Product form design knowledge pushing method based on crowd intelligence situation-trust relationship

Abstract

A product form design knowledge pushing method based on a crowd-sourcing situation-trust relationship comprises the following steps: constructing a multi-dimensional situation ontology and a product form design knowledge resource ontology by means of a structured ontology language, obtaining a nearest neighbor set of target designers by using situation similarity and a trust relationship, performing score completion according to adjacent persons, and pushing the product form design knowledge resource ontology to the target designers. Predicting a knowledge preference score of the target designer, and generating a knowledge pushing list; according to the method, by mining the multi-dimensional situation information generated by crowd intelligence interaction in the cloud environment and the cooperative trust relationship characteristics, a new methodological support is provided for accurate matching of knowledge services in the cloud environment, and the recommendation accuracy and reliability are improved.

Description

Technical Field

[0001] This invention belongs to the field of product form design knowledge push technology, specifically involving a product form design knowledge push method based on collective intelligence context-trust relationship. Background Technology

[0002] The paper "A Knowledge Recommendation Method for Topic Diversity in the Engineering Field," 2021, Vol. 27, pp. 214-227, discloses a knowledge recommendation method for topic diversity. This method addresses the problem that existing recommendation methods cannot simultaneously satisfy both accuracy and diversity in knowledge recommendation. It proposes a knowledge relevance network construction method that considers three similarities: content, context, and task, and then divides knowledge topic communities. Based on these communities, a user interest model is constructed to mine diverse user knowledge needs. Combining user group behavior data, the concept of user-topic expertise and its calculation method are proposed. User-topic expertise information is used to improve user-based collaborative filtering methods, and a post-filtering diversity strategy is proposed by combining contextual information and the user interest model to simultaneously ensure the contextual usability and topic diversity of the knowledge recommendation results. While the method described in the paper integrates contextual information and user interests, it does not consider the cold start problem caused by sparse or scarce user data, and the proposed contextual model is not applicable to product design in a cloud environment. Summary of the Invention

[0003] To overcome the challenges of insufficient historical data for effective recommendations when new designers join in a cloud environment, and the significant trust bias among designers regarding knowledge recommendations (i.e., a greater preference for recommendations from professionals with relevant backgrounds or domain authorities), this invention aims to provide a product form design knowledge recommendation method based on collective intelligence context and trust relationships. This method constructs a multi-dimensional contextual ontology and a product form design knowledge resource ontology using a structured ontology language. It then utilizes contextual similarity and trust relationships to obtain the nearest neighbor set of the target designer, and completes the list by rating neighbor individuals. This predicts the target designer's knowledge preference rating and generates a knowledge recommendation list. By mining the multi-dimensional contextual information and collaborative trust relationship features generated by collective intelligence interaction in the cloud environment, this method provides new methodological support for accurate matching of knowledge services in the cloud environment, improving the accuracy and reliability of recommendations.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A knowledge delivery method for product form design based on crowd intelligence context and trust relationship is characterized by the following steps:

[0006] Step 1: Construct a multi-dimensional contextual ontology model. Utilize OWL (Ontology Web Language) to build a multi-dimensional contextual ontology for product form design, with task context, designer context, and computational context as the main dimensions. During the knowledge push process, the cloud platform classifies and parses the multi-dimensional contextual information.

[0007] Step 2: Construct a product form design knowledge ontology model. Utilize OWL to build the knowledge ontology model, promote the standardized expression of product form design knowledge, and achieve efficient management and rapid dissemination.

[0008] Step 3: The comprehensive context similarity calculation method that integrates multi-dimensional context information weights is used. Domain experts evaluate the importance of different context elements and use the entropy weight method to calculate and determine the integrated multi-dimensional context information weights. Finally, the comprehensive multi-dimensional context similarity that integrates multi-dimensional context information weights is calculated.

[0009] Step 4, the trust relationship value calculation method, calculates the direct trust value through the historical interaction behavior between designers, and calculates the indirect trust value through the similarity of designers' situational ratings;

[0010] Step 5: Based on the knowledge score prediction of the collective intelligence context-trust relationship, combined with the comprehensive context similarity in Step 3 and the trust relationship value in Step 4, a set of neighboring designers of the target designer is selected. The top N designers with the highest similarity to the target designer are selected as the nearest neighbors of the target user to participate in the score prediction. The top K knowledge resources with the highest scores are generated into a recommendation list and recommended to the designers.

[0011] The specific steps for constructing the multidimensional contextual ontology model described in step 1 are as follows:

[0012] Construct a multi-dimensional contextual model for product design in a cloud environment, and build the ontology model using Protégé.

[0013] Ontology construction is performed using Protégé.

[0014] The multidimensional context model for product design in the cloud environment is defined as three categories: Task Context, Designer's Context, and Computational Context, as shown in Equation (1):

[0015] Multidimensional Context={Task Context, Designer's Context, Computational Context}(1)

[0016] Wherein: Task Context represents the task context, Designer's Context represents the designer context, and Computational Context represents the computational context.

[0017] Task Context:

[0018] The knowledge needs of designers are generally highly correlated with the product form design tasks they perform and the problems they encounter. Therefore, modeling the task context helps to effectively explore users' knowledge needs. Based on a summary of relevant research, the product form design stage can be divided into five key stages: requirements analysis, conceptual design, preliminary design, detailed design, and process design. Different stages have different design goals, each of which can be mapped to specific tasks and objects. Each task is executed and completed by designers using different design methods and tools. Therefore, the product design task context in a cloud environment is represented as follows:

[0019]

[0020] For details, please refer to Table 1:

[0021] Table 1 Description of Task Context Modeling

[0022]

[0023] Designer's Context:

[0024] The designer context describes the attributes and characteristics of the designer themselves. In a cloud environment, the knowledge needs of designers are uncertain and domain-specific. Their educational background, professional background, and skill level determine their knowledge level and learning ability. At the same time, past service experience, years of experience, and task roles also influence the designer's knowledge interests. Therefore, the designer context in a cloud environment is represented as follows:

[0025]

[0026] For details, please refer to Table 2:

[0027] Table 2 Description of Context Modeling by Designers

[0028]

[0029] Computational context:

[0030] The computing context can be represented by the data resources and websites used by designers when performing design activities. This content also reflects the designers' knowledge needs and can be recorded and analyzed based on big data from cloud platforms. Therefore, the computing context in the cloud environment can be represented as shown in equation (4):

[0031] Computer Context=<Data Resource, Web Resources, Computing Resources>(4)

[0032] See Table 3:

[0033] Table 3 Description of Computational Context Modeling

[0034]

[0035] The specific approach to constructing a product form design knowledge ontology model is as follows: Utilizing OWL to build a knowledge ontology model promotes the standardized expression of product form design knowledge, thereby achieving efficient management and rapid dissemination.

[0036] The product form design knowledge ontology model is represented as a quadruple KO:

[0037] KO={Entity, Attribute, Relation, Part}(5)

[0038] In equation (5): Entity represents the product form design knowledge resource entity, which can be formally represented as:

[0039]

[0040] `Id` is the unique identifier of the knowledge resource; `DesignObject` represents the target product object to which the knowledge is applied; `DisciplineArea` represents the subject area to which the knowledge belongs, including design, psychology, materials, management, mechanical engineering, and user research; `DesignStage` represents the application stage of the knowledge, including requirements analysis, conceptual design, preliminary design, detailed design, and process design; `KnowledgeType` represents the type of knowledge, including patent reports, 2D drawings, 3D models, design materials, color schemes, expert experience, data parameters, design rules, inspirational texts, design examples, and formulas; `KnowledgeSource` represents the source of the knowledge, including design manuals, national standards, market reports, journals, monographs, and user feedback; `KnowledgeContent` represents the specific description of the knowledge content.

[0041] Attribute is a set of attribute concepts that represent some attribute characteristics of knowledge entities, including domain attributes, stage attributes, object attributes, and product characteristic attributes. Among them, product characteristic attributes represent the characteristic information of design objects, including five subclasses: shape, color, material, interaction, and craftsmanship.

[0042] A relation is a set of relationships between concepts, representing all attribute relationships other than hierarchical relationships and autocorrelation relationships. Formalized as:<KO1,KO2,r> KO1, r represents the attribute relationship between concepts;

[0043] Part represents a hierarchical set of relationships between concepts, indicating the relationship between a parent class and its subclasses.

[0044] The method for calculating comprehensive context similarity using multi-dimensional contextual information weights is as follows: Domain experts evaluate the importance of different contextual elements, the entropy weight method is used to determine the weights of the multi-dimensional contextual information, and the similarities for text, attribute, numerical, vector, and fuzzy categories are calculated separately. These similarities are then fused to obtain the weighted comprehensive multi-dimensional context similarity. The specific solution is as follows:

[0045] 1) Information entropy calculation:

[0046]

[0047] In equations (7)-(8), Entropy j Let P represent the information entropy of the j-th context element. ij x′ represents the percentage of the i-th expert's rating for the j-th contextual element. ij It is the standardized score, ln(g) is the normalization factor, g is the total number of experts, and m is the total number of situational elements;

[0048] 2) Weight calculation:

[0049] D j =1-Entropy j (9)

[0050]

[0051] In equations (9)-(10), D j W represents the difference coefficient of the j-th contextual element. j This represents the weight of the j-th context element, where n is the total number of context elements;

[0052] 3) Calculation of comprehensive context similarity by incorporating context weights:

[0053]

[0054] In equation (11), SIM MC To achieve a comprehensive context similarity that integrates context weights, SIM(MC,MC') represents the ontology similarity between contexts MC and MC'.

[0055] The trust relationship value calculation method described in step 4 refers to the degree of mutual trust established between designers based on knowledge evaluation and shared interests. This is essentially a process of selecting more reliable users within a user's neighborhood. For example, if designer D... a For D b If users are interested in previously liked knowledge resources, it indicates a trust relationship. In the process of recommending knowledge related to product design, calculating the trust value between designers can enrich the surrounding information, ensure the reliability of recommendations, and solve the problem of data sparsity. This is achieved by comprehensively calculating the direct trust value T. im and indirect trust value T in To better reflect the trust relationship among designers, the specific approach is as follows: by analyzing the designers' rating behavior and the similarity of their contextualized ratings, direct trust values ​​and indirect trust values ​​are calculated separately, and a linear weighted method is used for comprehensive calculation. The specific solution is as follows:

[0056] 1) Direct Trust

[0057] The direct trust value indicates that there is a direct relationship between the target designer and other designers. If designers have jointly rated a project, they are considered to have a direct relationship. A designer trust matrix is ​​constructed based on the direct trust relationship between designers. The number 1 represents trust and 0 represents distrust. The designer trust matrix T is shown in Table 5.

[0058] Table 5 Designer Trust Matrix

[0059]

[0060] T(D a D b When ) = 1, it indicates that the designer D a With D b There is a direct trust relationship between them; conversely, when T(D) a D b When ) = 0, it indicates that the designer D a With D b There is no direct trust relationship between them. The method for calculating the direct trust value of users is shown in equations (12)-(13):

[0061]

[0062] In equations (12)-(13), L Da,Db Indicates designer D a With D b The shortest trust propagation distance, M represents the maximum trust propagation distance among all designers, m represents the total number of designer nodes, and Q represents the average in-degree and out-degree of designer nodes;

[0063] 2) Indirect Trust

[0064] In the initial stages of a cloud environment, direct trust values ​​are sparse. If only direct trust relationships based on shared rating behaviors between designers are considered, the trust relationship between users cannot be accurately determined when two designers do not share common rating knowledge. Designers' contextual rating metrics directly reflect their personal preferences. The closer the knowledge rating metrics of two designers are, the higher their similarity and therefore their indirect trust. Therefore, the Pearson correlation coefficient is used to calculate the similarity of contextual rating metrics between designer A and designer B to obtain their indirect trust.

[0065]

[0066] In equation (14), T in (a,b) represents the designer's indirect trust level, r a,kq,c This indicates that designer A, in context C, has access to knowledge resource K. q The rating, r b,kq,c This indicates that designer b, in context c, has access to knowledge resource K. q The rating, r a,c and r b,c These represent the evaluation values ​​of the knowledge resources jointly rated by designer a and designer b in the same context c.

[0067] 3) Comprehensive Trust Value Calculation

[0068] Based on equations (11) and (14), the comprehensive trust value of designer a and designer b is calculated using a linear weighted method. The calculation formula is as follows:

[0069] T combine (a,b)=θT in (a,b)+(1-θ)T im (D a D b (15)

[0070] In equation (15), θ is an adjustable weight coefficient. When θ is 0, the algorithm only calculates the direct trust between designers; when θ is 1, the algorithm only considers the indirect trust between designers.

[0071] Step 5, which describes the knowledge scoring prediction based on the collective intelligence context-trust relationship, is specifically implemented as follows:

[0072] 1) Calculation of neighboring personnel

[0073] Based on the comprehensive multidimensional contextual similarity and designer trust relationship value obtained in step 3, the designer trust value and multidimensional contextual similarity are fused to better calculate the comprehensive similarity value of the designer. The calculation formula is shown in equation (16):

[0074]

[0075] In equation (16), η is the designer's trust threshold; if D a and D b If the trust value is greater than η, it indicates that the result D a and D b It has a high level of trust, even D a and D b There are no similar knowledge ratings, but the two can completely trust and accept each other's preferred knowledge resources, if D a and D b If the trust value is less than η, it indicates that D a and D b If there are no similar knowledge scoring behaviors, the contextual similarity between the two needs to be further considered, where λ is the overall similarity weighting coefficient of the designers;

[0076] 2) Rating prediction

[0077] Based on the trust relationship value of the integrated designers and the comprehensive multi-dimensional contextual similarity, the top N designers with the highest similarity to the target designer are selected as the target designer's nearest neighbors to participate in the score prediction, and the top K knowledge items are recommended to the target designer. The score prediction is calculated as follows:

[0078]

[0079] In equation (17), P a,Kq Indicates the predictive designer D a Knowledge resource K i The rating, and These represent designers D. a and D b The average score, r b,kq Indicates neighboring person D a Knowledge resource K q The rating, N represents the designer D a The neighbor set.

[0080] The beneficial effects of this invention are:

[0081] This invention addresses the personalized recommendation problem for product form design knowledge delivery to cold-start users in a cloud environment, proposing a knowledge delivery method based on collective intelligence and trust relationships. A multi-dimensional contextual and product form design knowledge ontology model is constructed using Protégé software, achieving a standardized expression of multi-dimensional contextual information and product form design knowledge in the cloud environment, thus supporting efficient knowledge management and rapid delivery. By introducing the comprehensive similarity and trust relationship values ​​of the multi-dimensional contextual ontology, and defining a designer trust threshold to distinguish designers with different trust values, the method accurately identifies nearby designers. Based on this, a personalized knowledge delivery list is generated by predicting the knowledge preference scores of the target designers. This method effectively alleviates the data sparsity problem of cold-start users and significantly improves the accuracy and diversity of the recommendation system. Finally, experiments demonstrate that the proposed method has high recommendation accuracy and superior recommendation performance. Attached Figure Description

[0082] Figure 1 This is a diagram of the product form design knowledge service model in the cloud environment of this invention.

[0083] Figure 2 It is a knowledge push framework designed based on the product form of collective intelligence context and trust relationship.

[0084] Figure 3 This is a multi-dimensional contextual ontology model diagram of product form design in a cloud environment in the method of this invention.

[0085] Figure 4 This is a knowledge resource ontology model diagram of product form design in a cloud environment in the method of this invention.

[0086] Figure 5 This is a trust relationship diagram of designers in the method of this invention.

[0087] Figure 6 This is a line graph showing the impact of the designer's trust threshold η on MAE.

[0088] Figure 7 This is a line graph showing the impact of the designer's trust threshold η on RMSE.

[0089] Figure 8 This is a line graph showing the impact of the comprehensive similarity weighting coefficient λ on MAE.

[0090] Figure 9 This is a line graph showing the impact of the comprehensive similarity weighting coefficient λ on RMSE.

[0091] Figure 10 This is a flowchart of the push method of the present invention. Detailed Implementation

[0092] The present invention will now be described in detail with reference to specific embodiments.

[0093] See Figure 1 The knowledge delivery method based on the product form design of crowd intelligence context and trust relationship includes the following steps:

[0094] Analysis of product form design knowledge service models and problems in the cloud environment

[0095] Product form design in the cloud environment, as an emerging paradigm, fully leverages the advantages of the internet and big data. By integrating and optimizing distributed public knowledge resources, it improves design efficiency and yields diverse problem-solving approaches and a large number of innovative product form design solutions. It has now become the dominant paradigm for product design in many enterprises. The cloud platform breaks down the overall product form design task and assigns it to different designers within the cloud environment. Designers complete the task based on their own knowledge and experience or by utilizing externally provided knowledge. By aggregating extensive product form knowledge resources as a source of innovation for product form design solutions, the task is completed through an asynchronous and geographically dispersed collaborative model.

[0096] This invention studies a product form design knowledge service in a cloud environment. It utilizes cloud computing and virtual technology to uniformly virtualize and provide a service of massive and diverse knowledge resources, achieving resource sharing and complementary advantages, and offering personalized design knowledge services to users. The cloud design knowledge service model involves various cross-disciplinary and multi-professional knowledge, including standards and specifications, patent documents, document knowledge, case knowledge, expert experience, and model knowledge, exhibiting complexity, dynamism, and cross-fertility. Driven by the design context, designers searching and matching product design knowledge based on similarity algorithms, according to the task description information, optimize the knowledge matching set based on designer preferences, and proactively push product design knowledge that meets the designers' needs to them. See the diagram of the product form design knowledge service model in a cloud environment. Figure 1 .

[0097] The cloud environment resource pool stores a massive amount of product design knowledge resources. During the design process, the cloud platform divides design tasks into different sub-tasks and, based on the designers' task requirements, pushes targeted knowledge resources to assist in achieving product design goals and solutions. However, in the initial stage of cloud platform construction, the lack of sufficient designer behavior data and personalized feedback makes it impossible to effectively push design knowledge resources that match designers' needs. Furthermore, design knowledge is created and reused within specific contexts; knowledge resources can only realize their due value when placed within their specific design context. Product form design knowledge covers a wide range, from basic design principles, styling analysis, and material selection to cutting-edge design trends, user research, and innovative technologies; each type of knowledge has its specific application context. As designers participate in different tasks, their knowledge preferences vary, inevitably leading to diverse needs for product design knowledge. Therefore, effectively pushing knowledge to designers and increasing their trust in that knowledge becomes extremely important.

[0098] Therefore, this invention provides a solution for product form knowledge delivery in a cloud environment by integrating multi-dimensional contextual information and trust relationships. This method analyzes the knowledge usage behavior and rating information of designers in the cloud environment, uses contextual similarity and trust relationships to obtain the nearest neighbor set of the target designer, completes the rating based on the knowledge ratings of neighboring individuals, thereby predicting the target designer's knowledge preference rating, and generating a knowledge delivery list. This list delivers effective product design knowledge resources to designers, effectively improving the accuracy and trustworthiness of cold-start knowledge delivery. See the main steps for details. Figure 2 .

[0099] Step 1: Construct a multi-dimensional context model ontology model

[0100] Using OWL (Ontology Web Language), a multi-dimensional contextual ontology for product design is constructed, with task context, designer context, and computation context as the main dimensions. During the knowledge push process, the cloud platform classifies and parses the multi-dimensional contextual information (which is the information stored as the ontology). The multi-dimensional contextual information includes: the main dimensions of task context, designer context, and computation context, as well as multi-dimensional contexts including other sub-contexts.

[0101] Context, characterized by its objective existence, independence, and personalization, is a crucial factor distinguishing different designers and knowledge activities. It provides designers with personalized information while simultaneously filtering out useless information. The purpose of knowledge delivery is to enable designers to acquire the knowledge they need more efficiently. However, because knowledge is context-dependent, the value of the same knowledge can vary significantly across different design contexts. Therefore, considering the context of knowledge application is essential during the knowledge delivery process. Existing technologies generally summarize context into five dimensions: Who, When, Where, What, and How. This represents the designer's design activities at a specific time and place, using a specific method, as well as the content and results of those activities.

[0102] For product design in a cloud environment, the context determines or influences changes in designers' needs and their state during task completion. Its main characteristics include:

[0103] 1) Personnel relevance: A user's personal characteristics such as age, occupation, knowledge level, and interests will influence their knowledge preferences;

[0104] 2) Diversity: The contexts originate from designers, task characteristics, business processes, equipment resources, and the external environment, resulting in a complex and diverse composition with significant heterogeneity;

[0105] 3) Dynamism: It changes with the designers and tasks. However, in the cloud environment, the knowledge needs of designers are less affected by factors such as time and location, and are mainly closely related to user characteristics, task characteristics, and data and equipment resources.

[0106] To more accurately describe the design context in a cloud environment, this embodiment proposes a multi-dimensional context model for product design in a cloud environment, taking into account the characteristics of the cloud environment. The ontology model is constructed using Protégé. (See [link to documentation]). Figure 3 .

[0107] The multidimensional context model for product design in the cloud environment is defined as three categories: Task Context, Designer's Context, and Computational Context, as shown in Equation (1):

[0108] Multidimensional Context={Task Context, Designer's Context, Computational Context}(1)

[0109] in:

[0110] 1) Task Context

[0111] The knowledge needs of designers are generally highly correlated with the product form design tasks they perform and the problems they encounter. Therefore, modeling the task context helps to effectively explore users' knowledge needs. Based on a summary of relevant research, the product form design stage can be divided into five key stages: requirements analysis, conceptual design, preliminary design, detailed design, and process design. Each stage has different design goals, each of which can be mapped to specific tasks and objects. Each task is executed and completed by designers using different design methods and tools. Therefore, the product design task context in a cloud environment can be represented as follows:

[0112]

[0113] For details, please refer to Table 1:

[0114] Table 1 Description of Task Context Modeling

[0115]

[0116] 2) Designer's Context

[0117] The designer context describes the attributes and characteristics of the designer themselves. In a cloud environment, the knowledge needs of designers are uncertain and domain-specific. Their educational background, professional background, and skill level determine their knowledge level and learning ability. Meanwhile, past service experience, years of experience, and task roles also influence their knowledge interests. Therefore, the designer context in a cloud environment can be represented as:

[0118]

[0119] For details, please refer to Table 2:

[0120] Table 2 Description of Context Modeling by Designers

[0121]

[0122] 3) Computational Context

[0123] The computing context can be represented as the data resources and websites used by designers when performing design activities. These contents also reflect the designers' knowledge needs and can be recorded and analyzed based on big data from cloud platforms. Therefore, the computing context in the cloud environment can be represented as shown in equation (4):

[0124] Computer Context =<Data Resource,Web Resources,Computing Resources> (4)

[0125] See Table 3:

[0126] Table 3 Description of Computational Context Modeling

[0127]

[0128] Step 2: Construct a knowledge ontology model for product form design;

[0129] Construct a product form design knowledge ontology model, utilize OWL to build the knowledge ontology model, promote the standardized expression of product form design knowledge, and achieve efficient management and rapid dissemination;

[0130] The knowledge types involved in product form design include design drawings, patents, research reports, industry standards, product specifications, design tools, inspirational texts, symbolic elements, and color schemes. Furthermore, due to the characteristics of cloud design and the interdisciplinary nature of product form design, product form design knowledge is characterized by complexity, dynamism, and fragmentation. To improve the utilization rate of product form design knowledge, OWL is used to formally describe knowledge resources, and a knowledge ontology model is constructed using Protégé software. This promotes the standardized expression of product form design knowledge, enabling efficient management and rapid dissemination. See [link to relevant documentation]. Figure 4 .

[0131] The specific approach involves using OWL to formally describe knowledge resources, promoting the standardized expression of product form design knowledge, and achieving efficient management and rapid dissemination.

[0132] The product form design knowledge ontology is represented as a quadruple KO:

[0133] KO={Entity, Attribute, Relation, Part}(5)

[0134] In equation (5): Entity represents the product form design knowledge resource entity, which can be formally represented as:

[0135]

[0136] `Id` is the unique identifier of the knowledge resource; `DesignObject` represents the target product object to which the knowledge is applied; `DisciplineArea` represents the subject area to which the knowledge belongs, including design, psychology, materials, management, mechanical engineering, user research, etc.; `DesignStage` represents the application stage of the knowledge, including requirements analysis, conceptual design, preliminary design, detailed design, and process design; `KnowledgeType` represents the type of knowledge, including patent reports, 2D drawings, 3D models, design materials, color schemes, expert experience, data parameters, design rules, inspirational texts, design examples, formulas, etc.; `KnowledgeSource` represents the source of the knowledge, including design manuals, national standards, market reports, journals, monographs, user feedback, etc.; `KnowledgeContent` represents the specific description of the knowledge content.

[0137] Attribute is a set of attribute concepts that represent some attribute characteristics of knowledge entities, including domain attributes, stage attributes, object attributes, and product characteristic attributes. Among them, product characteristic attributes represent the characteristic information of design objects, including five subclasses: shape, color, material, interaction, and craftsmanship.

[0138] A relation is a set of relationships between concepts, representing all attribute relationships other than hierarchical relationships and autocorrelation relationships.

[0139] Formal representation:<KO1,KO2,r> ,in r represents the attribute relationship between concepts.

[0140] Part represents a hierarchical set of relationships between concepts, indicating the relationship between a parent class and its subclasses.

[0141] Step 3: A comprehensive context similarity calculation method that integrates context weights;

[0142] The comprehensive context similarity calculation method that integrates multi-dimensional context information weights involves domain experts evaluating the importance of different context elements, using the entropy weight method to calculate and determine the integrated multi-dimensional context information weights, and finally calculating the comprehensive multi-dimensional context similarity that integrates the multi-dimensional context information weights.

[0143] Multidimensional contextual information from different designers is fundamental to calculating contextual similarity. Designers' behaviors and knowledge preferences may differ across contexts, thus requiring the integration of contextual information to calculate similarity between designers. The contextual similarity calculation method proposed in this embodiment calculates ontology similarity SIM(MC,MC'), which includes text similarity, attribute similarity, numerical similarity, vector class similarity, and fuzzy class similarity. Furthermore, since contextual weights reflect the importance of different dimensions, experts evaluate the importance of different contextual elements, and the entropy weight method is used to determine their weights. Finally, the contextual similarity, incorporating contextual weights, is calculated. The specific formula for calculating contextual weights is as follows:

[0144] Information entropy calculation:

[0145]

[0146] In equations (7)-(8), Entropy j Let P represent the information entropy of the j-th context element. ij x′ represents the percentage of the i-th expert's rating for the j-th contextual element. ij It is the standardized score, ln(g) is the normalization factor, g is the total number of experts, and m is the total number of situational elements;

[0147] Weight calculation:

[0148] D j =1-Entropy j (9)

[0149]

[0150] In equations (9)-(10), D j W represents the difference coefficient of the j-th contextual element. j This represents the weight of the j-th context element, where n is the total number of context elements;

[0151] 3) Calculation of comprehensive context similarity by incorporating context weights:

[0152]

[0153] In equation (11), SIM MC To achieve a comprehensive context similarity that integrates context weights, SIM(MC,MC') represents the ontology similarity between contexts MC and MC'.

[0154] Step 4, the trust relationship value calculation method, calculates the direct trust value through the historical interaction behavior between designers, and calculates the indirect trust value through the similarity of designers' situational ratings;

[0155] Trust relationship value calculation;

[0156] Trust relationships refer to the degree of mutual trust established between designers based on factors such as knowledge evaluation and shared interests. Essentially, it's a process of selecting more reliable designers within a given field. For example, if designer D... a With D b If users are interested in previously liked knowledge resources, it indicates a trust relationship between them. In the process of recommending knowledge related to product form design, calculating the trust value between designers can enrich the surrounding information, ensure the reliability of the recommendations, and solve the problem of data sparsity. This paper calculates the direct trust value T through comprehensive calculation. im and indirect trust value T in To better illustrate the trust relationships among designers, please refer to the designer trust relationship diagram. Figure 5 The specific approach is as follows: By analyzing the designers' rating behavior and the similarity of their contextualized ratings, direct trust values ​​and indirect trust values ​​are calculated separately, and then a linear weighted method is used to synthesize them. The specific solution is as follows:

[0157] 1) Direct Trust

[0158] The direct trust value indicates that there is a direct relationship between the target designer and other designers. If designers have jointly rated a project, they are considered to have a direct relationship. A designer trust matrix is ​​constructed based on the direct trust relationship between designers. The number 1 represents trust and 0 represents distrust. The designer trust matrix T is shown in Table 5.

[0159] Table 5 Designer Trust Matrix

[0160]

[0161] T(D a D b When ) = 1, it indicates that the designer D a With D b There is a direct trust relationship between them; conversely, when T(D) a D b When ) = 0, it indicates that the designer D a With D b There is no direct trust relationship between them. The method for calculating the direct trust value of users is shown in equations (12)-(13):

[0162]

[0163] In equations (12)-(13), L Da,Db Indicates designer D a With D bThe shortest trust propagation distance, M represents the maximum trust propagation distance among all users, m represents the total number of designer nodes, and Q represents the average in-degree and out-degree of designer nodes;

[0164] 2) Indirect Trust

[0165] In the initial stages of a cloud environment, direct trust values ​​are sparse. If only direct trust relationships based on shared rating behaviors between designers are considered, the trust relationship between users cannot be accurately determined when two designers do not share common rating knowledge. Designers' contextual rating metrics directly reflect their personal preferences. The closer the knowledge rating metrics of two designers are, the higher their similarity and therefore their indirect trust. Therefore, the Pearson correlation coefficient is used to calculate the designer's D... a With D b The similarity of contextual ratings between the two measures yields the designer's indirect trust level:

[0166]

[0167] In equation (14), T in (a,b) represents the designer's indirect trust level, r a,kq,c This indicates that designer A, in context C, has access to knowledge resource K. q The rating, r b,kq,c This indicates that designer b, in context c, has access to knowledge resource K. q The rating, r a,c and r b,c These represent the evaluation values ​​of the knowledge resources jointly rated by designer a and designer b in the same context c.

[0168] 3) Comprehensive Trust Value Calculation

[0169] Based on equations (11) and (14), the comprehensive trust value (also known as the comprehensive trust level) of user a and user b is calculated using a linear weighting method. The calculation formula is as follows:

[0170] T combine (a,b)=θT in (a,b)+(1-θ)T im (D a D b (15)

[0171] In equation (15), θ is an adjustable weight coefficient. When θ is 0, the algorithm only calculates the direct trust between designers; when θ is 1, the algorithm only considers the indirect trust between designers.

[0172] Step 5: Based on the knowledge rating prediction of the collective intelligence context-trust relationship, combining the comprehensive context similarity from Step 3 and the trust relationship value from Step 4, a set of neighboring designers of the target designer is selected. The top N designers with the highest similarity to the target designer are selected as the target user's nearest neighbors to participate in the rating prediction, and the top K knowledge resources with the highest ratings are recommended to the designers (a recommendation list is generated from the top K knowledge resources with the highest ratings and then recommended to the designers). The specific implementation is as follows:

[0173] 1) Calculation of neighboring personnel

[0174] Based on the comprehensive multidimensional contextual similarity and designer trust relationship value obtained in step 3, the designer trust value and multidimensional contextual similarity are fused to better calculate the comprehensive similarity value of the designer. The calculation formula is shown in equation (16):

[0175]

[0176] In equation (16), η is the designer's trust threshold; if D a and D b If the trust value is greater than η, it indicates that the result D a and D b It has a high level of trust, even D a and D b There are no similar knowledge ratings, but the two can completely trust and accept each other's preferred knowledge resources, if D a and D b If the trust value is less than η, it indicates that D a and D b If there are no similar knowledge scoring behaviors, the contextual similarity between the two needs to be further considered, where λ is the overall similarity weighting coefficient of the designers;

[0177] 2) Rating prediction

[0178] Based on the trust relationship value of the integrated designers and the comprehensive multi-dimensional contextual similarity, the top N designers with the highest similarity to the target designer are selected as the target designer's nearest neighbors to participate in the score prediction, and the top K knowledge items are recommended to the target designer. The score prediction is calculated as follows:

[0179]

[0180] In equation (17), P a,Kq Indicates the predictive designer D a Knowledge resource K i The rating, and These represent designers D. a and D bThe average score, r b,kq Indicates neighboring person D a Knowledge resource K q The rating, N represents the designer D a The neighbor set.

[0181] Implementation Cases

[0182] The design of smart coffee machine products involves multiple aspects such as styling, CMF (Color, Material, Finish), and human-computer interaction, requiring collaboration from multiple disciplines. Furthermore, diverse market demands necessitate a balance between functional practicality, aesthetic value, and intelligent experience. Therefore, this embodiment uses a home-use smart coffee machine product form concept design activity as an example to analyze and verify the effectiveness of the proposed method. The entire case process simulates a cloud environment, using PCs to simulate dispersed designers. The platform provides unified management and operation of designers, knowledge resources, and the design process. The specific knowledge push process is as follows:

[0183] I. Contextual Data Acquisition by Target Designers

[0184] First, extract the multi-dimensional scenario instance MC1 of the target designer D1, as shown in Table 6-8:

[0185] Table 6 Examples of coffee machine task scenarios

[0186]

[0187] Table 7 Scenario Examples for Designers

[0188]

[0189] Table 8 Description of Computational Context Modeling

[0190]

[0191] II. Knowledge Push for Product Form Design Based on Collective Intelligence Context and Attribute Preferences

[0192] 1) Calculation of crowd intelligence situational similarity (referring to the total multidimensional situational similarity between groups)

[0193] Twenty experts evaluated the importance of situational elements using a Likert scale ranging from 1 to 5. The situational weights were calculated based on equations (7) to (10), as shown in Table (9).

[0194] Table 9 Contextual Weights

[0195]

[0196] Then, the weighted comprehensive situational similarity between the target designer D1 and other designs is calculated using equation (11), as shown in Table (10):

[0197] Table 10 Weighted Overall Contextual Similarity

[0198]

[0199] 2) Calculation of direct trust value

[0200] Based on the above calculations, the trust matrix was constructed by selecting the top 49 designers with the highest similarity to the D1 comprehensive situation, and their rating behavior was extracted. The direct trust between them was calculated according to formulas (12)-(13), as shown in Table (11).

[0201] Table 11 Direct Trust Level

[0202]

[0203] 3) Indirect Trust Calculation

[0204] To ensure the timeliness and representativeness of the ratings, this paper extracted the latest 100 viewed knowledge item ratings from each designer's recent rating records and calculated the indirect trust level among the 49 designers according to formula (14). See Table (12).

[0205] Table 12 Indirect Trust Level

[0206]

[0207] 4) Comprehensive Trust Calculation

[0208] Based on equations (12) and (14), the overall trust level between the target designer and other designers is calculated using a linear weighting method, as shown in Table (13):

[0209] Table 13 Overall Trust Level

[0210]

[0211] 5) Calculations by nearby designers

[0212] According to formula (16), the top 5 neighboring designers with the highest comprehensive similarity to the target user D1 are calculated, and the designers are ranked in order of similarity as follows: D9>D 21 >D3>D 16 >D 24 .

[0213] 6) Knowledge resource preference rating prediction

[0214] And according to Equation (17), the missing score of the target designer D1 is predicted, as shown in Table (14):

[0215] Table (14) D1 Knowledge Resource Preference Prediction Score

[0216]

[0217] 7) Generating push notification lists

[0218] Based on the above calculations, the top five knowledge resources with the highest ratings are output as the final push result, which is K. 91 >K 89 >K 43 >K 81 >K 96 The specific scoring is shown in Table (11):

[0219] Table 11 Final Push List and Scoring for Target Designers D1

[0220]

[0221] Recommendation Results Analysis

[0222] Recommendation accuracy is the most important metric in recommender systems. This embodiment uses Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) to measure the recommendation accuracy of the proposed algorithm. MAE and RMSE are the most commonly used metrics. MAE represents the average difference between the target designer's predicted score for knowledge resources in the training set and the actual score in the test set. RMSE represents the average of the sum of squares of the differences between the target designer's predicted value and the actual score in the test set. The smaller the MAE and RMSE values, the higher the algorithm's performance.

[0223] 1) The impact of the designer's trust threshold η

[0224] The size of the designer's trust threshold η affects the division of designer trust levels. An excessively high or low η value is detrimental to the designer's selection of nearest neighbors, thus impacting the accuracy of the final comprehensive similarity calculation. The user trust threshold ranges from 0 to 1, with increments of 0.1. The trends of MAE and RMSE values ​​as η increases are as follows... Figure 6-7 As shown.

[0225] Depend on Figure 6-7It can be seen that when the value of η is less than 0.5, the MAE and RMSE values ​​gradually decrease; when the value of η is greater than 0.5, the MAE and RMSE values ​​of the algorithm gradually increase. This is because when the designer trust threshold η is too low, some designers with lower trust values ​​will be included in the comprehensive similarity calculation, thus reducing accuracy. Conversely, when the user trust threshold is too high, although a set of designers with high trust can be filtered out, the sparse number of high-trust users results in less usable information. When the value of η is equal to 0.5, the algorithm's MAE and RMSE values ​​are minimized; therefore, the optimal value for the designer trust threshold η is 0.5.

[0226] 2) The influence of the overall similarity weight coefficient λ

[0227] A more accurate comprehensive similarity score can be obtained by weighting and combining the trust relationship between designers and the contextual similarity. When the value of λ is 0, the similarity between designers is the contextual similarity between designers; when the value of λ is 1, the similarity between designers is the trust value. The comprehensive similarity weight coefficient λ ranges from 0 to 1, with an increment of 0.1. As the value of λ increases, the trends of MAE and RMSE values ​​are as follows: Figure 8-9 As shown.

[0228] Depend on Figure 8-9 It can be seen that as the value of λ increases, the values ​​of MAE and RMSE generally show a trend of first decreasing and then increasing. When the value of λ is less than 0.5, the values ​​of MAE and RMSE gradually decrease. When the value of λ is greater than 0.5, the values ​​of MAE and RMSE gradually increase. When λ = 0.5, the values ​​of MAE and RMSE are the smallest, and the algorithm performs best. Therefore, the optimal value of the comprehensive similarity weight coefficient λ is 0.5.

[0229] 3) Comparative Analysis

[0230] To verify the effectiveness of the proposed method, a traditional collaborative filtering algorithm was used as a reference, with the Pearson coefficient employed as the similarity metric. The algorithm primarily addresses the "cold start" phenomenon in cloud environments; therefore, the main purpose of the testing is to assess the performance recommended by designers with few scoring items.

[0231] Table 12 shows that the MAE value of the traditional collaborative filtering (UCF) recommendation algorithm is 0.7337, while the MAE value of the proposed method is 0.6517, demonstrating higher recommendation accuracy. Furthermore, compared to the traditional collaborative filtering (UCF) algorithm, the proposed method improves recommendation accuracy by 11.18%. This is mainly because the traditional collaborative filtering algorithm relies on a user-item rating matrix, and its recommendation performance significantly decreases when data is sparse. The proposed method, however, incorporates dynamic contextual factors such as task context, personnel context, and computational context within a cloud environment, and expands the information dimension of similarity calculation using direct and indirect trust relationships among designers, thus maintaining high recommendation accuracy even with scarce data. Compared to existing research, the proposed method exhibits stronger adaptability in specific scenarios within the product form design field.

[0232] Table 12 Comparison of MAE and RMAE with traditional UCF

[0233]

Claims

1. A knowledge delivery method based on a product form design of collective intelligence context and trust relationship, characterized in that, Includes the following steps: Step 1: Construct a multi-dimensional contextual ontology model. Utilize OWL to construct a multi-dimensional contextual ontology for product form design, with task context, designer context, and computational context as the main dimensions. During the knowledge push process, the cloud platform classifies and parses the multi-dimensional contextual information. The multidimensional contextual information includes: the main dimensions of task context, designer context, and computational context, as well as multidimensional contexts including other sub-contexts; Step 2: Construct a product form design knowledge ontology model. Utilize OWL to build the knowledge ontology model, promote the standardized expression of product form design knowledge, and achieve efficient management and rapid dissemination. Step 3: The comprehensive context similarity calculation method that integrates multi-dimensional context information weights is used. Domain experts evaluate the importance of different context elements and use the entropy weight method to calculate and determine the integrated multi-dimensional context information weights. Finally, the comprehensive multi-dimensional context similarity that integrates multi-dimensional context information weights is calculated. Step 4, the trust relationship value calculation method, calculates the direct trust value through the historical interaction behavior between designers, and calculates the indirect trust value through the similarity of designers' situational ratings; Step 5: Based on the knowledge score prediction of the collective intelligence context-trust relationship, combined with the comprehensive context similarity in Step 3 and the trust relationship value in Step 4, a set of neighboring designers of the target designer is selected. The top N designers with the highest similarity to the target designer are selected as the nearest neighbors of the target user to participate in the score prediction. The top K knowledge resources with the highest scores are generated into a recommendation list and recommended to the designers.

2. The product form design knowledge push method based on crowd intelligence context-trust relationship as described in claim 1, characterized in that, The specific steps for constructing the multidimensional contextual ontology model described in step 1 are as follows: Construct a multi-dimensional contextual model for product design in a cloud environment, and build the ontology model using Protégé. The multi-dimensional contextual ontology model of product design in the cloud environment is set into three categories: task context, designer context, and computational context, as shown in equation (1): Multidimensional Context={Task Context, Designer's Context, ComputationalContext}(1) Where: Task Context represents the task context, and Designer's Context represents the designer's context. Computational context refers to the computing context. The product design task scenario in a cloud environment can be represented as: For details, please refer to Table 1: Table 1 Description of Task Context Modeling The designer's context in a cloud environment is represented as follows: For details, please refer to Table 2: Table 2 Description of Context Modeling by Designers The computing scenario in the cloud environment is represented by equation (4): Computer Context =<Data Resource,Web Resources,Computing Resources> (4) See Table 3: Table 3 Description of Computational Context Modeling 3. The product form design knowledge push method based on crowd intelligence context-trust relationship as described in claim 1, characterized in that, The specific approach to constructing a product form design knowledge ontology model is as follows: Utilizing OWL to build a knowledge ontology model promotes the standardized expression of product form design knowledge, thereby achieving efficient management and rapid dissemination. The product form design knowledge ontology model is represented as a quadruple KO: KO={Entity, Attribute, Relation, Part}(5) In equation (5): Entity represents the product form design knowledge resource entity, which can be formally represented as: Id is the unique identifier of the knowledge resource; DesignObject represents the target product object to which the knowledge is applied; DisciplineArea represents the subject area to which the knowledge belongs, including design, psychology, materials, management, mechanical engineering, and user research. DesignStage represents the application stage of knowledge, including requirements analysis, conceptual design, preliminary design, detailed design, and process design; KnowledgeType represents the type of knowledge, including patent reports, 2D drawings, 3D models, design materials, color schemes, expert experience, data parameters, design rules, inspirational texts, design examples, and formulas; KnowledgeSource represents the source of knowledge, including design manuals, national standards, market reports, journals, monographs, and user feedback; KnowledgeContent represents the specific description of the knowledge content. Attribute is a set of attribute concepts that represent some attribute characteristics of knowledge entities, including domain attributes, stage attributes, object attributes, and product characteristic attributes. Among them, product characteristic attributes represent the characteristic information of design objects, including five subclasses: shape, color, material, interaction, and craftsmanship. A relation is a set of relationships between concepts, representing all attribute relationships other than hierarchical relationships and autocorrelation relationships. Formalized as:<KO1,KO2,r> KO1, r represents the attribute relationship between concepts; Part represents a hierarchical set of relationships between concepts, indicating the relationship between a parent class and its subclasses.

4. The product form design knowledge push method based on crowd intelligence context-trust relationship as described in claim 1, characterized in that, The method for calculating the comprehensive context similarity based on the weighted multidimensional contextual information is as follows: The entropy weighting method is used to determine the weights of the multidimensional contextual information, and the similarities for text, attribute, numerical, vector, and fuzzy classes are calculated separately. These similarities are then fused to obtain the weighted comprehensive multidimensional context similarity. The specific solution is as follows: 1) Information entropy calculation: In equations (7)-(8), Entropy j Let P represent the information entropy of the j-th context element. ij x′ represents the percentage of the i-th expert's rating for the j-th contextual element. ij It is the standardized score, ln(g) is the normalization factor, g is the total number of experts, and m is the total number of situational elements; 2) Weight calculation: D j =1-Entropy j (9) In equations (9)-(10), D j W represents the difference coefficient of the j-th contextual element. j This represents the weight of the j-th context element, where n is the total number of context elements; 3) Calculation of comprehensive context similarity by incorporating context weights: In equation (11), SIM MC To achieve a comprehensive context similarity that integrates context weights, SIM(MC,MC') represents the ontology similarity between contexts MC and MC'.

5. The product form design knowledge push method based on crowd intelligence context-trust relationship as described in claim 1, characterized in that, The trust relationship value calculation method described in step 4 is as follows: By analyzing the designers' rating behavior and its contextualized rating similarity, direct trust value and indirect trust value are calculated separately, and then a linear weighted method is used for comprehensive calculation. The specific solution is as follows: 1) Direct Trust Value The direct trust value indicates that there is a direct relationship between the target designer and other designers. If designers have jointly rated a project, they are considered to have a direct relationship. A designer trust matrix is ​​constructed based on the direct trust relationship between designers. The number 1 represents trust and 0 represents distrust. The designer trust matrix T is shown in the table: Table 5 Designer Trust Matrix T(D a D b When ) = 1, it indicates that the designer D a With D b There is a direct trust relationship between them; conversely, when T(D) a D b When ) = 0, it indicates that the designer D a With D b There is no direct trust relationship between them. The method for calculating the direct trust value of users is shown in equations (12)-(13): T(D a D b ) In equations (12)-(13), L Da,Db Indicates designer D a With D b The shortest trust propagation distance, M represents the maximum trust propagation distance among all users, m represents the total number of user nodes, and Q represents the average in-degree and out-degree of user nodes; 2) Indirect Trust Value In the initial stages of a cloud environment, direct trust values ​​are sparse. If only direct trust relationships based on shared rating behaviors between designers are considered, the trust relationship between designers cannot be accurately determined when there is no shared rating knowledge between them. A designer's contextual rating metric directly reflects their personal preferences. The closer the knowledge rating metrics of two designers are, the higher their similarity, and therefore, the higher their indirect trust. Therefore, the Pearson correlation coefficient is used to calculate the similarity of contextual rating metrics between designer A and designer B to obtain their indirect trust. In equation (14), T in (a,b) represents the designer's indirect trust level, r a,kq,c This indicates that designer A, in context C, has access to knowledge resource K. q The rating, r b,kq,c This indicates that designer b, in context c, has access to knowledge resource K. q The rating, r a,c and r b,c These represent the evaluation values ​​of the knowledge resources jointly rated by designer a and designer b in the same context c. 3) Comprehensive Trust Value Calculation Based on equations (11) and (14), the comprehensive trust value of engineer a and designer b is calculated using a linear weighting method. The calculation formula is as follows: T combine (a,b)=θT in (a,b)+(1-θ)T im (D a ,D b ) (15) In equation (15), θ is an adjustable weight coefficient. When θ is 0, the algorithm only calculates the direct trust between designers; when θ is 1, the algorithm only considers the indirect trust between designers.

6. The knowledge push method for cloud-based product form design based on crowd intelligence context-trust relationship as described in claim 1, characterized in that, Step 5, which describes the knowledge scoring prediction based on the collective intelligence context-trust relationship, is specifically implemented as follows: 1) Calculation of neighboring personnel Based on the comprehensive multidimensional contextual similarity and designer trust relationship value obtained in step 3, the designer trust value and multidimensional contextual similarity are fused to better calculate the comprehensive similarity value of the designer. The calculation formula is shown in equation (16): In equation (16), η is the designer's trust threshold; if D a and D b If the trust value is greater than η, it indicates that the result D a and D b It has a high level of trust, even D a and D b There are no similar knowledge ratings, but the two can completely trust and accept each other's preferred knowledge resources, if D a and D b If the trust value is less than η, it indicates that D a and D b If there are no similar knowledge scoring behaviors, the contextual similarity between the two needs to be further considered, where λ is the overall similarity weighting coefficient of the designers; 2) Rating prediction Based on the trust relationship value of the integrated designers and the comprehensive multi-dimensional contextual similarity, the top N designers with the highest similarity to the target designer are selected as the target designer's nearest neighbors to participate in the score prediction, and the top K knowledge items are recommended to the target designer. The score prediction is calculated as follows: In equation (17), P a,Kq Indicates the predictive designer D a Knowledge resource K i The rating, and These represent designers D. a and D b The average score, r b,kq Indicates neighboring person D a Knowledge resource K q The rating, N represents the designer D a The neighbor set.

Images