Commodity position distribution method and device and virtual exhibition hall

Abstract

The embodiment of the invention provides a commodity position distribution method and device and a virtual exhibition hall, and relates to the technical field of data processing. The method comprises the following steps: acquiring attribute information of to-be-put-on-shelf commodities; inputting the attribute information of the to-be-put-on-shelf commodities into a classification model, classifying the to-be-put-on-shelf commodities through the classification model, and obtaining placement position information corresponding to a classification result; and according to the placement position information, distributing a shelving position for the to-be-shelved commodity. Therefore, according to the commodity position distribution method provided by the embodiment of the invention, the attribute information (such as brands, types, sizes, display requirements and the like) of the to-be-shelved commodities is input into the classification model, and the shelving positions can be automatically distributed to the to-be-shelved commodities according to the placement position information output by the classification model, so that the time of manual decision making and distribution can be reduced, and the accuracy of commodity position distribution is improved. The commodity shelving efficiency is improved, the space utilization of the virtual exhibition hall is more reasonable and efficient, and the user experience is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a commodity location allocation method, device and virtual exhibition hall. Background Art

[0002] With the continuous development of science and technology, augmented reality (AR) technology and virtual reality (VR) technology have provided new possibilities for product display and marketing interaction in the e-commerce field. Virtual exhibition halls built with AR and VR technologies can not only reduce the site rental and physical exhibition hall construction costs of e-commerce companies, but also provide consumers with an immersive interactive experience.

[0003] In the related art, in order to build a virtual exhibition hall, the relevant technicians need to first use 3D modeling software to create a virtual exhibition hall model, and then complete the allocation of the shelf positions of the goods in the virtual exhibition hall model. For example, the relevant technicians need to manually allocate the placement of the goods in the virtual exhibition hall according to the brand, type, size and display requirements of the goods.

[0004] However, once the number or type of goods in the virtual exhibition hall changes, relevant technical personnel will be required to manually adjust the display area of ​​the goods again, which not only consumes a lot of time and energy of relevant technical personnel and reduces the efficiency of goods location allocation, but may also affect users' browsing experience and purchasing decisions. Summary of the invention

[0005] Based on the above problems, the present application provides a product location allocation method, device and virtual exhibition hall, which can improve the user experience while improving the efficiency of product listing.

[0006] The embodiments of the present application disclose the following technical solutions:

[0007] In a first aspect, the present application discloses a method for allocating commodity locations, the method comprising:

[0008] Get the attribute information of the products to be put on the shelves;

[0009] Inputting the attribute information of the commodities to be put on the shelves into a classification model, classifying the commodities to be put on the shelves by using the classification model, and obtaining placement position information corresponding to the classification result;

[0010] According to the placement position information, a shelf position is allocated for the commodity to be put on the shelf.

[0011] Optionally, the classification model includes a first fully connected layer, an encoder layer, a maximum pooling layer, and an activation function layer connected in sequence;

[0012] The step of inputting the attribute information of the commodities to be put on the shelves into a classification model, classifying the commodities to be put on the shelves by using the classification model, and obtaining placement position information corresponding to the classification result includes:

[0013] The attribute information of the goods to be put on the shelves is input into the classification model, the target vector of the attribute information of the goods to be put on the shelves is obtained through the first fully connected layer, the fusion result of the target vector is obtained through the encoder layer, the fusion result is subjected to dimensionality reduction processing through the maximum pooling layer to obtain abstract features, the classification result corresponding to the abstract features is determined through the activation function layer, and the placement position information corresponding to the classification result is obtained.

[0014] Optionally, the formula for obtaining the fusion result of the target vector is as follows:

[0015]

[0016] Where PSsa represents the attention mechanism; Q is the query vector, which is part of the target vector y'; K is the key vector, which is part of the target vector y'; V is the value vector, which is the content corresponding to the target vector y'; the Softmax function is used to normalize the calculated attention scores to obtain the weight distribution; d is the dimension of the target vector y'; Vy' is the value corresponding to the target vector y'; y attention The final attention output is obtained by weighted summing of the value vector Vy' according to the normalized attention score.

[0017] Optionally, the attribute information is an attribute text; and inputting the attribute information of the commodity to be put on the shelf into a classification model includes:

[0018] Segmenting the attribute text according to the semantic features of the attribute text to obtain a word sequence of the attribute text;

[0019] Convert the word sequence into a word vector;

[0020] According to the plurality of word vectors, a fine-grained information matrix is ​​obtained, wherein each row of the fine-grained information matrix represents a set of word vectors corresponding to the attribute text;

[0021] The fine-grained information matrix is ​​input into a classification model.

[0022] Optionally, the method further comprises:

[0023] Get a virtual showroom model;

[0024] According to the shelf location, the product image and description information of the product to be put on the shelf are displayed in the virtual exhibition hall model.

[0025] In a second aspect, the present application discloses a commodity location allocation device, the device comprising: an information acquisition module, a model processing module and a location allocation module;

[0026] The information acquisition module is used to acquire the attribute information of the product to be put on the shelf;

[0027] The model processing module is used to input the attribute information of the commodities to be put on the shelves into the classification model, classify the commodities to be put on the shelves by using the classification model, and obtain placement position information corresponding to the classification result;

[0028] The position allocation module is used to allocate a shelf position for the commodity to be put on the shelf according to the placement position information.

[0029] Optionally, the classification model includes a first fully connected layer, an encoder layer, a maximum pooling layer, and an activation function layer connected in sequence;

[0030] The model processing module is specifically used to: input the attribute information of the goods to be put on the shelves into the classification model, obtain the target vector of the attribute information of the goods to be put on the shelves through the first fully connected layer, obtain the fusion result of the target vector through the encoder layer, perform dimensionality reduction processing on the fusion result through the maximum pooling layer to obtain abstract features, determine the classification result corresponding to the abstract feature through the activation function layer, and obtain the placement position information corresponding to the classification result.

[0031] Optionally, the formula for obtaining the fusion result of the target vector is as follows:

[0032]

[0033] Where PSsa represents the attention mechanism; Q is the query vector, which is part of the target vector y'; K is the key vector, which is part of the target vector y'; V is the value vector, which is the content corresponding to the target vector y'; the Softmax function is used to normalize the calculated attention scores to obtain the weight distribution; d is the dimension of the target vector y'; Vy' is the value corresponding to the target vector y'; y attention The final attention output is obtained by weighted summing of the value vector Vy' according to the normalized attention score.

[0034] Optionally, the attribute information is attribute text; the model processing module is specifically used to: segment the attribute text according to the semantic features of the attribute text to obtain a word sequence of the attribute text; convert the word sequence into a word vector; obtain a fine-grained information matrix based on multiple word vectors, wherein each row of the fine-grained information matrix represents a set of word vectors corresponding to the attribute text; and input the fine-grained information matrix into a classification model.

[0035] Optionally, the device further comprises: a model acquisition module and a commodity display module;

[0036] The model acquisition module is used to acquire the virtual exhibition hall model;

[0037] The commodity display module is used to display the commodity images and description information of the commodities to be put on the shelves in the virtual exhibition hall model according to the shelf positions.

[0038] In a third aspect, the present application discloses a virtual exhibition hall, which includes a commodity location allocation device, and the commodity location allocation device is used to execute the commodity location allocation method described in the first aspect.

[0039] Compared with the prior art, this application has the following beneficial effects:

[0040] The embodiment of the present application provides a method, device and virtual exhibition hall for allocating commodity positions, the method comprising: obtaining attribute information of commodities to be put on the shelves; inputting the attribute information of commodities to be put on the shelves into a classification model, classifying the commodities to be put on the shelves through the classification model, and obtaining placement information corresponding to the classification result; and allocating a shelf position for the commodities to be put on the shelves according to the placement information. Thus, the commodity position allocation method provided by the embodiment of the present application can automatically allocate a shelf position for the commodities to be put on the shelves according to the placement information output by the classification model by inputting the attribute information of the commodities to be put on the shelves (such as brand, type, size, display requirements, etc.) into the classification model, which can not only reduce the time of manual decision-making and allocation, and improve the efficiency of commodity shelving, but also make the space utilization of the virtual exhibition hall more reasonable and efficient, and improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] 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 or the description of the prior art 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 labor.

[0042] Figure 1 A flowchart of a method for allocating commodity locations provided in an embodiment of the present application;

[0043] Figure 2 A flowchart of another method for allocating commodity locations provided in an embodiment of the present application;

[0044] Figure 3 A schematic diagram of the correspondence between a word sequence and a word vector provided in an embodiment of the present application;

[0045] Figure 4A flowchart of an informer classification model training method provided in an embodiment of the present application;

[0046] Figure 5 A schematic diagram of an informer classification model structure provided in an embodiment of the present application;

[0047] Figure 6 A schematic diagram of a commodity location allocation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] As described above, in order to build a virtual exhibition hall, relevant technicians need to first use 3D modeling software to create a virtual exhibition hall model, and then complete the allocation of product shelf locations in the virtual exhibition hall model. For example, relevant technicians need to manually allocate product placement locations in the virtual exhibition hall based on the brand, type, size, and display requirements of the product.

[0049] However, once the number or type of goods in the virtual exhibition hall changes, relevant technical personnel will be required to manually adjust the display area of ​​the goods again, which not only consumes a lot of time and energy of relevant technical personnel and reduces the efficiency of goods location allocation, but may also affect users' browsing experience and purchasing decisions.

[0050] After research, the inventors have proposed a product location allocation method, device and virtual exhibition hall. The product location allocation method provided in the embodiment of the present application can automatically allocate shelf locations for the products to be put on the shelves according to the placement position information output by the classification model by inputting the attribute information of the products to be put on the shelves (such as brand, type, size, display requirements, etc.) into the classification model. This not only reduces the time for manual decision-making and allocation, improves the efficiency of product shelving, but also makes the space utilization of the virtual exhibition hall more reasonable and efficient, thereby improving the user experience.

[0051] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings 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. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] See also Figure 1 , which is a flow chart of a commodity location allocation method provided in an embodiment of the present application. The method includes:

[0053] S101: Obtaining attribute information of the product to be put on the shelf.

[0054] The attribute information of the products to be listed includes the basic attributes of the products to be listed (including the name, brand, model, color, material, etc. of the products to be listed), specification attributes (such as physical properties such as size, weight, capacity, etc.), price attributes (such as selling price, promotional price, discount information, etc.), and category attributes (such as electronic products, household items, clothing, etc.).

[0055] In some specific implementations, the attribute information of the above-mentioned goods to be put on the shelves can be obtained from the supplier, usually in the form of a spreadsheet or a data file. Alternatively, the attribute information of the above-mentioned goods to be put on the shelves can also be obtained from an e-commerce platform or other third-party data source. This application does not limit the specific acquisition method.

[0056] For example, assuming that the product to be listed is a smart phone, the acquired attribute information may include: name: XYZ smart phone; model: XYZ-123; size: 160mm x 75mm x 8mm; current price: 2999 yuan; original price: 3599 yuan; sales volume: 10,000; category: electronic products > smart phones.

[0057] S102: Input the attribute information of the commodities to be put on the shelves into the classification model, classify the commodities to be put on the shelves by the classification model, and obtain placement position information corresponding to the classification result.

[0058] The classification model is a machine learning model that has been trained with a large amount of data. It can classify products according to the input attribute information and determine the corresponding placement information based on the classification results.

[0059] Specifically, the attribute information of the goods to be put on the shelves is input into the trained classification model, so that the classification model performs classification prediction based on the input attribute information to obtain the classification results of the goods to be put on the shelves. Among them, the classification result can be multiple category labels, which together represent the classification results of the goods to be put on the shelves. Subsequently, once the goods to be put on the shelves are classified, the placement information of the goods to be put on the shelves can be determined according to the pre-defined mapping rules. Among them, the mapping rule is a rule that converts the classification result into specific placement information. For example, the "electronic products" category may be mapped to area A of the exhibition hall, while the "household items" category may be mapped to area B of the exhibition hall. The placement information can be a specific area, shelf, number of layers, position number, etc. By applying these mapping rules, the placement position of each product to be put on the shelves can be accurately determined.

[0060] For example, if the above attribute information of the smartphone is input into the trained classification model, the classification model can classify the smartphone into the category of "electronic products - special price" according to the attribute information. Then, according to the pre-defined mapping rules, the placement location corresponding to the "electronic products" category is Area A of the exhibition hall, and the placement location corresponding to the "special price" category is the first row of containers. Then, the classification model can determine the placement location information corresponding to the smartphone as: the first row of containers in Area A of the exhibition hall.

[0061] S103: Allocate a shelf position for the product to be put on the shelf according to the placement position information.

[0062] It should be noted that as sales activities progress, the shelf position can be continuously optimized based on factors such as the sales situation of the product, customer purchasing habits, and sales strategies.

[0063] In summary, the embodiment of the present application provides a method for allocating commodity positions, the method comprising: obtaining attribute information of commodities to be put on the shelves; inputting the attribute information of commodities to be put on the shelves into a classification model, classifying the commodities to be put on the shelves through the classification model, and obtaining placement information corresponding to the classification result; and allocating a shelf position for the commodities to be put on the shelves according to the placement information. Therefore, the method for allocating commodity positions provided by the embodiment of the present application can automatically allocate a shelf position for the commodities to be put on the shelves according to the placement information output by the classification model by inputting the attribute information of the commodities to be put on the shelves (such as brand, type, size, display requirements, etc.) into the classification model, which can not only reduce the time of manual decision-making and allocation, and improve the efficiency of commodity shelving, but also make the space utilization of the virtual exhibition hall more reasonable and efficient, and improve the user experience.

[0064] See also Figure 2 , which is a flow chart of another commodity location allocation method provided by an embodiment of the present application. The method includes:

[0065] S201: Generate a virtual exhibition hall model.

[0066] First, generate an initialized virtual exhibition hall model, including at least one of setting the scene size of the virtual exhibition hall, determining the camera position, determining the lighting conditions and setting the background. Among them, setting the scene size of the virtual exhibition hall refers to defining the overall size (including length, width and height) of the virtual exhibition hall model, which determines the user's range of activities in the virtual exhibition hall model. Determining the camera position refers to determining the initial position and direction of the camera. Determining the lighting conditions refers to configuring the light source (including main light source, auxiliary light source and ambient light, etc.) in the virtual exhibition hall model. The lighting conditions directly affect the visual effect and atmosphere of the virtual exhibition hall model. Setting the background refers to selecting or creating a suitable background image or color to enhance the realism and aesthetics of the virtual exhibition hall model.

[0067] Secondly, load the sub-models such as display racks, display cabinets, large screens, and decorations into the initialized virtual exhibition hall model to obtain an updated virtual exhibition hall model. It should be noted that in the process of loading the above sub-models, the sub-models need to be placed in the specified positions in the initialized virtual exhibition hall model, and the angles and orientations of the sub-models need to be adjusted to ensure a reasonable and beautiful layout. In addition, it is also necessary to check whether there is overlap or collision between the sub-models, and make adjustments if necessary to avoid visual confusion.

[0068] Subsequently, the appearance and texture of the updated virtual exhibition hall model are adjusted to conform to the design style and activity type, and the virtual exhibition hall model after the second update is obtained. In one example, a suitable material (such as metal, wood, glass, etc.) can be selected for the updated virtual exhibition hall model to reflect its physical properties. In another example, a texture map can also be applied to the surface of the updated virtual exhibition hall model to increase details and realism. In another example, the reflectivity, transparency, roughness and other properties of the material can also be adjusted as needed to achieve the best visual effect.

[0069] Next, interactive elements are added to the virtual exhibition hall model after the second update to provide interactive functions, and the virtual exhibition hall model after the third update is obtained. In one example, buttons such as "Back", "Next Page", "View Details", etc. can be added at any location to facilitate user navigation and operation. In another example, labels can also be added to exhibits to display information such as name, price, description, etc. to enhance user understanding. In another example, navigation paths and arrows can also be set to guide users to move around the exhibition hall. In another example, feedback mechanisms for interactive elements can also be configured, such as sound effects and animations when clicking buttons, to enhance user experience.

[0070] Then, animation effects or other special effects are added to the virtual exhibition hall model after three updates to enhance the display effect and user experience, and the virtual exhibition hall model after four updates is obtained. In one example, animations such as rotation, translation, and scaling can be set to make the exhibits more attractive. In another example, special effects such as particle systems, light and shadow effects, and dynamic backgrounds can also be added to increase the dynamics and visual impact of the scene.

[0071] Finally, based on the virtual exhibition hall model after four updates, the user interface of the virtual exhibition hall model is generated, such as creating menus, setting toolbars, setting status bars, providing interactive design, etc., to provide users with control and feedback on the scene.

[0072] S202: Generate a corresponding fine-grained information matrix according to the attribute text of the product to be put on the shelf.

[0073] First, the attribute text of the product to be listed is converted into a machine-processable word sequence. Specifically, irrelevant characters, punctuation marks, stop words, etc. in the attribute text of the product to be listed are first removed, and meaningful words in the attribute text are retained; then, according to the semantic features of the retained attribute text, the retained attribute text is segmented into multiple words (tokens) to form a corresponding word sequence. For example, "XYZ smartphone has very good performance" is segmented into "XYZ smartphone", "performance", and "very good".

[0074] Second, convert the word sequence into a word vector. Figure 3 , which is a schematic diagram of the correspondence between a word sequence and a word vector provided in an embodiment of the present application. For each word ω in the word sequence, the trained Word2Vec model is used to convert it into a word vector of fixed length m Thus, the semantic relationship between words in the word sequence is captured. Among them, the Word2Vec model is a model pre-trained using a large amount of attribute text (such as product descriptions, comments, etc.).

[0075] Finally, multiple word vectors {v_1, v_2, ..., v_n} are arranged in order and integrated to form a fine-grained information matrix y(v). The dimension of the fine-grained information matrix y(v) is nxm, and each row of the fine-grained information matrix represents a set of word vectors corresponding to an attribute text.

[0076] S203: Send the fine-grained information matrix to the informer classification model, classify the to-be-shelf commodities corresponding to the fine-grained information matrix through the informer classification model, and obtain placement position information corresponding to the classification result.

[0077] See also Figure 4 , which is a flow chart of an informer classification model training method provided in an embodiment of the present application. The informer classification model training method is as follows:

[0078] A1: Obtain a sample data set, where the sample data set includes attribute text and placement information of sample products.

[0079] A2: Determine the fine-grained information matrix corresponding to the sample product based on the attribute text of the sample product.

[0080] A3: Use the fine-grained information matrix corresponding to the sample products as the input of the machine learning model, and the placement information of the sample products as the output of the machine learning model to train the machine learning model and obtain the informer classification model.

[0081] It should be noted that in this step, it is necessary to select an appropriate machine learning algorithm (such as a neural network, decision tree, etc.) and set the model parameters of the machine learning model. By iteratively adjusting the model parameters to minimize the prediction error, the machine learning model gradually learns to predict the placement of the product from its fine-grained information matrix. Ultimately, the trained machine learning model is called an informer classification model, which can accept a new fine-grained information matrix as input and predict its possible placement.

[0082] After obtaining the fine-grained information matrix corresponding to the products to be put on the shelves, the fine-grained information matrix can be input into the informer classification model constructed by the above method. Figure 5 , which is a schematic diagram of an informer classification model structure provided in an embodiment of the present application. The informer classification model includes a first fully connected layer, an encoder layer, a maximum pooling layer, and an activation function layer connected in sequence.

[0083] Specifically, first, the fine-grained information matrix is ​​input into the first fully connected layer of the informer classification model so that the first fully connected layer transforms the fine-grained information matrix into a unified The target vector of dimension is as shown in formula (1):

[0084] y′=Fully(y(v)) ( 1 )

[0085] Among them, y(v) is the fine-grained information matrix, Fully represents the fully connected layer operation, and y' is the unified The target vector of dimension .

[0086] Secondly, the Informer encoder layer based on the ProbSparse Self-Attention mechanism is used to unify the converted The target vector y' of dimension is fused, as shown in formula (2):

[0087]

[0088] Where PSsa represents the attention mechanism; Q is the query vector, which is part of the target vector y'; K is the key vector, which is part of the target vector y'; V is the value vector, which is the content corresponding to the target vector y'; the Softmax function is used to normalize the calculated attention scores to obtain the weight distribution; d is the dimension of the target vector y'; Vy' is the value corresponding to the target vector y'; y attention The final attention output is obtained by weighted summing of the value vector Vy' according to the normalized attention score.

[0089] It should be noted that, unlike the traditional self-attention mechanism method, the probabilistic sparse self-attention mechanism shown in formula (3) can reduce the quadratic computational complexity and memory consumption of the self-attention mechanism while maintaining the performance of the informer classification model.

[0090]

[0091] Among them, M(q i , K) is the calculation formula of the attention score, which represents the query vector q i and the correlation between the key vector group K; L K is the number of elements in the key vector group K; k j is the jth key vector; d is the dimension of the vector.

[0092] The core idea of ​​formula (3) is to introduce the concept of probabilistic sparsity in the process of calculating the attention score. By comparing the sum of the inner products of the query vector and all key vectors with their mean, the correlation strength between the query vector and the specific key vector is determined, which helps to reduce the computational complexity, especially when processing large-scale data, and can find key information more efficiently.

[0093] Subsequently, the fusion result is subjected to dimensionality reduction processing through the maximum pooling layer, thereby reducing the sequence length and further extracting the abstract features in the fusion result. Specifically, the maximum pooling layer design can be shown in the following formula (4):

[0094]

[0095] Among them, y pooling is the output of the maximum pooling layer, MaxPool represents the maximum pooling operation, ELU represents the activation function, Conv1D represents the 1D convolution kernel, and y attention Indicates the fusion result.

[0096] The core idea of ​​formula (4) is to firstly attention A 1D convolution operation is performed, followed by the introduction of nonlinearity through the ELU activation function, and finally a maximum pooling operation is performed to reduce the dimension and extract important features.

[0097] Finally, the classification result corresponding to the abstract feature is determined by the Softmax activation function of the activation function layer, and the placement position information corresponding to the classification result is obtained.

[0098] It should be noted that in actual applications, the informer classification model outputs the placement information of the goods to be put on the shelves. Therefore, the informer classification model cancels the traditional mask mechanism, so that the informer classification model can directly use the fine-grained information matrix for classification.

[0099] It should also be noted that a second fully connected layer can be added after the maximum pooling layer of the informer classification model and before the activation function layer. The purpose of adding the second fully connected layer is to slow down the dimensionality reduction speed of the informer classification model, better retain the feature information output by the informer classification model, reduce the information loss of the informer classification model, and improve the classification accuracy of the informer classification model.

[0100] S204: Allocate a shelf position for the product to be put on the shelf according to the placement position information.

[0101] S205: Displaying product images and description information of the products to be put on the shelves in the virtual exhibition hall model according to the shelf locations.

[0102] At a location corresponding to the shelf location (which may be a specific shelf number, display area, etc.) of the virtual exhibition hall model constructed in S201, the product image and description information of the product to be put on the shelf are displayed. The product image may be a high-definition picture provided by a supplier or an automatically generated image.

[0103] After displaying the product images and description information of the products to be put on the shelves in the virtual exhibition hall model, consumers can log in to the virtual exhibition hall through mobile phones, PCs and other terminals, and interact with all the products displayed in the virtual exhibition hall. The interaction methods may include the following:

[0104] First, consumers can select products on their mobile phones or PCs through virtual showrooms and use the virtual try-on or trial functions. By dragging or clicking, consumers can place products in the virtual scene and try on clothes or use products on their own virtual images to experience the effects and applicability of the products.

[0105] Second, consumers can browse product information through mobile phones, PCs or other terminals. This information may include product name, description, price, material, color, size, etc. Customers can click on product icons or related tags to view more detailed product information.

[0106] Third, in the virtual exhibition hall, consumers can use mobile phones or PCs for interactive navigation. For example, they can move and browse in the virtual scene by clicking or dragging to better understand the displayed products and scene layout.

[0107] Fourth, consumers can communicate and consult with online customer service or sales staff through real-time communication functions. For example, they can send messages, voice or video calls on mobile phones or PCs to obtain more information about the product or ask questions.

[0108] Fifth, virtual showrooms can provide interactive animations and special effects, allowing consumers to interact with the products in the virtual showroom in a more vivid and interesting way. For example, consumers can click a button or trigger a specific operation to start product display animation or special effects, thereby enhancing consumer participation and experience.

[0109] It should be noted that this application does not limit the specific interaction method.

[0110] S206: When it is detected that the virtual exhibition hall model after displaying the products to be put on the shelves fails the review, an alarm indication is triggered to enable relevant technical personnel to change the display position of the products to be put on the shelves in the virtual exhibition hall model.

[0111] Before reviewing the virtual exhibition hall model after displaying the products to be put on the shelves, it is necessary to first formulate a set of review standards for evaluating the display effect of the products to be put on the shelves in the virtual exhibition hall model, such as the rationality of the display location, etc. It should be noted that this application does not limit the specific review standards.

[0112] Regarding the review criteria for the rationality of display locations, in one example, if multiple identical or similar products are placed in adjacent display locations, causing the display area to appear crowded and lacking in diversity, an alarm indication needs to be triggered so that relevant technical personnel can reallocate the display locations of the products to ensure that different types of products can be evenly distributed, thereby increasing the diversity and attractiveness of the display. In another example, if a high-selling product is placed in a display area for low-selling products, causing the sales of the high-selling product to drop, an alarm indication needs to be triggered so that relevant technical personnel can adjust the high-selling product to a display area that is more in line with its positioning to increase sales.

[0113] When it is detected that the display of the virtual exhibition hall model fails to pass the review, an alarm is triggered to inform the relevant technicians that there is a problem with the display position. The alarm can be notified to the relevant technicians in a variety of ways, such as email, SMS, system notification, etc. The technicians re-evaluate and adjust the display position of the products to be put on the shelves in the virtual exhibition hall model according to the alarm prompt.

[0114] In other specific implementations, the review standard may also be a review standard for visual effects, or a review standard for user experience. For the review standard for visual effects, in one example, if the product image is blurry or has a low resolution, which affects the customer's perception, an alarm indication needs to be triggered so that the relevant technical personnel can replace the high-quality product image to ensure that the image is clear and beautiful. For the review standard for user experience, in one example, if the product description information is incomplete or inaccurate, and the customer cannot obtain enough information to make a purchase decision, an alarm indication needs to be triggered so that the relevant technical personnel can complete the product description information, including name, brand, model, price, functional features, etc., to ensure that the information is comprehensive and accurate.

[0115] S207: Update the informer classification model according to the shelf position corresponding to the changed display position.

[0116] Get the new shelf position from the virtual exhibition hall model adjusted by the technician, update the informer classification model according to the new shelf position, and evaluate the performance of the updated informer classification model through cross-validation and other methods to ensure its accuracy and reliability. After updating the informer classification model, save the updated informer classification model for subsequent use.

[0117] To sum up, the embodiment of the present application provides a method for allocating product locations. The method provided by the embodiment of the present application inputs the attribute information of the products to be put on the shelves (such as brand, type, size, display requirements, etc.) into the classification model, and can automatically allocate shelf locations for the products to be put on the shelves according to the placement position information output by the classification model. This not only reduces the time for manual decision-making and allocation, improves the efficiency of putting products on the shelves, but also makes the space utilization of the virtual exhibition hall more reasonable and efficient, thereby improving the user experience.

[0118] See also Figure 6 , which is a schematic diagram of a commodity location allocation device provided in an embodiment of the present application. The commodity location allocation device 600 includes: an information acquisition module 601, a model processing module 602 and a location allocation module 603;

[0119] The information acquisition module 601 is used to acquire the attribute information of the product to be put on the shelf;

[0120] Model processing module 602, used to input the attribute information of the goods to be put on the shelves into the classification model, classify the goods to be put on the shelves through the classification model, and obtain placement position information corresponding to the classification result;

[0121] The position allocation module 603 is used to allocate a shelf position for the commodity to be put on the shelf according to the placement position information.

[0122] In some specific implementations, the classification model includes a first fully connected layer, an encoder layer, a maximum pooling layer, and an activation function layer connected in sequence;

[0123] The model processing module 602 is specifically used to: input the attribute information of the goods to be put on the shelves into the classification model, obtain the target vector of the attribute information of the goods to be put on the shelves through the first fully connected layer, obtain the fusion result of the target vector through the encoder layer, perform dimensionality reduction processing on the fusion result through the maximum pooling layer to obtain abstract features, determine the classification results corresponding to the abstract features through the activation function layer, and obtain the placement position information corresponding to the classification results.

[0124] In some specific implementations, the formula for obtaining the fusion result of the target vector is as follows:

[0125]

[0126] Where PSsa represents the attention mechanism; Q is the query vector, which is part of the target vector y'; K is the key vector, which is part of the target vector y'; V is the value vector, which is the content corresponding to the target vector y'; the Softmax function is used to normalize the calculated attention scores to obtain the weight distribution; d is the dimension of the target vector y'; Vy' is the value corresponding to the target vector y'; y attention The final attention output is obtained by weighted summing of the value vector Vy' according to the normalized attention score.

[0127] In some specific implementations, the attribute information is attribute text; the model processing module 602 is specifically used to: segment the attribute text according to the semantic features of the attribute text to obtain the word sequence of the attribute text; convert the word sequence into a word vector; obtain a fine-grained information matrix based on multiple word vectors, wherein each row of the fine-grained information matrix represents a set of word vectors corresponding to an attribute text; and input the fine-grained information matrix into the classification model.

[0128] In some specific implementations, the commodity location allocation device 600 further includes: a model acquisition module and a commodity display module;

[0129] A model acquisition module is used to acquire a virtual exhibition hall model;

[0130] The product display module is used to display product images and description information of the products to be put on the shelves in the virtual exhibition hall model according to the shelf location.

[0131] To sum up, the embodiment of the present application provides a product location allocation device. The product location allocation device provided in the embodiment of the present application inputs the attribute information of the product to be put on the shelf (such as brand, type, size, display requirements, etc.) into the classification model, and can automatically allocate shelf positions for the products to be put on the shelf according to the placement position information output by the classification model. This not only reduces the time for manual decision-making and allocation, improves the efficiency of product shelving, but also makes the space utilization of the virtual exhibition hall more reasonable and efficient, thereby improving the user experience.

[0132] The embodiment of the present application also provides a corresponding virtual exhibition hall, which includes a commodity position allocation device, which is used to execute the commodity position allocation method of the first aspect. The virtual exhibition hall can automatically allocate shelf positions for the commodities to be put on the shelves according to the placement position information output by the classification model by inputting the attribute information of the commodities to be put on the shelves (such as brand, type, size, display requirements, etc.) into the classification model, which can not only reduce the time of manual decision-making and allocation, improve the efficiency of commodity shelving, but also make the space utilization of the virtual exhibition hall more reasonable and efficient, and improve the user experience.

[0133] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device 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 indicated 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 paying creative labor.

[0134] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for allocating commodity positions, characterized in that: The method comprises: Get the attribute information of the product to be put on the shelf; Inputting the attribute information of the commodities to be put on the shelves into a classification model, classifying the commodities to be put on the shelves by using the classification model, and obtaining placement position information corresponding to the classification result; According to the placement position information, a shelf position is allocated for the commodity to be put on the shelf.

2. The method according to claim 1, characterized in that The classification model includes a first fully connected layer, an encoder layer, a maximum pooling layer and an activation function layer connected in sequence; The step of inputting the attribute information of the commodities to be put on the shelves into a classification model, classifying the commodities to be put on the shelves by using the classification model, and obtaining placement position information corresponding to the classification result includes: The attribute information of the goods to be put on the shelves is input into the classification model, the target vector of the attribute information of the goods to be put on the shelves is obtained through the first fully connected layer, the fusion result of the target vector is obtained through the encoder layer, the fusion result is subjected to dimensionality reduction processing through the maximum pooling layer to obtain abstract features, the classification result corresponding to the abstract features is determined through the activation function layer, and the placement position information corresponding to the classification result is obtained.

3. The method according to claim 2, characterized in that The formula for obtaining the fusion result of the target vector is as follows: Among them, PSsa represents the attention mechanism; Q is the query vector, which is part of the target vector y'; K is the key vector, which is part of the target vector y'; V is the value vector, which is the content corresponding to the target vector y'; the Softmax function is used to normalize the calculated attention scores to obtain the weight distribution; d is the dimension of the target vector y'; Vy ’ is the value corresponding to the target vector y'; y attention is the value vector Vy based on the normalized attention score ’ The weighted sum is performed to obtain the final attention output.

4. The method according to claim 1, characterized in that The attribute information is an attribute text; the step of inputting the attribute information of the commodity to be put on the shelf into a classification model includes: Segmenting the attribute text according to the semantic features of the attribute text to obtain a word sequence of the attribute text; Convert the word sequence into a word vector; According to the plurality of word vectors, a fine-grained information matrix is ​​obtained, wherein each row of the fine-grained information matrix represents a set of word vectors corresponding to the attribute text; The fine-grained information matrix is ​​input into a classification model.

5. The method according to claim 1, characterized in that The method further comprises: Get a virtual showroom model; According to the shelf location, the product image and description information of the product to be put on the shelf are displayed in the virtual exhibition hall model.

6. A commodity location allocation device, characterized in that: The device comprises: an information acquisition module, a model processing module and a position allocation module; The information acquisition module is used to acquire the attribute information of the product to be put on the shelf; The model processing module is used to input the attribute information of the commodities to be put on the shelves into the classification model, classify the commodities to be put on the shelves by using the classification model, and obtain placement position information corresponding to the classification result; The position allocation module is used to allocate a shelf position for the commodity to be put on the shelf according to the placement position information.

7. The device according to claim 6, characterized in that The classification model includes a first fully connected layer, an encoder layer, a maximum pooling layer and an activation function layer connected in sequence; The model processing module is specifically used to: input the attribute information of the goods to be put on the shelves into the classification model, obtain the target vector of the attribute information of the goods to be put on the shelves through the first fully connected layer, obtain the fusion result of the target vector through the encoder layer, perform dimensionality reduction processing on the fusion result through the maximum pooling layer to obtain abstract features, determine the classification result corresponding to the abstract feature through the activation function layer, and obtain the placement position information corresponding to the classification result.

8. The device according to claim 7, characterized in that The formula for obtaining the fusion result of the target vector is as follows: Among them, PSsa represents the attention mechanism; Q is the query vector, which is part of the target vector y'; K is the key vector, which is part of the target vector y'; V is the value vector, which is the content corresponding to the target vector y'; the Softmax function is used to normalize the calculated attention scores to obtain the weight distribution; d is the dimension of the target vector y'; Vy ’ is the value corresponding to the target vector y'; y attention is the value vector Vy based on the normalized attention score ’ The weighted sum is performed to obtain the final attention output.

9. The device according to claim 6, characterized in that The attribute information is attribute text; the model processing module is specifically used to: segment the attribute text according to the semantic features of the attribute text to obtain the word sequence of the attribute text; convert the word sequence into a word vector; obtain a fine-grained information matrix based on multiple word vectors, wherein each row of the fine-grained information matrix represents a set of word vectors corresponding to the attribute text; and input the fine-grained information matrix into a classification model.

10. A virtual exhibition hall, characterized in that: The virtual exhibition hall comprises a commodity position allocation device, and the commodity position allocation device is used to execute the commodity position allocation method according to any one of claims 1 to 5.

Images