WO2023124793A1 - 图像推送方法和装置 - Google Patents

图像推送方法和装置 Download PDF

Info

Publication number
WO2023124793A1
WO2023124793A1 PCT/CN2022/136500 CN2022136500W WO2023124793A1 WO 2023124793 A1 WO2023124793 A1 WO 2023124793A1 CN 2022136500 W CN2022136500 W CN 2022136500W WO 2023124793 A1 WO2023124793 A1 WO 2023124793A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
design
pushed
features
user
Prior art date
Application number
PCT/CN2022/136500
Other languages
English (en)
French (fr)
Inventor
徐立峰
周维斯
王勇
黄勇尤
Original Assignee
北京沃东天骏信息技术有限公司
北京京东世纪贸易有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京沃东天骏信息技术有限公司, 北京京东世纪贸易有限公司 filed Critical 北京沃东天骏信息技术有限公司
Publication of WO2023124793A1 publication Critical patent/WO2023124793A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/53—Querying
    • G06F16/535—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

  • the embodiments of the present disclosure relate to the field of computer technology, and in particular to an image pushing method and device.
  • multi-modal creative design is playing an increasingly important role in information push, and there have also been effects of applying multi-modal creative design to improve the efficiency and accuracy of information push.
  • Existing implementation schemes mainly include two types.
  • One is post-investment tuning. This method is mainly to launch multiple groups of creative designs under the grayscale traffic, and then predict the optimal delivery ratio of each creative design based on the grayscale effect through the Bayesian algorithm to carry out the full volume. delivery.
  • the other is pre-cast prediction, which is mainly based on multiple sets of creative design images, extracting high-level features of the images through image embedding, and then performing predictive modeling to determine the creative design image with the highest score for delivery .
  • Embodiments of the present disclosure provide an image pushing method and device.
  • One or more embodiments of the present disclosure provide an image push method, the method includes: acquiring the visual features of the image to be pushed, wherein the visual features include design features and target visual features, and the target visual features are obtained by vector embedding; according to The visual feature determines the user's preference for the image to be pushed; in response to determining that the preference meets a preset condition, the image to be pushed is pushed to a terminal corresponding to the user.
  • an image push device which includes: a visual feature acquisition unit configured to acquire visual features of an image to be pushed, wherein the visual features include design features and target visual features, and the target The visual features are obtained by vector embedding; the preference determination unit is configured to determine the user's preference for the image to be pushed according to the visual features; the push unit is configured to respond to the determination that the preference meets the preset condition, and push the user's corresponding terminal It's time to push the image.
  • One or more embodiments of the present disclosure provide an electronic device, which includes: one or more processors; a storage device for storing one or more programs; when one or more programs are used by one or more processors, so that one or more processors implement the above image pushing method.
  • One or more embodiments of the present disclosure provide a computer-readable medium on which a computer program is stored, and when the computer program is executed by a processor, the above image pushing method is realized.
  • FIG. 1 is an exemplary system architecture diagram to which an embodiment of the present disclosure can be applied;
  • Fig. 2 is a flowchart of an embodiment of the image pushing method according to the present disclosure
  • FIG. 3 is a flowchart of another embodiment of an image pushing method according to an embodiment of the present disclosure.
  • FIG. 4 is a flowchart of another embodiment of an image pushing method according to an embodiment of the present disclosure.
  • FIG. 5 is a schematic diagram of an application scenario of an image pushing method according to an embodiment of the present disclosure.
  • Fig. 6 is a schematic structural diagram of an embodiment of an image pushing device according to the present disclosure.
  • FIG. 7 is a schematic structural diagram of an electronic device suitable for implementing an embodiment of the present disclosure.
  • FIG. 1 shows an exemplary architecture 100 to which an embodiment of the image pushing method or image pushing apparatus of the present disclosure can be applied.
  • a system architecture 100 may include terminal devices 101 , 102 , 103 , a network 104 and a server 105 .
  • the network 104 is used as a medium for providing communication links between the terminal devices 101 , 102 , 103 and the server 105 .
  • Network 104 may include various connection types, such as wires, wireless communication links, or fiber optic cables, among others.
  • the terminal devices 101, 102, 103 interact with the server 105 via the network 104 to receive or send messages and the like.
  • Various client applications may be installed on the terminal devices 101, 102, 103. For example, browser applications, search applications, shopping applications, instant messaging tools, social applications, image processing applications, and so on.
  • the terminal devices 101, 102, and 103 may be hardware or software.
  • the terminal devices 101, 102, 103 When the terminal devices 101, 102, 103 are hardware, they may be various electronic devices, including but not limited to smart phones, tablet computers, e-book readers, laptop computers, desktop computers and so on.
  • the terminal devices 101, 102, 103 When the terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above. It may be implemented as multiple software or software modules (for example, multiple software or software modules for providing distributed services), or as a single software or software module. No specific limitation is made here.
  • the server 105 may be a server that provides various services, for example, a backend server that provides service support for client applications installed on the terminal devices 101 , 102 , 103 .
  • the server 105 can obtain the visual features of the image to be pushed, and determine the user's preference for the image to be pushed according to the visual feature. Push the image to be pushed.
  • the above-mentioned visual features of the image to be pushed may be directly stored locally in the server 105, and the server 105 may directly extract and process the visual features of the image to be pushed stored locally. At this time, there may be no terminal device 101, 102, 103 and network 104.
  • the image pushing method provided by the embodiments of the present disclosure is generally executed by the server 105, and correspondingly, the image pushing device is generally set in the server 105.
  • image push applications may also be installed in terminal devices 101, 102, and 103, and terminal devices 101, 102, and 103 may also determine the user's preference for an image to be pushed based on the visual characteristics of the image to be pushed based on the image push application. degree of preference, and when the degree of preference meets the preset condition, push the image to be pushed to the terminal corresponding to the user.
  • the image pushing method can be executed by the terminal devices 101 , 102 , 103 , and correspondingly, the image pushing apparatus can also be set in the terminal devices 101 , 102 , 103 .
  • server 105 and network 104 may not exist in exemplary system architecture 100 .
  • the server 105 may be hardware or software.
  • the server 105 can be implemented as a distributed server cluster composed of multiple servers, or as a single server.
  • the server 105 is software, it can be implemented as multiple software or software modules (for example, multiple software or software modules for providing distributed services), or can be implemented as a single software or software module. No specific limitation is made here.
  • terminal devices, networks and servers in Fig. 1 are only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers.
  • FIG. 2 shows a process 200 of an embodiment of the image pushing method according to the present disclosure.
  • the image push method includes the following steps:
  • Step 201 acquire the visual features of the image to be pushed.
  • the images to be pushed may be images of various types and contents.
  • the image to be pushed may be an introduction image of an item, and the like.
  • the image to be pushed may be a promotional image of a certain service or activity, and the like.
  • the visual feature of the image to be pushed means the feature presented by the image to be pushed from a visual angle. Visual features may include design features and targeted visual features.
  • the design feature can be used to represent the design feature of the image to be pushed.
  • the design features may include various aspects of design features according to actual application scenarios or application requirements. For example, including but not limited to: layout, background decoration, style, color system, etc.
  • the target visual features can be obtained by using vector embedding (Embedding), and specifically can be obtained by using various vector embedding methods for images.
  • target visual features can be represented by feature vectors.
  • the visual features of the image to be pushed can be determined based on various methods. For example, design features can be pre-annotated by designers.
  • the target visual features can be extracted by using convolutional neural network and so on. As an example, based on the existing classification model and network structure such as ResNet, several blocks in the shallow layer can be reserved, and then the fully connected layer and the predicted output layer can be connected to build a model and train. At this time, the output of the fully connected layer can be used as the target visual feature .
  • the executing subject of the image pushing method can obtain the visual features of the image to be pushed from a local or other storage device. It should be noted that the visual features of the image to be pushed may be determined by the person executing the image pushing method, or may be determined by other electronic devices.
  • Step 202 Determine the user's preference for the image to be pushed according to the visual features.
  • the user's preference degree of the image to be pushed may represent the degree of interest of the user in the image to be pushed.
  • the degree of preference can be expressed in various specific ways according to actual application scenarios.
  • the degree of preference can be represented by data indicators such as click-through rate and browsing time.
  • various methods may be used to determine the user's preference for the image to be pushed according to the visual characteristics of the image to be pushed.
  • the user's preference for the image to be pushed can be determined according to the visual features of images with high user preference in the statistical historical period, and then according to the similarity between the statistically obtained visual features and the visual features of the image to be pushed.
  • the user's preference for the image to be pushed is positively correlated with the determined similarity.
  • various data indicator prediction methods may be used to determine the user's preference for the image to be pushed.
  • various existing click-through rate prediction models are used to determine the user's click-through rate of the image to be pushed according to the visual features of the image to be pushed.
  • the user's preference for pushing images is positively correlated with the determined click-through rate.
  • Step 203 in response to determining that the degree of preference meets the preset condition, push the image to be pushed to a terminal corresponding to the user.
  • the preset conditions can be flexibly set according to actual application requirements.
  • the preset condition may include that the determined preference degree is not less than a preset threshold.
  • the preset condition may include that the sorting position corresponding to the determined preference degree is not smaller than the preset position.
  • the ranking position may refer to a position in a ranking formed by sorting the preference degrees corresponding to a plurality of preset images to be pushed.
  • the preference degree meets the preset condition, it may indicate that the user is more interested in the pushed image.
  • the image to be pushed can be further pushed to the terminal corresponding to the user (such as the terminal used by the user, etc.), so that image push based on design features can be realized, and image push is avoided by only using the target visual features obtained by vector embedding. The user's preference requirements for the design features of the image are ignored.
  • the design features of the image to be pushed may be determined through the following steps:
  • Step 1 Analyze the design elements of the image to be pushed.
  • design elements include but are not limited to: color system, font, style, graphic layout, etc.
  • various analysis methods may be used to determine the design elements of the image to be pushed.
  • a design document of an image to be pushed may be preset by a designer, and each design element may be recorded in the design document. At this time, each design element of the image to be pushed can be analyzed by reading the design document of the image to be pushed.
  • Step 2 Identify the semantics represented by the design elements, obtain the recognition results, and determine the obtained recognition results as design features.
  • the semantics expressed by the design elements may refer to the intentions expressed by the designers through the design elements. For example, different semantics such as solemnity or ease can be expressed through different color systems.
  • the semantic recognition results can be obtained, and then the obtained semantic recognition results can be determined as design features, so that the design intention of the image to be pushed can be transformed into interpretable design features.
  • various recognition methods can be used to identify the semantics represented by design elements according to actual application scenarios. For example, the correspondence between design elements and semantics can be pre-labeled by designers. At this point, the semantics represented by the design elements can be determined by querying the preset corresponding relationship.
  • the semantics represented by the color system of the background can be determined through the following steps:
  • Step 1 Count the color values of each pixel contained in the background, and filter the top ten color values corresponding to the largest number of pixels.
  • Step 2 Screen out the color values whose saturation is not lower than the preset saturation threshold and whose brightness is not lower than the preset brightness threshold from the ten color values, so as to exclude invalid colors or neutral colors with too low saturation or too low brightness.
  • Step 3 Clustering the filtered color values to obtain a clustering result.
  • Step 4 Use the preset color value-semantic correspondence to query the semantics represented by the clusters in the clustering results.
  • the feature extraction and representation based on image embedding cannot analyze the creative design features such as color, emotional tendency and style contained in the image, but the creative design features are more able to convey the visual atmosphere and design mentality. Therefore, The push image predicted based only on the feature extraction and representation of image embedding may not meet the user's current emotional appeal and mental communication. Based on this, by combining the feature extraction of image embedding and the corresponding design features of the image, the matching degree between the image and the user can be more comprehensively measured, which helps to improve the accuracy of the pushed image and further enhance the user experience.
  • the design features may include at least one of the following: design content features and design performance features.
  • design content features may refer to design features presented in terms of content.
  • design content features may include but not limited to: Logo (trademark/logo), objects presented in images, characters presented in images (such as copywriting), various controls in images (such as buttons, etc.) and so on.
  • Design performance features may refer to design features in terms of visual performance.
  • design representation features include, but are not limited to: color scheme, layout, drawing of objects presented in an image, background decoration, style, and the like.
  • the design features of the images to be pushed can be visualized, which can be used to express the design features of the images to be pushed more comprehensively and conveniently.
  • design content characteristics can be determined through the following steps:
  • Step 1 Analyze the design content elements of the image to be pushed based on semantic segmentation.
  • the set of design content elements can be preset according to requirements, and at this time, the design content elements belonging to the preset set of design content elements can be identified from the image to be pushed.
  • a pre-trained design content element recognition model based on semantic segmentation can be used to identify each design content element contained in the image to be pushed.
  • an object area, a character area, a logo area, a space area, etc. in the image to be pushed may be identified.
  • Step 2 Identify the semantics represented by the elements of the design content, obtain a recognition result, and determine the obtained recognition result as a feature of the design content.
  • Semantic segmentation can be used to accurately identify each design element, thereby more accurately representing the design features of the image to be pushed.
  • design performance characteristics can be determined by the following steps:
  • Step 1 Analyze the design file of the image to be pushed to obtain the design expression elements of the image to be pushed.
  • the design file of the image to be pushed may refer to various materials formed during the design process of the image to be pushed. For example, if the image to be pushed is made by using a certain design tool or application, the file formed by using the design tool or application can be used as the design file of the image to be pushed.
  • design files may use different parsing methods. For example, the corresponding number of layers, layer types, layer sizes, etc. can be read through the design file first, and then the content of each layer can be specifically analyzed to determine the design elements of the image to be pushed.
  • Step 2 Identify the semantics represented by the design expression elements, obtain the recognition results, and determine the obtained recognition results as the design expression features.
  • the original design semantics can be understood by parsing and identifying the design file of the image to be pushed, so as to represent the design features more accurately.
  • the target visual features can be obtained by processing the image to be pushed, or can be obtained by processing the image template corresponding to the image to be pushed by using vector embedding.
  • a convolutional neural network is used to extract a feature vector from an image template corresponding to an image to be pushed.
  • the image template may refer to a design template of the image to be pushed, may include various design elements, and may not include specific content of each design element.
  • the content of the design element indicated by the design feature of the image to be pushed can be supplemented in the design template, and the image obtained after the supplement is used as the image to be pushed, and The obtained image to be pushed is pushed to the user.
  • the method provided by the above-mentioned embodiments of the present disclosure determines the user’s preference for the image to be pushed by combining the design features of the image to be pushed and the target visual features obtained by image embedding, and then pushes the image based on the preference, which can avoid the above-mentioned post-selection adjustment.
  • Optimization needs to carry out the process of grayscale implementation and delivery, and on the basis of using the high-level features of the image, it further combines the color emotional tendency, style and other creative design semantics expressed by the design features of the image to predict the preference, which is helpful to improve
  • the accuracy of the image push results makes the pushed images better match the user's needs in terms of design.
  • FIG. 3 it shows a process 300 of another embodiment of the image pushing method.
  • the flow 300 of the image pushing method includes the following steps:
  • Step 301 acquire the visual features of the image to be pushed.
  • Step 302 acquiring associated features of the image to be pushed.
  • the associated feature may refer to various features associated with the image to be pushed except the visual feature.
  • the associated features may include at least one of the following: attribute features, text features.
  • the text features may represent the features of the characters in the image to be pushed.
  • the attribute feature may include at least one of the following: the attribute feature of the user, and the attribute feature of the object described by the image to be pushed.
  • the attribute characteristics of the user may include the user's inherent attributes, behavioral attributes, and so on.
  • the objects described by the images to be pushed may be various types of objects. Different types of objects can have different attribute characteristics.
  • the object described in the image to be pushed may be the item.
  • the attribute characteristics of the object are the attribute characteristics of the item (such as composition, origin, shelf life, etc.).
  • the image to be pushed is a promotion image of a certain service
  • the object described in the image to be pushed may be the service.
  • the attribute characteristics of the object are the attribute characteristics of the service (such as start time, end time, location, etc.).
  • the associated feature of the image to be pushed can be determined in various ways. For example, attribute characteristics can be determined by collecting relevant attribute data. As another example, for text features, characters can be extracted from the image to be pushed based on methods such as OCR (Optical Character Recognition, Optical Character Recognition), and then the corresponding text features can be extracted using methods such as NLP (Natural Language Processing, Natural Language Processing) .
  • OCR Optical Character Recognition, Optical Character Recognition
  • NLP Natural Language Processing
  • LSTM Long Short-Term Memory
  • Embedding result of the output of the last fully connected layer can be used as a text feature.
  • the execution subject can obtain the associated features of the images to be pushed from local or other data sources. It should be noted that the associated feature of the image to be pushed may be determined by the above-mentioned execution subject, or may be determined by other electronic devices. The visual features and associated features of the image to be pushed may be determined by the same electronic device, or may be determined by different electronic devices.
  • Step 303 Determine the user's preference for the image to be pushed according to the visual features and associated features.
  • the user's preference for the image to be pushed can be determined by integrating the visual features and associated features of the image to be pushed. Specifically, various methods for determining the degree of preference may be used according to actual application requirements.
  • the user's first degree of preference for the image to be pushed may be determined first according to the visual characteristics of the image to be pushed, and then the user's second degree of preference for the image to be pushed may be determined according to the associated characteristics of the image to be pushed. Then, the weighted sum of the first preference degree and the second preference degree is fused, and the fusion result is determined as the user's preference degree of the image to be pushed.
  • various neural network models based on deep learning can be used to determine the user's preference for the image to be pushed according to the visual features and associated features of the image to be pushed.
  • the preference determination model can be pre-trained, with the visual features and associated features of the image to be pushed as input, and the user's preference for the image to be pushed as output, so that the user's preference for the image to be pushed can be obtained by using the preference determination model.
  • Step 304 in response to determining that the degree of preference meets the preset condition, push the image to be pushed to a terminal corresponding to the user.
  • the target visual features and text features of the image to be pushed can be realized based on the neural network model, and the preference prediction and push prediction can be realized based on the neural network model.
  • training data can be extracted by collecting historical behavior records of users to train related neural network models. Different neural network models can be trained independently, and multi-modal data fusion and error correction can also be performed during the training process to achieve multi-task training and improve the training effect of each neural network model. For example, you can first train the neural network model based on vector embedding to extract feature vectors. At this time, the extracted feature vectors can represent the target visual features, and at the same time train the neural network model for extracting text features, and then use these models as pre-training models for joint training. Image push prediction model or preference prediction model, etc.
  • the method provided by the above-mentioned embodiments of the present disclosure combines the feature data of three different modalities of the visual feature, text feature and attribute feature of the image to be pushed to push the image.
  • the attributes of the object described by the image, the user attribute , visual features (such as the object display picture presented in the image, etc.), copywriting, etc. are several aspects that are relatively easy to attract users. Therefore, using the feature data of three different modalities of visual features, text features and attribute features Consider various factors that affect the push effect in the image push scene, so as to help improve the effect of image push.
  • FIG. 4 it shows a process 400 of another embodiment of the image pushing method.
  • the flow 400 of the image pushing method includes the following steps:
  • Step 401 acquire the visual features of the image to be pushed.
  • Step 402 Determine the user's preference for the image to be pushed according to the visual features.
  • Step 403 in response to determining that the degree of preference meets the preset condition, push the image to be pushed to a terminal corresponding to the user.
  • Step 404 acquiring the user's operation information on the image to be pushed.
  • the operation information may be used to represent the operation performed by the user.
  • the operation indicated by the operation information may refer to various operations performed by the user after receiving the image to be pushed.
  • the user's operations may include: click/unclick, browse/unbrowse, favorite/unfavorite, comment/uncomment, and so on.
  • the terminal corresponding to the user may collect the user's operation information on the image to be pushed, and then acquire the user's operation information on the image to be pushed from the terminal corresponding to the user.
  • Step 405 build a knowledge map according to the operation information.
  • the knowledge graph can be used to record user preference information for visual features.
  • the preference information may represent a user's preference.
  • the user's preference for the pushed image can be learned according to the user's operation information, and then recorded to build a knowledge graph.
  • the content of knowledge map records can be flexibly set according to actual application requirements.
  • the user clicks on an image to be pushed for presenting an item the user ID, item ID, design element, semantics and design intent of the design element may be recorded.
  • the design intention may refer to the design intention of the designer when designing the image to be pushed.
  • a knowledge graph can be constructed to record the correspondence between items, users, design elements, design semantics, and scene intent.
  • the semantics of saturation can specifically include low saturation, medium saturation, and high saturation, and the sequential changes of these three saturations can correspond to changes in scene intent from young to mature. Therefore, if it is determined that the user prefers low-saturated design semantics according to the user's operation information, it can be recorded in the knowledge graph to indicate that the user prefers a young design style.
  • the construction of a knowledge map is usually a process of continuous iteration and precipitation. Therefore, in practical applications, the knowledge map can be continuously updated according to the results of continuous image push.
  • image push based on post-investment optimization cannot analyze the creative design, making it impossible to accumulate reusable knowledge for each creative design prediction.
  • image push based on image embedding since image embedding itself is a high-level feature, it usually participates in the calculation in the form of vectors, making it impossible to know which factors or features contribute to the better creative design that is screened out, so it is also impossible to filter out The creative design after the creative design direction.
  • the methods provided by the above-mentioned embodiments of the present disclosure can realize the extraction of interpretable low-level design features of images to be pushed by using design file parsing or semantic segmentation-based parsing and design semantic recognition, thereby solving the problem of image embedding.
  • the unexplainable problems brought about, and by building a knowledge map, knowledge can be recorded, precipitated and transformed according to the effect of each push, and then the constructed knowledge map can be used to assist in subsequent user-oriented image push, image search and other applications Scenes.
  • FIG. 5 is a schematic application scenario 500 of the image pushing method according to this embodiment.
  • the corresponding design features 5011 can be determined based on algorithms such as design file parsing and semantic segmentation, and the corresponding relationship between the design template and its corresponding design features can be recorded 504.
  • the convolutional neural network can be used to extract the feature vector of the design template as the target visual feature 5012 , so as to obtain the visual feature 501 composed of the design feature 5011 and the target visual feature 5012 .
  • the text feature 502 corresponding to the design template can be extracted based on technologies such as OCR and NLP, and the attribute feature 503 composed of the user attribute feature 5031 and the item attribute feature 5032 can be obtained based on statistical analysis in advance.
  • the visual features 501 , text features 502 and attribute features 503 can be input into the pre-trained click-through rate prediction model 505 to obtain the click-through rate 506 corresponding to the design template. Then, according to the click-through rates corresponding to each design template, the design template with a higher click-through rate can be selected, and the design features corresponding to the design template can be added to the design template to form an image and pushed to the user.
  • the present disclosure provides an embodiment of an image push device, which corresponds to the method embodiment shown in FIG. 2 , and can be specifically applied to in various electronic devices.
  • the image push device 600 includes a visual feature acquisition unit 601 , a preference degree determination unit 602 and a push unit 603 .
  • the visual feature acquisition unit 601 is configured to acquire the visual features of the image to be pushed, wherein the visual features include design features and target visual features, and the target visual features are obtained by vector embedding;
  • the preference determination unit 602 is configured to, according to the visual features, Determine the user's preference for the image to be pushed;
  • the pushing unit 603 is configured to push the image to be pushed to a terminal corresponding to the user in response to determining that the preference meets a preset condition.
  • the specific processing of the visual feature acquisition unit 601, the preference determination unit 602, and the push unit 603 and the technical effects brought about by them can refer to step 201 in the corresponding embodiment in FIG. 2 , step 202 and step 203 related descriptions will not be repeated here.
  • the design features are determined through the following steps: analyzing the design elements of the image to be pushed; identifying the semantics represented by the design elements, obtaining the recognition result, and determining the obtained recognition result as the design feature .
  • the design features include at least one of the following: design content features and design performance features.
  • the design content features are determined through the following steps: analyzing the design content elements of the image to be pushed based on semantic segmentation; identifying the semantics represented by the design content elements, obtaining the recognition result, and the obtained The recognition result is determined as the design content feature.
  • the design representation features are determined through the following steps: analyzing the design file of the image to be pushed to obtain the design representation elements of the image to be pushed; identifying the semantics represented by the design representation elements to obtain the recognition result , and determine the obtained recognition results as design performance characteristics.
  • the image pushing device further includes: an associated feature acquiring unit (not shown in the figure), configured to acquire an associated feature of an image to be pushed, wherein the associated feature includes at least the following One: attribute feature, text feature, the attribute feature includes at least one of the following: the attribute feature of the user, the attribute feature of the object described by the image to be pushed; and the preference determination unit 602 is further configured to determine according to the visual feature and the associated feature The user's preference for pushed images.
  • an associated feature acquiring unit (not shown in the figure), configured to acquire an associated feature of an image to be pushed, wherein the associated feature includes at least the following One: attribute feature, text feature, the attribute feature includes at least one of the following: the attribute feature of the user, the attribute feature of the object described by the image to be pushed
  • the preference determination unit 602 is further configured to determine according to the visual feature and the associated feature The user's preference for pushed images.
  • the above-mentioned image push device further includes: an operation information acquisition unit (not shown in the figure), configured to acquire user operation information on the image to be pushed; a construction unit configured According to the operation information, a knowledge map is constructed, wherein the knowledge map is used to record the user's preference information for visual features.
  • an operation information acquisition unit (not shown in the figure), configured to acquire user operation information on the image to be pushed
  • a construction unit configured According to the operation information, a knowledge map is constructed, wherein the knowledge map is used to record the user's preference information for visual features.
  • the visual feature acquisition unit acquires the visual features of the image to be pushed, wherein the visual features include design features and target visual features, and the target visual features are obtained by vector embedding; the preference determination unit according to the visual features , to determine the user's preference for the image to be pushed; the push unit pushes the image to be pushed to the terminal corresponding to the user in response to determining that the preference meets the preset condition, and can further combine the design of the image on the basis of using the high-order features of the image Prediction of creative design semantics such as color emotional tendency and style expressed by features can help improve the accuracy of image push results, making the pushed images better match the needs of users in terms of design.
  • FIG. 7 it shows a schematic structural diagram of an electronic device (such as the server in FIG. 1 ) 700 suitable for implementing embodiments of the present disclosure.
  • the server shown in FIG. 7 is only an example, and should not limit the functions and scope of use of the embodiments of the present disclosure.
  • an electronic device 700 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) Various appropriate actions and processes are executed by programs in the memory (RAM) 703 . In the RAM 703, various programs and data necessary for the operation of the electronic device 700 are also stored.
  • the processing device 701, ROM 702, and RAM 703 are connected to each other through a bus 704.
  • An input/output (I/O) interface 705 is also connected to the bus 704 .
  • the following devices can be connected to the I/O interface 705: input devices 706 including, for example, a touch screen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; including, for example, a liquid crystal display (LCD), speakers, vibration an output device 707 such as a computer; a storage device 708 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 709.
  • the communication means 709 may allow the electronic device 700 to communicate with other devices wirelessly or by wire to exchange data. While FIG. 7 shows electronic device 700 having various means, it should be understood that implementing or having all of the means shown is not a requirement. More or fewer means may alternatively be implemented or provided. Each block shown in FIG. 7 may represent one device, or may represent multiple devices as required.
  • embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, where the computer program includes program codes for executing the methods shown in the flowcharts.
  • the computer program may be downloaded and installed from a network via communication means 709, or from storage means 708, or from ROM 702.
  • the processing device 701 When the computer program is executed by the processing device 701, the above-mentioned functions defined in the methods of the embodiments of the present disclosure are performed.
  • the computer-readable medium described in the embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the above two.
  • a 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 thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable 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.
  • 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, apparatus, or device.
  • a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code therein. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • a computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can transmit, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device .
  • Program code embodied on a computer readable medium may be transmitted by any appropriate medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.
  • the above-mentioned computer-readable medium may be included in the above-mentioned electronic device, or may exist independently without being incorporated into the electronic device.
  • the above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed by the electronic device, the electronic device: acquires the visual features of the image to be pushed, wherein the visual features include design features and target visual features feature, the target visual feature is obtained by vector embedding; according to the visual feature, the user's preference degree of the image to be pushed is determined; in response to determining that the preference degree meets the preset condition, the image to be pushed is pushed to the terminal corresponding to the user.
  • Computer program code for carrying out operations of embodiments of the present disclosure may be written in one or more programming languages, or combinations thereof, including object-oriented programming languages—such as Java, Smalltalk, C++, Also included are conventional procedural programming languages - such as the "C" language or similar programming languages.
  • the program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server.
  • the remote computer may be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (for example, using an Internet service provider to connected via the Internet).
  • LAN local area network
  • WAN wide area network
  • Internet service provider for example, using an Internet service provider to connected via the Internet.
  • each block in a flowchart or block diagram may represent a module, program segment, or portion of code that contains one or more logical functions for implementing specified executable instructions.
  • the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved.
  • each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations can be implemented by a dedicated hardware-based system that performs the specified functions or operations , or may be implemented by a combination of dedicated hardware and computer instructions.
  • the units involved in the embodiments described in the present disclosure may be implemented by software or by hardware.
  • the described units may also be set in a processor, for example, it may be described as: a processor includes a visual feature acquisition unit, a preference determination unit, and a push unit.
  • a processor includes a visual feature acquisition unit, a preference determination unit, and a push unit.
  • the names of these units do not constitute a limitation of the unit itself in some cases, for example, the visual feature acquisition unit can also be described as "a unit that acquires the visual features of the image to be pushed, wherein the visual features include design features and the target visual feature, the target visual feature is obtained by vector embedding".

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Multimedia (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • User Interface Of Digital Computer (AREA)
  • Collating Specific Patterns (AREA)

Abstract

本公开的实施例公开了图像推送方法和装置。该方法包括：获取待推送图像的视觉特征，其中，视觉特征包括设计特征和目标视觉特征，目标视觉特征利用向量嵌入得到；根据视觉特征，确定用户对待推送图像的偏好度；响应于确定偏好度符合预设条件，向用户对应的终端推送该待推送图像。

Description

图像推送方法和装置
本公开要求于2021年12月27日提交的申请号为202111610765.3、发明名称为“图像推送方法和装置”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
技术领域
本公开的实施例涉及计算机技术领域，具体涉及图像推送方法和装置。
背景技术
随着多媒体终端的应用普及，多模态创意设计在信息推送中发挥着越来越重要的作用，由此也出现了应用多模态创意设计以提升信息推送效率和准确度等效果。
现有的实现方案主要包括两种。一种是投后调优，这种方式主要是先在灰度流量下进行多组创意设计投放，然后再基于灰度效果通过贝叶斯算法等预测各创意设计的最优投放比例以进行全量投放。另一种是投前预测，这种方式主要是基于多组创意设计图像，通过图像Embedding的方式抽取图像的高阶特征，然后再进行预测建模，以确定得分最高的创意设计图像以进行投放。
发明内容
本公开的实施例提出了图像推送方法和装置。
本公开的一个或多个实施例提供了一种图像推送方法，该方法包括：获取待推送图像的视觉特征，其中，视觉特征包括设计特征和目标视觉特征，目标视觉特征利用向量嵌入得到；根据视觉特征，确定用户对待推送图像的偏好度；响应于确定偏好度符合预设条件，向用户对应的终端推送该待推送图像。
本公开的一个或多个实施例提供了一种图像推送装置，该装置包括：视 觉特征获取单元，被配置成获取待推送图像的视觉特征，其中，视觉特征包括设计特征和目标视觉特征，目标视觉特征利用向量嵌入得到；偏好度确定单元，被配置成根据视觉特征，确定用户对待推送图像的偏好度；推送单元，被配置成响应于确定偏好度符合预设条件，向用户对应的终端推送该待推送图像。
本公开的一个或多个实施例提供了一种电子设备，该电子设备包括：一个或多个处理器；存储装置，用于存储一个或多个程序；当一个或多个程序被一个或多个处理器执行，使得一个或多个处理器实现以上图像推送方法。
本公开的一个或多个实施例提供了一种计算机可读介质，其上存储有计算机程序，该计算机程序被处理器执行时实现以上图像推送方法。
附图说明
通过阅读参照以下附图所作的对非限制性实施例所作的详细描述，本公开的其它特征、目的和优点将会变得更明显：
图1是本公开的一个实施例可以应用于其中的示例性系统架构图；
图2是根据本公开的图像推送方法的一个实施例的流程图；
图3是根据本公开的实施例的图像推送方法的又一个实施例的流程图；
图4是根据本公开的实施例的图像推送方法的再一个实施例的流程图；
图5是根据本公开的实施例的图像推送方法的一个应用场景的示意图；
图6是根据本公开的图像推送装置的一个实施例的结构示意图；
图7是适于用来实现本公开的实施例的电子设备的结构示意图。
具体实施方式
下面结合附图和实施例对本公开作进一步的详细说明。可以理解的是，此处所描述的具体实施例仅仅用于解释相关发明，而非对该发明的限定。另外还需要说明的是，为了便于描述，附图中仅示出了与有关发明相关的部分。
需要说明的是，在不冲突的情况下，本公开中的实施例及实施例中的特 征可以相互组合。下面将参考附图并结合实施例来详细说明本公开。
图1示出了可以应用本公开的图像推送方法或图像推送装置的实施例的示例性架构100。
如图1所示，系统架构100可以包括终端设备101、102、103，网络104和服务器105。网络104用以在终端设备101、102、103和服务器105之间提供通信链路的介质。网络104可以包括各种连接类型，例如有线、无线通信链路或者光纤电缆等等。
终端设备101、102、103通过网络104与服务器105交互，以接收或发送消息等。终端设备101、102、103上可以安装有各种客户端应用。例如，浏览器类应用、搜索类应用、购物类应用、即时通讯工具、社交类应用、图像处理类应用等等。
终端设备101、102、103可以是硬件，也可以是软件。当终端设备101、102、103为硬件时，可以是各种电子设备，包括但不限于智能手机、平板电脑、电子书阅读器、膝上型便携计算机和台式计算机等等。当终端设备101、102、103为软件时，可以安装在上述所列举的电子设备中。其可以实现成多个软件或软件模块(例如用来提供分布式服务的多个软件或软件模块)，也可以实现成单个软件或软件模块。在此不做具体限定。
服务器105可以是提供各种服务的服务器，例如为终端设备101、102、103上安装的客户端应用提供服务支持的后端服务器。服务器105可以获取待推送图像的视觉特征，并根据视觉特征确定用户对待推送图像的偏好度，若偏好度符合预设条件，可以进一步向用户对应的终端(如终端设备101、102、103等)推送该待推送图像。
需要说明的是，上述待推送图像的视觉特征可以直接存储在服务器105的本地，服务器105可以直接提取本地所存储的待推送图像的视觉特征并进行处理，此时，可以不存在终端设备101、102、103和网络104。
需要说明的是，本公开的实施例所提供的图像推送方法一般由服务器 105执行，相应地，图像推送装置一般设置于服务器105中。
还需要指出的是，终端设备101、102、103中也可以安装有图像推送类应用，终端设备101、102、103也可以基于图像推送类应用根据待推送图像的视觉特征确定用户对待推送图像的偏好度，并在偏好度符合预设条件时，向用户对应的终端推送该待推送图像。此时，图像推送方法可以由终端设备101、102、103执行，相应地，图像推送装置也可以设置于终端设备101、102、103中。此时，示例性系统架构100可以不存在服务器105和网络104。
需要说明的是，服务器105可以是硬件，也可以是软件。当服务器105为硬件时，可以实现成多个服务器组成的分布式服务器集群，也可以实现成单个服务器。当服务器105为软件时，可以实现成多个软件或软件模块(例如用来提供分布式服务的多个软件或软件模块)，也可以实现成单个软件或软件模块。在此不做具体限定。
应该理解，图1中的终端设备、网络和服务器的数目仅仅是示意性的。根据实现需要，可以具有任意数目的终端设备、网络和服务器。
继续参考图2，其示出了根据本公开的图像推送方法的一个实施例的流程200。该图像推送方法包括以下步骤：
步骤201，获取待推送图像的视觉特征。
在本实施例中，待推送图像可以是各种类型、各种内容的图像。例如，待推送图像可以是某个物品的介绍图等。又例如，待推送图像可以是某个服务或活动的推广图等等。待推送图像的视觉特征表示待推送图像在视觉角度呈现出的特征。视觉特征可以包括设计特征和目标视觉特征。
其中，设计特征可以用于表示待推送图像在设计方面的特征。设计特征可以根据实际的应用场景或应用需求包括多种方面的设计特征。例如，包括但不限于：布局、背景装饰、风格、色系等等。
目标视觉特征可以利用向量嵌入(Embedding)得到，具体可以利用各 种针对图像的向量嵌入方法得到。一般地，目标视觉特征可以使用特征向量进行表示。
待推送图像的视觉特征可以基于各种方法确定。例如，设计特征可以由设计人员预先标注。目标视觉特征可以利用卷积神经网络等提取得到。作为示例，可以基于现有的分类模型和ResNet等网络结构，保留浅层的若干Block，再连接全连接层和预测输出层构建模型并训练，此时可以将全连接层的输出作为目标视觉特征。
图像推送方法的执行主体(如图1所示的服务器105等)可以从本地或其它存储设备获取待推送图像的视觉特征。需要说明的是，待推送图像的视觉特征可以由图像推送方法的执行主体确定，也可以由其它电子设备确定。
步骤202，根据视觉特征，确定用户对待推送图像的偏好度。
在本实施例中，用户对待推送图像的偏好度可以表示用户对待推送图像的感兴趣程度。具体地，偏好度可以根据实际的应用场景采用各种具体的表示方式。例如，偏好度可以采用点击率、浏览时长等数据指标来表示。
根据不同的应用场景，可以采用各种不同的方法根据待推送图像的视觉特征确定用户对待推送图像的偏好度。例如，可以根据统计历史时段内用户偏好度较高的图像的视觉特征，然后根据统计得到的视觉特征与待推送图像的视觉特征的相似度确定用户对待推送图像的偏好度。一般地，用户对待推送图像的偏好度与确定的相似度正相关。
又例如，在使用点击率等数据指标表示偏好度的情况下，可以采用各种数据指标预测方法确定用户对待推送图像的偏好度。作为示例，利用现有的各种点击率预测模型根据待推送图像的视觉特征确定用户对待推送图像的点击率。一般地，用户对待推送图像的偏好度与确定的点击率正相关。
步骤203，响应于确定偏好度符合预设条件，向用户对应的终端推送该待推送图像。
在本实施例中，预设条件可以根据实际的应用需求灵活设置。例如，预 设条件可以包括确定的偏好度不小于预设阈值。又例如，预设条件可以包括确定的偏好度对应的排序位置不小于预设位置。其中，排序位置可以指在预设的多个待推送图像分别对应的偏好度进行排序形成的排序中的位置。
若偏好度符合预设条件，可以表示用户对待推送图像比较感兴趣。此时，可以进一步向用户对应的终端(如用户所使用的终端等)推送该待推送图像，从而可以实现基于设计特征的图像推送，避免仅采用利用向量嵌入得到的目标视觉特征进行图像推送，忽略了用户对图像的设计特征的偏好需求。
在本实施例的一些可选的实现方式中，待推送图像的设计特征可以通过如下步骤确定：
步骤一、解析待推送图像的设计元素。
在本步骤中，一般地，待推送图像的设计可以由其具有的各种设计元素来体现。不同的设计可以包括各种不同的设计元素。对于同一个设计，不同的划分方式也会产生不同的设计元素。例如，设计元素包括但不限于：色系、字体、样式、图文布局等等。
具体地，可以采用各种解析方法确定待推送图像具有的设计元素。例如，可以由设计人员预先设置待推送图像的设计文档，并在设计文档中记录各设计元素。此时，可以通过读取待推送图像的设计文档解析待推送图像具有的各设计元素。
步骤二、识别设计元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计特征。
在本步骤中，设计元素所表示的语义可以指设计人员期望通过设计元素所表达的意图。例如，通过不同的色系可以表达庄重或轻松等不同的语义。
通过识别设计元素所表示的语义，可以得到语义识别结果，进而可以将得到的语义识别结果确定为设计特征，从而可以将待推送图像的设计意图转化为可解释的设计特征。
具体地，可以根据实际的应用场景采用各种不同的识别方法来识别设计 元素所表示的语义。例如，可以由设计人员预先标注设计元素和语义之间的对应关系。此时，可以通过查询预设的对应关系确定设计元素所表示的语义。
又例如，以背景的色系语义识别为示例，可以通过如下步骤确定背景的色系所表示的语义：
步骤一、统计背景所包含的各像素点的色值，以及筛选对应像素点数量最多的前十种色值。
步骤二、从十种色值中筛选饱和度不低于预设饱和度阈值且亮度不低于预设亮度阈值的色值，以排除饱和度过低或亮度过低的无效色或中立色。
步骤三、对筛选出的色值进行聚类，得到聚类结果。
步骤四、利用预设的色值-语义对应关系，查询聚类结果中的聚类簇所表示的语义。
现有技术中，仅基于图像Embedding化的特征提取和表示方式无法解析出图像包含的色系情感倾向、风格等创意设计特征，而创意设计特征是比较能传递视觉氛围和设计心智的，因此，仅基于图像Embedding化的特征提取和表示方式预测出的推送图像可能并不符合用户当前的情感诉求和心智传达。基于此，通过结合图像Embedding化的特征提取和图像对应的设计特征可以更全面的衡量图像和用户之间的匹配度，从而有助于提升推送图像的准确度，进而提升用户体验。
在本实施例的一些可选的实现方式中，设计特征可以包括以下至少一项：设计内容特征、设计表现特征。
其中，设计内容特征可以指呈现于内容方面的设计特征。例如，设计内容特征可以包括但不限于：Logo(商标/徽标)、图像中呈现的对象、图像中呈现的字符(如文案)、图像中的各种控件(如按钮等)等等。
设计表现特征可以指视觉表现方面的设计特征。例如，设计表现特征包括但不限于：色系、布局、图像中呈现的对象的图、背景装饰、风格等等。
通过从内容和表现两方面解析并识别待推送图像的设计特征，可以对待 推送图像的设计特征进行具象化，其能够通过这种方式比较全面且便捷地表示待推送图像的设计特征。
可选地，设计内容特征可以通过如下步骤确定：
步骤一、基于语义分割解析待推送图像的设计内容元素。
在本步骤中，可以采用现有的各种语义分割(Semantic Segmentation)方法，从待推送图像中识别其具有的各设计内容元素。
在一些情况下，可以根据需求预先设置设计内容元素集，此时，可以从待推送图像中识别其具有的、属于预设的设计内容元素集中的设计内容元素。
例如，可以利用预先训练的、基于语义分割实现的设计内容元素识别模型，从待推送图像中识别其具有的各设计内容元素。作为示例，可以识别待推送图像中的对象区域、字符区域、Logo区域、空间区域等等。
步骤二、识别设计内容元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计内容特征。
该步骤的语义识别可以参考上文中的相关描述，在此不再赘述。
利用语义分割可以准确识别各设计元素，从而更准确地表示待推送图像的设计特征。
可选地，设计表现特征可以通过如下步骤确定：
步骤一、解析待推送图像的设计文件，得到待推送图像的设计表现元素。
在本步骤中，待推送图像的设计文件可以指待推送图像的设计过程形成的各种资料。例如，待推送图像是利用某设计工具或应用制作的，则可以将利用该设计工具或应用形成的文件作为待推送图像的设计文件。
不同类型的设计文件可以采用不用的解析方法。例如，可以先通过设计文件读取对应的图层数量、图层类别和图层尺寸等，然后再具体解析各图层的内容以确定待推送图像的设计元素。
步骤二、识别设计表现元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计表现特征。
该步骤的语义识别可以参考上文中的相关描述，在此不再赘述。
通过对待推送图像的设计文件的解析和识别可以了解原始的设计语义，从而更准确地表示设计特征。
在一些情况下，目标视觉特征可以对待推送图像进行处理得到，也可以利用向量嵌入对待推送图像对应的图像模板进行处理得到。例如，利用卷积神经网络从待推送图像对应的图像模板提取特征向量。其中，图像模板可以指待推送图像的设计模板，可以包括各设计元素，可以不包括各设计元素具体的内容。
此时，在确定用户对待推送图像的偏好度符合预设条件之后，可以在设计模板中补充待推送图像的设计特征指示的设计元素的内容，并将补充后得到的图像作为待推送图像，以及向用户推送得到的待推送图像。
现有的基于投后调优的图像推送，由于每次进行创意设计投放时可能缺乏足够的先验知识，因此通常需要先进行灰度实验再进行投放调优。但灰度实验期间通常采用流量均分的方式，导致无法实现流量使用效率最优化，容易出现较差的创意设计占用较多的流量比例的情况。
本公开的上述实施例提供的方法通过结合待推送图像的设计特征和利用图像Embedding化得到的目标视觉特征确定用户对待推送图像的偏好度，进而基于偏好度进行图像推送，可以避免上述投后调优需要进行灰度实现投放等过程，而且在利用图像的高阶特征的基础上，进一步结合图像的设计特征所表达的色系情感倾向、风格等创意设计语义进行偏好度预测，有助于提升图像推送结果的准确度，使得推送的图像在设计方面更匹配用户的需求。
进一步参考图3，其示出了图像推送方法的又一个实施例的流程300。该图像推送方法的流程300，包括以下步骤：
步骤301，获取待推送图像的视觉特征。
步骤302，获取待推送图像的关联特征。
在本实施例中，关联特征可以指除视觉特征之外，与待推送图像相关联的各种特征。关联特征可以包括如下至少一项：属性特征、文本特征。
其中，文本特征可以表示待推送图像中的字符的特征。属性特征可以包括以下至少一项：用户的属性特征、待推送图像所描述的对象的属性特征。用户的属性特征可以包括用户的固有属性、行为属性等等。待推送图像所描述的对象可以是各种类型的对象。不同类型的对象可以具有不同的属性特征。
例如，待推送图像为某物品的介绍图，则待推送图像所描述的对象可以为该物品。此时，对象的属性特征即为该物品的属性特征(如成分、产地、保质期等等)。又例如，待推送图像为某服务的推广图，则待推送图像所描述的对象可以为该服务。此时，对象的属性特征即为该服务的属性特征(如开始时间、结束时间、地点等等)。
待推送图像的关联特征可以采用各种不同的方法确定。例如，对于属性特征，可以通过采集相关属性数据进行确定。又例如，对于文本特征，可以基于OCR(Optical Character Recognition，光学字符识别)等方法从待推送图像中提取其中的字符，然后利用NLP(Natural Language Processing，自然语言处理)等方法提取对应的文本特征。
例如，可以构建并训练长短期记忆网络(LSTM，Long Short-Term Memory)，以最后全连接层的输出的Embedding结果作为文本特征。
执行主体可以从本地或其他数据源获取待推送图像的关联特征。需要说明的是，待推送图像的关联特征可以由上述执行主体确定，也可以由其它电子设备确定。待推送图像的视觉特征和关联特征可以由同一电子设备确定，也可以由不同的电子设备确定。
步骤303，根据视觉特征和关联特征，确定用户对待推送图像的偏好度。
在本实施例中，可以综合待推送图像的视觉特征和关联特征，确定用户对待推送图像的偏好度。具体地，可以根据实际的应用需求采用各种各不同的偏好度确定方法。
例如，可以先根据待推送图像的视觉特征确定用户对待推送图像的第一偏好度，然后根据待推送图像的关联特征确定用户对待推送图像的第二偏好度。然后对第一偏好度和第二偏好度进行加权和等方式进行融合，并将融合结果确定为用户对待推送图像的偏好度。
又例如，可以利用各种基于深度学习的神经网络模型，根据待推送图像的视觉特征和关联特征，确定用户对待推送图像的偏好度。作为示例，可以预先训练偏好度确定模型，以待推送图像的视觉特征和关联特征作为输入，以用户对待推送图像的偏好度作为输出，从而利用偏好度确定模型得到用户对待推送图像的偏好度。
步骤304，响应于确定偏好度符合预设条件，向用户对应的终端推送该待推送图像。
在一些情况下，待推送图像的目标视觉特征、文本特征都可以基于神经网络模型来实现，以及偏好度预测和推送预测等都可以基于神经网络模型来实现。具体地，可以通过采集用户的历史行为记录等提取训练数据，以训练相关的神经网络模型。各个不同的神经网络模型可以独立训练，也可以在训练过程中进行多模态数据融合和校差等以实现多任务训练，以提升各神经网络模型的训练效果。例如，可以先训练基于向量嵌入提取特征向量的神经网络模型，此时，提取的特征向量可以表示目标视觉特征，同时训练提取文本特征的神经网络模型，然后将这些模型作为预训练模型，联合训练图像推送预测模型或偏好度预测模型等等。
本公开的上述实施例提供的方法通过结合待推送图像的视觉特征、文本特征和属性特征三种不同模态的特征数据，以进行图像推送，一般地，图像所描述的对象的属性、用户属性、视觉特征(如图像中呈现的对象展示图等)、文案等是比较容易吸引用户的几个方面，因此，利用视觉特征、文本特征和属性特征三种不同模态的特征数据可以全方面的考虑图像推送场景中影响推送效果的各种因素，从而有助于提升图像推送的效果。
进一步参考图4，其示出了图像推送方法的再一个实施例的流程400。该图像推送方法的流程400，包括以下步骤：
步骤401，获取待推送图像的视觉特征。
步骤402，根据视觉特征，确定用户对待推送图像的偏好度。
步骤403，响应于确定偏好度符合预设条件，向用户对应的终端推送该待推送图像。
步骤404，获取用户针对待推送图像的操作信息。
在本实施例中，操作信息可以用于表示用户执行的操作。操作信息指示的操作可以指用户在接收到该待推送图像之后，执行的各种操作。例如，用户的操作可以包括：点击/未点击、浏览/未浏览、收藏/未收藏、评论/未评论等等。
一般地，可以由用户对应的终端采集用户针对待推送图像的操作信息，然后从用户对应的终端获取用户针对待推送图像的操作信息。
步骤405，根据操作信息，构建知识图谱。
在本实施例中，知识图谱可以用于记录用户针对视觉特征的偏好信息。偏好信息可以表示用户的偏好。一般地，可以根据用户的操作信息了解用户对推送的图像的偏好，进而进行记录以构建知识图谱。知识图谱记录的内容具体可以根据实际的应用需求灵活设置。
作为示例，若用户点击了用于呈现某物品的待推送图像，则可以记录用户标识、物品标识、设计元素、设计元素的语义和设计意图。其中，设计意图可以指设计人员设计待推送图像时的设计意图。此时，知识图谱的结构可以如表1所示：
如表1所示，可以构建知识图谱来记录物品、用户、设计元素、设计语义、场景意图之间的对应关系。以饱和度作为示例，饱和度的语义具体可以包括低饱和、中饱和、和高饱和，这三种饱和度的顺序变化可以对应场景意 图从年轻到成熟的变化。因此，若根据用户的操作信息，确定用户偏好低饱和的设计语义，则可以在知识图谱中进行记录，以表示用户喜欢年轻的设计风格。
需要说明的是，知识图谱的构建通常是一个不断迭代和沉淀的过程，因此，可以在实际应用中根据持续地图像推送结果不断地更新知识图谱。
表1
Figure PCTCN2022136500-appb-000001
现有的基于投后调优的图像推送，由于无法对创意设计进行解析，使得每次的创意设计预测都无法沉淀可复用的知识。基于图像Embedding的图像推送，由于图像Embedding本身是高阶特征，通常是以向量的形式参与计算，使得无法了解筛选出的较佳的创意设计是哪些因素或特征贡献的，从而也无法根据筛选出的创意设计指导之后的创意设计。
针对这些情况，本公开的上述实施例提供的方法利用设计文件解析或基于语义分割的解析和设计语义识别，可以实现对待推送图像的可解释的低阶设计特征的抽取，从而解决图像Embedding化所带来的不可解释的问题，而且通过构建知识图谱可以根据每次的推送效果进行知识记录、沉淀和转化，进而可以利用构建的知识图谱辅助于后续针对用户的图像推送、图像搜索等各种应用场景。
下面参见图5，图5是根据本实施例的图像推送方法的一个示意性的应用场景500。在图5的应用场景中，对于指定物品的每个设计模板，可以先基于设计文件解析和语义分割等算法确定对应的设计特征5011，并记录该设计模板与其对应的设计特征之间的对应关系504。同时，可以利用卷积神经网络提取该设计模板的特征向量来作为目标视觉特征5012，从而得到由设计特征5011和目标视觉特征5012组成的视觉特征501。
此外，可以基于对OCR和NLP等技术提取该设计模板对应的文本特征502，以及预先基于统计分析获取用户属性特征5031和物品属性特征5032组成的属性特征503。
之后，可以将视觉特征501、文本特征502和属性特征503输入至预先训练的点击率预测模型505，得到该设计模板对应的点击率506。然后可以根据各设计模板分别对应的点击率，选择点击率较大的设计模板，并在设计模板中添加该设计模板对应的设计特征形成图像推送至用户。
进一步参考图6，作为对上述各图所示方法的实现，本公开提供了图像推送装置的一个实施例，该装置实施例与图2所示的方法实施例相对应，该 装置具体可以应用于各种电子设备中。
如图6所示，本实施例提供的图像推送装置600包括视觉特征获取单元601、偏好度确定单元602和推送单元603。其中，视觉特征获取单元601被配置成获取待推送图像的视觉特征，其中，视觉特征包括设计特征和目标视觉特征，目标视觉特征利用向量嵌入得到；偏好度确定单元602被配置成根据视觉特征，确定用户对待推送图像的偏好度；推送单元603被配置成响应于确定偏好度符合预设条件，向用户对应的终端推送该待推送图像。
在本实施例中，图像推送装置600中：视觉特征获取单元601、偏好度确定单元602和推送单元603的具体处理及其所带来的技术效果可分别参考图2对应实施例中的步骤201、步骤202和步骤203的相关说明，在此不再赘述。
在本实施例的一些可选的实现方式中，设计特征通过如下步骤确定：解析待推送图像的设计元素；识别设计元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计特征。
在本实施例的一些可选的实现方式中，设计特征包括以下至少一项：设计内容特征、设计表现特征。
在本实施例的一些可选的实现方式中，设计内容特征通过如下步骤确定：基于语义分割解析待推送图像的设计内容元素；识别设计内容元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计内容特征。
在本实施例的一些可选的实现方式中，设计表现特征通过如下步骤确定：解析待推送图像的设计文件，得到待推送图像的设计表现元素；识别设计表现元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计表现特征。
在本实施例的一些可选的实现方式中，上述图像推送装置还包括：关联特征获取单元(图中未示出)，被配置成获取待推送图像的关联特征，其中，关联特征包括以下至少一项：属性特征、文本特征，属性特征包括以下至少 一项：用户的属性特征、待推送图像描述的对象的属性特征；以及偏好度确定单元602进一步被配置成根据视觉特征和关联特征，确定用户对待推送图像的偏好度。
在本实施例的一些可选的实现方式中，上述图像推送装置还包括：操作信息获取单元(图中未示出)，被配置成获取用户针对待推送图像的操作信息；构建单元，被配置成根据操作信息，构建知识图谱，其中，知识图谱用于记录用户针对视觉特征的偏好信息。
本公开的上述实施例提供的装置，通过视觉特征获取单元获取待推送图像的视觉特征，其中，视觉特征包括设计特征和目标视觉特征，目标视觉特征利用向量嵌入得到；偏好度确定单元根据视觉特征，确定用户对待推送图像的偏好度；推送单元响应于确定偏好度符合预设条件，向用户对应的终端推送该待推送图像，可以在利用图像的高阶特征的基础上，进一步结合图像的设计特征所表达的色系情感倾向、风格等创意设计语义进行偏好度预测，有助于提升图像推送结果的准确度，使得推送的图像在设计方面更匹配用户的需求。
下面参考图7，其示出了适于用来实现本公开的实施例的电子设备(例如图1中的服务器)700的结构示意图。图7示出的服务器仅仅是一个示例，不应对本公开的实施例的功能和使用范围带来任何限制。
如图7所示，电子设备700可以包括处理装置(例如中央处理器、图形处理器等)701，其可以根据存储在只读存储器(ROM)702中的程序或者从存储装置708加载到随机访问存储器(RAM)703中的程序而执行各种适当的动作和处理。在RAM 703中，还存储有电子设备700操作所需的各种程序和数据。处理装置701、ROM 702以及RAM 703通过总线704彼此相连。输入/输出(I/O)接口705也连接至总线704。
通常，以下装置可以连接至I/O接口705：包括例如触摸屏、触摸板、 键盘、鼠标、摄像头、麦克风、加速度计、陀螺仪等的输入装置706；包括例如液晶显示器(LCD)、扬声器、振动器等的输出装置707；包括例如磁带、硬盘等的存储装置708；以及通信装置709。通信装置709可以允许电子设备700与其他设备进行无线或有线通信以交换数据。虽然图7示出了具有各种装置的电子设备700，但是应理解的是，并不要求实施或具备所有示出的装置。可以替代地实施或具备更多或更少的装置。图7中示出的每个方框可以代表一个装置，也可以根据需要代表多个装置。
特别地，根据本公开的实施例，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709从网络上被下载和安装，或者从存储装置708被安装，或者从ROM 702被安装。在该计算机程序被处理装置701执行时，执行本公开的实施例的方法中限定的上述功能。
需要说明的是，本公开的实施例所述的计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质或者是上述两者的任意组合。计算机可读存储介质例如可以是——但不限于——电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子可以包括但不限于：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机访问存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本公开的实施例中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。而在本公开的实施例中，计算机可读信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的 数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。计算机可读信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读信号介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。计算机可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于：电线、光缆、RF(射频)等等，或者上述的任意合适的组合。
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，使得该电子设备：获取待推送图像的视觉特征，其中，视觉特征包括设计特征和目标视觉特征，目标视觉特征利用向量嵌入得到；根据视觉特征，确定用户对待推送图像的偏好度；响应于确定偏好度符合预设条件，向用户对应的终端推送该待推送图像。
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实施例的操作的计算机程序代码，所述程序设计语言包括面向对象的程序设计语言—诸如Java、Smalltalk、C++，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络——包括局域网(LAN)或广域网(WAN)——连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。
附图中的流程图和框图，图示了按照本公开各种实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段、或代码的一部分，该模块、程序段、或代码的一部分包含一个或多个用于实现规定的逻辑功能的可执行 指令。也应当注意，在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个接连地表示的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或操作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
描述于本公开的实施例中所涉及到的单元可以通过软件的方式实现，也可以通过硬件的方式来实现。所描述的单元也可以设置在处理器中，例如，可以描述为：一种处理器包括视觉特征获取单元、偏好度确定单元和推送单元。其中，这些单元的名称在某种情况下并不构成对该单元本身的限定，例如，视觉特征获取单元还可以被描述为“获取待推送图像的视觉特征的单元，其中，视觉特征包括设计特征和目标视觉特征，目标视觉特征利用向量嵌入得到”。
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，本公开的实施例中所涉及的发明范围，并不限于上述技术特征的特定组合而成的技术方案，同时也应涵盖在不脱离上述发明构思的情况下，由上述技术特征或其等同特征进行任意组合而形成的其它技术方案。例如上述特征与本公开的实施例中公开的(但不限于)具有类似功能的技术特征进行互相替换而形成的技术方案。

Claims (16)

  1. 一种图像推送方法，包括：
    获取待推送图像的视觉特征，其中，所述视觉特征包括设计特征和目标视觉特征，所述目标视觉特征利用向量嵌入得到；
    根据所述视觉特征，确定用户对所述待推送图像的偏好度；以及
    响应于确定所述偏好度符合预设条件，向所述用户对应的终端推送所述待推送图像。
  2. 根据权利要求1所述的方法，其中，所述设计特征通过如下步骤确定：
    解析所述待推送图像的设计元素；以及
    识别所述设计元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计特征。
  3. 根据权利要求2所述的方法，其中，所述设计特征包括以下至少一项：设计内容特征、设计表现特征。
  4. 根据权利要求3所述的方法，其中，所述设计内容特征通过如下步骤确定：
    基于语义分割解析所述待推送图像的设计内容元素；以及
    识别所述设计内容元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计内容特征。
  5. 根据权利要求3所述的方法，其中，所述设计表现特征通过如下步骤确定：
    解析所述待推送图像的设计文件，得到所述待推送图像的设计表现元 素；以及
    识别所述设计表现元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计表现特征。
  6. 根据权利要求1所述的方法，其中，所述方法还包括：
    获取所述待推送图像的关联特征，其中，所述关联特征包括以下至少一项：属性特征、文本特征，所述属性特征包括以下至少一项：所述用户的属性特征、所述待推送图像描述的对象的属性特征；以及
    所述根据所述视觉特征，确定用户对所述待推送图像的偏好度，包括：
    根据所述视觉特征和所述关联特征，确定用户对所述待推送图像的偏好度。
  7. 根据权利要求1-6之一所述的方法，其中，在向所述用户对应的终端推送所述待推送图像之后，所述方法还包括：
    获取所述用户针对所述待推送图像的操作信息；以及
    根据所述操作信息，构建知识图谱，其中，所述知识图谱用于记录所述用户针对视觉特征的偏好信息。
  8. 一种图像推送装置，其中，所述装置包括：
    视觉特征获取单元，被配置成获取待推送图像的视觉特征，其中，所述视觉特征包括设计特征和目标视觉特征，所述目标视觉特征利用向量嵌入得到；
    偏好度确定单元，被配置成根据所述视觉特征，确定用户对所述待推送图像的偏好度；以及
    推送单元，被配置成响应于确定所述偏好度符合预设条件，向所述用户对应的终端推送所述待推送图像。
  9. 根据权利要求8所述的装置，其中，所述设计特征通过如下步骤确定：
    解析所述待推送图像的设计元素；以及
    识别所述设计元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计特征。
  10. 根据权利要求9所述的装置，其中，所述设计特征包括以下至少一项：设计内容特征、设计表现特征。
  11. 根据权利要求10所述的装置，其中，所述设计内容特征通过如下步骤确定：
    基于语义分割解析所述待推送图像的设计内容元素；以及
    识别所述设计内容元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计内容特征。
  12. 根据权利要求10所述的装置，其中，所述设计表现特征通过如下步骤确定：
    解析所述待推送图像的设计文件，得到所述待推送图像的设计表现元素；以及
    识别所述设计表现元素所表示的语义，得到识别结果，以及将得到的识别结果确定为设计表现特征。
  13. 根据权利要求8所述的装置，其中，所述装置还包括：
    关联特征获取单元，被配置成获取所述待推送图像的关联特征，其中，所述关联特征包括以下至少一项：属性特征、文本特征，所述属性特征包括以下至少一项：所述用户的属性特征、所述待推送图像描述的对象的属性特征；以及
    所述偏好度确定单元进一步被配置成根据所述视觉特征和所述关联特征，确定用户对所述待推送图像的偏好度。
  14. 根据权利要求8-13之一所述的装置，其中，所述装置还包括：
    操作信息获取单元，被配置成在向所述用户对应的终端推送所述待推送图像之后，获取所述用户针对所述待推送图像的操作信息；以及
    构建单元，被配置成根据所述操作信息，构建知识图谱，其中，所述知识图谱用于记录所述用户针对视觉特征的偏好信息。
  15. 一种电子设备/终端/服务器，包括：
    一个或多个处理器；
    存储装置，其上存储有一个或多个程序；
    当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如权利要求1-7中任一所述的方法。
  16. 一种计算机可读介质，其上存储有计算机程序，其中，该程序被处理器执行时实现如权利要求1-7中任一所述的方法。
PCT/CN2022/136500 2021-12-27 2022-12-05 图像推送方法和装置 WO2023124793A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111610765.3 2021-12-27
CN202111610765.3A CN114265948A (zh) 2021-12-27 2021-12-27 图像推送方法和装置

Publications (1)

Publication Number Publication Date
WO2023124793A1 true WO2023124793A1 (zh) 2023-07-06

Family

ID=80830161

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/136500 WO2023124793A1 (zh) 2021-12-27 2022-12-05 图像推送方法和装置

Country Status (2)

Country Link
CN (1) CN114265948A (zh)
WO (1) WO2023124793A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114265948A (zh) * 2021-12-27 2022-04-01 北京沃东天骏信息技术有限公司 图像推送方法和装置

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104915354A (zh) * 2014-03-12 2015-09-16 深圳市腾讯计算机系统有限公司 多媒体文件推送方法及装置
CN110413869A (zh) * 2018-12-05 2019-11-05 北京京东尚科信息技术有限公司 用于推送信息的方法和装置
CN110825968A (zh) * 2019-11-04 2020-02-21 腾讯科技（深圳）有限公司 信息推送方法、装置、存储介质和计算机设备
CN111144933A (zh) * 2019-12-13 2020-05-12 北京三快在线科技有限公司 商品推荐方法、装置、电子设备及存储介质
CN111949819A (zh) * 2019-05-15 2020-11-17 北京字节跳动网络技术有限公司 用于推送视频的方法和装置
CN113449193A (zh) * 2021-07-08 2021-09-28 中国工商银行股份有限公司 基于多分类图像的信息推荐方法及装置
CN114265948A (zh) * 2021-12-27 2022-04-01 北京沃东天骏信息技术有限公司 图像推送方法和装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104915354A (zh) * 2014-03-12 2015-09-16 深圳市腾讯计算机系统有限公司 多媒体文件推送方法及装置
CN110413869A (zh) * 2018-12-05 2019-11-05 北京京东尚科信息技术有限公司 用于推送信息的方法和装置
CN111949819A (zh) * 2019-05-15 2020-11-17 北京字节跳动网络技术有限公司 用于推送视频的方法和装置
CN110825968A (zh) * 2019-11-04 2020-02-21 腾讯科技（深圳）有限公司 信息推送方法、装置、存储介质和计算机设备
CN111144933A (zh) * 2019-12-13 2020-05-12 北京三快在线科技有限公司 商品推荐方法、装置、电子设备及存储介质
CN113449193A (zh) * 2021-07-08 2021-09-28 中国工商银行股份有限公司 基于多分类图像的信息推荐方法及装置
CN114265948A (zh) * 2021-12-27 2022-04-01 北京沃东天骏信息技术有限公司 图像推送方法和装置

Also Published As

Publication number Publication date
CN114265948A (zh) 2022-04-01

Similar Documents

Publication Publication Date Title
US11574470B2 (en) Suggested actions for images
TWI629644B (zh) 用於自影像偵測與辨認文字之非暫態電腦可讀取儲存媒體、方法與系統
CN110134931B (zh) 媒介标题生成方法、装置、电子设备及可读介质
US20180004397A1 (en) Systems and Methods of Providing Content Selection
CN114375435A (zh) 增强物理活动表面上的有形内容
WO2021190174A1 (zh) 信息确定方法、装置、计算机设备及存储介质
WO2017124116A1 (en) Searching, supplementing and navigating media
CN108416003A (zh) 一种图片分类方法和装置、终端、存储介质
CN113010703B (zh) 一种信息推荐方法、装置、电子设备和存储介质
CN111381909A (zh) 一种页面展示方法、装置、终端设备及存储介质
CN107209775A (zh) 用于搜索图像的方法和装置
US20210209289A1 (en) Method and apparatus for generating customized content based on user intent
CN110633423B (zh) 目标账号识别方法、装置、设备及存储介质
US11934953B2 (en) Image detection apparatus and operation method thereof
WO2024051609A1 (zh) 广告创意数据选取方法及装置、模型训练方法及装置、设备、存储介质
CN113204691B (zh) 一种信息展示方法、装置、设备及介质
CN113515942A (zh) 文本处理方法、装置、计算机设备及存储介质
WO2023016349A1 (zh) 一种文本输入方法、装置、电子设备和存储介质
CN111897950A (zh) 用于生成信息的方法和装置
WO2023124793A1 (zh) 图像推送方法和装置
US20190227634A1 (en) Contextual gesture-based image searching
US11653071B2 (en) Responsive video content alteration
US20230315990A1 (en) Text detection method and apparatus, electronic device, and storage medium
US8928929B2 (en) System for generating tag layouts
CN114995691A (zh) 一种文档处理方法、装置、设备和介质

Legal Events

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

Ref document number: 22914048

Country of ref document: EP

Kind code of ref document: A1