Packaging design method and device based on target GNN model, equipment and medium

Abstract

The application provides a packaging design method and device based on a target GNN model, equipment and medium, which are used in the technical field of packaging design and can solve the problems of poor efficiency and quality of existing packaging design. The method comprises the following steps: performing feature recognition on a to-be-designed image, determining a target design vector according to each image vector and a corresponding target text vector; calculating a material similarity value and a process similarity value according to the target GNN model, and determining a target design scheme according to the material similarity value and the process similarity value; calculating a score loss value according to a user score value and a predicted score value corresponding to each target design scheme, and calculating a divergence value according to a preset material, a preset process, a plurality of historical design schemes and the target design scheme; performing weighted calculation on the score loss value and the divergence value to obtain a target loss value, and adjusting parameters of the target GNN model according to the target loss value and a preset loss threshold; in this way, the quality and efficiency of the design are improved.

Description

Technical Field

[0001] This application relates to the field of packaging design technology, and in particular to packaging design methods, apparatus, equipment and media based on target GNN models. Background Technology

[0002] Packaging and printing design deeply integrates the visual creativity, functional requirements, and production implementation of product packaging. Through the combination of graphics, colors, materials, and processes, it creates packaging solutions that combine protection, brand communication, and user experience, thereby enhancing the product's advantages and improving its competitiveness.

[0003] Existing packaging design methods generally involve pre-setting the packaging box structure, relying on rule libraries for process recommendations, and failing to integrate multimodal image analysis; or matching materials through user tags, which easily overlooks the details of the design drawings.

[0004] However, existing packaging design methods suffer from low design efficiency and quality. Summary of the Invention

[0005] This application provides a packaging design method, apparatus, equipment, and medium based on a target GNN model to address the problems of low design efficiency and quality in existing packaging design methods.

[0006] Firstly, this application provides a packaging design method based on a target GNN model, the method comprising:

[0007] Feature recognition is performed on the image to be designed to obtain each image feature and the image vector corresponding to the image feature. Based on each image vector and its corresponding target text vector, the target design vector is determined.

[0008] Based on the target GNN model, the material similarity value between the target design vector and each material node vector, as well as the process similarity value between the target design vector and each process node vector, are calculated. Based on the material similarity value and the process similarity value, multiple target design schemes are determined. The target design scheme includes the target material and the target process.

[0009] The scoring loss value is calculated based on the user rating and predicted rating value corresponding to each target design scheme, and the divergence value is calculated based on the preset materials, preset processes, multiple historical design schemes and target design schemes; the predicted rating value is the value determined by the target GNN model.

[0010] Based on preset weights, the scoring loss value and divergence value are weighted and calculated to obtain the target loss value. Then, based on the target loss value and the preset loss threshold, the parameters of the target GNN model are adjusted to obtain the adjusted target GNN model, so that the target loss value calculated based on the adjusted target GNN model is less than the preset loss threshold.

[0011] In some embodiments of this application, the target design vector is determined based on each image vector and its corresponding target text vector, including:

[0012] The text to be designed is semantically segmented to obtain multiple sub-texts, and the vector corresponding to each sub-text is determined to obtain multiple text vectors;

[0013] The contrastive learning mechanism based on the CLIP model determines the target text vector corresponding to each image vector in the text vector, and associates the image vector and the target text vector to obtain the target vector.

[0014] Based on the attention mechanism of the CLIP model, vector fusion is performed on all target vectors and text vectors to obtain the target design vector.

[0015] In some embodiments of this application, before calculating the material similarity value between the target design vector and each material node vector, and the process similarity value with each process node vector, based on the target GNN model, the method further includes:

[0016] Based on the CLIP model, sample images, and corresponding sample text, the corresponding sample design vector is determined. Then, based on the preset design database, the positive and negative sample schemes corresponding to the sample design vector are determined. The positive sample scheme is the sample scheme that is associated with the sample design vector, and the negative sample scheme is the sample scheme that is not associated with the sample design vector.

[0017] Determine the positive sample vector corresponding to the positive sample scheme and the negative sample vector corresponding to the negative sample scheme, and calculate the positive similarity value between the sample design vector and the positive sample vector and the negative similarity value between the sample design vector and the negative sample vector based on the initial GNN model.

[0018] Based on the difference between negative and positive similarity values, the corresponding loss value is determined. Then, based on the loss value and a preset threshold, the model parameters of the initial GNN model are adjusted to obtain the target GNN model.

[0019] In some embodiments of this application, the model parameters of the initial GNN model are adjusted according to the loss value and a preset threshold to obtain the target GNN model, including:

[0020] By comparing the loss value with the preset threshold, the corresponding comparison results are obtained;

[0021] If the comparison result shows that the loss value is less than the preset threshold, the model parameters of the initial GNN model are adjusted according to the loss value to obtain the target GNN model, so that the loss value calculated based on the target GNN model is not less than the preset threshold.

[0022] If the comparison result shows that the loss value is not less than the preset threshold, then the initial GNN model is determined to be the target GNN model.

[0023] In some embodiments of this application, multiple target design schemes are determined based on material similarity values ​​and process similarity values, including:

[0024] According to the preset correspondence rules, at least one process node corresponding to each material node is determined, and each material node and its corresponding process nodes are determined as target design nodes; a material node is the node corresponding to the material node vector, and a process node is the node corresponding to the process node vector.

[0025] The target material similarity value and target process similarity value corresponding to each target design node are added together to obtain multiple design similarity values. The design similarity values ​​are then sorted from largest to smallest to obtain a design similarity value sequence. The target material similarity value is the material similarity value corresponding to the material node in the target design node, and the target process similarity value is the process similarity value corresponding to the process node in the target design node.

[0026] A preset number of design similarity values ​​are determined sequentially in the design similarity value sequence to obtain multiple target design similarity values. Based on the target material node and target process node corresponding to each target design similarity value, the corresponding target material and target process are determined.

[0027] Based on the target materials and target processes, multiple corresponding target design schemes are determined.

[0028] In some embodiments of this application, the corresponding rating loss value is calculated based on the user rating value and predicted rating value corresponding to each target design scheme, including:

[0029] Identify and output multiple target design schemes;

[0030] Based on the output feedback results corresponding to the target design schemes, determine the user rating values ​​corresponding to each target design scheme, and based on the target GNN model, determine the predicted rating values ​​corresponding to each target design scheme.

[0031] Based on the user rating and the corresponding predicted rating, calculate the error value for each target design scheme, and calculate the average value of all error values ​​to obtain the rating loss value.

[0032] In some embodiments of this application, the divergence value is calculated based on preset materials, preset processes, multiple historical design schemes, and a target design scheme, including:

[0033] Based on the pre-set design database, pre-set materials, pre-set processes, and multiple historical design schemes are determined. Based on the pre-set materials and pre-set processes, target material schemes and target process schemes are determined in the historical design schemes and target design schemes, respectively. The target material scheme is a scheme that includes the pre-set materials, and the target process scheme is a scheme that includes the pre-set processes.

[0034] Calculate the material divergence value based on the target material quantity corresponding to the target design scheme and the historical design scheme, and calculate the process divergence value based on the target process quantity corresponding to the target design scheme and the historical design scheme; the target material quantity is the quantity corresponding to the target material scheme, and the target process quantity is the quantity corresponding to the target process scheme.

[0035] Based on the preset weight values ​​corresponding to the target material quantity and the target process quantity, the target material quantity and the target process quantity are weighted and summed to obtain the divergence value.

[0036] Secondly, this application provides a packaging design apparatus based on a target GNN model, the apparatus comprising:

[0037] The recognition module is used to perform feature recognition on the design image to obtain each image feature and the image vector corresponding to the image feature, and to determine the target design vector based on each image vector and its corresponding target text vector.

[0038] The determination module is used to calculate the material similarity value between the target design vector and each material node vector, and the process similarity value between the target design vector and each process node vector, based on the target GNN model, and to determine multiple target design schemes based on the material similarity value and the process similarity value; the target design scheme includes the target material and the target process;

[0039] The calculation module is used to calculate the score loss value based on the user score and predicted score corresponding to each target design scheme, and to calculate the divergence value based on the preset materials, preset processes, multiple historical design schemes and target design schemes; the predicted score value is a value determined based on the target GNN model.

[0040] The adjustment module is used to perform weighted calculation of the scoring loss value and divergence value according to preset weights to obtain the target loss value, and adjust the parameters of the target GNN model according to the target loss value and the preset loss threshold to obtain the adjusted target GNN model, so that the target loss value calculated based on the adjusted target GNN model is less than the preset loss threshold.

[0041] Thirdly, this application provides a computer device, including: a processor, and a memory communicatively connected to the processor;

[0042] The memory stores the instructions that the computer executes;

[0043] The processor executes computer execution instructions stored in memory to implement the method of this application.

[0044] Fourthly, this application provides a computer-readable storage medium storing program code, which, when executed by a processor, is used to implement the method of this application.

[0045] This application provides a packaging design method, apparatus, device, and medium based on a target GNN model. The method involves feature recognition of the image to be designed to obtain various image features and their corresponding image vectors. Based on these image vectors and their corresponding target text vectors, a target design vector is determined. According to the target GNN model, the material similarity value between the target design vector and each material node vector, and the process similarity value with each process node vector, are calculated. Multiple target design schemes are determined based on these material and process similarity values. Each target design scheme includes a target material and a target process. A rating loss value is calculated based on the user rating and predicted rating values ​​corresponding to each target design scheme. A divergence value is calculated based on preset materials, preset processes, multiple historical design schemes, and the target design scheme. The predicted rating value is a value determined according to the target GNN model. The rating loss value and divergence value are weighted according to preset weights to obtain a target loss value. The parameters of the target GNN model are adjusted based on the target loss value and a preset loss threshold to obtain an adjusted target GNN model, ensuring that the target loss value calculated using the adjusted model is less than the preset loss threshold.

[0046] Thus, by performing feature recognition on the design image and generating image vectors, the target design vector is determined by combining it with the target text vector. This enables accurate identification of image features based on the image, and further feature fusion of image features and corresponding text features facilitates subsequent matching between materials and processes, while avoiding a disconnect between the actual design and user needs. Furthermore, by using a GNN model to calculate the similarity value between the target design vector and the material and process node vectors, and generating solutions containing the target materials and processes, this approach leverages the associative reasoning capabilities of graph neural networks to efficiently match the optimal material and process combination, while also providing multiple solution options. This addresses the issues of low efficiency in matching materials and processes and the limited number of solutions generated. The problem is that by calculating the rating loss value, the model's fit to user preferences can be quantified, while the divergence value can constrain the consistency between the generated solutions and the company's long-term style and basic requirements. By combining these two factors, the loss value can be calculated, ensuring that the design meets the user's real-time needs while avoiding deviation from the company's established strategy. By pre-setting weights, the rating loss value and divergence value are weighted to obtain the target loss value, and the GNN model parameters are adjusted accordingly to keep the loss value below the threshold. This enables continuous iterative optimization of the model, so that subsequent design solutions can continuously improve their accuracy in meeting user needs, matching corporate style, and adapting to materials and processes, thereby improving the efficiency, accuracy, and quality of packaging design. Attached Figure Description

[0047] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0048] Figure 1 A flowchart illustrating a packaging design method based on a target GNN model provided in this application embodiment;

[0049] Figure 2 A flowchart illustrating another packaging design method based on a target GNN model provided in this application embodiment;

[0050] Figure 3 A schematic diagram of the structure of a packaging design device based on a target GNN model provided in an embodiment of this application;

[0051] Figure 4 This is a structural block diagram of an apparatus for performing a packaging design method based on a target GNN model according to an embodiment of this application. Detailed Implementation

[0052] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0053] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0054] Figure 1 This is a flowchart illustrating a packaging design method based on a target GNN model, provided as an embodiment of this application. Figure 1 As shown, this packaging design method based on a target GNN model may include the following steps:

[0055] S110. Perform feature recognition on the image to be designed to obtain each image feature and the image vector corresponding to the image feature, and determine the target design vector based on each image vector and its corresponding target text vector.

[0056] Among them, the image to be designed refers to the original image material that needs to be used for the current packaging printing design. It is usually a two-dimensional design drawing, such as a flat rendering of the product packaging, a design draft containing background patterns, logos, or decorative elements, etc., so as to provide the visual basis information for packaging design, so that the corresponding packaging design can be carried out according to the image to be designed, thereby meeting the user's real-time design needs.

[0057] Image features refer to visual regions with specific semantics that are identified and segmented from an image to be designed. This can be achieved by using the YOLO detection model to transform the original image into quantifiable visual elements, so that image information can be converted into features that can be calculated for matching. For example, the background texture type can be identified as specific features in the image, such as a red matte gradient or a gold frosted finish.

[0058] The target text vector refers to the high-dimensional vector that transforms the user's textual requirements for packaging design into a vector, thereby realizing the vectorized expression of the text so that subsequent calculations can be performed based on the vector.

[0059] The target design vector is a comprehensive vector obtained by fusing the image vectors corresponding to various image features with the corresponding target text vectors. It contains both the visual information of the image and the semantic requirements of the text. It can be directly input into the subsequent GNN model to calculate the similarity with the material node vector and the process node vector, thereby ensuring the consistency of the final design scheme in terms of both vision and function.

[0060] Based on this, by determining the image vector corresponding to the image to be designed and the text vector corresponding to the design text, the image features and text features are combined through the fusion of image vectors and text vectors. This ensures that the final generated target solution can take into account both image and text requirements, thereby meeting the user's design needs and improving the quality of the design.

[0061] S120. Based on the target GNN model, calculate the material similarity value between the target design vector and each material node vector, and the process similarity value between the target design vector and each process node vector. Based on the material similarity value and the process similarity value, determine multiple target design schemes. The target design scheme includes the target material and the target process.

[0062] In this context, the target GNN model refers to a GNN model that has undergone specific pre-training, enabling it to map the input target vector to an embedding space and calculate the similarity between this vector and all node vectors in the knowledge base; the GNN model refers to a Graph Neural Network model, an algorithmic framework based on deep learning for processing graph-structured data.

[0063] Material node vectors refer to the high-dimensional vector representations of various packaging materials related to design. By encoding the properties of materials, such as material, thickness, color, scratch resistance, cost, etc., the generated vectors are used to quantify the characteristics of materials, thereby transforming the properties of physical materials into a vector form that can be computed by computers.

[0064] Material similarity value refers to the similarity score between the target design vector and a certain material node vector. It is usually a value between 0 and 1 and can be calculated by algorithms such as cosine similarity and Euclidean distance. The higher the value, the higher the matching degree between the material and the design requirements.

[0065] Process node vectors refer to the high-dimensional vector representations of various printing processes in the process library. They are generated by encoding process attributes such as the desired effect, applicable materials, cost, and precision. They are used to quantify the characteristics of the process so as to transform the abstract process effect into a computable vector.

[0066] The process similarity value refers to the similarity score between the target design vector and a certain process node vector. It is usually a value between 0 and 1 and can be calculated by algorithms such as cosine similarity and Euclidean distance. The higher the value, the higher the matching degree between the process and the design requirements.

[0067] The target design scheme refers to the combination of materials and processes selected based on material similarity and process similarity, such as 275g photoresist paper, screen printing crystal oil, and embossing. It includes specific material selection, process type, and parameters, thereby transforming abstract design requirements into a feasible production solution.

[0068] Target materials refer to the specific packaging materials included in the target design scheme, and target processes refer to the specific packaging processes included in the target design scheme.

[0069] Based on this, a target GNN model that has been pre-trained is used to calculate the similarity value between the target design vector and multiple pre-determined material node vectors and process node vectors, so as to determine the target material and target process that best match the target design vector according to the magnitude of the similarity value.

[0070] S130. Calculate the score loss value based on the user score and predicted score corresponding to each target design scheme, and calculate the divergence value based on the preset materials, preset processes, multiple historical design schemes and target design schemes; the predicted score value is the value determined based on the target GNN model.

[0071] Among them, the user rating score refers to the actual evaluation score of the generated target design scheme by users, reflecting the user's satisfaction with the scheme, providing an intuitive basis for subsequent adjustment of model preferences, avoiding the generation of models from deviating from user needs, and thus improving design quality.

[0072] The predicted score refers to the score that the target GNN model predicts in advance for the target design scheme based on the relationship between the design scheme and user preferences that it has learned. It is used to measure the accuracy of the model's understanding of user preferences, thereby enabling the model to predict the popularity of the scheme among users.

[0073] Rating loss value refers to the loss metric calculated by quantifying the difference between the predicted rating value and the user's rating value. The commonly used calculation method is mean squared error.

[0074] Pre-set materials refer to packaging materials that are pre-defined in the technical solution and meet the company's basic needs or hard constraints. These are usually standard materials provided by the company's long-term cooperative suppliers. Pre-set processes are packaging processes that are pre-defined in the technical solution and meet the company's basic needs or hard constraints.

[0075] Historical design schemes refer to a collection of packaging design schemes that have been completed previously or have received positive user feedback.

[0076] The divergence value is an index calculated by quantifying the difference between the distribution of materials and processes in the target design scheme and the distribution of materials and processes in the historical design scheme. This prevents the model from deviating from its long-term style in order to cater to a single score. In other words, it calculates the divergence value by calculating the schemes containing preset materials and processes in the historical design scheme and the target design scheme.

[0077] Based on this, after generating the target design scheme, the loss value is calculated by determining the different rating values ​​of the scheme between the user and the model, and the frequency of occurrence of materials and processes in the historical schemes and the current target scheme is calculated to calculate the divergence value. This allows for further adjustment of the parameters of the target GNN model based on the loss value and the divergence value, thereby improving the quality of the subsequently generated schemes.

[0078] S140. According to the preset weights, the scoring loss value and the divergence value are weighted and calculated to obtain the target loss value. Based on the target loss value and the preset loss threshold, the parameters of the target GNN model are adjusted to obtain the adjusted target GNN model, so that the target loss value calculated based on the adjusted target GNN model is less than the preset loss threshold.

[0079] The preset weights are pre-determined weights corresponding to the rating loss value and the divergence value, which are used to balance the influence of the rating loss value and the divergence value on the total loss.

[0080] The target loss value is the overall loss value obtained by multiplying the rating loss value and its corresponding weight value, and multiplying the divergence value and its corresponding weight value, and then adding the two multipliers together. It is a comprehensive quantitative evaluation of the model's prediction bias and style deviation. The smaller the value, the better the model's overall performance in both meeting user needs and corporate constraints.

[0081] The preset loss threshold refers to the pre-set critical value for judging whether the target GNN model has reached the optimization target. When the target loss value is less than the threshold, it means that the overall performance of the model meets the target; if it is greater than the threshold, the parameters need to be adjusted.

[0082] Based on this, the scoring loss value can characterize the different scoring values ​​of the target design scheme between the user and the model, and the divergence value can characterize the distribution rate of the preset materials and preset processes in the historical design scheme and the target design scheme, respectively. The target design scheme itself is determined by the target GNN. Therefore, by weighting the scoring loss value and the divergence value, the target loss value is obtained. Based on the target loss value, the model parameters of the target GNN model are adjusted so that the adjusted GNN model can better meet the actual needs of the user.

[0083] Based on the feasible implementation of S110 described above, this application further provides a method for determining a target design vector based on each image vector and its corresponding target text vector, including:

[0084] The text to be designed is semantically segmented to obtain multiple sub-texts, and the vector corresponding to each sub-text is determined to obtain multiple text vectors;

[0085] The contrastive learning mechanism based on the CLIP model determines the target text vector corresponding to each image vector in the text vector, and associates the image vector and the target text vector to obtain the target vector.

[0086] Based on the attention mechanism of the CLIP model, vector fusion is performed on all target vectors and text vectors to obtain the target design vector.

[0087] The text to be designed refers to the user's textual description requirements for packaging printing design, including key information such as function, style, and scenario.

[0088] CLIP (Contrastive Language-Image Pretraining) is a cross-modal pretraining model that is trained on large-scale image and text data to achieve semantic association between images and text. The model can understand the correspondence between the visual content of an image and the semantic description of the text, such as matching the text about a red gift box with an image of a red gift box.

[0089] A target vector refers to a single image vector and its matching target text vector, which together form a composite vector. For example, when an image vector with a red background is associated with a target text vector for the theme of wedding, a fused vector containing the visual red color and the semantic meaning of wedding is formed.

[0090] Based on this, after determining the image vectors corresponding to the image features, the text can be segmented according to the semantics of the text to be designed, and each sub-text can be expressed as a vector to obtain the corresponding text vector. Furthermore, through the CLIP model, cross-modal semantic association between text and image can be achieved through contrastive learning, thereby processing both image and text simultaneously and fusing them to obtain the target design vector.

[0091] Based on the feasible implementation of S120 described above, this application further provides a method that, before calculating the material similarity value between the target design vector and each material node vector, and the process similarity value with each process node vector according to the target GNN model, includes:

[0092] Based on the CLIP model, sample images, and corresponding sample text, the corresponding sample design vector is determined. Then, based on the preset design database, the positive and negative sample schemes corresponding to the sample design vector are determined. The positive sample scheme is the sample scheme that is associated with the sample design vector, and the negative sample scheme is the sample scheme that is not associated with the sample design vector.

[0093] Determine the positive sample vector corresponding to the positive sample scheme and the negative sample vector corresponding to the negative sample scheme, and calculate the positive similarity value between the sample design vector and the positive sample vector and the negative similarity value between the sample design vector and the negative sample vector based on the initial GNN model.

[0094] Based on the difference between negative and positive similarity values, the corresponding loss value is determined. Then, based on the loss value and a preset threshold, the model parameters of the initial GNN model are adjusted to obtain the target GNN model.

[0095] Among them, sample images refer to historical packaging design images used to train the model; sample text refers to the text descriptions corresponding to the sample images; and sample design vectors refer to the composite vectors obtained by fusing the image vectors of the sample images and the text vectors of the sample texts through the CLIP model.

[0096] The pre-built design database refers to a pre-constructed database containing a large number of historical packaging design schemes. Each scheme records information such as materials, processes, and corresponding design requirements.

[0097] A positive sample solution refers to a historical design solution in a pre-defined design database that is strongly correlated with the sample design vector. For example, if the sample design vector is "wedding, red", the positive sample solution is "wedding gift box with red background and hot stamping", which is a valid solution that meets the sample design requirements.

[0098] Negative sample schemes refer to historical design schemes in the pre-set design database that are unrelated or weakly related to the sample design vector. For example, if the sample design vector is "wedding, red", the negative sample scheme is "blue background + ordinary offset printing industrial packaging scheme", which is an invalid scheme that does not meet the sample design requirements.

[0099] The preset threshold refers to the critical value set during model training to determine whether the difference between positive and negative similarity values ​​meets the standard.

[0100] Based on this, to achieve specific pre-training of the GNN model, we can determine sample data, i.e., sample design vectors, and identify positive sample schemes that are related to the sample design vectors, as well as negative sample schemes that are not related. Positive sample schemes represent schemes corresponding to the current design requirements, while negative sample schemes are schemes that do not correspond to the design requirements. By determining the corresponding positive and negative sample vectors, and based on the initial GNN model, we can calculate the positive similarity value between the sample design vector and the positive sample vector, and the negative similarity value between the sample design vector and the negative sample vector. Correspondingly, the positive similarity value can be understood as meeting the user's requirements. The similarity value is calculated, and the negative similarity value can be understood as the similarity value that does not meet the user's needs. In practical applications, in order to improve the computational accuracy of the GNN model and thus improve the quality of packaging design, the more similar the output of the GNN model is to the user's needs, the better. Therefore, by calculating the difference between the negative and positive similarity values, the corresponding loss value is obtained. The loss value reflects the difference between the sample vector generated by the GNN model and the positive and negative solutions determined by the user. By adjusting the loss value to be higher than the preset threshold, the GNN model is optimized and the target GNN model is obtained.

[0101] Based on the feasible implementation of S120 described above, this application further provides a method for adjusting the model parameters of an initial GNN model according to a loss value and a preset threshold to obtain a target GNN model, including:

[0102] By comparing the loss value with the preset threshold, the corresponding comparison results are obtained;

[0103] If the comparison result shows that the loss value is less than the preset threshold, the model parameters of the initial GNN model are adjusted according to the loss value to obtain the target GNN model, so that the loss value calculated based on the target GNN model is not less than the preset threshold.

[0104] If the comparison result shows that the loss value is not less than the preset threshold, then the initial GNN model is determined to be the target GNN model.

[0105] Based on this, after obtaining the loss value, since the loss value can represent the difference between positive and negative similarity values, that is, the similarity difference between schemes generated by the GNN model, a preset threshold is determined, and the loss value and the preset threshold are compared, so as to determine the target GNN model that meets the actual needs based on different comparison results.

[0106] Based on the feasible implementation of S120 described above, this application further provides a method for determining multiple target design schemes based on material similarity values ​​and process similarity values, including:

[0107] According to the preset correspondence rules, at least one process node corresponding to each material node is determined, and each material node and its corresponding process nodes are determined as target design nodes; a material node is the node corresponding to the material node vector, and a process node is the node corresponding to the process node vector.

[0108] The target material similarity value and target process similarity value corresponding to each target design node are added together to obtain multiple design similarity values. The design similarity values ​​are then sorted from largest to smallest to obtain a design similarity value sequence. The target material similarity value is the material similarity value corresponding to the material node in the target design node, and the target process similarity value is the process similarity value corresponding to the process node in the target design node.

[0109] A preset number of design similarity values ​​are determined sequentially in the design similarity value sequence to obtain multiple target design similarity values. Based on the target material node and target process node corresponding to each target design similarity value, the corresponding target material and target process are determined.

[0110] Based on the target materials and target processes, multiple corresponding target design schemes are determined.

[0111] Among them, the preset corresponding rules refer to the pre-set rules that constrain the relationship between material nodes and process nodes. These rules are used to screen technically feasible and effective combinations of materials and processes, thereby eliminating invalid combinations.

[0112] A target design node refers to a combination of material nodes and process nodes that are associated through preset corresponding rules. It is the smallest unit representing the matching relationship between materials and processes. For example, the material node of 275g photoresist paper and the process node of screen printing crystal oil are combined to form the corresponding target design node.

[0113] The target material similarity value refers to the similarity value between the material node in the target design node and the target design vector; the target process similarity value refers to the similarity value between the process node in the target design node and the target design vector.

[0114] The preset quantity refers to the number of target design schemes that are pre-set and ultimately required to be output.

[0115] The target material node refers to the material node in the target design node that corresponds to the target design similarity value; the target process node refers to the process node in the target design node that corresponds to the target design similarity value.

[0116] Target material refers to the actual packaging material corresponding to the target material node; target process refers to the actual packaging process corresponding to the target process node.

[0117] Based on this, in practical applications, the materials and processes in packaging design need to meet actual correspondence rules, that is, the selected materials and processes must be compatible. Therefore, by first determining the processes corresponding to each material, materials and processes that meet the correspondence are selected to obtain target design nodes. Since the similarity value represents the similarity between the target design vector and each material node vector, as well as each process node vector, and since the material node vector and process node correspond to the material node and process node respectively, the design similarity value corresponding to the target design node can be calculated by adding the material nodes and process nodes contained in each target design node. The design similarity value represents the similarity between the materials and processes and the user's actual design requirements (i.e., the target design vector). Therefore, based on the values ​​from large to small, materials and processes that are highly similar to the target design vector are determined, resulting in multiple target design schemes.

[0118] Based on the feasible implementation of S130 described above, this application further provides a method for calculating a rating loss value based on the user rating value and predicted rating value corresponding to each target design scheme, including:

[0119] Identify and output multiple target design schemes;

[0120] Based on the output feedback results corresponding to the target design schemes, determine the user rating values ​​corresponding to each target design scheme, and based on the target GNN model, determine the predicted rating values ​​corresponding to each target design scheme.

[0121] Based on the user rating and the corresponding predicted rating, calculate the error value for each target design scheme, and calculate the average value of all error values ​​to obtain the rating loss value.

[0122] The output feedback results refer to the subjective evaluations given by users of the target design solution after the solution is output, based on dimensions such as visual effect, functional adaptability, production feasibility, and cost acceptability. For example, it can include user rating values, such as quantitative scores of 0-10 points, and can also cover qualitative feedback, such as written suggestions and solution selection results.

[0123] Based on this, by outputting multiple target design schemes, user ratings for each scheme are determined according to user feedback, thus identifying the user's design preferences for each scheme. Furthermore, based on the target GNN model, predicted ratings for each scheme are generated. The error between the user ratings and the predicted ratings determines whether the target GNN model's predictions meet actual requirements. The average of all error values ​​is calculated to obtain the rating loss. In practical applications, the loss can be calculated using the Mean Squared Error (MSE) function. MSE is the most commonly used loss function in regression problems, used to measure the difference between the model's predicted values ​​and the true values; that is, the model's prediction accuracy is evaluated by calculating the average of the squared prediction errors.

[0124] Based on the feasible implementation of S130 described above, this application further provides a method for calculating divergence values ​​based on preset materials, preset processes, multiple historical design schemes, and a target design scheme, including:

[0125] Based on the pre-set design database, pre-set materials, pre-set processes, and multiple historical design schemes are determined. Based on the pre-set materials and pre-set processes, target material schemes and target process schemes are determined in the historical design schemes and target design schemes, respectively. The target material scheme is a scheme that includes the pre-set materials, and the target process scheme is a scheme that includes the pre-set processes.

[0126] Calculate the material divergence value based on the target material quantity corresponding to the target design scheme and the historical design scheme, and calculate the process divergence value based on the target process quantity corresponding to the target design scheme and the historical design scheme; the target material quantity is the quantity corresponding to the target material scheme, and the target process quantity is the quantity corresponding to the target process scheme.

[0127] Based on the preset weight values ​​corresponding to the target material quantity and the target process quantity, the target material quantity and the target process quantity are weighted and summed to obtain the divergence value.

[0128] Material divergence refers to the divergence index calculated by quantifying the number of schemes containing preset materials in the target design scheme (i.e., the number of target materials) and the distribution difference between these two schemes and the number of schemes containing preset materials in historical design schemes (i.e., the number of historical target materials). Common methods for calculation include KL divergence and cosine distance. For example, if there are 100 historical design schemes, and 80 of them contain the preset material 275g photolithography paper, then the historical target material accounts for 80%. If there are 5 target design schemes, and 2 of them contain the preset material, then the target material accounts for 40%. The material divergence value is used to measure the degree of deviation between the 40% and 80% distributions. The smaller the value, the closer the two distributions are, indicating that the target scheme is more in line with historical material preferences.

[0129] The process divergence value is a divergence index calculated by quantifying the number of schemes containing preset processes in the target design scheme, i.e., the number of target processes, and the distribution difference between the number of schemes containing preset processes in historical design schemes, i.e., the number of historical target processes.

[0130] Therefore, in practical applications, to prevent the model from deviating from its long-term style to cater to a single score, such as a company specializing in wedding products whose packaging should maintain its preferred gold foil tradition, a preset database can be established to determine preset materials and processes. Based on multiple randomly selected historical design schemes and multiple target design schemes that include preset materials or processes, material divergence values ​​and process divergence values ​​can be calculated. Since the target design schemes are generated based on the target GNN model, the model parameters can be adjusted by adjusting the divergence value obtained after weighted summation of the material divergence value and process divergence value.

[0131] Please refer to Figure 2 , Figure 2 A flowchart illustrating another packaging design method based on a target GNN model provided in this application embodiment; as shown Figure 2 As shown, by inputting a two-dimensional design drawing and requirement text, the region segmentation unit under the multimodal analysis module uses YOLO to segment elements such as background, logo, and decoration. The image-text alignment unit generates a semantic vector Vd using the CLIP model, and the knowledge matching unit calculates similarity using GNN. Then, the solution generation module enters the solution generation module. The material and process combination unit outputs the Top-K solutions, the cost calculation unit calls local knowledge values, and then uses the Blender physics engine to perform 3D effect rendering, thereby outputting a three-dimensional rendering and cost list. User rating feedback is collected, and the GNN weights are updated based on the feedback through the closed-loop learning unit.

[0132] In some embodiments of this application, feature recognition is performed on the image to be designed to obtain various image features and corresponding image vectors. Based on each image vector and its corresponding target text vector, a target design vector is determined. According to the target GNN model, the material similarity value between the target design vector and each material node vector, and the process similarity value between the target design vector and each process node vector, are calculated. Based on the material similarity value and the process similarity value, multiple target design schemes are determined. Each target design scheme includes a target material and a target process. Based on the user rating value and predicted rating value corresponding to each target design scheme, a rating loss value is calculated. Based on preset materials, preset processes, multiple historical design schemes, and the target design scheme, a divergence value is calculated. The predicted rating value is a value determined according to the target GNN model. Based on preset weights, the rating loss value and the divergence value are weighted to obtain a target loss value. Based on the target loss value and a preset loss threshold, the parameters of the target GNN model are adjusted to obtain an adjusted target GNN model, so that the target loss value calculated based on the adjusted target GNN model is less than the preset loss threshold.

[0133] Thus, by performing feature recognition on the design image and generating image vectors, the target design vector is determined by combining it with the target text vector. This enables accurate identification of image features based on the image, and further feature fusion of image features and corresponding text features facilitates subsequent matching between materials and processes, while avoiding a disconnect between the actual design and user needs. Furthermore, by using a GNN model to calculate the similarity value between the target design vector and the material and process node vectors, and generating solutions containing the target materials and processes, this approach leverages the associative reasoning capabilities of graph neural networks to efficiently match the optimal material and process combination, while also providing multiple solution options. This addresses the issues of low efficiency in matching materials and processes and the limited number of solutions generated. The problem is that by calculating the rating loss value, the model's fit to user preferences can be quantified, while the divergence value can constrain the consistency between the generated solutions and the company's long-term style and basic requirements. By combining these two factors, the loss value can be calculated, ensuring that the design meets the user's real-time needs while avoiding deviation from the company's established strategy. By pre-setting weights, the rating loss value and divergence value are weighted to obtain the target loss value, and the GNN model parameters are adjusted accordingly to keep the loss value below the threshold. This enables continuous iterative optimization of the model, so that subsequent design solutions can continuously improve their accuracy in meeting user needs, matching corporate style, and adapting to materials and processes, thereby improving the efficiency, accuracy, and quality of packaging design.

[0134] Figure 3 This is a schematic diagram of a packaging design device 300 based on a target GNN model, provided as an embodiment of this application. Figure 3As shown, the packaging design device 300 based on a target GNN model includes: an identification module 310, a determination module 320, a calculation module 330, and an adjustment module 340; wherein:

[0135] The recognition module 310 is used to perform feature recognition on the design image to obtain each image feature and the image vector corresponding to the image feature, and to determine the target design vector based on each image vector and its corresponding target text vector.

[0136] The determination module 320 is used to calculate the material similarity value between the target design vector and each material node vector, and the process similarity value between the target design vector and each process node vector, based on the target GNN model, and to determine multiple target design schemes based on the material similarity value and the process similarity value; the target design scheme includes target material and target process;

[0137] The calculation module 330 is used to calculate the score loss value based on the user score value and predicted score value corresponding to each target design scheme, and to calculate the divergence value based on the preset materials, preset processes, multiple historical design schemes and target design schemes; the predicted score value is a value determined based on the target GNN model.

[0138] The adjustment module 340 is used to perform weighted calculation of the scoring loss value and divergence value according to the preset weights to obtain the target loss value, and adjust the parameters of the target GNN model according to the target loss value and the preset loss threshold to obtain the adjusted target GNN model, so that the target loss value calculated according to the adjusted target GNN model is less than the preset loss threshold.

[0139] In this embodiment of the application, the identification module 310 can also be specifically used for:

[0140] The text to be designed is semantically segmented to obtain multiple sub-texts, and the vector corresponding to each sub-text is determined to obtain multiple text vectors;

[0141] The contrastive learning mechanism based on the CLIP model determines the target text vector corresponding to each image vector in the text vector, and associates the image vector and the target text vector to obtain the target vector.

[0142] Based on the attention mechanism of the CLIP model, vector fusion is performed on all target vectors and text vectors to obtain the target design vector.

[0143] In this embodiment of the application, the determining module 320 may also be specifically used for:

[0144] Based on the CLIP model, sample images, and corresponding sample text, the corresponding sample design vector is determined. Then, based on the preset design database, the positive and negative sample schemes corresponding to the sample design vector are determined. The positive sample scheme is the sample scheme that is associated with the sample design vector, and the negative sample scheme is the sample scheme that is not associated with the sample design vector.

[0145] Determine the positive sample vector corresponding to the positive sample scheme and the negative sample vector corresponding to the negative sample scheme, and calculate the positive similarity value between the sample design vector and the positive sample vector and the negative similarity value between the sample design vector and the negative sample vector based on the initial GNN model.

[0146] Based on the difference between negative and positive similarity values, the corresponding loss value is determined. Then, based on the loss value and a preset threshold, the model parameters of the initial GNN model are adjusted to obtain the target GNN model.

[0147] In this embodiment of the application, the determining module 320 may also be specifically used for:

[0148] By comparing the loss value with the preset threshold, the corresponding comparison results are obtained;

[0149] If the comparison result shows that the loss value is less than the preset threshold, the model parameters of the initial GNN model are adjusted according to the loss value to obtain the target GNN model, so that the loss value calculated based on the target GNN model is not less than the preset threshold.

[0150] If the comparison result shows that the loss value is not less than the preset threshold, then the initial GNN model is determined to be the target GNN model.

[0151] In this embodiment of the application, the determining module 320 may also be specifically used for:

[0152] According to the preset correspondence rules, at least one process node corresponding to each material node is determined, and each material node and its corresponding process nodes are determined as target design nodes; a material node is the node corresponding to the material node vector, and a process node is the node corresponding to the process node vector.

[0153] The target material similarity value and target process similarity value corresponding to each target design node are added together to obtain multiple design similarity values. The design similarity values ​​are then sorted from largest to smallest to obtain a design similarity value sequence. The target material similarity value is the material similarity value corresponding to the material node in the target design node, and the target process similarity value is the process similarity value corresponding to the process node in the target design node.

[0154] A preset number of design similarity values ​​are determined sequentially in the design similarity value sequence to obtain multiple target design similarity values. Based on the target material node and target process node corresponding to each target design similarity value, the corresponding target material and target process are determined.

[0155] Based on the target materials and target processes, multiple corresponding target design schemes are determined.

[0156] In this embodiment of the application, the calculation module 330 can also be specifically used for:

[0157] Identify and output multiple target design schemes;

[0158] Based on the output feedback results corresponding to the target design schemes, determine the user rating values ​​corresponding to each target design scheme, and based on the target GNN model, determine the predicted rating values ​​corresponding to each target design scheme.

[0159] Based on the user rating and the corresponding predicted rating, calculate the error value for each target design scheme, and calculate the average value of all error values ​​to obtain the rating loss value.

[0160] In this embodiment of the application, the calculation module 330 can also be specifically used for:

[0161] Based on the pre-set design database, pre-set materials, pre-set processes, and multiple historical design schemes are determined. Based on the pre-set materials and pre-set processes, target material schemes and target process schemes are determined in the historical design schemes and target design schemes, respectively. The target material scheme is a scheme that includes the pre-set materials, and the target process scheme is a scheme that includes the pre-set processes.

[0162] Calculate the material divergence value based on the target material quantity corresponding to the target design scheme and the historical design scheme, and calculate the process divergence value based on the target process quantity corresponding to the target design scheme and the historical design scheme; the target material quantity is the quantity corresponding to the target material scheme, and the target process quantity is the quantity corresponding to the target process scheme.

[0163] Based on the preset weight values ​​corresponding to the target material quantity and the target process quantity, the target material quantity and the target process quantity are weighted and summed to obtain the divergence value.

[0164] Figure 4 This is a schematic diagram of the structure of an apparatus for performing a packaging design method based on a target GNN model according to an embodiment of this application. Figure 4 As shown, the device 400 includes:

[0165] The device 400 may include a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, a communication component 403, and other components. The processor 401, memory 402, and communication component 403 are connected via a bus 404.

[0166] In the specific implementation process, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to execute the above-described packaging design method based on a target GNN model.

[0167] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0168] Furthermore, the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0169] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0170] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0171] In some embodiments, a computer program product is also provided, comprising a computer program or instructions that, when executed by a processor, implement the steps in any of the above-described packaging design methods based on a target GNN model.

[0172] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0173] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0174] Therefore, embodiments of this application provide a computer-readable storage medium storing a plurality of program codes that can be loaded by a processor to execute steps in any of the packaging design methods based on a target GNN model provided in embodiments of this application.

[0175] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0176] According to one aspect of this application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium.

[0177] Since the instructions stored in the storage medium can execute the steps in any of the packaging design methods based on the target GNN model provided in the embodiments of this application, the beneficial effects that any of the packaging design methods based on the target GNN model provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.

[0178] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the appended claims.

[0179] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A packaging design method based on a target GNN model, characterized by, The method comprises: performing feature recognition on the to-be-designed image to obtain each image feature and an image vector corresponding to the image feature, and determining a target design vector according to each image vector and a target text vector corresponding thereto; calculating, according to a target GNN model, a material similarity value of the target design vector and each material node vector and a process similarity value of the target design vector and each process node vector, and determining a plurality of target design schemes according to the material similarity value and the process similarity value; the target design scheme comprises a target material and a target process; wherein the determining a plurality of target design schemes according to the material similarity value and the process similarity value comprises: determining at least one process node corresponding to each material node according to a preset correspondence rule, and determining each material node and each process node corresponding thereto as a target design node; the material node is a node corresponding to the material node vector, and the process node is a node corresponding to the process node vector; adding a target material similarity value corresponding to each target design node and a target process similarity value to obtain a plurality of design similarity values, and sorting the design similarity values from large to small to obtain a design similarity value sequence; the target material similarity value is the material similarity value corresponding to the material node in the target design node, and the target process similarity value is the process similarity value corresponding to the process node in the target design node; determining a plurality of target design similarity values in sequence from a preset number of design similarity values in the design similarity value sequence, and determining a corresponding target material and a target process according to a target material node and a target process node corresponding to each target design similarity value; determining a plurality of corresponding target design schemes according to each target material and the target process; calculating a score loss value according to a user score value and a predicted score value corresponding to each target design scheme, and calculating a divergence value according to a preset material, a preset process, a plurality of historical design schemes and the target design scheme; the predicted score value is a value determined according to the target GNN model; weighting the score loss value and the divergence value according to a preset weight to obtain a target loss value, and adjusting parameters of the target GNN model according to the target loss value and a preset loss threshold to obtain an adjusted target GNN model, so that the target loss value calculated according to the adjusted target GNN model is less than the preset loss threshold.

2. The method of claim 1, wherein, The determining a target design vector according to each image vector and a target text vector corresponding thereto comprises: performing semantic division on the to-be-designed text to obtain a plurality of subtexts, and determining a vector corresponding to each subtext to obtain a plurality of text vectors; determining a target text vector corresponding to each image vector in the text vector based on a contrast learning mechanism of a CLIP model, and associating the image vector and the target text vector to obtain a target vector; According to the attention mechanism of the CLIP model, vector fusion is performed on all the target vectors and the text vector to obtain the target design vector.

3. The method of claim 1, wherein, Before the step of calculating the material similarity value of the target design vector and each material node vector and the process similarity value of the target design vector and each process node vector according to the target GNN model, the method further comprises: According to the CLIP model, a sample image and corresponding sample text, a corresponding sample design vector is determined, and according to a preset design database, a positive sample scheme corresponding to the sample design vector and a negative sample scheme corresponding to the sample design vector are determined; the positive sample scheme is a sample scheme associated with the sample design vector, and the negative sample scheme is a sample scheme not associated with the sample design vector; A positive sample vector corresponding to the positive sample scheme and a negative sample vector corresponding to the negative sample scheme are determined, and according to an initial GNN model, a positive similarity value of the sample design vector and the positive sample vector and a negative similarity value of the sample design vector and the negative sample vector are calculated; According to the difference between the negative similarity value and the positive similarity value, a corresponding loss value is determined, and according to the loss value and a preset threshold, the model parameters of the initial GNN model are adjusted to obtain the target GNN model.

4. The method of claim 3, wherein, The step of adjusting the model parameters of the initial GNN model according to the loss value and the preset threshold to obtain the target GNN model comprises: Comparing the loss value and the preset threshold to obtain a corresponding comparison result; If the comparison result is that the loss value is less than the preset threshold, the model parameters of the initial GNN model are adjusted according to the loss value to obtain the target GNN model, so that the loss value calculated according to the target GNN model is not less than the preset threshold; If the comparison result is that the loss value is not less than the preset threshold, the initial GNN model is determined as the target GNN model.

5. The method of claim 1, wherein, The step of calculating a score loss value according to the user score value and the predicted score value corresponding to each target design scheme comprises: A plurality of target design schemes are determined and outputted; According to the output feedback result corresponding to the target design scheme, the user score value corresponding to each target design scheme is determined, and according to the target GNN model, the predicted score value corresponding to each target design scheme is determined; According to the user score value and the corresponding predicted score value, an error value corresponding to each target design scheme is calculated, and an average value corresponding to all error values is calculated to obtain the score loss value.

6. The method of claim 1, wherein, The step of calculating a divergence value according to a preset material, a preset process, a plurality of historical design schemes and the target design scheme comprises: According to a preset design database, the preset material, the preset process and a plurality of the historical design schemes are determined, and according to the preset material and the preset process, a target material scheme and a target process scheme in the historical design scheme and the target design scheme are respectively determined; the target material scheme is a scheme containing the preset material, and the target process scheme is a scheme containing the preset process; According to the target material quantity corresponding to the target design scheme and the historical design scheme respectively, a material divergence value is calculated, and according to the target process quantity corresponding to the target design scheme and the historical design scheme respectively, a process divergence value is calculated; the target material quantity is the quantity corresponding to the target material scheme, and the target process quantity is the quantity corresponding to the target process scheme; According to the preset weight value corresponding to the target material quantity and the target process quantity respectively, the target material quantity and the target process quantity are weighted and summed to obtain the divergence value.

7. A packaging design device based on a target GNN model, characterized by, The device comprises: The recognition module is configured to perform feature recognition on the to-be-designed image to obtain each image feature and an image vector corresponding to the image feature, and determine a target design vector according to each image vector and a target text vector corresponding thereto; The determination module is configured to calculate a material similarity value of the target design vector and each material node vector and a process similarity value of the target design vector and each process node vector according to a target GNN model, and determine a plurality of target design schemes according to the material similarity value and the process similarity value; the target design scheme comprises a target material and a target process; wherein, the determination of the plurality of target design schemes according to the material similarity value and the process similarity value comprises: According to a preset corresponding rule, at least one process node corresponding to each material node is determined, and each material node and each process node corresponding thereto is determined as a target design node; the material node is a node corresponding to the material node vector, and the process node is a node corresponding to the process node vector; The target material similarity value corresponding to each target design node and the target process similarity value are added to obtain a plurality of design similarity values, and the design similarity values are sorted from large to small to obtain a design similarity value sequence according to the numerical value; the target material similarity value is the material similarity value corresponding to the material node in the target design node, and the target process similarity value is the process similarity value corresponding to the process node in the target design node; A plurality of target design similarity values are obtained by sequentially determining a preset number of design similarity values in the design similarity value sequence, and a target material and a target process corresponding to each target design similarity value are determined; According to each target material and the target process, a plurality of corresponding target design schemes are determined. The computing module is configured to calculate a score loss value according to a user score value and a predicted score value corresponding to each target design scheme, and calculate a divergence value according to a preset material, a preset process, a plurality of historical design schemes and the target design scheme; the predicted score value is a numerical value determined according to the target GNN model; The adjusting module is configured to perform weighted calculation on the score loss value and the divergence value according to a preset weight to obtain a target loss value, and adjust parameters of the target GNN model according to the target loss value and a preset loss threshold to obtain an adjusted target GNN model, so that the target loss value calculated according to the adjusted target GNN model is less than the preset loss threshold.

8. A computer device, comprising: Comprise: One or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to perform the method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores program code, and the program code can be called and executed by the processor to perform the method of any one of claims 1 to 6.

Images