different representative processes and corresponding figures in this article can be applied together and / or combined in different ways. Therefore, the individual functionality, features, and concepts described with respect to the different example environments, devices, components, figures, and processes in this article can be used in any suitable combination, and are not limited to the specific combinations represented by the examples listed in this specification.
[0051] Training data generation
[0052] Figure 2 A system 200 in an example implementation for generating training data to train a machine learning model 128 is depicted. Figure 3 Depicted is a process 300 in an example implementation that uses an exploration / exploitation technique to generate training data and train a machine learning model 128 to select digital images from a plurality of digital images depicting variations of an object to be used to generate digital content.
[0053] The following discussion describes techniques that can be implemented using the previously described systems and devices. Aspects of each of the processes can be implemented in hardware, firmware, software, or a combination thereof. The processes are shown as a set of blocks that specify operations performed by one or more devices and are not necessarily limited to the order shown for performing the operations by the respective blocks. In portions of the following discussion, reference will be made to Figures 1 to 3 .
[0054] First, in this example, Figure 1 The digital content generation system 112 receives a request 110 for digital content 114 (eg, a web page). The communication module 108 is configured as a web browser of the client device 104 to generate a request 110 for a particular web page. The request 110 includes a user ID 202 associated with a user of the client device 104.
[0055] The user ID 202 is then passed as input to the profile collection module 204 of the digital image selection module 120. The profile collection module 204 is configured to obtain a user profile 206 from a storage device 208 based on the user ID 202, which is associated with the request 110 for digital content including a digital image of an object (box 302). The storage device 208 can be maintained locally at the service provider system 102 and / or remotely by a third-party system or client device 104. The user profile 206 is configured to describe user interactions with digital images, including which digital images are exposed to the corresponding user ID and the results of the exposure, such as conversions. The user profile 206 can also describe characteristics of the corresponding user, such as demographic data (e.g., age, gender) and other information related to the user ID 202, such as a corresponding geographic location, IP address, etc.
[0056] The explore / exploit determination module 210 is then utilized to determine whether to explore or exploit the user behavior associated with the user ID 202 in response to the request 110 (block 304). The determination to explore the user behavior involves selecting the digital image in order to learn more about the user behavior regarding the object depicted in the digital image, e.g., the user's preference for the manner in which the object is depicted. On the other hand, the determination to exploit the user behavior is made in order to maximize the likelihood of obtaining a desired result when exposed to the object via the digital image, e.g., recommending an item of interest, a conversion, etc.
[0057] To make the determination in the illustrated example, the Epsilon-Greedy module 212 is employed by the exploration / exploitation determination module 210. For example, Epsilon may be defined as a value between zero and one, such as 0.1. This value indicates the percentage of user IDs and associated user behaviors to be explored, and therefore indicates the remaining percentage of user IDs and associated user behaviors to be developed. The value of Epsilon may be user-specified, automatically specified without user intervention based on heuristics, etc. The tradeoff between exploration and exploitation allows training data to be generated by the digital image selection module 120, which captures new trends in user behavior and is therefore kept accurate and up-to-date. Other techniques may also be employed by the exploration / exploitation determination module 210 to make the determination.
[0058] Thus, in response to the exploration determination, the exploration module 214 employs the random image selection module 216 to randomly select a digital image from a plurality of digital images that has variations in the way an object is depicted one after another (block 306). On the other hand, in response to the exploitation determination, the exploitation module 218 is operable to select a digital image from a plurality of digital images that has variations in the way an object is depicted using the machine learning model 128 (block 308). In the following discussion, reference is made to Figure 6 and Figure 7 Further discussion of the operation of the machine learning model 128 for selecting digital images of objects is described.
[0059] The selected digital image is included as part of the digital content, and the result of the user interaction with the digital image is transmitted to the training data generation module 220. For example, the selected digital image may capture an object that is an item for sale as part of a web page. Therefore, the result is whether a conversion occurred, which is transmitted to the training data generation module 220. As previously described, the conversion can correspond to various actions, such as whether an interaction with the digital image occurred (e.g., hovering or "clicking"), whether a corresponding product (e.g., object) or service was added to a shopping cart, whether a corresponding product or service was purchased, etc. Without departing from the spirit and scope of the current subject matter, other results are contemplated in addition to conversions, such as launching a corresponding digital video by selecting a digital image as representing a video.
[0060] The training data generation module 220 is then used to generate training data 222 (illustrated as being stored in a storage device 224) that can be used to train the machine learning model 128. To this end, the training data generation module 220 generates an interaction event 226 as corresponding to the request 110. The interaction event 226 includes the user ID 202, the user profile 206 associated with the user ID 202, the image ID 228 of the selected digital image, the image metadata 230, and the result data 232 that describes the result of including the selected digital image as part of the digital content (block 310). For example, the result data 232 can describe whether a result did occur or did not occur, such as a conversion or other action.
[0061] As previously described, the image metadata 230 may include features 234 extracted from the selected digital image using machine learning. This may be stored as part of the training data 222 when generated, or later by taking the digital image corresponding to the image ID 228 and processing the image using feature extraction as described above. The image metadata 230 may also include object metadata 236. The object metadata 236 includes information about the object captured by the digital image (e.g., product category, description, color, size, image tags, etc.), which may be obtained from text associated with the corresponding digital image (e.g., title, tags, description) or elsewhere. As further described in the following sections, the training data 222 is then used to generate a machine learning model (box 312).
[0062] Machine Learning Model Training
[0063] Figure 4 Describes the use Figure 2 System 400 in an example implementation of training data to train a machine learning model 128. Figure 5 Depicted is a process 500 in an example implementation using training data having interaction events, the training data for interaction events comprising: a user profile; image metadata comprising image features extracted from a digital image; and result data for training a machine learning model 128.
[0064] The following discussion describes techniques that can be implemented using the previously described systems and devices. Aspects of each of the processes can be implemented in hardware, firmware, software, or a combination thereof. The processes are shown as a set of blocks that specify operations performed by one or more devices and are not necessarily limited to the order shown for performing the operations by the respective blocks. In portions of the following discussion, reference will be made to Figures 4 to 5 .
[0065] This example discussion continues from the previous section and thus begins with receiving training data 222 by the machine learning module 126 of the digital image selection module 120. The image metadata 230 may also include object metadata 236. The training data 222 includes object metadata 236 having information about the object captured by the digital image, such as product category, description, color, size, image tags, etc.
[0066] Based on observed user interactions with digital images of objects, training data 222 as described in the previous example is collected as a plurality of interaction events 226. As part thereof, each of the interaction events 226 in the training data 222 includes a user profile 206 associated with the user ID 202, image metadata 230 having image features 234 extracted from a corresponding digital image of the plurality of digital images using machine learning, and result data 232 describing the result of including the selected digital image as part of the digital content (block 502). Thus, the object metadata 236 may be characteristics that are specific to the interaction event 226 (e.g., how the object is captured in the digital image) and / or common to the plurality of digital images, such as a color that is shared between the digital images.
[0067] The machine learning module 126 is then employed to generate a machine learning model 128, which is trained based on the object metadata 236, the user profile 206, the image metadata 230, and the loss function 402 based on the outcome data 232 (block 504). The loss function 402 is a function that maps the values of one or more explanatory variables (e.g., features) to a real number representing a cost associated with an event, and in an optimization design, the loss function 402 is minimized in order to train the machine learning model 128. In classification, for example, the loss function 402 is a penalty for incorrect classifications, such as whether the outcome described in the output data 232 did occur or did not occur.
[0068] Thus, the object metadata 236, the user profile 206, and the image features 234 are processed by the embedding layer 404 of the machine learning model 128 to generate, for example, a training prediction 406 for each of the interaction events 226. The training predictions are used together with the result data 232 as part of the loss function 402 to back-propagate the comparison results of the training predictions 406 and the result data 232 in order to set parameter values within the machine learning model 128 (e.g., neurons and corresponding connections within the neural network) to train the machine learning model 128.
[0069] In this way, the machine learning model 128 learns an embedding space for different images of an object, which can be used to determine similarities between digital images, thus solving the conventional cold start and counterfactual challenges as previously described. This training can be performed to initially generate the machine learning model 128 as well as to generate updated versions of the machine learning model 128, for example to capture changing trends in user behavior regarding the way objects are represented in digital images. As further described in the following sections, the generated machine learning model 128 can then be used to select digital images.
[0070] Digit image selection using machine learning models
[0071] Figure 6 Describes the use Figure 4 System 600 in an example implementation in which a machine learning model 128 trained in the method selects a digital image from a plurality of digital images depicting changes of an object. Figure 7 Depicted is a process 700 in an example implementation of generating digital content 114 having digital images selected by a machine learning model 128 based on a user profile 206 and image metadata 230.
[0072] The following discussion describes techniques that can be implemented using the previously described systems and devices. Aspects of each of the processes can be implemented in hardware, firmware, software, or a combination thereof. The processes are shown as a set of blocks that specify operations performed by one or more devices and are not necessarily limited to the order shown for performing the operations by the respective blocks. In portions of the following discussion, reference will be made to Figures 4 to 5 .
[0073] In this example, the digital content 114 is generated using a trained machine learning model 128 as described in the previous section. First, a user ID 202 associated with a request 110 for digital content 114 including a digital image of an object is received (block 702). For example, the digital content may be configured as a web page and include the digital image as part of the web page. Other examples are also contemplated, such as thumbnails for representing digital videos.
[0074] In response, a user profile 206 associated with the user ID 202 is obtained by the profile collection module 204 from a storage device 208, which may be local to the service provider system 102 or removed from the service provider system 102 (block 704). The user profile 206 describes various characteristics associated with the user ID 202. This may include characteristics of the associated user, such as demographic information (e.g., age and gender), characteristics of how access was obtained by the user ID 202 (e.g., device type, network connection), location, etc. The user profile 206 may also describe past user interactions with the corresponding digital image, such as results of the interactions with the digital image.
[0075] Also obtain multiple digital images associated with object ID 130 (box 706), and these multiple digital images include the variation of the way that depicts the object one by one.For example, image collection module 602 can locate object ID 130, and object ID 130 corresponds to the item of digital content to be generated.Continue to refer to previous example, digital content can be configured as a web page with a part for depicting an object, such as a product or service for sale in an e-commerce website.Therefore, the web page includes object ID 130, and this object ID is associated with the digital image 604 that depicts this object.Then, make the selection about which digital image in multiple digital images 604 is to be included in the web page.In this way, in this example, only make this selection based on object rather than based on whole digital content, and thus form prediction with improved accuracy, because prediction is not distorted by "which other content" is included in digital content.
[0076] To this end, the user profile 206 and the digital image 604 are passed to the machine learning module 126. The machine learning module 126 is then configured to generate a plurality of prediction scores 606 for the plurality of digital images 604. Each prediction score is generated by the machine learning model 128 based on the user profile 206 and features extracted from the corresponding digital image in the plurality of digital images 604 (box 708). For example, the machine learning model 128 may include an embedding layer 404 to generate image metadata having image features extracted from each of the digital images 604. These image features are processed by the machine learning model 128 using machine learning together with the user profile to generate a prediction score 606 for each of the digital images 604. The prediction score indicates a probability (e.g., between zero and one) that the corresponding result will occur based on including the selected digital image as part of the digital content 114. For example, the prediction score 606 may indicate the likelihood of a conversion, such as indicating that the digital image is selected to launch the corresponding digital image, initiate a purchase of a good or service corresponding to an object in the digital image, etc.
[0077] The prediction scores 606 are then passed as input to the prediction selection module 608 by the machine learning module 126. The prediction selection module 608 is configured to select a digital image from the plurality of digital images 604 based on the plurality of prediction scores (block 710). For example, the prediction selection module 608 may select a digital image that is most likely to achieve a desired result (e.g., a conversion) based on the prediction scores 606. The predictions 610 are then passed to the digital content generation module 612 to generate the digital content 114 with the selected digital image 604(n) including the object (block 712), for example, to include the digital image 604(n) as part of a web page.
[0078] In this way, the techniques described herein overcome challenges, limitations, and computational inefficiencies of conventional techniques. This includes: handling counterfactual problems, cold start problems (and therefore, computational resources available weeks before conventional techniques), and processing the image content itself to learn patterns in digital images, which is not possible with conventional techniques.
[0079] Example Systems and Devices
[0080] Figure 8 An example system is illustrated generally at 800 that includes an example computing device 802 that represents one or more computing systems and / or devices that may implement the various techniques described herein. This is illustrated by including the digital image selection module 120. The computing device 802 may be, for example, a server of a service provider, a device associated with a client (e.g., a client device), a system on a chip, and / or any other suitable computing device or computing system.
[0081] As illustrated, the example computing device 802 includes a processing system 804, one or more computer-readable media 806, and one or more I / O interfaces 808 that are communicatively coupled to each other. Although not shown, the computing device 802 may also include a system bus or other data and command delivery system that couples the various components to each other. The system bus may include any one or combination of different bus structures, such as a memory bus or memory controller, a peripheral bus, a universal serial bus, and / or a processor or local bus utilizing any of a variety of bus architectures. Various other examples are also contemplated, such as control lines and data lines.
[0082] Processing system 804 represents functionality for performing one or more operations using hardware. Thus, processing system 804 is illustrated as including hardware elements 810 that may be configured as processors, functional blocks, and the like. This may include implementation in hardware as a dedicated integrated circuit or other logic device formed using one or more semiconductors. Hardware elements 810 are not limited by the materials from which they are formed or the processing mechanisms employed therein. For example, a processor may be composed of (multiple) semiconductors and / or transistors (e.g., electronic integrated circuits (ICs)). In this context, processor executable instructions may be electronic executable instructions.
[0083] Computer-readable storage media 806 is illustrated as including memory / storage 812. Memory / storage 812 represents memory / storage capacity associated with one or more computer-readable media. Memory / storage 812 may include volatile media (such as random access memory (RAM)) and / or non-volatile media (such as read-only memory (ROM), flash memory, optical disks, magnetic disks, etc.). Memory / storage 812 may include fixed media (e.g., RAM, ROM, fixed hard drive, etc.) and removable media (e.g., flash memory, removable hard drive, optical disk, etc.). As further described below, computer-readable media 806 may be configured in various other ways.
[0084] Input / output interface(s) 808 represent functionality for allowing a user to enter commands and information into computing device 802, and also allowing information to be presented to the user and / or other components or devices using various input / output devices. Examples of input devices include a keyboard, a cursor control device (e.g., a mouse), a microphone, a scanner, touch functionality (e.g., a capacitive or other sensor configured to detect physical touch), a camera (e.g., which may employ visible wavelengths or invisible wavelengths (such as infrared frequencies) to recognize movement as gestures that do not involve touch), and the like. Examples of output devices include a display device (e.g., a monitor or projector), a speaker, a printer, a network card, a tactile response device, and the like. Thus, computing device 802 may be configured in various ways as further described below to support user interaction.
[0085] Various techniques may be described herein in the general context of software, hardware elements, or program modules. Typically, these modules include routines, programs, objects, elements, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The terms "module," "functionality," and "component" as used herein typically represent software, firmware, hardware, or a combination thereof. The features of the techniques described herein are platform-independent, which means that these techniques may be implemented on various commercial computing platforms with various processors.
[0086] An implementation of the described modules and techniques may be stored on or transmitted on some form of computer-readable media. Computer-readable media may include various media that may be accessed by computing device 802. By way of example and not limitation, computer-readable media may include “computer-readable storage media” and “computer-readable signal media”.
[0087] "Computer-readable storage media" may refer to media and / or devices that enable persistent storage and / or non-transient storage of information as compared to signal transmission alone, carrier waves, or the signal itself. Therefore, computer-readable storage media refers to non-signal bearing media. Computer-readable storage media include hardware (such as volatile and non-volatile media, removable and non-removable media) and / or storage devices implemented with methods or technologies suitable for storing information such as computer-readable instructions, data structures, program modules, logical elements / circuits, or other data. Examples of computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical storage device, hard disk, cassette, magnetic tape, magnetic disk storage device or other magnetic storage device, or other storage device, tangible medium or article of manufacture suitable for storing desired information and accessible by a computer.
[0088] "Computer-readable signal media" may refer to signal-bearing media that is configured to transmit instructions to the hardware of the computing device 802, such as via a network. Signal media may typically embody computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, data signal, or other transport mechanism. Signal media also include any information delivery media. The term "modulated data signal" refers to a signal that has one or more of its characteristics set or changed in such a manner as to encode the information in the signal. By way of example, and not limitation, communication media include wired media, such as a wired network or direct wired connection; and wireless media, such as acoustic, RF, infrared, and other wireless media.
[0089] As previously described, hardware elements 810 and computer-readable media 806 represent modules, programmable device logic, and / or fixed device logic implemented in hardware, which may be employed in some embodiments to implement at least some aspects of the techniques described herein, such as executing one or more instructions. Hardware may include components of an integrated circuit or system on a chip, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a complex programmable logic device (CPLD), and other implementations in silicon or other hardware. In this context, hardware may operate as a processing device that performs program tasks defined by instructions and / or logic implemented by hardware and hardware for storing instructions to be executed, such as the computer-readable storage medium described previously.
[0090] The aforementioned combinations may also be used to implement the various techniques described herein. Therefore, software, hardware, or executable modules may be implemented as one or more instructions and / or logic (these instructions and / or logic are implemented on some form of computer-readable storage medium) and / or implemented by one or more hardware elements 810. Computing device 802 may be configured to implement specific instructions and / or functions corresponding to software and / or hardware modules. Therefore, an implementation of a module executable by computing device 802 as software may be achieved at least in part with hardware (e.g., by using a computer-readable storage medium and / or hardware element 810 of a processing system 804). Instructions and / or functions may be executable / operable by one or more articles (e.g., one or more computing devices 802 and / or processing systems 804) to implement the techniques, modules, and examples described herein.
[0091] The techniques described herein can be supported by various configurations of computing device 802 and are not limited to the specific examples of the techniques described herein. The functionality can also be implemented in whole or in part using a distributed system, such as through a "cloud" 814 via a platform 816 as described below.
[0092] The cloud 814 includes and / or represents a platform 816 for resources 818. The platform 816 abstracts the underlying functionality of the hardware (e.g., servers) and software resources of the cloud 814. The resources 818 may include applications and / or data that may be utilized when executing computer processing on a server remote from the computing device 802. The resources 818 may also include services provided over the Internet and / or over a subscriber network, such as a cellular network or a Wi-Fi network.
[0093] The platform 816 can abstract resources and functionality to connect the computing device 802 with other computing devices. The platform 816 can also be used to abstract resource scaling to provide a corresponding level of scaling to meet demand for resources 818 implemented via the platform 816. Thus, in an interconnected device embodiment, the implementation of the functionality described herein can be distributed throughout the system 800. For example, the functionality can be implemented in part on the computing device 802 and implemented via the platform 816 that abstracts the functionality of the cloud 814.
[0094] in conclusion
[0095] Although the invention has been described in language specific to structural features and / or methodological acts, it should be understood that the invention defined in the appended claims is not necessarily limited to the specific features or acts described. Instead, the specific features and acts are disclosed as example forms of implementing the claimed invention.
Claims
1. A method implemented by at least one computing device in a digital media digital content generation environment for processing object changes as part of a digital image selection, the method comprising: receiving, by the at least one computing device, a user ID associated with a request for digital content; obtaining, by the at least one computing device, a plurality of digital images having variations in the manner in which an object is depicted and a user profile associated with the user ID; generating, by the at least one computing device, a plurality of prediction scores for the plurality of digital images, each prediction score generated by a machine learning model based on the user profile and image metadata, the image metadata comprising features extracted from a corresponding digital image in the plurality of digital images; selecting, by the at least one computing device, a digital image from the plurality of digital images based on the plurality of prediction scores; as well as The digital content is generated by the at least one computing device to have the selected digital image depicting the object.
2. The method of claim 1, wherein said generating comprises: The features are extracted from the corresponding digital images as embeddings using a convolutional neural network.
3. The method of claim 1, wherein the image metadata further describes characteristics of the object, the characteristics including a product category or object description from text associated with the corresponding digital image. The method of claim 1 , the user profile having user information including demographic information or location information.
5. The method of claim 1, wherein a prediction score in the plurality of prediction scores indicates a probability of obtaining a result resulting from including the corresponding digital image as part of the digital content. The method of claim 5 , wherein the result is a conversion.
7. The method of claim 1, wherein the digital content is a web page and the plurality of digital images include corresponding differences in the manner in which the object is depicted.
8. The method of claim 7, wherein the object is an article of clothing and the corresponding difference is a human model wearing the article of clothing.
9. The method of claim 1, wherein the digital content is a digital video and the selected digital image is configured as a thumbnail selectable to launch the digital video.
10. The method of claim 1, wherein the machine learning model is a single convolutional neural network trained using a plurality of training digital images, and the plurality of prediction scores are generated based on the plurality of digital images using the single convolutional neural network.
11. A system in a digital media machine learning model training environment that processes object variations as part of a digital image selection, the system comprising: an exploration / exploitation determination module implemented at least in part in hardware of the computing device to make an exploration determination or an exploitation determination in response to a request for digital content to explore or exploit a user behavior associated with the user ID; an exploration module implemented at least in part in hardware of the computing device to randomly select a digital image from a plurality of digital images that depict, one after the other, a change in an object in response to the exploration determination; an exploitation module implemented at least in part in hardware of the computing device to select, in response to the exploitation determination, a digital image from the plurality of digital images based on a machine learning model, the plurality of digital images depicting the change in the object; a training data generation module implemented at least in part in hardware of the computing device to generate an interaction event as part of the training data for each of the requests, the interaction event comprising a user profile associated with the user ID, result data describing a result of including the selected digital image as part of the digital content, and image metadata having features extracted from the selected digital image using machine learning; as well as A machine learning module is implemented at least in part in hardware of the computing device to generate a machine learning model using the training data.
12. The system of claim 11, wherein the features are extracted from the selected digital image using a convolutional neural network.
13. The system of claim 11, wherein the training data further describes characteristics of the object, the characteristics including a product category or object description from text associated with the corresponding digital image.
14. The system of claim 11, further comprising: A profile collection module is implemented at least in part in hardware of the computing device to obtain the user profile based on the user ID associated with the request for the digital content.
15. The system of claim 11, further comprising: An image collection module is implemented at least in part in hardware of the computing device to obtain the plurality of digital images based on an object ID associated with the digital content.
16. The system according to claim 11, wherein the exploration / exploitation determination module adopts Epsilon-greedy exploration technology.
17. A system in a digital media machine learning model training environment, the system comprising: A component for receiving training data, the training data comprising: object metadata describing objects included in the plurality of digital images; and A plurality of interaction events, each of the plurality of interaction events comprising result data, a user profile and image metadata, the image metadata having features extracted from a corresponding digital image of the plurality of digital images using machine learning; and a component for generating a machine learning model trained using machine learning based on the object metadata, user profile and image metadata and a loss function based on the result data.
18. The system of claim 17, wherein the features are extracted as embeddings from the corresponding digital images using a convolutional neural network.
19. The system of claim 17, wherein the image metadata further describes characteristics of the object, the characteristics including a product category or object description from text associated with the corresponding digital image.
20. The system of claim 17, wherein the user profile has user information including demographic information or location information.
