Information processing method, device and electronic equipment
The neural network model extracts product object pictures and user feature vectors, predicts click-through rates, and solves the problem of low quality of product object information recommendation in the existing technology, achieving more efficient resource utilization and user experience improvement.
Patent Information
- Application Number
- CN201911163055.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-11-25
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2039-11-25
AI Technical Summary
In the prior art, the quality of product object information recommendation is not high, resulting in waste of resources and a low proportion of user clicks or purchases.
A neural network model is used to extract features of product object images, generate high-dimensional feature vectors, combine user feature vectors, predict click-through rates through recommendation algorithms, and optimize display of recommended information.
It improves the click-through rate of recommendation information, reduces resource waste, and improves the user experience through more accurate feature expression and recommendation results.
Smart Images

Figure CN112837108B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of information processing technology, and in particular to information processing methods, devices and electronic equipment. Background Art
[0002] In the commodity object information service system, the recommended information of commodity objects is often provided to users in a variety of ways. For example, in the client homepage, the "Guess You Like" section can be used to provide users with recommended commodity object information in the form of information flow. There can be many reference information in the specific recommendation algorithm. For example, the system can locate the buyer's style preferences based on the consumer's usual browsing habits, browsing records, favorite products and other information, and recommend similar commodity objects to consumers based on this, and can display them through pictures of commodity objects. In addition, the system can also locate the consumption level of the crowd, and when recommending, it can determine the appropriate crowd based on the attributes and price of the commodity object and other information to achieve the expected effect of the user; furthermore, it can also comprehensively integrate the popularity of the commodity object (including click-through rate, collection rate, add-to-cart rate, etc.) and other information to recommend better quality commodity objects to consumer users, and so on.
[0003] In short, the existing method of providing recommendation information mainly selects specific recommended commodity objects for users based on the category, price, popularity and other information of the commodity objects. However, in specific implementation, the proportion of such recommended information actually clicked or purchased by users may not be high, and most of the recommended information may be ignored because it does not really meet the user's preferences, thus wasting various resources such as computing and network that have been occupied.
[0004] Therefore, how to further improve the quality of recommendation information and reduce resource waste has become a technical problem that needs to be solved by those skilled in the art. Summary of the invention
[0005] The present application provides an information processing method, device and electronic device, which can improve the quality of recommended information and reduce resource waste.
[0006] This application provides the following solutions:
[0007] An information processing method, comprising:
[0008] Obtain at least one neural network model for extracting features from a product object image;
[0009] Inputting the target product object image into the neural network model, obtaining the output result of the output layer of the neural network model, and the activation value of at least one intermediate layer; the activation value of the intermediate layer represents the feature information extracted from the product object image by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer;
[0010] According to the output result of the output layer and the activation value of the intermediate layer, feature values in multiple dimensions are determined, and a feature vector of the commodity object image is generated.
[0011] A method for determining user characteristic information, comprising:
[0012] Determine image information of multiple commodity objects associated with the user's historical behavior information;
[0013] Obtain at least one neural network model for extracting features from a product object image;
[0014] Inputting pictures of multiple commodity objects into the neural network model respectively, obtaining an output result of an output layer of the neural network model and an activation value of at least one intermediate layer; the activation value of the intermediate layer represents feature information extracted from the commodity object pictures by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0015] By counting the output results of the output layer and the activation values of the intermediate layer respectively corresponding to the pictures of the multiple commodity objects, feature values on multiple dimensions are determined to generate a feature vector of the user.
[0016] A commodity object recommendation method, comprising:
[0017] Determine the target recommendation user of the product object and the source database of the product object recommendation information;
[0018] Acquire feature vector information of the user and feature vector information of multiple commodity objects in the source database, wherein the feature vector of the commodity object includes feature values of a picture of the commodity object in multiple dimensions, and the feature values include: an output result of an output layer after the picture of the commodity object is input into at least one neural network model for extracting features from the picture of the commodity object, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the picture of the commodity object by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0019] The feature vector information of the user and the feature vector information of the plurality of commodity objects are input into a recommendation algorithm to predict click rate information obtained after the commodity objects are recommended to the user.
[0020] A commodity object recommendation method, comprising:
[0021] Determine the image information of the product object to be recommended;
[0022] Acquire feature vector information of the commodity object, wherein the feature vector of the commodity object includes feature values of the image in multiple dimensions, wherein the feature values include: an output result of an output layer after the image of the commodity object is input into at least one neural network model for extracting features from the commodity object image, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the image of the commodity object by the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0023] Obtain feature vector information of multiple users;
[0024] The feature vector information of the commodity object and the feature vector information of the multiple users are input into a recommendation algorithm to predict click rate information obtained after the commodity object to be recommended is recommended to the user.
[0025] A method for providing a commodity object, comprising:
[0026] In the process of providing similar commodity object information according to the target commodity object, feature vector information of the target commodity object and feature vector information of multiple commodity objects in the source database are obtained; wherein the feature vector of the commodity object includes feature values of the commodity object image in multiple dimensions, and the feature values include: the output result of the output layer after the commodity object image is input into at least one neural network model for extracting features from the commodity object image, and the activation value of at least one intermediate layer; the activation value of the intermediate layer is the feature information extracted from the commodity object image by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons;
[0027] By performing a similarity comparison between the feature vector information of the target commodity object and the feature vector information of multiple commodity objects in a source database, commodity object information having a similarity to the designated commodity object that meets the condition is provided.
[0028] A method for providing commodity object information, comprising:
[0029] Receiving an access request to a target page, wherein the target page is associated with a plurality of resource locations for displaying information of a plurality of commodity objects, wherein the commodity objects are associated with a plurality of different pictures;
[0030] Acquire feature vector information corresponding to the plurality of different images, and feature vector information of the visitor user; wherein the feature vector corresponding to the image includes feature values of the image in multiple dimensions, and the feature values include: the output result of the output layer after the image is input into at least one neural network model for extracting features from the commodity object image, and the activation value of at least one intermediate layer; the activation value of the intermediate layer is the feature information extracted from the commodity object image by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the upper layer;
[0031] Inputting the feature vector information of the visitor user and the feature vector information corresponding to the multiple pictures of the commodity object into the prediction algorithm respectively, so as to predict the click rate information obtained after the multiple pictures of the commodity object are displayed to the user respectively;
[0032] According to the prediction result, the pictures whose click rates meet the conditions are displayed as representative pictures of the corresponding commodity objects in the resource position corresponding to the target page.
[0033] An information recommendation method, comprising:
[0034] Determine the images associated with the target product object;
[0035] By performing feature analysis on the image, determining display scene category information corresponding to the commodity object in the image;
[0036] Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in;
[0037] Determine the target user according to the display scene category information corresponding to the image and the display scene category information that the user is interested in;
[0038] The picture is recommended to a client associated with the target user.
[0039] An information recommendation method, comprising:
[0040] Determine the video associated with the target product object;
[0041] Extracting at least one frame of image from the video, and determining display scene category information corresponding to the commodity object in the video by performing feature analysis on the image;
[0042] Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in;
[0043] Determine a target user according to the display scene category information corresponding to the video and the display scene category information that the user is interested in;
[0044] The video is recommended to a client associated with the target user.
[0045] An information processing device, comprising:
[0046] A model acquisition unit, used to acquire at least one neural network model for extracting features from a commodity object image;
[0047] An analysis unit, configured to input a target product object image into the neural network model, obtain an output result of an output layer of the neural network model, and an activation value of at least one intermediate layer; the activation value of the intermediate layer represents feature information extracted from the product object image by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0048] The image feature vector generating unit is used to determine the feature values in multiple dimensions according to the output result of the output layer and the activation value of the intermediate layer, and to generate the feature vector of the commodity object image.
[0049] A device for determining user characteristic information, comprising:
[0050] A picture information determination unit, used to determine picture information of multiple commodity objects associated with the user's historical behavior information;
[0051] A model acquisition unit, used to acquire at least one neural network model for extracting features from a commodity object image;
[0052] An analysis unit, configured to input pictures of a plurality of commodity objects into the neural network model respectively, and obtain an output result of an output layer of the neural network model, and an activation value of at least one intermediate layer; the activation value of the intermediate layer represents feature information extracted from the commodity object pictures by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0053] The user feature vector generating unit is used to determine feature values in multiple dimensions by counting the output results of the output layer and the activation values of the intermediate layer respectively corresponding to the pictures of the multiple commodity objects, so as to generate the feature vector of the user.
[0054] A commodity object information recommendation device, comprising:
[0055] An information determination unit, used to determine a target recommendation user for a commodity object and a source database for commodity object recommendation information;
[0056] a feature vector acquisition unit, used to acquire feature vector information of the user and feature vector information of multiple commodity objects in the source database, wherein the feature vector of the commodity object includes feature values of a picture of the commodity object in multiple dimensions, wherein the feature values include: an output result of an output layer after the picture of the commodity object is input into at least one neural network model for extracting features from the picture of the commodity object, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the picture of the commodity object by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0057] The prediction unit is used to input the feature vector information of the user and the feature vector information of the multiple commodity objects into a recommendation algorithm to predict click rate information obtained after the commodity objects are recommended to the user.
[0058] A commodity object recommendation device, comprising:
[0059] A picture information determination unit, used to determine the picture information of the commodity object to be recommended;
[0060] A commodity feature vector acquisition unit, used to acquire feature vector information of the commodity object, wherein the feature vector of the commodity object includes feature values of the image in multiple dimensions, wherein the feature values include: an output result of an output layer after the image of the commodity object is input into at least one neural network model for extracting features from the commodity object image, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the image of the commodity object by the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0061] A user feature vector acquisition unit, for acquiring feature vector information of multiple users;
[0062] The prediction unit is used to input the feature vector information of the commodity object and the feature vector information of the multiple users into the recommendation algorithm to predict the click rate information obtained after the commodity object to be recommended is recommended to the user.
[0063] A device for providing a commodity object, comprising:
[0064] A feature vector acquisition unit is used to acquire feature vector information of a target product object and feature vector information of multiple product objects in a source database in a process of providing similar product object information according to a target product object; wherein the feature vector of a product object includes feature values of a product object image in multiple dimensions, and the feature values include: an output result of an output layer after the product object image is input into at least one neural network model for extracting features from the product object image, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the product object image by the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0065] The similar commodity object information providing unit is used to provide commodity object information that meets the condition of similarity with the designated commodity object by performing similarity comparison between the feature vector information of the target commodity object and the feature vector information of multiple commodity objects in the source database.
[0066] A device for providing commodity object information, comprising:
[0067] An access request receiving unit, used for receiving an access request for a target page, wherein the target page is associated with a plurality of resource locations for displaying information of a plurality of commodity objects, wherein the commodity objects are associated with a plurality of different pictures;
[0068] A feature vector acquisition unit is used to acquire feature vector information corresponding to the plurality of different pictures, and feature vector information of a visitor user; wherein the feature vector corresponding to the picture includes feature values of the picture in multiple dimensions, and the feature values include: an output result of an output layer after the picture is input into at least one neural network model for extracting features from a picture of a commodity object, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the picture of the commodity object by the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0069] A prediction unit, used to input the feature vector information of the visitor user and the feature vector information corresponding to the multiple pictures of the commodity object into a prediction algorithm, so as to predict the click rate information obtained after the multiple pictures of the commodity object are displayed to the user;
[0070] The representative image determination unit is used to display the image with a click rate that meets the conditions as the representative image of the corresponding commodity object in the resource position corresponding to the target page according to the prediction result.
[0071] An information recommendation device, comprising:
[0072] A picture determination unit, used to determine a picture associated with a target product object;
[0073] A display scene determination unit, configured to determine display scene category information corresponding to the commodity object in the picture by performing feature analysis on the picture;
[0074] A user characteristic information acquisition unit, used to determine a user set and acquire user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the user is interested in;
[0075] A target user determination unit, configured to determine a target user according to the display scene category information corresponding to the image and the display scene category information in which the user is interested;
[0076] The picture recommendation unit is used to recommend the picture to a client associated with the target user.
[0077] An information recommendation device, comprising:
[0078] A video determination unit, used to determine a video associated with a target product object;
[0079] A display scene determination unit, configured to extract at least one frame of image from the video, and determine display scene category information corresponding to the commodity object in the video by performing feature analysis on the image;
[0080] A user characteristic information acquisition unit, used to determine a user set and acquire user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the user is interested in;
[0081] A target user determination unit, configured to determine a target user according to the display scene category information corresponding to the video and the display scene category information in which the user is interested;
[0082] The video recommendation unit is used to recommend the video to a client associated with the target user.
[0083] An electronic device, comprising:
[0084] one or more processors; and
[0085] A memory associated with the one or more processors, the memory is used to store program instructions, and when the program instructions are read and executed by the one or more processors, the methods described above are executed.
[0086] According to the specific embodiments provided in this application, this application discloses the following technical effects:
[0087] Through the embodiments of the present application, it is possible to extract features from commodity object images through a neural network model, and not only can the output results of the model output layer be used as feature values on specific dimensions in the image feature vector, but the activation values of the model middle layer can also be used as feature values on more dimensions in the feature vector. In this way, a high-dimensional feature vector can be generated for a commodity object image, which can include not only feature values on dimensions that humans can understand and define (output results of the output layer), but also feature values on dimensions that humans may not be able to understand or accurately define (activation values of the middle layer), so that this feature vector can more completely express the characteristics of the image. Furthermore, in scenarios such as commodity object information recommendation, more accurate recommendation results can be obtained, the click-through rate of recommended information can be improved, and resource waste can be reduced.
[0088] Of course, any product implementing the present application does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0089] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0090] Figure 1 It is a schematic diagram of the system architecture provided by the embodiment of the present application;
[0091] Figure 2 is a flow chart of the first method provided in an embodiment of the present application;
[0092] Figure 3 is a schematic diagram of a neural network structure provided in an embodiment of the present application;
[0093] Figure 4 is a flow chart of the second method provided in an embodiment of the present application;
[0094] Figure 5is a flowchart of the third method provided in an embodiment of the present application;
[0095] Figure 6 is a flowchart of the fourth method provided in an embodiment of the present application;
[0096] Figure 7 is a flowchart of the fifth method provided in an embodiment of the present application;
[0097] Figure 8 is a flowchart of the sixth method provided in an embodiment of the present application;
[0098] Fig. 9 is a flowchart of the seventh method provided in an embodiment of the present application;
[0099] Fig.10 is a flowchart of an eighth method provided in an embodiment of the present application;
[0100] Fig.11 is a schematic diagram of a first device provided in an embodiment of the present application;
[0101] Fig.12 is a schematic diagram of a second device provided in an embodiment of the present application;
[0102] Fig.13 is a schematic diagram of a third device provided in an embodiment of the present application;
[0103] Fig.14 is a schematic diagram of a fourth device provided in an embodiment of the present application;
[0104] Fig.15 is a schematic diagram of a fifth device provided in an embodiment of the present application;
[0105] Fig.16 is a schematic diagram of a sixth device provided in an embodiment of the present application;
[0106] Fig.17 is a schematic diagram of a seventh device provided in an embodiment of the present application;
[0107] Fig.18 is a schematic diagram of an eighth device provided in an embodiment of the present application;
[0108] Fig.19 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0109] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.
[0110] First of all, it should be noted that the inventor of this application found in the process of implementing this application that pictures of commodity objects usually have some characteristics. For example, there will be respective main color information in the pictures; in addition, for commodity objects such as clothing, the shooting scenes may also be different, some may be "street shots" conducted outdoors, some may be "studio shots" conducted in indoor studios, some have human models to show the usage status, and some are detailed displays without models. In addition, in the case of human models, the display range, human posture, expression, etc. of the human models may also be different. Different features convey different tones to users, and different users may have different degrees of interest in various different features. Therefore, when providing some recommended commodity object information to users, whether the recommended commodity object can obtain user clicks and other operations is not only related to the basic information such as the category and brand of the commodity object, but also closely related to whether the picture of the recommended commodity object meets the user's interest points. For example, a user prefers khaki color for a windbreaker, and when browsing the product information, most of the pictures of the product clicked by the user have the following characteristics: the pictures are taken in an indoor studio and shown by a human model, the human model is full-body, etc. If the above information can be effectively used to recommend product information, the recommended information will have a higher probability of obtaining user click-through rate.
[0111] Therefore, in specific implementation, some image processing algorithms can be used to extract features from product object images. For example, the shooting scene of the image, the presence or absence of human models or props, and even the display range of human models, human posture, expression, etc. can all be trained through deep learning and other methods. Then, the corresponding features in the image can be extracted through the trained model. In this way, a feature vector of the product object image can be generated, and then in the specific recommendation algorithm, the specific recommendation scheme can be determined by calculating the feature vector of the product object image and the user feature vector. The method of describing the features of the product object image through this feature vector and inputting it into the recommendation algorithm for calculation can improve the click-through rate of the recommended information to a certain extent.
[0112] However, when extracting image features in the above manner, the number of features that can be obtained is usually limited, and accordingly, the dimension of the feature vector of the generated commodity object is usually in the order of dozens or tens. Moreover, the specific features are usually some relatively vivid features, including the shooting scene, the presence or absence of a human model or prop, and even the display range of the human model, human posture, expression, etc., as mentioned above, and may also include the main color tone and brightness of the picture, etc. These vivid features are usually defined according to human understanding, but in fact, a picture usually contains more features, which may be relatively abstract features, and may even be beyond the scope of human understanding, discovery or definition. If these abstract features are also added to the feature vector of the commodity object picture to generate a higher-dimensional feature vector, the features of the picture can be expressed more completely and richly. Furthermore, when this high-dimensional feature vector is input into a specific recommendation algorithm for calculation, it is conducive to obtaining more refined and accurate recommendation information and obtaining a better click-through rate for the recommended information.
[0113] To achieve the above purpose, the embodiment of the present application can extract the features of the commodity object picture through a neural network model. Since the neural network usually includes multiple layers, each layer can include multiple neurons (wherein, the number of neurons in the input layer and the output layer is often fixed, and the middle layer can be freely specified, for example: for a three-layer neural network, the input layer has 3 input neurons, the output layer has 2 neurons, the middle layer can have 4 neurons, or 5 neurons, etc.). Among them, the specific output result of the output layer is usually determined according to the training direction specified in the training process. For example, a neural network is used to identify the shooting scene of a certain picture. After a specific picture is input into the neural network, the output result of the output layer is whether the shooting scene corresponding to the picture is a street shot or a studio shot, or what are the probabilities of various. But at the same time, the neural network also has some intermediate layers, and the activation value of each neuron in these intermediate layers can be various features extracted from the picture, or, by the upper layer of neurons, the features obtained by weighted summation and nonlinear transformation. These features may be abstract features that humans cannot understand or define, but because these features can affect the output result of the output layer, they are also meaningful. To this end, in an embodiment of the present application, in the process of extracting features from a commodity object image using a neural network, in addition to determining the feature values of the commodity object in one or more dimensions based on the output results of the output layer, the output values of some or all of the intermediate layers (usually referred to as activation values) can also be obtained, and the activation values of the intermediate layers can also be used as the feature values of the commodity object in more dimensions. In this way, the feature vector actually generated for the commodity object can include feature values in some dimensions that are more figurative and can be understood by humans, as well as some more abstract feature values. In other words, abstract features can be used as a supplement to image features, thereby obtaining a high-dimensional feature vector for the commodity object. This high-dimensional feature vector can then be input into a specific recommendation algorithm to obtain a more refined and accurate recommendation result.
[0114] In terms of system architecture, see Figure 1, the embodiment of the present application may mainly involve the server and client of the commodity object information service system. Among them, the server can generate specific recommendation information and then send it to the client of the corresponding user, so that the user can browse the specific recommendation information through the client. Specifically, the server can also provide a feature extraction module, in which multiple neural network models can be stored, which can be used to extract the features contained in the commodity object picture from different angles. For example, neural network A is used to identify the main area and background area of the commodity object in the picture, neural network B is used to identify the shooting scene category of the picture, and so on. Specifically, when it is necessary to extract the feature information of the picture of a certain commodity object, the same picture can be input into multiple neural networks for feature extraction. In addition, the feature extraction module can also include a feature vector generation module, which can obtain the output results of the final output layer and the activation values of one or more intermediate layers from the above-mentioned multiple neural networks, and then these output results and activation values can be used as feature values in multiple dimensions to generate a feature vector of a specific picture. The feature vector can be input into a specific recommendation algorithm module to obtain a specific recommendation result and provide it to the client. Of course, in actual applications, the high-dimensional feature vectors about product objects generated by the solution provided in the embodiments of the present application can also be applied to other scenarios outside the recommendation system. For example, when placing product object information on a page, it is possible to determine which specific image to use as a representative image to obtain a higher user click-through rate, etc., based on the above-mentioned feature vector information.
[0115] The specific implementation scheme provided in the embodiments of the present application is described in detail below.
[0116] Embodiment 1
[0117] First, this embodiment provides a method for information processing. Figure 2 , the method may specifically include:
[0118] S201: Obtain at least one neural network model for extracting features from a product object image;
[0119] In an embodiment of the present application, first, a plurality of neural network models can be obtained, and such models can be models having the function of extracting features from commodity object images. Among them, the neural network model has corresponding model categories according to different feature recognition objects. The reason why there are multiple models is that for the same picture, the features that need to be extracted may be multifaceted, and these features may not be completely mutually exclusive. Therefore, it is difficult to extract these features through the same model. To this end, multiple different neural network models can be used to achieve feature extraction of commodity object images from multiple different angles. For example, it can specifically include a model for distinguishing the area where the commodity object theme is located and the background area, and it can also include a model for identifying whether the shooting scene belongs to street shooting or studio shooting, and it can also include a model for identifying the display range of human models, and so on. In short, for the same picture that needs to be extracted, it can be input into multiple different neural network models respectively to extract features from different angles.
[0120] In a specific implementation, the neural network model may be some existing models, for example, there may be models for recognizing human faces in the prior art, etc. Or, in more cases, it may be obtained through training with training samples, wherein the training samples may include multiple commodity object image information, and annotation information for annotating the commodity object images according to factors affecting the click-through rate of the images. That is, the model is trained with specific images and annotation information, so that the trained model can be used to identify the features corresponding to the annotation information.
[0121] For example, the specific annotation information may include: the area corresponding to the main body of the commodity object in the training sample image and the area corresponding to the background. At this time, after the image of the specific commodity object is input into the trained model, the output result of the output layer of the neural network model may include: the category information that multiple pixels in the target commodity object image belong to the main body of the commodity object or the background. Specifically, assuming that the image of a certain commodity object is a 100×100 pixel image, the output of the model may be a 100×100 matrix, and the value of each element in the matrix may be 1 or 0, wherein 1 may indicate that the corresponding pixel belongs to the main body of the commodity object, and 0 indicates that the corresponding pixel belongs to the background, and so on. Since the main color tone of the main part of the commodity object and other information may be a factor affecting the user click rate in the specific implementation, after obtaining the above output result, the pixel attributes of each pixel in the element category may be determined according to the output result, the target pixel attribute is determined by comparing the pixel attributes of each pixel, and then the feature value of the data object image in at least one dimension is determined according to the target pixel attribute. For example, in a specific implementation method, the color attribute information of multiple pixels belonging to the main body of the commodity object in the target commodity object image can be determined based on the output result, and the main color tone information of the main part of the commodity object can be determined. Then, the main color tone information of the main part of the commodity object can be determined as the feature value on one dimension in the feature vector.
[0122] Alternatively, in another way, the proportion of the main part and the background part of the commodity object in the target commodity object image, and / or the number information of connected areas can also be determined. In this way, the proportion and / or the number information of the connected areas can also be determined as the feature values on some dimensions in the feature vector, and so on.
[0123] In addition, the annotation information corresponding to the training sample may also be: the shooting scene category information corresponding to the training sample image; in this case, the output result of the output layer of the neural network model includes: the category information to which the shooting scene of the target commodity object image belongs. Specifically, the shooting scene category information may also have different classification methods from a variety of different angles, for example, it may include indoor shooting or outdoor shooting, shooting of the commodity object in use by models or props, or shooting of the details of the commodity object without models or props. Among them, different neural network models can be used for feature extraction under different classification methods.
[0124] Furthermore, for training sample images that use models to display the usage status of commodity objects, specific annotation information may also include: feature information of the model. Specifically, the feature information of the model can also be divided into multiple categories from different perspectives, for example, including: whether the target commodity image includes the full body / half body of the model, or whether the face image of the model is included, or the posture feature information of the model, or the expression feature information of the model, etc. Similarly, different neural network models can be used for feature extraction under different division methods.
[0125] S202: Input the target product object image into the neural network model to obtain the output result of the output layer of the neural network model and the activation value of at least one intermediate layer; the activation value of the intermediate layer represents the feature information extracted from the product object image by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer;
[0126] After obtaining multiple neural network models, these models can be used to extract features from images of specific target commodity objects. Among them, the target commodity object can be a specific commodity object that needs to be recommended to the user, for example, a new product newly launched by a merchant, etc. Alternatively, it can also be a commodity object in a pre-set commodity pool when it is necessary to provide users with recommendation information such as "Guess you like", etc. In short, for each specific commodity object that may be recommended, the image of the commodity object can be input into multiple neural network models respectively to extract features to generate a feature vector. Among them, the image of the commodity object can specifically be a picture used as a representative image of the commodity object. The so-called representative image means that when the commodity object is placed on a page for display, a picture used to represent the commodity object is displayed in the resource position of the page.
[0127] In the embodiment of the present application, after a specific picture is input into the neural network model, the specific features obtained may include not only the output result of the output layer of the neural network model, but also the activation value of the intermediate layer. Figure 3As shown, for a neural network model, it usually includes multiple layers, each layer includes multiple neurons. Specifically, in a convolutional neural network, each layer can be called a convolution layer, and a convolution layer usually includes several feature planes (featureMap), and each feature plane is composed of some neurons arranged in a rectangular shape. Among them, in the process of feature extraction through the neural network model, each neuron in the middle layer is also extracting features of the picture respectively. The features extracted by these neurons may be some abstract features that humans cannot understand, but because they are features related to the feature judgment of the final output, they will also have a certain reference value. For this reason, in an embodiment of the present application, in addition to obtaining the output result of the output layer of the neural network model, the activation value of the middle layer can also be obtained, and these activation values can also be used as part of describing the features of the picture.
[0128] That is to say, in an embodiment of the present application, for a specific image of a commodity object that needs to be recommended or processed, it can be input into multiple neural network models, and for each neural network model, the output result of the specific output layer and the activation value of at least one intermediate layer can be obtained.
[0129] Among them, according to the different categories of specific neural network models, the output value of the specific output layer and the activation value of the intermediate layer are also different. For example, in one case, the output result of the output layer of the neural network model may include: the element category feature information of the image pixels in the target commodity object picture, and the element category includes the commodity object body or background. At this time, the activation value of the intermediate layer includes: in the process of obtaining the element category feature information, the feature information extracted from the target commodity object picture by the intermediate layer of the neural network model, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons. Alternatively, the output result of the output layer of the neural network model includes: the category feature information of the shooting scene of the target commodity object picture; at this time, the activation value of the intermediate layer includes: in the process of obtaining the category feature information of the shooting scene, the feature information extracted from the target commodity object picture by the intermediate layer of the neural network model, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons. Alternatively, the output result of the output layer of the neural network model may also include: for training sample pictures that show the usage status of the commodity object through the model character, the feature information of the model character; at this time, the activation value of the middle layer includes: in the process of obtaining the feature information of the model character, the feature information extracted from the target commodity object picture by the middle layer of the neural network model, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer.
[0130] S203: Determine feature values in multiple dimensions according to the output result of the output layer and the activation value of the intermediate layer, and generate a feature vector of the product object image.
[0131] After obtaining the output results of the specific neural network model and the activation values of the intermediate layer, they can be determined as feature values on multiple dimensions, and then the feature vector of the commodity object picture can be generated. Among them, for different pictures, when generating feature vectors, the specific dimension information can be corresponding. For example, the first dimension can represent the brightness information of the picture, the second dimension represents the main color information of the main part of the commodity object in the picture, the third dimension represents the shooting scene category of the picture, and so on; and in some dimensions that are ranked later, there may be some abstract features, and the dimension identifiers of these features can be replaced by numbers, such as the 500th dimension, the 700th dimension, and so on. Of course, these dimensions represented by numbers can specifically correspond to the activation values of the same neuron in the same intermediate layer of the same neural network. Through the above-mentioned dimensional alignment, it is convenient to perform operations between the feature vectors of different pictures, for example, calculating the similarity of pictures.
[0132] In addition, in the specific implementation, the output results of the output layer and the activation values of the intermediate layer can also be screened. For example, after obtaining the output values and the activation values of the intermediate layer from multiple models, the specific number of features obtained may be very large. At this time, these features can be first screened to filter out some useless features, and then the feature values in multiple dimensions can be determined based on the screening results to improve the accuracy of the algorithm.
[0133] It should be noted that, in the embodiment of the present application, the specific commodity object image may include a static image (photo, etc.) associated with the commodity object, or may also include an image extracted from a video associated with the commodity object, etc. For the former, when specifically recommending to the user, the image of the commodity object may be recommended to the client associated with the user, and for the latter, after generating a feature vector from the image extracted from the video and matching it with the user, the video of the commodity object may be recommended to the client associated with the user.
[0134] In summary, through the embodiments of the present application, it is possible to extract features from commodity object images through a neural network model, and not only can the output results of the model output layer be used as feature values on specific dimensions in the image feature vector, but the activation values of the model middle layer can also be used as feature values on more dimensions in the feature vector. In this way, a high-dimensional feature vector can be generated for the commodity object image, which may include not only feature values on dimensions that humans can understand and define, but also feature values on dimensions that humans may not be able to understand or accurately define, so that this feature vector can more completely express the characteristics of the image. Furthermore, in scenarios such as commodity object information recommendation, more accurate recommendation results can be obtained, the click-through rate of recommended information can be improved, and resource waste can be reduced.
[0135] It should be noted that the embodiments of the present application may involve the use of user data. In actual applications, user-specific personal data can be used in the scheme described herein within the scope permitted by applicable laws and regulations, subject to the requirements of applicable laws and regulations of the country where the user is located (for example, with the user's explicit consent, effective notification to the user, etc.).
[0136] Embodiment 2
[0137] The above-mentioned embodiment 1 provides a method for obtaining a feature vector of a specific commodity object image, and in other application scenarios, a similar scheme can also be used to obtain a user feature vector. Among them, the so-called user feature vector refers to a vector used to describe user features, and the user here can refer to a consumer user, a buyer user, etc. in a commodity object information service system. In scenarios such as recommending commodity object information to users, or recommending other users with common preferences to users, it is usually necessary to input the user feature vector into a specific matching algorithm for calculation. Among them, the user feature vector may include feature values on multiple dimensions, and specific dimensions may include basic feature values on dimensions such as the user's age, gender, and occupation. In addition, according to the information of commodity objects related to the user's historical behavior (browsing, collection, purchase, purchase, etc.), the commonalities of the commodity objects of interest to the user can be extracted, and the information on these dimensions can also be used as feature values on specific dimensions in the user feature vector to describe the user's features.
[0138] Among them, when extracting specific commodity object feature information from commodity objects associated with user historical behavior information, in addition to including basic information such as the category and brand of the commodity object, some visual features of the pictures associated with the commodity object can also be included, such as the shooting scene category mentioned above, whether it contains human models, etc. When obtaining these visual features, they can also be obtained through a neural network model, and the output results of the output layer of the neural network model and the activation values of the intermediate layer can also be obtained, thereby generating a high-dimensional feature vector for the user. It can include feature values on dimensions that humans can understand and define, and it can also include feature values on dimensions that humans cannot understand or define exactly, so as to achieve a more comprehensive and detailed description of user characteristics. So that the commodity object information recommended to the user or the information of other users is more accurate.
[0139] For details, see Figure 4 The second embodiment provides a method for determining user characteristic information, which may specifically include:
[0140] S401: Acquire image information of multiple commodity objects associated with the user's historical behavior information;
[0141] The image of a commodity object may specifically refer to a representative image of the commodity object, that is, an image used as the header image of the commodity object in the detail page of the commodity object, or an image displayed in the resource position when the commodity object is placed on a specific commodity object list page. If a user clicks on a commodity object from among many commodity objects, the factor that attracts the user to the commodity object may be not only its category, brand, etc., but also the representative image of the commodity object that arouses the user's interest. Therefore, the characteristics of such commodity object images can also be used as part of user characteristics to describe user characteristics.
[0142] S402: Obtain at least one neural network model for extracting features from a product object image;
[0143] S403: Inputting pictures of multiple commodity objects into the neural network model respectively, obtaining an output result of the output layer of the neural network model and an activation value of at least one intermediate layer; the activation value of the intermediate layer represents feature information extracted from the commodity object pictures by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in the previous layer;
[0144] S404: Determine feature values in multiple dimensions by counting output results of the output layer and activation values of the intermediate layer respectively corresponding to the images of the multiple commodity objects, so as to generate a feature vector of the user.
[0145] After obtaining the output results of the output layer corresponding to the images of multiple product objects and the activation value of the intermediate layer respectively, statistics can be performed, for example, the number or frequency of occurrence of various features can be counted to determine the user's interest in the features on the corresponding dimensions, and then the information such as the interest level corresponding to these dimensions is used as feature values to generate a specific user feature vector.
[0146] Embodiment 3
[0147] The above-mentioned embodiment 1 provides a method for obtaining a feature vector of a specific commodity object image, and after determining the feature vector of a specific commodity object image, it can be applied to a specific commodity object recommendation and other scenarios. Specifically, the embodiment 2 provides one of the specific application scenarios, which can be a scenario of recommendation information such as "Guess you like" provided through the client homepage, etc. That is to say, when a user visits the client homepage, the homepage can provide a "Guess you like" section, in which multiple resource positions can be provided, and each resource position can be associated with specific recommended commodity object information. The commodity object information recommended here is usually determined by matching the characteristics of the current visitor user with the commodity objects in the commodity object library. Among them, the characteristics of the commodity object can be described with the commodity object feature vector, which can include the characteristics corresponding to the specific commodity object image, and the characteristics of the commodity object image can include some visualized features, and can also include abstract features obtained through the activation value of the middle layer of the neural network, so as to generate a high-dimensional feature vector of the commodity object. Then, the feature vector of the commodity object and the user vector are input into the recommendation algorithm for matching operation to determine the commodity object that is specifically suitable for recommendation to the current user. The user vector may be expressed by a vector in the prior art, or by a high-dimensional feature vector in the second embodiment, and so on.
[0148] For details, see Figure 5 The third embodiment provides a method for recommending commodity objects, which may specifically include:
[0149] S501: Determine the target recommendation user of the commodity object and the source database of the commodity object recommendation information;
[0150] The specific target recommendation user may be the user who wants to obtain the recommended information of the commodity object, and there may be many specific situations. For example, it may be a user who initiates an access request to a target page. If the target page contains sections such as "Guess you like", the access request initiated to the target page may also be a request initiated by the user to obtain the recommended information. The specific source database of the recommended information may also be determined according to the specific application scenario. For example, it may be the full database of the system, or it may be a small-scale database of some businesses or merchants, etc.
[0151] S502: Acquire feature vector information of the user and feature vector information of multiple commodity objects in the source database, wherein the feature vector of the commodity object includes feature values of a picture of the commodity object in multiple dimensions, and the feature values include: an output result of an output layer after the picture of the commodity object is input into at least one neural network model for extracting features from the picture of the commodity object, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the picture of the commodity object by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0152] S503: Input the feature vector information of the user and the feature vector information of the plurality of commodity objects into a recommendation algorithm to predict click rate information obtained after the commodity objects are recommended to the user.
[0153] Among them, the specific recommendation algorithm is not the focus of the embodiments of the present application, so it will not be described in detail here.
[0154] After obtaining a specific click-through rate prediction result, at least one commodity object whose click-through rate meets the condition may be determined from the multiple commodity objects so as to be recommended to the user.
[0155] Embodiment 4
[0156] This fourth embodiment provides another specific recommendation scenario. Specifically, given a specific commodity object, one or more users who are most likely to click on the commodity object can be predicted from a large number of users, and the commodity object can be recommended to the user. This situation can mainly occur in scenarios such as merchants' new product recommendations. For example, a merchant has newly listed a commodity object and needs to select some users for targeted recommendations. At this time, the feature vector of the commodity object can be matched with the user feature vectors of multiple users to determine the probability information of the commodity object being clicked by the user when it is recommended to each user. Then determine which specific users to recommend to. For details, see Figure 6The fourth embodiment provides a commodity object recommendation method, which may specifically include:
[0157] S601: Determine the image information of the commodity object to be recommended;
[0158] S602: Acquire feature vector information of the commodity object, wherein the feature vector of the commodity object includes feature values of the image in multiple dimensions, wherein the feature values include: an output result of an output layer after the image of the commodity object is input into at least one neural network model for extracting features from the commodity object image, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the image of the commodity object by the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0159] S603: Acquire feature vector information of multiple users;
[0160] S604: Input the feature vector information of the commodity object and the feature vector information of the multiple users into a recommendation algorithm to predict click rate information obtained after the commodity object to be recommended is recommended to the user.
[0161] In specific implementation, the commodity object information to be recommended may also be recommended to users whose predicted click rate values meet the conditions according to the predicted click rate information.
[0162] Embodiment 5
[0163] In addition to matching the feature vector of the commodity object with the feature vector of the user to achieve the recommendation of a specific commodity object to the user, the feature vector of the commodity object generated by the embodiment of the present application can also be applied to scenarios such as similarity calculation between commodity objects. For example, in a scenario, information of multiple commodity objects can be displayed in multiple resource positions in some commodity object list pages, wherein, in each resource position, operation options such as "find similar" can be used, and the user can obtain information of more other commodity objects similar to the commodity object in the resource position by clicking the operation option. At this time, it will involve similarity calculation between the feature vector of the commodity object in the resource position and the feature vector of more commodity objects to determine the similarity between the commodity objects. Alternatively, when the current user is performing operations such as browsing the details page for a certain commodity object, adding to a collection to be purchased, or collecting, the system may also provide a "watch again" function, providing more commodity object information similar to the commodity object operated by the user for the user to choose, etc. At this time, it will also involve similarity comparison between commodity objects, which can also be determined by the operation between the feature vectors of the commodity objects. As for the feature vector of the commodity object, the solution provided in the embodiment of the present application can also be adopted to implement it in the form of a high-dimensional vector, which may include abstract features obtained by the activation value of the intermediate layer of the neural network model. Specifically, this embodiment 5 provides a method for providing a commodity object, see Figure 7 , the method may specifically include:
[0164] S701: In the process of providing similar commodity object information according to the target commodity object, feature vector information of the target commodity object and feature vector information of multiple commodity objects in the source database are obtained; wherein the feature vector of the commodity object includes feature values of the commodity object image in multiple dimensions, and the feature values include: the output result of the output layer after the commodity object image is input into at least one neural network model for extracting features from the commodity object image, and the activation value of at least one intermediate layer; the activation value of the intermediate layer is the feature information extracted from the commodity object image by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer;
[0165] Specifically, in the process of displaying a target page including at least one commodity object information, a request for obtaining similar commodity object information for a specified commodity object is received, and the specified commodity object is determined as the target commodity object. Alternatively, an operation request for browsing a detail page of a specified commodity object, or adding the specified commodity object to a to-be-purchased collection, or adding the specified commodity object to favorites is received, and the specified commodity object is determined as the target commodity object.
[0166] S702: Provide commodity object information that meets the similarity requirement with the designated commodity object by performing a similarity comparison between the feature vector information of the target commodity object and the feature vector information of multiple commodity objects in the source database.
[0167] Embodiment 6
[0168] In addition to being used in scenarios such as recommending commodity objects or providing information about similar commodity objects, the high-dimensional feature vector information of a specific commodity object image can also be applied to the optimization of a representative image of a commodity object. Specifically, since the representative image of a commodity object can usually be displayed in a resource position of a specific page, and the characteristics of the image may affect the user's click. For example, for a certain user, if one of the images of a commodity object is displayed in a resource position, the commodity object may not be clicked, but if another image is replaced, the commodity object may be clicked, and so on. In view of this situation, in an embodiment of the present application, when a specific commodity object is placed in a target page, multiple optional images can be associated with the commodity object, wherein which image is specifically displayed as a representative image in a specific resource position can be determined according to the characteristics of a specific visitor user. In other words, after receiving a specific visitor user's access request to the target page, the user's feature vector can be matched with the feature vectors of each image of the commodity object to calculate which image is displayed, and a higher click probability can be obtained for the commodity object, and then this image can be displayed as a representative image of the commodity object in a specific resource position. In this way, the product object representative layer can be "one thousand faces for one thousand people", thereby improving the click-through rate of the product object. Among them, the feature vector of the specific image can be a high-dimensional feature vector containing some abstract features generated in the embodiment of the present application, and the feature vector of the user can be low-dimensional or high-dimensional, so as to achieve more accurate matching.
[0169] Specifically, this embodiment 6 provides a method for providing commodity object information, see Figure 8 , the method may specifically include:
[0170] S801: receiving an access request to a target page, wherein the target page is associated with a plurality of resource locations for displaying information of a plurality of commodity objects, wherein the commodity objects are associated with a plurality of different pictures;
[0171] There may be many specific target pages, for example, the target page may be a client home page, or a home page of some activity venues, a channel home page, and so on.
[0172] S802: Acquire feature vector information corresponding to the plurality of different images, and feature vector information of the visitor user; wherein the feature vector corresponding to the image includes feature values of the image in multiple dimensions, and the feature values include: the output result of the output layer after the image is input into at least one neural network model for extracting features from the commodity object image, and the activation value of at least one intermediate layer; the activation value of the intermediate layer is the feature information extracted from the commodity object image by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer;
[0173] S803: Inputting the feature vector information of the visitor user and the feature vector information corresponding to the multiple pictures of the commodity object into a prediction algorithm respectively, so as to predict the click rate information obtained after the multiple pictures of the commodity object are displayed to the user respectively;
[0174] S804: According to the prediction result, the pictures whose click rates meet the conditions are displayed as representative pictures of the corresponding product objects in the resource position corresponding to the target page.
[0175] Among them, for the parts not described in detail in Examples 2 to 6, please refer to the description in the aforementioned Example 1, and no further details will be given here.
[0176] Embodiment 7
[0177] As mentioned above, through a neural network model or other algorithm model, the image of the commodity object can be analyzed to determine the display scene category information associated with the image of the commodity object, for example, it can include catwalk scenes, coffee drinking scenes, indoor studio scenes, outdoor street scenes, sports scenes, etc. For the user, when obtaining specific user feature information, the display scene information that the user is interested in can also be obtained. For example, by analyzing the images of the commodity objects that the user has browsed historically, the display scene information corresponding to these images can be analyzed to determine the display scene information that the user is interested in, etc. Therefore, when recommending information to the user, it can also be combined with the display scene category information associated with this specific commodity object image for recommendation. Specifically, in this embodiment seven, an information recommendation method is provided, see. Fig. 9 , the method may specifically include:
[0178] S901: Determine the image associated with the target product object;
[0179] S902: Determine display scene category information corresponding to the commodity object in the picture by performing feature analysis on the picture;
[0180] Among them, when performing feature analysis on the image, a pre-trained neural network model or a common algorithm model can be used, which will not be described in detail here.
[0181] S903: Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in;
[0182] Regarding the acquisition of user characteristics, there are corresponding records in the above-mentioned embodiment 1, which will not be repeated here.
[0183] S904: Determine a target user according to the display scene category information corresponding to the image and the display scene category information that the user is interested in;
[0184] S905: Recommend the picture to a client associated with the target user.
[0185] Embodiment 8
[0186] This eighth embodiment is similar to the seventh embodiment, except that the specific recommended object is no longer the image of the commodity object, but the video of the commodity object. This video can have a variety of sources, for example, it can include videos shot and uploaded by merchant users, or it can also be a video captured in a live broadcast scene, etc. In this case, after determining the video to be recommended, several frames of images can be extracted from the video, and by analyzing the image features, the display scene category information corresponding to the video can be determined. Then, the video can be recommended to users who are interested in the scene category. For details, see Fig.10 The eighth embodiment specifically provides an information recommendation method, which may specifically include:
[0187] S1001: Determine the video associated with the target product object;
[0188] S1002: extracting at least one frame of image from the video, and determining display scene category information corresponding to the commodity object in the video by performing feature analysis on the image;
[0189] S1003: Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in;
[0190] S1004: determining a target user according to the display scene category information corresponding to the video and the display scene category information in which the user is interested;
[0191] S1005: Recommend the video to a client associated with the target user.
[0192] Corresponding to the first embodiment, the present application embodiment also provides an information processing device, see Fig.11 , the device may specifically include:
[0193] The model acquisition unit 1101 is used to acquire at least one neural network model for extracting features from a commodity object image;
[0194] The analyzing unit 1102 is used to input the target product object image into the neural network model to obtain the output result of the output layer of the neural network model and the activation value of at least one intermediate layer; the activation value of the intermediate layer represents the feature information extracted from the product object image by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer;
[0195] The image feature vector generating unit 1103 is used to determine feature values in multiple dimensions according to the output result of the output layer and the activation value of the intermediate layer, and to generate a feature vector of the product object image.
[0196] The neural network model is obtained by training with training samples, wherein the training samples include information of multiple commodity object pictures and labeling information for labeling the commodity object pictures according to the click-through rate of the pictures; the labeling information includes: labeling information for distinguishing the commodity object area and the background area in the training sample pictures, or the shooting scene category information corresponding to the training sample pictures, or the feature information of the model person included in the training sample pictures.
[0197] Specifically, the output result of the output layer of the neural network model includes: element category feature information of image pixels in the target product object picture, and the element category includes the product object body or background; at this time, the activation value of the middle layer includes: in the process of obtaining the element category feature information, the feature information extracted from the target product object picture by the middle layer of the neural network model, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer.
[0198] In a specific implementation, the device may further include:
[0199] a pixel attribute determination subunit, configured to determine the pixel attribute of each pixel in the element category according to the output result;
[0200] A target pixel attribute determination subunit, used to determine the target pixel attribute by comparing the pixel attributes of each pixel;
[0201] At this time, the picture feature vector generating unit may be specifically used to determine a feature value of the data object picture in at least one dimension according to the target pixel attribute.
[0202] Specifically, the pixel attribute determination subunit may be used to: determine color attribute information of a plurality of pixels corresponding to a commodity object body in the target commodity object image according to the output result;
[0203] The target pixel attribute determination subunit can be specifically used to: determine the main color tone information of the main part of the commodity object;
[0204] The image feature vector generating unit may be specifically configured to: determine the main color tone information of the main part of the commodity object as a feature value on one dimension in the feature vector.
[0205] Alternatively, in another manner, the target pixel attribute determination subunit may be specifically used to: determine the proportions of the main part and the background part of the commodity object in the target commodity object image, and / or the quantity information of the connected regions;
[0206] The image feature vector generating unit may be specifically configured to: determine the proportion and / or the quantity information of the connected regions as feature values on some dimensions in the feature vector.
[0207] In addition, the output result of the output layer of the neural network model includes: the category feature information of the shooting scene of the target commodity object picture; at this time, the activation value of the intermediate layer includes: in the process of obtaining the category feature information of the shooting scene, the feature information extracted from the target commodity object picture by the intermediate layer of the neural network model, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer. Among them, the shooting scene category information: indoor shooting or outdoor shooting, shooting of the commodity object in use by a model or prop, or shooting of the details of the commodity object without a model or prop.
[0208] Alternatively, the output result of the output layer of the neural network model includes: for the training sample pictures showing the usage status of the commodity object through the model character, the feature information of the model character; the activation value of the intermediate layer includes: in the process of obtaining the feature information of the model character, the feature information extracted from the target commodity object picture by the intermediate layer of the neural network model, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons. Among them, the feature information of the model character includes: whether the target commodity object picture includes the whole body / half body of the model character, or whether it includes the face image of the model character, or the posture feature information of the model character, or the expression feature information of the model character.
[0209] In a specific implementation, the device may further include:
[0210] A feature screening unit, used for screening the output result of the output layer and the activation value of the intermediate layer;
[0211] The image feature vector generating unit may be specifically used to determine feature values in multiple dimensions according to the screening result.
[0212] Corresponding to the second embodiment, the present application embodiment also provides a device for determining user characteristic information, see Fig.12 , the device may include:
[0213] The picture information determining unit 1201 is used to determine the picture information of multiple commodity objects associated with the user's historical behavior information;
[0214] A model acquisition unit 1202 is used to acquire at least one neural network model for extracting features from a commodity object image;
[0215] The analyzing unit 1203 is used to input the pictures of multiple commodity objects into the neural network model respectively, and obtain the output result of the output layer of the neural network model and the activation value of at least one intermediate layer; the activation value of the intermediate layer represents the feature information extracted from the commodity object pictures by the intermediate layer of the neural network model in the process of obtaining the output result;
[0216] The user feature vector generating unit 1204 is used to determine feature values in multiple dimensions by counting the output results of the output layer and the activation values of the intermediate layer respectively corresponding to the pictures of the multiple commodity objects, so as to generate the feature vector of the user.
[0217] Corresponding to the third embodiment, the present application embodiment also provides a commodity object information recommendation device, see Fig.13 , the device may include:
[0218] An information determination unit 1301 is used to determine a target recommendation user for a commodity object and a source database for commodity object recommendation information;
[0219] A feature vector acquisition unit 1302 is used to acquire feature vector information of the user and feature vector information of multiple commodity objects in the source database, wherein the feature vector of the commodity object includes feature values of a picture of the commodity object in multiple dimensions, and the feature values include: an output result of an output layer after the picture of the commodity object is input into at least one neural network model for extracting features from the picture of the commodity object, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the picture of the commodity object by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0220] The prediction unit 1303 is used to input the feature vector information of the user and the feature vector information of the multiple commodity objects into a recommendation algorithm to predict click rate information obtained after the commodity objects are recommended to the user.
[0221] In a specific implementation, the device may further include:
[0222] The recommendation unit is used to determine at least one commodity object whose click rate meets the condition from the multiple commodity objects according to the predicted click rate information, so as to recommend it to the user.
[0223] Corresponding to the fourth embodiment, the present application embodiment also provides a commodity object recommendation device, see Fig.14 , the device may include:
[0224] The picture information determining unit 1401 is used to determine the picture information of the commodity object to be recommended;
[0225] The commodity feature vector acquisition unit 1402 is used to acquire feature vector information of the commodity object, wherein the feature vector of the commodity object includes feature values of the image in multiple dimensions, wherein the feature values include: an output result of an output layer after the image of the commodity object is input into at least one neural network model for extracting features from the commodity object image, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the image of the commodity object by the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in the previous layer;
[0226] A user feature vector acquisition unit 1403, for acquiring feature vector information of multiple users;
[0227] The prediction unit 1404 is used to input the feature vector information of the commodity object and the feature vector information of the multiple users into a recommendation algorithm to predict click rate information obtained after the commodity object to be recommended is recommended to the user.
[0228] When implementing it specifically, it may also include:
[0229] The prediction unit is used to recommend the commodity object information to be recommended to users whose predicted click rate values meet the conditions according to the predicted click rate information.
[0230] Corresponding to the fifth embodiment, the present application embodiment further provides a device for providing a commodity object, see Fig.15 , the device may include:
[0231] The feature vector acquisition unit 1501 is used to acquire feature vector information of the target product object and feature vector information of multiple product objects in the source database in the process of providing similar product object information according to the target product object; wherein the feature vector of the product object includes feature values of the product object image in multiple dimensions, and the feature values include: the output result of the output layer after the product object image is input into at least one neural network model for extracting features from the product object image, and the activation value of at least one intermediate layer; the activation value of the intermediate layer is the feature information extracted from the product object image by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons;
[0232] The similar commodity object information providing unit 1502 is configured to provide commodity object information having a similarity to the designated commodity object by performing a similarity comparison between the feature vector information of the target commodity object and the feature vector information of multiple commodity objects in the source database.
[0233] In addition, the device may also include:
[0234] The first request receiving unit is used to receive a request for obtaining similar commodity object information for a designated commodity object during displaying a target page including at least one commodity object information, and determine the designated commodity object as the target commodity object.
[0235] Alternatively, the second request receiving unit is used to receive an operation request for browsing the detail page of a specified commodity object, or adding the commodity object to a to-be-purchased set, or a favorite, and determine the specified commodity object as the target commodity object.
[0236] Corresponding to the sixth embodiment, the present application embodiment also provides a device for providing commodity object information, see Fig.16 , the device may include:
[0237] The access request receiving unit 1601 is used to receive an access request for a target page, wherein the target page is associated with multiple resource locations for displaying information of multiple commodity objects, wherein the commodity objects are associated with multiple different pictures;
[0238] The feature vector acquisition unit 1602 is used to acquire feature vector information corresponding to the multiple different pictures, and feature vector information of the visitor user; wherein the feature vector corresponding to the picture includes feature values of the picture in multiple dimensions, and the feature values include: the output result of the output layer after the picture is input into at least one neural network model for extracting features from the product object picture, and the activation value of at least one intermediate layer; the activation value of the intermediate layer is the feature information extracted from the product object picture by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons;
[0239] The prediction unit 1603 is used to input the feature vector information of the visitor user and the feature vector information corresponding to the multiple pictures of the commodity object into the prediction algorithm to predict the click rate information obtained after the multiple pictures of the commodity object are displayed to the user;
[0240] The representative image determining unit 1604 is used to display the image with a click rate that meets the condition as the representative image of the corresponding commodity object in the resource position corresponding to the target page according to the prediction result.
[0241] Corresponding to the seventh embodiment, the present application embodiment further provides an information recommendation device, see Fig.17 , the device may include:
[0242] The picture determination unit 1701 is used to determine the picture associated with the target product object;
[0243] A display scene determination unit 1702 is used to determine display scene category information corresponding to the commodity object in the picture by performing feature analysis on the picture;
[0244] A user characteristic information acquisition unit 1703 is used to determine a user set and acquire user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the user is interested in;
[0245] A target user determination unit 1704 is used to determine a target user according to the display scene category information corresponding to the picture and the display scene category information that the user is interested in;
[0246] The picture recommendation unit 1705 is used to recommend the picture to the client associated with the target user.
[0247] Corresponding to the eighth embodiment, the present application embodiment also provides an information recommendation device, see Fig.18 , the device may include:
[0248] The video determination unit 1801 is used to determine the video associated with the target product object;
[0249] A display scene determination unit 1802 is used to extract at least one frame of image from the video, and determine display scene category information corresponding to the commodity object in the video by performing feature analysis on the image;
[0250] A user characteristic information acquisition unit 1803 is used to determine a user set and acquire user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes information on display scene categories that the users are interested in;
[0251] A target user determination unit 1804 is used to determine a target user according to the display scene category information corresponding to the video and the display scene category information that the user is interested in;
[0252] The video recommendation unit 1805 is used to recommend the video to the client associated with the target user.
[0253] In addition, an embodiment of the present application further provides an electronic device, including:
[0254] one or more processors; and
[0255] A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the following operations are performed:
[0256] Obtain at least one neural network model for extracting features from a product object image;
[0257] Inputting the target product object image into the neural network model, obtaining the output result of the output layer of the neural network model, and the activation value of at least one intermediate layer; the activation value of the intermediate layer represents the feature information extracted from the product object image by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer;
[0258] According to the output result of the output layer and the activation value of the intermediate layer, feature values in multiple dimensions are determined, and a feature vector of the commodity object image is generated.
[0259] or,
[0260] Determine image information of multiple commodity objects associated with the user's historical behavior information;
[0261] Obtain at least one neural network model for extracting features from a product object image;
[0262] Inputting pictures of multiple commodity objects into the neural network model respectively, obtaining an output result of an output layer of the neural network model and an activation value of at least one intermediate layer; the activation value of the intermediate layer represents feature information extracted from the commodity object pictures by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0263] By counting the output results of the output layer and the activation values of the intermediate layer respectively corresponding to the pictures of the multiple commodity objects, feature values on multiple dimensions are determined to generate a feature vector of the user.
[0264] or,
[0265] Determine the target recommendation user of the product object and the source database of the product object recommendation information;
[0266] Acquire feature vector information of the user and feature vector information of multiple commodity objects in the source database, wherein the feature vector of the commodity object includes feature values of a picture of the commodity object in multiple dimensions, and the feature values include: an output result of an output layer after the picture of the commodity object is input into at least one neural network model for extracting features from the picture of the commodity object, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the picture of the commodity object by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0267] The feature vector information of the user and the feature vector information of the plurality of commodity objects are input into a recommendation algorithm to predict click rate information obtained after the commodity objects are recommended to the user.
[0268] or,
[0269] Determine the image information of the product object to be recommended;
[0270] Obtaining feature vector information of the commodity object, wherein the feature vector of the commodity object includes feature values of the image in multiple dimensions, wherein the feature values include: an output result of an output layer after the image of the commodity object is input into at least one neural network model for extracting features from the commodity object image, and an activation value of at least one intermediate layer; the activation value of the intermediate layer is feature information extracted from the image of the commodity object by the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in an upper layer;
[0271] Obtain feature vector information of multiple users;
[0272] The feature vector information of the commodity object and the feature vector information of the multiple users are input into a recommendation algorithm to predict click rate information obtained after the commodity object to be recommended is recommended to the user.
[0273] or,
[0274] In the process of providing similar commodity object information according to the target commodity object, feature vector information of the target commodity object and feature vector information of multiple commodity objects in the source database are obtained; wherein the feature vector of the commodity object includes feature values of a picture of the commodity object in multiple dimensions, and the feature values include: the output result of the output layer after the picture of the commodity object is input into at least one neural network model for extracting features from the commodity object picture, and the activation value of at least one intermediate layer; the activation value of the intermediate layer is the feature information extracted from the picture of the commodity object by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of neurons in the previous layer;
[0275] By performing a similarity comparison between the feature vector information of the target commodity object and the feature vector information of multiple commodity objects in a source database, commodity object information having a similarity to the designated commodity object that meets the condition is provided.
[0276] or,
[0277] Receiving an access request to a target page, wherein the target page is associated with a plurality of resource locations for displaying information of a plurality of commodity objects, wherein the commodity objects are associated with a plurality of different pictures;
[0278] Obtaining feature vector information corresponding to the plurality of different images, and feature vector information of the visitor user; wherein the feature vector corresponding to the image includes feature values of the image in multiple dimensions, and the feature values include: the output result of the output layer after the image is input into at least one neural network model for extracting features from the commodity object image, and the activation value of at least one intermediate layer; the activation value of the intermediate layer is the feature information extracted from the commodity object image by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons;
[0279] Inputting the feature vector information of the visitor user and the feature vector information corresponding to the multiple pictures of the commodity object into the prediction algorithm respectively, so as to predict the click rate information obtained after the multiple pictures of the commodity object are displayed to the user respectively;
[0280] According to the prediction result, the pictures whose click rates meet the conditions are displayed as representative pictures of the corresponding commodity objects in the resource position corresponding to the target page.
[0281] or,
[0282] Determine the images associated with the target product object;
[0283] By performing feature analysis on the image, determining display scene category information corresponding to the commodity object in the image;
[0284] Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in;
[0285] Determine the target user according to the display scene category information corresponding to the image and the display scene category information that the user is interested in;
[0286] The picture is recommended to a client associated with the target user.
[0287] or,
[0288] Determine the video associated with the target product object;
[0289] Extracting at least one frame of image from the video, and determining display scene category information corresponding to the commodity object in the video by performing feature analysis on the image;
[0290] Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in;
[0291] Determine a target user according to the display scene category information corresponding to the video and the display scene category information that the user is interested in;
[0292] The video is recommended to a client associated with the target user.
[0293] in, Fig.19 The electronic device architecture is shown as an example, which may include a processor 1910, a video display adapter 1911, a disk drive 1912, an input / output interface 1913, a network interface 1914, and a memory 1920. The processor 1910, the video display adapter 1911, the disk drive 1912, the input / output interface 1913, the network interface 1914, and the memory 1920 may be communicatively connected via a communication bus 1930.
[0294] Among them, the processor 1910 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solution provided in this application.
[0295] The memory 1920 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1920 can store an operating system 1921 for controlling the operation of the electronic device 1900, and a basic input and output system (BIOS) for controlling the low-level operation of the electronic device 1900. In addition, a web browser 1923, a data storage management system 1924, and an information processing system 1925, etc. can also be stored. The above-mentioned information processing system 1925 can be an application program that specifically implements the operations of the aforementioned steps in the embodiment of the present application. In short, when the technical solution provided in the present application is implemented by software or firmware, the relevant program code is stored in the memory 1920 and is called and executed by the processor 1910.
[0296] The input / output interface 1913 is used to connect the input / output module to realize information input and output. The input / output module can be configured as a component in the device (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.
[0297] The network interface 1914 is used to connect to a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).
[0298] The bus 1930 comprises a pathway for transmitting information between the various components of the device (eg, the processor 1910, the video display adapter 1911, the disk drive 1912, the input / output interface 1913, the network interface 1914, and the memory 1920).
[0299] It should be noted that, although the above device only shows a processor 1910, a video display adapter 1911, a disk drive 1912, an input / output interface 1913, a network interface 1914, a memory 1920, a bus 1930, etc., in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include components necessary for implementing the solution of the present application, and does not necessarily include all the components shown in the figure.
[0300] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.
[0301] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.
[0302] The above is a detailed introduction to the information processing method, device and electronic device provided by the present application. The principles and implementation methods of the present application are described in detail using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present application. At the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present application.
Claims
1. An information processing method, characterized in that: include: Obtain multiple neural network models for feature extraction of product object images; The multiple neural network models include: a model for distinguishing the area where the subject of the commodity object is located and the background area, a model for identifying whether the shooting scene is a street shot or a studio shot, and whether there are human models or props, and a model for identifying whether the human model in the picture includes the whole body / half body, a face image or the expression features of the human model; Inputting the target product object image into the multiple neural network models, obtaining the output results of the multiple output layers of the multiple neural network models as visualized feature values, and the activation values of the multiple intermediate layers as abstract feature values; the activation values of the intermediate layers represent the feature information extracted from the target product object image by the intermediate layer of the neural network model in the process of obtaining the output results, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the upper layer; According to the output results of multiple output layers and the activation values of multiple intermediate layers, the abstract feature values are used as a supplement to the visualized feature values to generate multiple higher-dimensional feature vectors, thereby generating feature vectors of multiple angles of the product object image.
2. The method according to claim 1, characterized in that The neural network model is obtained by training with training samples, wherein the training samples include information of a plurality of commodity object pictures and labeling information for labeling the commodity object pictures according to the click-through rate of the pictures; The annotation information includes: annotation information for distinguishing the commodity object area and the background area in the training sample image, or shooting scene category information corresponding to the training sample image, or feature information of the model included in the training sample image.
3. The method according to claim 1, characterized in that: The output result of the output layer of the neural network model includes: element category feature information of image pixels in the target commodity object picture, and the element category includes the commodity object body or background; The activation value of the middle layer includes: in the process of obtaining the element category feature information, the feature information extracted from the target product object image by the middle layer of the neural network model, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer.
4. The method according to claim 3, characterized in that Also includes: Determine pixel attributes of each pixel in the element category according to the output result; Determine the target pixel attribute by comparing the pixel attributes of each pixel; The method of using the abstract feature values as a supplement to the visualized feature values according to the output results of the multiple output layers and the activation values of the multiple intermediate layers to generate multiple feature vectors with higher dimensions includes: Determine a feature value of the commodity object image in at least one dimension according to the target pixel attribute.
5. The method according to claim 4, characterized in that Determining the pixel attribute of each pixel in the element category according to the output result includes: Determine, according to the output result, color attribute information of a plurality of pixels corresponding to a main body of the commodity object in the target commodity object image; Determining the target pixel attribute by comparing the pixel attributes of each pixel includes: Determine the main color tone information of the main part of the commodity object; The determining, according to the target pixel attribute, a feature value of the commodity object image in at least one dimension includes: The main color tone information of the main part of the commodity object is determined as a feature value on one dimension in the feature vector.
6. The method according to claim 3, characterized in that The target pixel attributes are determined by comparing the pixel attributes of each pixel, including: Determine the proportion of the main part and the background part of the commodity object in the target commodity object image, and / or the number of connected areas; The determining, according to the target pixel attribute, a feature value of the commodity object image in at least one dimension includes: The proportion and / or the quantity information of the connected regions are determined as eigenvalues on some dimensions in the eigenvector.
7. The method according to claim 1, characterized in that The output result of the output layer of the neural network model includes: the category feature information to which the shooting scene of the target commodity object picture belongs; The activation value of the middle layer includes: in the process of obtaining the category feature information to which the shooting scene belongs, the feature information extracted from the target product object picture by the middle layer of the neural network model, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer.
8. The method according to claim 7, characterized in that The shooting scene category information includes: indoor shooting or outdoor shooting, shooting of the commodity object in use by a model or props, or shooting of the details of the commodity object without a model or props.
9. The method according to claim 1, characterized in that: The output result of the output layer of the neural network model includes: for a training sample picture showing the usage status of a commodity object through a model character, feature information of the model character; The activation value of the middle layer includes: in the process of obtaining the feature information of the model, the feature information extracted from the target product object image by the middle layer of the neural network model, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer.
10. The method according to claim 9, characterized in that The characteristic information of the model character includes: whether the target commodity object image includes the full body / half body of the model character, or whether the face image of the model character is included, or the posture characteristic information of the model character, or the expression characteristic information of the model character.
11. The method according to any one of claims 1 to 10, characterized in that: Also includes: Filter the output results of multiple output layers and the activation values of multiple intermediate layers; The generating of multiple higher-dimensional feature vectors includes: Generate multiple higher-dimensional feature vectors based on the screening results.
12. A method for determining user characteristic information, characterized in that: include: Determine image information of multiple commodity objects associated with the user's historical behavior information; Acquire multiple neural network models for extracting features from the commodity object image; the multiple neural network models include: a model for distinguishing the area where the commodity object subject is located and the background area, a model for identifying whether the shooting scene is a street shot or a studio shot, and whether there are human models or props, and a model for identifying whether the human model in the image includes the full body / half body, a face image, or the expression features of the human model; Inputting pictures of multiple commodity objects into the multiple neural network models respectively, obtaining output results of multiple output layers of the multiple neural network models as visualized feature values, and activation values of multiple intermediate layers as abstract feature values; the activation values of the intermediate layers represent feature information extracted from the commodity object pictures by the intermediate layers of the neural network model in the process of obtaining the output results, or feature information obtained by weighted summation and nonlinear transformation of neurons in the upper layer; By counting the output results of multiple output layers and the activation values of multiple intermediate layers corresponding to the images of the multiple commodity objects, the abstract feature values are used as a supplement to the visualized feature values to generate multiple higher-dimensional feature vectors for generating feature vectors from multiple angles of the user.
13. A method for recommending a commodity object, characterized in that: include: Determine the target recommendation user of the product object and the source database of the product object recommendation information; Acquire feature vector information of the user and feature vector information of multiple commodity objects in the source database, wherein the feature vector of the commodity object includes feature vectors of multiple angles of the commodity object picture, and the feature vectors of the multiple angles are generated by multiple higher-dimensional feature vectors, wherein the multiple higher-dimensional feature vectors include: after the picture of the commodity object is input into multiple neural network models for feature extraction of the commodity object picture, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the higher-dimensional feature vector is generated by supplementing the visualized feature values with the abstract feature values; the activation value of the intermediate layer is the feature information extracted from the picture of the commodity object by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human models in the picture include full body / half body, human face image or expression features of human models; The feature vector information of the user and the feature vector information of the plurality of commodity objects are input into a recommendation algorithm to predict click rate information obtained after the commodity objects are recommended to the user.
14. The method according to claim 13, characterized in that Also includes: According to the predicted click rate information, at least one commodity object whose click rate meets the condition is determined from the multiple commodity objects to recommend to the user.
15. A method for recommending a commodity object, characterized in that: include: Determine the image information of the product object to be recommended; Acquire feature vector information of the commodity object, wherein the feature vector of the commodity object includes feature vectors of multiple angles of the image, and the feature vectors of the multiple angles are generated by multiple feature vectors of higher dimensions, wherein the multiple feature vectors of higher dimensions include: after the image of the commodity object is input into multiple neural network models for extracting features from the commodity object image, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the feature vector of higher dimensions is generated by supplementing the visualized feature values with the abstract feature values; the activation value of the intermediate layer is feature information extracted from the image of the commodity object by the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in the upper layer; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human models in the image include full body / half body, human face image or expression features of human models; Obtain feature vector information of multiple users; The feature vector information of the commodity object and the feature vector information of the multiple users are input into a recommendation algorithm to predict click rate information obtained after the commodity object to be recommended is recommended to the user.
16. The method according to claim 15, characterized in that Also includes: According to the predicted click rate information, the commodity object information to be recommended is recommended to users whose predicted click rate values meet the conditions.
17. A method for providing a commodity object, characterized in that: include: In the process of providing similar commodity object information according to the target commodity object, feature vector information of the target commodity object and feature vector information of multiple commodity objects in the source database are obtained; wherein the feature vector of the commodity object includes feature vectors of multiple angles of the commodity object picture, and the feature vectors of the multiple angles are generated by multiple higher-dimensional feature vectors, and the multiple higher-dimensional feature vectors include: after the picture of the commodity object is input into multiple neural network models for feature extraction of the commodity object picture, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the higher-dimensional feature vector is generated by supplementing the visualized feature values with the abstract feature values; the activation value of the intermediate layer is the feature information extracted from the picture of the commodity object by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human model in the picture includes full body / half body, human face image or expression features of human models; By performing a similarity comparison between the feature vector information of the target commodity object and the feature vector information of multiple commodity objects in a source database, commodity object information having a similarity to the designated commodity object that meets the conditions is provided.
18. The method according to claim 17, characterized in that The method also includes: In the process of displaying a target page including at least one commodity object information, a request for obtaining similar commodity object information for a designated commodity object is received, and the designated commodity object is determined as the target commodity object.
19. The method according to claim 17, characterized in that The method also includes: An operation request for browsing a detail page of a designated commodity object, or adding the commodity object to a to-be-purchased set, or adding the commodity object to a favorite is received, and the designated commodity object is determined as the target commodity object.
20. A method for providing commodity object information, characterized in that: include: Receiving an access request to a target page, wherein the target page is associated with a plurality of resource locations for displaying information of a plurality of commodity objects, wherein the commodity objects are associated with a plurality of different pictures; Acquire feature vector information corresponding to the plurality of different pictures, and feature vector information of the visitor user; wherein the feature vector corresponding to the picture includes feature vectors of multiple angles of the picture, and the feature vectors of the multiple angles are generated by multiple feature vectors of higher dimensions, and the multiple feature vectors of higher dimensions include: after the picture is input into multiple neural network models for feature extraction of the commodity object picture, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the higher dimensional feature vector is generated by supplementing the visualized feature value with the abstract feature value; the activation value of the intermediate layer is the feature information extracted from the commodity object picture by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human model in the picture includes the full body / half body, human face image or the expression features of the human model; Inputting the feature vector information of the visitor user and the feature vector information corresponding to the multiple pictures of the commodity object into the prediction algorithm respectively, so as to predict the click rate information obtained after the multiple pictures of the commodity object are displayed to the user respectively; According to the prediction result, the pictures whose click rates meet the conditions are displayed as representative pictures of the corresponding commodity objects in the resource position corresponding to the target page.
21. An information recommendation method, characterized in that: include: Determine the images associated with the target product object; By performing feature analysis on the image, the display scene category information corresponding to the commodity object in the image is determined; wherein, when performing feature analysis on the image, the image is input into a plurality of neural network models, and the output results of a plurality of output layers of the plurality of neural network models are obtained as visualized feature values, and the activation values of a plurality of intermediate layers are obtained as abstract feature values; the activation values of the intermediate layers represent the feature information extracted from the commodity object image by the intermediate layer of the neural network model in the process of obtaining the output results, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; according to the output results of the plurality of output layers and the activation values of the plurality of intermediate layers, the abstract feature values are used as a supplement to the visualized feature values to generate a plurality of feature vectors with higher dimensions, and then feature vectors of a plurality of angles of the image are generated; the plurality of neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, and whether there are human models or props, and a model for identifying whether the human models in the image include full body / half body, human face image or expression features of human models; Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in; Determine the target user according to the display scene category information corresponding to the image and the display scene category information that the user is interested in; The picture is recommended to a client associated with the target user.
22. An information recommendation method, characterized in that: include: Determine the video associated with the target product object; Extract at least one frame of image from the video, and determine the display scene category information corresponding to the commodity object in the video by performing feature analysis on the image; wherein, when performing feature analysis on the image, input the image into multiple neural network models, obtain the output results of multiple output layers of the multiple neural network models as visualized feature values, and the activation values of multiple intermediate layers as abstract feature values; the activation value of the intermediate layer represents the feature information extracted from the commodity object image by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; according to the output results of the multiple output layers and the activation values of the multiple intermediate layers, use the abstract feature value as a supplement to the visualized feature value to generate multiple higher-dimensional feature vectors, and then generate feature vectors of multiple angles of the image; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human model in the picture includes full body / half body, human face image or expression features of human models; Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in; Determine a target user according to the display scene category information corresponding to the video and the display scene category information that the user is interested in; The video is recommended to a client associated with the target user.
23. An information processing device, characterized in that: include: A model acquisition unit, used to acquire multiple neural network models for extracting features from commodity object images; The multiple neural network models include: a model for distinguishing the area where the subject of the commodity object is located and the background area, a model for identifying whether the shooting scene is a street shot or a studio shot, and whether there are human models or props, and a model for identifying whether the human model in the picture includes the whole body / half body, a face image or the expression features of the human model; An analysis unit, configured to input the target product object image into the multiple neural network models, obtain output results of multiple output layers of the multiple neural network models as visualized feature values, and activation values of multiple intermediate layers as abstract feature values; the activation value of the intermediate layer represents feature information extracted from the target product object image by the intermediate layer of the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in the upper layer; The image feature vector generating unit is used to use the abstract feature value as a supplement to the visualized feature value according to the output results of the multiple output layers and the activation values of the multiple intermediate layers to generate multiple higher-dimensional feature vectors, thereby generating feature vectors of multiple angles of the commodity object image.
24. A device for determining user characteristic information, characterized in that: include: A picture information determination unit, used to determine picture information of multiple commodity objects associated with the user's historical behavior information; A model acquisition unit, used to acquire multiple neural network models for feature extraction of commodity object pictures; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is a street shot or a studio shot, and whether there are human models or props, and a model for identifying whether the human model in the picture includes a full body / half body, a face image, or the expression features of the human model; An analysis unit, used to input the pictures of multiple commodity objects into the multiple neural network models respectively, obtain the output results of the multiple output layers of the multiple neural network models as visualized feature values, and the activation values of the multiple intermediate layers as abstract feature values; the activation values of the intermediate layers represent the feature information extracted from the commodity object pictures by the intermediate layers of the neural network model in the process of obtaining the output results, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the upper layer; The user feature vector generating unit is used to generate multiple feature vectors of higher dimensions by counting the output results of multiple output layers and the activation values of multiple intermediate layers respectively corresponding to the pictures of the multiple commodity objects, and using the abstract feature values as a supplement to the visualized feature values, so as to generate feature vectors of multiple angles for the user.
25. A commodity object information recommendation device, characterized in that: include: An information determination unit, used to determine a target recommendation user for a commodity object and a source database for commodity object recommendation information; A feature vector acquisition unit is used to acquire feature vector information of the user and feature vector information of multiple commodity objects in the source database, wherein the feature vector of the commodity object includes feature vectors of multiple angles of the commodity object picture, and the feature vectors of the multiple angles are generated by multiple feature vectors of higher dimensions, wherein the multiple feature vectors of higher dimensions include: after the picture of the commodity object is input into multiple neural network models for feature extraction of the commodity object picture, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the feature vector of higher dimension is generated by supplementing the visualized feature values with the abstract feature values; the activation value of the intermediate layer is the feature information extracted from the picture of the commodity object by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the upper layer; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human models in the picture include full body / half body, human face image or expression features of human models; The prediction unit is used to input the feature vector information of the user and the feature vector information of the multiple commodity objects into a recommendation algorithm to predict click rate information obtained after the commodity objects are recommended to the user.
26. A commodity object recommendation device, characterized in that: include: A picture information determination unit, used to determine the picture information of the commodity object to be recommended; A commodity feature vector acquisition unit is used to acquire feature vector information of the commodity object, wherein the feature vector of the commodity object includes feature vectors of multiple angles of the image, and the feature vectors of the multiple angles are generated by multiple feature vectors of higher dimensions, wherein the multiple feature vectors of higher dimensions include: after the image of the commodity object is input into multiple neural network models for extracting features from the commodity object image, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the higher dimensional feature vector is generated by supplementing the visualized feature values with the abstract feature values; the activation value of the intermediate layer is the feature information extracted from the commodity object image by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the upper layer; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human models in the image include full body / half body, human face image or expression features of human models; A user feature vector acquisition unit, used to acquire feature vector information of multiple users; The prediction unit is used to input the feature vector information of the commodity object and the feature vector information of the multiple users into the recommendation algorithm to predict the click rate information obtained after the commodity object to be recommended is recommended to the user.
27. A device for providing a commodity object, characterized in that: include: The feature vector acquisition unit is used to acquire the feature vector information of the target product object and the feature vector information of multiple product objects in the source database in the process of providing similar product object information according to the target product object; wherein the feature vector of the product object includes feature vectors of multiple angles of the product object image, and the feature vectors of the multiple angles are generated by multiple higher-dimensional feature vectors, and the multiple higher-dimensional feature vectors include: after the product object image is input into multiple neural network models for feature extraction of the product object image, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature value; the feature vector of higher dimension is generated by supplementing the abstract feature value with the visualized feature value; the activation value of the middle layer is the feature information extracted from the picture of the commodity object by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the upper layer; the multiple neural network models include: a model for distinguishing the area where the subject of the commodity object is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human model in the picture includes the whole body / half body, a face image or the expression features of the human model; The similar commodity object information providing unit is used to provide commodity object information that meets the conditions of similarity with the designated commodity object by performing similarity comparison between the feature vector information of the target commodity object and the feature vector information of multiple commodity objects in the source database.
28. A device for providing commodity object information, characterized in that: include: An access request receiving unit, used for receiving an access request for a target page, wherein the target page is associated with a plurality of resource locations for displaying information of a plurality of commodity objects, wherein the commodity objects are associated with a plurality of different pictures; A feature vector acquisition unit is used to acquire feature vector information corresponding to the plurality of different pictures, and feature vector information of a visitor user; wherein the feature vector corresponding to the picture includes feature vectors of multiple angles of the picture, and the feature vectors of the multiple angles are generated by multiple feature vectors of higher dimensions, and the multiple feature vectors of higher dimensions include: after the picture is input into multiple neural network models for feature extraction of commodity object pictures, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the higher dimensional feature vector is generated by supplementing the visualized feature values with the abstract feature values; the activation value of the intermediate layer is the feature information extracted from the commodity object picture by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human models in the picture include full body / half body, human face image or expression features of human models; A prediction unit, used to input the feature vector information of the visitor user and the feature vector information corresponding to the multiple pictures of the commodity object into a prediction algorithm, so as to predict the click rate information obtained after the multiple pictures of the commodity object are displayed to the user; The representative image determination unit is used to display the image with a click rate that meets the conditions as the representative image of the corresponding commodity object in the resource position corresponding to the target page according to the prediction result.
29. An information recommendation device, characterized in that: include: A picture determination unit, used to determine a picture associated with a target product object; A display scene determination unit is used to determine the display scene category information corresponding to the commodity object in the picture by performing feature analysis on the picture; wherein, when performing feature analysis on the picture, the picture is input into a plurality of neural network models, and the output results of the output layers of the plurality of neural network models are obtained as visualized feature values, and the activation values of the plurality of intermediate layers are obtained as abstract feature values; the activation values of the intermediate layers represent the feature information extracted from the commodity object picture by the intermediate layer of the neural network model in the process of obtaining the output results, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the upper layer; according to the output results of the plurality of output layers and the activation values of the plurality of intermediate layers, the abstract feature values are used as a supplement to the visualized feature values to generate a plurality of feature vectors with higher dimensions, and then feature vectors of multiple angles of the picture are generated; the plurality of neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, and whether there are human models or props, and a model for identifying whether the human models in the picture include full body / half body, human face image or expression features of human models; A user characteristic information acquisition unit, used to determine a user set and acquire user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the user is interested in; A target user determination unit, configured to determine a target user according to the display scene category information corresponding to the image and the display scene category information in which the user is interested; The picture recommendation unit is used to recommend the picture to a client associated with the target user.
30. An information recommendation device, characterized in that: include: A video determination unit, used to determine a video associated with a target product object; A display scene determination unit is used to extract at least one frame of image from the video, and determine the display scene category information corresponding to the commodity object in the video by performing feature analysis on the image; wherein, when performing feature analysis on the image, the image is input into multiple neural network models, and the output results of multiple output layers of the multiple neural network models are obtained as visualized feature values, and the activation values of multiple intermediate layers are obtained as abstract feature values; the activation value of the intermediate layer represents the feature information extracted from the commodity object image by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; according to the output results of multiple output layers and the activation values of multiple intermediate layers, the abstract feature value is used as a supplement to the visualized feature value to generate multiple higher-dimensional feature vectors, and then generate feature vectors of multiple angles of the image; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human model in the picture includes the full body / half body, human face image or the expression features of the human model; A user characteristic information acquisition unit, used to determine a user set and acquire user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the user is interested in; A target user determination unit, configured to determine a target user according to the display scene category information corresponding to the video and the display scene category information in which the user is interested; The video recommendation unit is used to recommend the video to a client associated with the target user.
31. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the following operations are performed: Acquire multiple neural network models for extracting features from the commodity object image; the multiple neural network models include: a model for distinguishing the area where the commodity object subject is located and the background area, a model for identifying whether the shooting scene is a street shot or a studio shot, and whether there are human models or props, and a model for identifying whether the human model in the image includes the full body / half body, a face image, or the expression features of the human model; Inputting the target product object image into the multiple neural network models, obtaining the output results of the multiple output layers of the multiple neural network models as visualized feature values, and the activation values of the multiple intermediate layers as abstract feature values; the activation values of the intermediate layers represent the feature information extracted from the target product object image by the intermediate layer of the neural network model in the process of obtaining the output results, or the feature information obtained by weighted summation and nonlinear transformation of the neurons in the previous layer; According to the output results of multiple output layers and the activation values of multiple intermediate layers, the abstract feature values are used as a supplement to the visualized feature values to generate multiple higher-dimensional feature vectors, thereby generating feature vectors of multiple angles of the product object image.
32. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the following operations are performed: Determine image information of multiple commodity objects associated with the user's historical behavior information; Acquire multiple neural network models for extracting features from the commodity object image; the multiple neural network models include: a model for distinguishing the area where the commodity object subject is located and the background area, a model for identifying whether the shooting scene is a street shot or a studio shot, and whether there are human models or props, and a model for identifying whether the human model in the image includes the full body / half body, a face image, or the expression features of the human model; Inputting pictures of multiple commodity objects into the multiple neural network models respectively, obtaining output results of multiple output layers of the multiple neural network models as visualized feature values, and activation values of multiple intermediate layers as abstract feature values; the activation values of the intermediate layers represent feature information extracted from the commodity object pictures by the intermediate layers of the neural network model in the process of obtaining the output results, or feature information obtained by weighted summation and nonlinear transformation of neurons in the upper layer; By counting the output results of multiple output layers and the activation values of multiple intermediate layers corresponding to the images of the multiple commodity objects, the abstract feature values are used as a supplement to the visualized feature values to generate multiple higher-dimensional feature vectors for generating feature vectors from multiple angles of the user.
33. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the following operations are performed: Determine the target recommendation user of the product object and the source database of the product object recommendation information; Acquire feature vector information of the user and feature vector information of multiple commodity objects in the source database, wherein the feature vector of the commodity object includes feature vectors of multiple angles of the commodity object picture, and the feature vectors of the multiple angles are generated by multiple higher-dimensional feature vectors, wherein the multiple higher-dimensional feature vectors include: after the picture of the commodity object is input into multiple neural network models for feature extraction of the commodity object picture, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the higher-dimensional feature vector is generated by supplementing the visualized feature values with the abstract feature values; the activation value of the intermediate layer is the feature information extracted from the picture of the commodity object by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human models in the picture include full body / half body, human face image or expression features of human models; The feature vector information of the user and the feature vector information of the plurality of commodity objects are input into a recommendation algorithm to predict click rate information obtained after the commodity objects are recommended to the user.
34. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the following operations are performed: Determine the image information of the product object to be recommended; Obtaining feature vector information of the commodity object, wherein the feature vector of the commodity object includes feature vectors of multiple angles of the image, and the feature vectors of the multiple angles are generated by multiple feature vectors of higher dimensions, wherein the multiple feature vectors of higher dimensions include: after the image of the commodity object is input into multiple neural network models for extracting features from the commodity object image, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the feature vector of higher dimensions is generated by supplementing the visualized feature values with the abstract feature values; the activation value of the intermediate layer is feature information extracted from the image of the commodity object by the neural network model in the process of obtaining the output result, or feature information obtained by weighted summation and nonlinear transformation of neurons in the upper layer; the multiple neural network models include: a model for distinguishing the area where the subject of the commodity object is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human models in the image include full body / half body, human face image or expression features of human models; Obtain feature vector information of multiple users; The feature vector information of the commodity object and the feature vector information of the multiple users are input into a recommendation algorithm to predict click rate information obtained after the commodity object to be recommended is recommended to the user.
35. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the following operations are performed: In the process of providing similar commodity object information according to the target commodity object, feature vector information of the target commodity object and feature vector information of multiple commodity objects in the source database are obtained; wherein the feature vector of the commodity object includes feature vectors of multiple angles of the commodity object picture, and the feature vectors of the multiple angles are generated by multiple higher-dimensional feature vectors, and the multiple higher-dimensional feature vectors include: after the picture of the commodity object is input into multiple neural network models for feature extraction of the commodity object picture, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the higher-dimensional feature vector is generated by supplementing the visualized feature values with the abstract feature values; the activation value of the intermediate layer is the feature information extracted from the picture of the commodity object by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human models in the picture include full body / half body, human face image or expression features of human models; By performing a similarity comparison between the feature vector information of the target commodity object and the feature vector information of multiple commodity objects in a source database, commodity object information having a similarity to the designated commodity object that meets the conditions is provided.
36. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the following operations are performed: Receiving an access request to a target page, wherein the target page is associated with a plurality of resource locations for displaying information of a plurality of commodity objects, wherein the commodity objects are associated with a plurality of different pictures; Obtain feature vector information corresponding to the plurality of different pictures, and feature vector information of the visitor user; wherein the feature vector corresponding to the picture includes feature vectors of multiple angles of the picture, and the feature vectors of the multiple angles are generated by multiple feature vectors of higher dimensions, and the multiple feature vectors of higher dimensions include: after the picture is input into multiple neural network models for feature extraction of commodity object pictures, the output results of multiple output layers are used as visualized feature values, and the activation values of multiple intermediate layers are used as abstract feature values; the higher dimensional feature vector is generated by supplementing the visualized feature value with the abstract feature value; the activation value of the intermediate layer is the feature information extracted from the commodity object picture by the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human model in the picture includes the full body / half body, human face image or the expression features of the human model; Inputting the feature vector information of the visitor user and the feature vector information corresponding to the multiple pictures of the commodity object into the prediction algorithm respectively, so as to predict the click rate information obtained after the multiple pictures of the commodity object are displayed to the user respectively; According to the prediction result, the pictures whose click rates meet the conditions are displayed as representative pictures of the corresponding commodity objects in the resource position corresponding to the target page.
37. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the following operations are performed: Determine the images associated with the target product object; By performing feature analysis on the image, the display scene category information corresponding to the commodity object in the image is determined; wherein, when performing feature analysis on the image, the image is input into a plurality of neural network models, and the output results of a plurality of output layers of the plurality of neural network models are obtained as visualized feature values, and the activation values of a plurality of intermediate layers are obtained as abstract feature values; the activation values of the intermediate layers represent the feature information extracted from the commodity object image by the intermediate layer of the neural network model in the process of obtaining the output results, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; according to the output results of the plurality of output layers and the activation values of the plurality of intermediate layers, the abstract feature values are used as a supplement to the visualized feature values to generate a plurality of feature vectors with higher dimensions, and then feature vectors of a plurality of angles of the image are generated; the plurality of neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, and whether there are human models or props, and a model for identifying whether the human models in the image include full body / half body, human face image or expression features of human models; Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in; Determine the target user according to the display scene category information corresponding to the image and the display scene category information that the user is interested in; The picture is recommended to a client associated with the target user.
38. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the following operations are performed: Determine the video associated with the target product object; Extract at least one frame of image from the video, and determine the display scene category information corresponding to the commodity object in the video by performing feature analysis on the image; wherein, when performing feature analysis on the image, input the image into multiple neural network models, obtain the output results of multiple output layers of the multiple neural network models as visualized feature values, and the activation values of multiple intermediate layers as abstract feature values; the activation value of the intermediate layer represents the feature information extracted from the commodity object image by the intermediate layer of the neural network model in the process of obtaining the output result, or the feature information obtained by weighted summation and nonlinear transformation of the upper layer of neurons; according to the output results of the multiple output layers and the activation values of the multiple intermediate layers, use the abstract feature value as a supplement to the visualized feature value to generate multiple higher-dimensional feature vectors, and then generate feature vectors of multiple angles of the image; the multiple neural network models include: a model for distinguishing the area where the commodity object theme is located and the background area, a model for identifying whether the shooting scene is street shooting or studio shooting, whether there are human models or props, and a model for identifying whether the human model in the picture includes full body / half body, human face image or expression features of human models; Determine a user set, and obtain user characteristic information corresponding to a plurality of users in the user set, wherein the user characteristic information includes display scene category information that the users are interested in; Determine a target user according to the display scene category information corresponding to the video and the display scene category information that the user is interested in; The video is recommended to a client associated with the target user.
