A retrieval method and retrieval device based on Kansei Engineering knowledge graph

Abstract

The application discloses a retrieval method and retrieval device based on an ergonomics knowledge graph. The method establishes a design element space and an emotional image space, and provides a specific implementation mode for establishing the design element space and a specific implementation mode for establishing the emotional image space. Based on the connection between the design element space and the emotional image space, the two are organically combined to construct an ergonomics knowledge graph, so as to describe ergonomics knowledge in the form of a triple of entities and the relationship between the entities. Compared with related technologies, the technical scheme of the application can represent ergonomics knowledge in a simpler and easier-to-implement manner, thereby realizing effective management of ergonomics knowledge. By constructing the ergonomics knowledge graph, it is helpful to make ergonomics knowledge more easily applicable. By using the scalability and explainable reasoning capability of the knowledge graph, the retrieval demand of designers or consumers can be better responded to, and the demand for continuous enrichment and improvement of ergonomics knowledge can be met.

Description

Technical Field

[0001] This application relates to the field of Kansei Engineering knowledge representation technology, and in particular to a retrieval method and retrieval device based on Kansei Engineering knowledge graphs. Background Technology

[0002] Kansei Engineering is an important demand transformation technology that originated in Japan. It aims to establish a connection between consumers' emotional perception of a product and its design elements, thereby providing a replicable and interpretable design experience. Kansei Engineering can help designers select appropriate design elements for specific emotional design goals during the design process, and it can also determine whether a product design meets specific emotional needs of consumers during the evaluation process.

[0003] Kansei Engineering is the only widely recognized technology capable of addressing consumers' emotional needs. It has a long history of practical application in industry, assisting manufacturers in product development and playing a crucial role in the design of product features such as shape, color, material, and texture. However, despite extensive industrial practice, effective methods for representing and managing Kansei Engineering knowledge remain lacking. Furthermore, due to the complexity, ambiguity, and uncertainty of human emotional experiences, the knowledge structure of Kansei Engineering is also complex, making its representation difficult. Therefore, how to represent Kansei Engineering knowledge in a simpler and more easily implemented way, and how to effectively manage it, are the primary technical challenges that need to be addressed for the further widespread application of Kansei Engineering. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a retrieval method and apparatus based on a Kansei Engineering knowledge graph. The aim is to express and manage Kansei Engineering knowledge in a simple and easy-to-implement manner, thereby making Kansei Engineering knowledge easier to apply.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] In a first aspect, this application provides a method for constructing a knowledge graph of sensible engineering, the method comprising:

[0007] Obtain a core emotional vocabulary database for the target product category. The core emotional vocabulary database includes multiple core emotional words, which are emotional words that have a positive impact on consumers' behavior of purchasing products in the target product category.

[0008] The design elements of the target product category are deconstructed according to multiple preset deconstruction levels to obtain deconstruction results; the deconstruction results include the design elements deconstructed at each of the multiple deconstruction levels.

[0009] Based on the design elements in the deconstruction results and the relationship between different design elements in the deconstruction results at the deconstruction level, a design element space of the Kansei Engineering knowledge graph is constructed.

[0010] Based on the ability of product images of the target product category to represent the multiple core emotional terms, a first set of emotional terms and a second set of emotional terms are constructed based on the multiple core emotional terms; wherein, the emotional terms in the first set of emotional terms have the ability to represent one or more emotional terms in the second set of emotional terms.

[0011] Based on the first set of sensory vocabulary, the second set of sensory vocabulary, and the relationship between the sensory vocabulary in the first set of sensory vocabulary and the sensory vocabulary in the second set of sensory vocabulary, a sensory image space of the sensory engineering knowledge graph is constructed.

[0012] Based on the correlation between the design elements in the design element space and the sensory vocabulary in the sensory image space, a connection is constructed between the design element space and the sensory image space. Based on the design element space, the sensory image space, and the connection between the design element space and the sensory image space, a sensory engineering knowledge graph of the target product category is obtained.

[0013] In one optional implementation, the step of constructing a first set of emotional terms and a second set of emotional terms based on the expressive power of product images of the target product category for the multiple core emotional terms includes:

[0014] Based on the product images of the target product category and the multiple core emotional terms, a first semantic difference scale is constructed; the first semantic difference scale is used to collect first evaluation results on the performance of the product images of the target product category in relation to the multiple core emotional terms;

[0015] Cluster analysis is performed based on multiple first evaluation results to obtain N groups of emotional vocabulary groups; where N is an integer greater than 1.

[0016] For each group of sensory vocabulary, a representative sensory vocabulary is determined, or a representative sensory vocabulary and at least one secondary sensory vocabulary are determined; wherein, the relationship between the representative sensory vocabulary and the secondary sensory vocabulary determined for the same group of sensory vocabulary is a representative relationship;

[0017] The first set of emotional vocabulary is constructed using representative emotional vocabulary determined for each of the N sets of emotional vocabulary, and the second set of emotional vocabulary is constructed using secondary emotional vocabulary determined for each of the N sets of emotional vocabulary.

[0018] In one optional implementation, for a target emotional vocabulary group among the N groups of emotional vocabulary groups, a representative emotional vocabulary is determined for the target emotional vocabulary group, or a representative emotional vocabulary and at least one secondary emotional vocabulary are determined, including:

[0019] In the target emotional vocabulary group, a core emotional vocabulary word that encompasses the semantics of a predetermined proportion or more of the core emotional vocabulary words in the target emotional vocabulary group is determined as the representative emotional vocabulary word corresponding to the target emotional vocabulary group, and the other words in the target emotional vocabulary group other than the representative emotional vocabulary word are designated as the secondary emotional vocabulary words corresponding to the target emotional vocabulary group; or,

[0020] Based on the semantics of each core emotional word in the target emotional vocabulary group, an emotional word that does not belong to the target emotional vocabulary group is determined as the representative emotional word corresponding to the target emotional vocabulary group, and each core emotional word in the target emotional vocabulary group is taken as the secondary emotional word corresponding to the target emotional vocabulary group.

[0021] In one alternative implementation, the method further includes:

[0022] For each group of emotional vocabulary, representative and secondary emotional vocabulary words are collected from the corpus, and synonyms not included in the core emotional vocabulary corpus are collected as supplementary emotional vocabulary words. Among them, the relationship between the synonyms collected based on the representative emotional vocabulary words and the representative emotional vocabulary words is a similarity relationship, and the relationship between the synonyms collected based on the secondary emotional vocabulary words and the secondary emotional vocabulary words is a similarity relationship.

[0023] Construct a third set of sensory vocabulary based on the collected supplementary sensory vocabulary;

[0024] The construction of the sensory image space of the sensory engineering knowledge graph based on the first sensory vocabulary set, the second sensory vocabulary set, and the association between sensory words in the first sensory vocabulary set and sensory words in the second sensory vocabulary set includes:

[0025] Construct representative sensory word entities one by one based on the names of representative sensory words in the first sensory word set; construct secondary sensory word entities one by one based on the names of secondary sensory words in the second sensory word set; and construct supplementary sensory word entities one by one based on the names of supplementary sensory words in the third sensory word set.

[0026] Based on the representational relationship between representative and secondary sensory words, construct the representational relationship between the corresponding representative sensory word entities and secondary sensory word entities; based on the similarity relationship between representative and supplementary sensory words, construct the similarity relationship between the corresponding representative sensory word entities and supplementary sensory word entities; based on the similarity relationship between secondary and supplementary sensory words, construct the similarity relationship between the corresponding secondary sensory word entities and supplementary sensory word entities.

[0027] In one optional implementation, before constructing the connection between the design element space and the sensory image space based on the relevance between design elements in the design element space and sensory words in the sensory image space, the method further includes:

[0028] The product images of the target product category are encoded based on the deconstruction results to obtain an x-dimensional encoding vector. The deconstruction results include x design elements, and the x-dimensional vector corresponds one-to-one with the x design elements. If the value of the x-dimensional encoding vector in the i-th dimension is the first value, it means that the encoded product image reflects the design element corresponding to the i-th dimension. If the value of the x-dimensional encoding vector in the i-th dimension is the second value, it means that the encoded product image does not reflect the design element corresponding to the i-th dimension.

[0029] Based on the encoded product images and the sensory words in the first sensory word set, a second semantic difference scale is constructed; the second semantic difference scale is used to collect second evaluation results on the ability of product images to represent the sensory words in the first sensory word set;

[0030] Based on the second evaluation result and the x-dimensional encoding vector, a multiple linear regression analysis was performed to obtain a regression equation with the emotional words in the first set of emotional words as the dependent variable and x design elements as independent variables.

[0031] Extract the regression coefficients corresponding to the x design elements that serve as independent variables in the regression equation; the magnitude of the regression coefficients reflects the correlation between the corresponding design elements and the emotional vocabulary.

[0032] In one optional implementation, constructing the connection between the design element space and the sensory image space based on the relevance between design elements in the design element space and sensory words in the sensory image space includes:

[0033] For a target emotional word in the first set of emotional words, the design element corresponding to the regression coefficient that is greater than or equal to a preset correlation threshold in the regression equation with the target emotional word as the dependent variable and x design elements as independent variables is determined as the relevant design element of the target emotional word; the target emotional word is any emotional word in the first set of emotional words.

[0034] Based on the relevant design elements of each sensory word in the first set of sensory words, the connection between the design element space and the sensory image space is constructed.

[0035] A second aspect of this application provides a retrieval method based on a knowledge graph of kanji engineering, the method comprising:

[0036] Receive search keywords and search purpose information provided by the user;

[0037] Based on the search keywords, sensory words are matched in the sensory image space of the sensory engineering knowledge graph; the sensory engineering knowledge graph is constructed according to the method provided in the first aspect;

[0038] In the aforementioned Kansei Engineering knowledge graph, design elements associated with the matched Kansei terms are identified;

[0039] Based on the identified design elements, search results matching the search objective are provided to the user.

[0040] In one optional implementation, if the search objective indicated by the search objective information is a design element, the step of providing the user with search results matching the search objective information based on the determined design element includes:

[0041] The identified design elements will be provided to users as search results.

[0042] In one optional implementation, if the search objective information indicates a product image as the search objective, providing the user with search results matching the search objective information based on the determined design elements includes:

[0043] The query can deconstruct individual products containing the identified design elements;

[0044] The product images of the individual products retrieved will be provided to the user as search results.

[0045] In one optional implementation, the sensory image space is constructed based on the names and relationships between the representative sensory words in the first sensory vocabulary set, the secondary sensory words in the second sensory vocabulary set, and the supplementary sensory words in the third sensory vocabulary set; wherein, the supplementary sensory words in the third sensory vocabulary set are obtained by searching for synonyms based on the representative sensory words in the first sensory vocabulary set and the secondary sensory words in the second sensory vocabulary set.

[0046] The process of matching sensory words in the sensory image space of the sensory engineering knowledge graph based on the search keywords includes:

[0047] Based on the search keywords, representative emotional words are first matched in the emotional image space of the emotional engineering knowledge graph. If a representative emotional word is successfully matched, the matching of emotional words for the search keywords in the emotional image space is stopped.

[0048] If a match for a representative emotional word fails, the matching of secondary emotional words continues in the emotional image space of the emotional engineering knowledge graph. If a secondary emotional word is successfully matched, the matching of emotional words for the search keyword in the emotional image space is stopped.

[0049] If a match for a secondary sensory term fails, the matching process continues in the sensory imagery space of the sensory engineering knowledge graph to find supplementary sensory terms. If a supplementary sensory term is successfully matched, the matching process stops in the sensory imagery space for the search keywords.

[0050] A third aspect of this application provides an apparatus for constructing a knowledge graph of sensible engineering, the apparatus comprising:

[0051] The vocabulary acquisition module is used to acquire the core emotional vocabulary of the target product category. The core emotional vocabulary includes multiple core emotional words, which are emotional words that have a positive impact on consumers' behavior of purchasing products of the target product category.

[0052] The design element deconstruction module is used to deconstruct the design elements of the target product category according to multiple preset deconstruction levels to obtain deconstruction results; the deconstruction results include the design elements deconstructed at each of the multiple deconstruction levels.

[0053] The design element space construction module is used to construct the design element space of the Kansei Engineering knowledge graph based on the design elements in the deconstruction results and the relationship between different design elements in the deconstruction results at the deconstruction level.

[0054] The emotional vocabulary evaluation module is used to construct a first emotional vocabulary set and a second emotional vocabulary set based on the performance of product images of the target product category on multiple core emotional vocabulary words; wherein, the emotional vocabulary words in the first emotional vocabulary set have representativeness for one or more emotional vocabulary words in the second emotional vocabulary set.

[0055] The sensory imagery space construction module is used to construct the sensory imagery space of the sensory engineering knowledge graph based on the first sensory vocabulary set, the second sensory vocabulary set, and the association between the sensory vocabulary in the first sensory vocabulary set and the sensory vocabulary in the second sensory vocabulary set.

[0056] The spatial association module is used to construct a connection between the design element space and the sensory image space based on the correlation between the design elements in the design element space and the sensory vocabulary in the sensory image space, and to obtain a sensory engineering knowledge graph of the target product category based on the design element space, the sensory image space, and the connection between the design element space and the sensory image space.

[0057] In one optional implementation, the sensory vocabulary evaluation module includes:

[0058] The scale construction unit is used to construct a first semantic difference scale based on the product images of the target product category and the multiple core emotional words; the first semantic difference scale is used to collect first evaluation results on the performance ability of the product images of the target product category to the multiple core emotional words;

[0059] The clustering analysis unit is used to perform clustering analysis based on multiple first evaluation results to obtain N groups of sensory vocabulary groups; where N is an integer greater than 1.

[0060] The vocabulary type determination unit is used to determine a representative emotional word for each group of emotional vocabulary, or to determine a representative emotional word and at least one secondary emotional word; wherein, the relationship between the representative emotional word and the secondary emotional word determined for the same group of emotional vocabulary is a representative relationship.

[0061] The vocabulary set construction unit is used to construct the first emotional vocabulary set using representative emotional words determined for each of the N emotional vocabulary groups, and to construct the second emotional vocabulary set using secondary emotional words determined for each of the N emotional vocabulary groups.

[0062] In one optional implementation, for the target lexical group among the N groups of sensual lexical groups, the lexical type determination unit is specifically used for:

[0063] In the target emotional vocabulary group, a core emotional vocabulary word that encompasses the semantics of a predetermined proportion or more of the core emotional vocabulary words in the target emotional vocabulary group is determined as the representative emotional vocabulary word corresponding to the target emotional vocabulary group, and the other words in the target emotional vocabulary group other than the representative emotional vocabulary word are designated as the secondary emotional vocabulary words corresponding to the target emotional vocabulary group; or,

[0064] Based on the semantics of each core emotional word in the target emotional vocabulary group, an emotional word that does not belong to the target emotional vocabulary group is determined as the representative emotional word corresponding to the target emotional vocabulary group, and each core emotional word in the target emotional vocabulary group is taken as the secondary emotional word corresponding to the target emotional vocabulary group.

[0065] In one alternative implementation, the apparatus further includes:

[0066] The vocabulary supplementation module is used to collect synonyms not included in the core vocabulary database from the corpus for the representative and secondary emotional vocabulary of each emotional vocabulary group, as supplementary emotional vocabulary; wherein, the relationship between the synonyms collected based on the representative emotional vocabulary and the representative emotional vocabulary is a similarity relationship, and the relationship between the synonyms collected based on the secondary emotional vocabulary and the secondary emotional vocabulary is a similarity relationship.

[0067] The vocabulary set construction module is used to construct a third set of sensory vocabulary based on the collected supplementary sensory vocabulary;

[0068] The sensory imagery space construction module includes:

[0069] The vocabulary entity construction unit is used to construct representative emotional vocabulary entities one by one according to the names of representative emotional vocabulary in the first set of emotional vocabulary, construct secondary emotional vocabulary entities one by one according to the names of secondary emotional vocabulary in the second set of emotional vocabulary, and construct supplementary emotional vocabulary entities one by one according to the names of supplementary emotional vocabulary in the third set of emotional vocabulary.

[0070] The entity relation construction unit is used to construct the representation relationship between the representative emotional word entity and the secondary emotional word entity based on the representation relationship between the representative emotional word and the secondary emotional word; to construct the similarity relationship between the representative emotional word and the supplementary emotional word entity based on the similarity relationship between the representative emotional word and the supplementary emotional word; and to construct the similarity relationship between the secondary emotional word and the supplementary emotional word entity based on the similarity relationship between the secondary emotional word and the supplementary emotional word.

[0071] In one alternative implementation, the apparatus further includes:

[0072] The image encoding module is used to encode product images of the target product category based on the deconstruction result, obtaining an x-dimensional encoding vector. The deconstruction result includes x design elements, and the x-dimensional vector corresponds one-to-one with the x design elements. If the value of the x-dimensional encoding vector in the i-th dimension is a first value, it means that the encoded product image reflects the design element corresponding to the i-th dimension. If the value of the x-dimensional encoding vector in the i-th dimension is a second value, it means that the encoded product image does not reflect the design element corresponding to the i-th dimension.

[0073] The scale construction module is used to construct a second semantic difference scale based on the encoded product image and the sensory words in the first sensory word set; the second semantic difference scale is used to collect a second evaluation result on the ability of the product image to express the sensory words in the first sensory word set;

[0074] The regression analysis module is used to perform multiple linear regression analysis based on the second evaluation result and the x-dimensional encoding vector to obtain a regression equation with the emotional words in the first set of emotional words as the dependent variable and x design elements as independent variables.

[0075] The regression coefficient extraction module is used to extract the regression coefficients corresponding to the x design elements that are independent variables in the regression equation; the magnitude of the regression coefficients reflects the correlation between the corresponding design elements and the emotional vocabulary.

[0076] In one optional implementation, the spatial association module includes:

[0077] The relevant design element determination unit is used to determine, for the target emotional word in the first set of emotional words, the design element corresponding to the regression coefficient that is greater than or equal to a preset correlation threshold in the regression equation with the target emotional word as the dependent variable and x design elements as independent variables, as the relevant design element of the target emotional word; the target emotional word is any emotional word in the first set of emotional words;

[0078] The spatial connection construction unit is used to construct the connection between the design element space and the sensory image space based on the relevant design elements of each sensory word in the determined first set of sensory words.

[0079] A fourth aspect of this application provides a retrieval device based on a knowledge graph of kinetic engineering, the device comprising:

[0080] The information receiving module is used to receive search keywords and search purpose information provided by the user;

[0081] The vocabulary matching module is used to match sensory words in the sensory image space of the sensory engineering knowledge graph based on the search keywords; the sensory engineering knowledge graph is constructed according to the construction method of sensory engineering knowledge graph;

[0082] The design element determination module is used to determine the design elements associated with the matched sensory vocabulary in the sensory engineering knowledge graph.

[0083] The search results providing module is used to provide users with search results that match the search objective information based on the determined design elements.

[0084] In one optional implementation, if the search objective indicated by the search objective information is a design element, the search result providing module is specifically used for:

[0085] The identified design elements will be provided to users as search results.

[0086] In one optional implementation, if the search purpose indicated by the search purpose information is product images, the search result providing module is specifically used for:

[0087] The query can deconstruct individual products containing the identified design elements;

[0088] The product images of the individual products retrieved will be provided to the user as search results.

[0089] In one optional implementation, the sensory image space is constructed based on the names and relationships between the representative sensory words in the first sensory vocabulary set, the secondary sensory words in the second sensory vocabulary set, and the supplementary sensory words in the third sensory vocabulary set; wherein, the supplementary sensory words in the third sensory vocabulary set are obtained by searching for synonyms based on the representative sensory words in the first sensory vocabulary set and the secondary sensory words in the second sensory vocabulary set.

[0090] The vocabulary matching module includes: a first matching unit, a second matching unit, and a third matching unit;

[0091] The first matching unit is used to first match representative emotional words in the emotional image space of the emotional engineering knowledge graph based on the search keywords. If the first matching unit successfully matches a representative emotional word, the word matching module stops matching emotional words for the search keywords in the emotional image space.

[0092] The second matching unit is used to continue matching secondary sensory words in the sensory image space of the sensory engineering knowledge graph if the matching of representative sensory words fails. If the second matching unit successfully matches a secondary sensory word, the word matching module stops matching sensory words for the search keyword in the sensory image space.

[0093] The third matching unit is used to continue matching supplementary sensory words in the sensory image space of the sensory engineering knowledge graph if the matching of secondary sensory words fails. If the third matching unit successfully matches supplementary sensory words, the word matching module stops matching sensory words for the search keywords in the sensory image space.

[0094] Compared with the prior art, this application has the following beneficial effects:

[0095] This application proposes establishing a design element space and a sensory image space, and provides specific implementation methods for establishing both. Based on the connection between the design element space and the sensory image space, the two are organically combined to construct a sensory ergonomics knowledge graph. Sensory ergonomics knowledge is described using triples of entities and relationships between entities. Compared to related technologies, this application's technical solution can represent sensory ergonomics knowledge in a simpler and easier-to-implement way, thereby achieving effective management of sensory ergonomics knowledge. The technical solution provided in this application, by constructing a sensory ergonomics knowledge graph, helps to make sensory ergonomics knowledge easier to apply. Furthermore, knowledge graphs are a machine- and human-understandable means of knowledge representation. By utilizing the scalability and interpretable reasoning capabilities of knowledge graphs, it can better respond to the retrieval needs of designers or consumers and meet the needs for the continuous enrichment and improvement of sensory ergonomics knowledge. Attached Figure Description

[0096] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0097] Figure 1 A flowchart illustrating a method for constructing a knowledge graph of kinesiology provided in this application embodiment;

[0098] Figure 2 A structural schematic diagram of a design element space provided in an embodiment of this application;

[0099] Figure 3A schematic diagram illustrating the determination of representative sensory words and secondary sensory words in an embodiment of this application;

[0100] Figure 4 A schematic diagram illustrating another method for determining representative sensory words and secondary sensory words, provided for an embodiment of this application;

[0101] Figure 5 A schematic diagram of the structure of a sensory image space provided in an embodiment of this application;

[0102] Figure 6 A flowchart illustrating a retrieval method based on a knowledge graph of kinetic engineering, provided for an embodiment of this application;

[0103] Figure 7 A schematic diagram of the structure of a device for constructing a sensory engineering knowledge graph provided in an embodiment of this application;

[0104] Figure 8 This is a schematic diagram of the structure of a retrieval device based on a knowledge graph of kinetic engineering, provided in an embodiment of this application. Detailed Implementation

[0105] As described earlier, related technologies lack effective methods for representing and managing Kansei Engineering knowledge. Furthermore, due to the complexity, ambiguity, and uncertainty of human sensory experience, the knowledge structure of Kansei Engineering is also complex, making its representation difficult. For example, one related technology discloses a method for establishing a Kansei Engineering product knowledge base, using the Function-Behavior-Structure (FBS) model as theoretical support to deconstruct the product under study at three levels: function, behavior, and structure, to obtain the intra-level connections between each level; and to establish inter-level connections between the product's function, behavior, and structure to obtain the Kansei Engineering knowledge base. However, this technology does not provide specific instructions on the internal structure of the sensory image space, nor does it clarify the specific implementation method for constructing the Kansei Engineering knowledge base. In summary, although the internal structure of the design element space is deeply deconstructed based on the FBS model, the internal structure of the sensory image space and the connections between the two spaces are not sufficiently explained. Moreover, no reasonable solution is provided for the retrieval failures that may result from the ambiguity in the expressions of designers and consumers. Furthermore, this technology only mentions using the Python programming language to build a knowledge base for Kansei Engineering, without providing any details about the more specific code or execution methods.

[0106] This application aims to provide a means for effectively representing and managing sensual engineering knowledge based on knowledge graphs. It not only outlines the construction methods and internal structures of the design element space and the sensual image space, but also elucidates the connection between them. Based on this, it clarifies the specific implementation method for constructing a sensual engineering knowledge base using knowledge graph tools, thereby assisting designers or consumers in efficiently handling sensual needs and providing search results.

[0107] To facilitate understanding, before formally introducing the technical solution, we will first explain some technical terms that may appear in this article:

[0108] Knowledge graph: A knowledge representation and management technology that represents real-world concepts and the relationships between them in the form of triples (entity, relation, entity).

[0109] Kansei engineering is a design methodology that establishes a connection between consumers' emotional perception of a product and the product's design elements, thereby assisting in the design development or evaluation process.

[0110] Design elements: External design features that, after deconstructing a product, are obtained at different levels and can influence the consumer's emotional experience.

[0111] Kansei image: A certain emotional experience evoked by a product design element, such as the "rustic" feeling evoked by the wooden casing of a speaker.

[0112] Kansei word: A verbal expression of a certain emotional experience evoked by product design elements, such as the word "rustic" mentioned above.

[0113] Kansei Engineering Knowledge Base: Various forms of knowledge bases used to represent, store, and retrieve Kansei Engineering knowledge.

[0114] Kansei Engineering Knowledge Graph: A knowledge base for Kansei Engineering built upon knowledge graph tools.

[0115] Sensory Image Space: The collection of all sensory terms in the sensory engineering knowledge graph. This invention involves three types of sensory image entities: "representative sensory terms", "secondary sensory terms", and "supplementary sensory terms".

[0116] Design Element Space: The collection of all design elements in the Kansei Engineering Knowledge Graph. This invention involves four types of design element entities: “components”, “modules”, “attributes”, and “levels”.

[0117] Figure 1This is a flowchart illustrating a method for constructing a sensory engineering knowledge graph, as provided in an embodiment of this application. Figure 1 As shown, the construction method of the Kansei Engineering knowledge graph includes the following steps:

[0118] S101. Obtain a core vocabulary database of the target product category.

[0119] In this embodiment, when constructing a Kansei Engineering knowledge graph, a product category is used as the granularity for constructing the Kansei Engineering knowledge graph. For example, for product category A, a corresponding Kansei Engineering knowledge graph a is constructed; for product category B, a corresponding Kansei Engineering knowledge graph b is constructed. In this embodiment, the construction process of the Kansei Engineering knowledge graph is described using a target product category as an example. The target product category can be one of many product categories for which a Kansei Engineering knowledge graph needs to be constructed. For example, the target product category can be furniture.

[0120] In this embodiment, each product category can have a corresponding core emotional vocabulary library. Taking the target product category as an example, its core emotional vocabulary library includes multiple core emotional terms. These core emotional terms are emotional terms that positively influence consumers' behavior in purchasing products of the target product category. In other words, the core emotional vocabulary library of the target product category is constructed based on an analysis of its influence on purchasing products of that target product category.

[0121] In one alternative implementation, for the target product category, a wide range of emotional terms can be collected through methods such as desk research, expert interviews, online questionnaires, or web crawling. Then, consumer surveys or expert methods are used to screen and retain 15-20 emotional terms that are relatively important to the target product category. The criterion for selection is that the emotional imagery represented by the emotional term will influence consumer purchasing behavior. These retained emotional terms are called core emotional terms and are added to the core emotional term library for the target product category.

[0122] S102. Deconstruct the design elements of the target product category according to multiple preset deconstruction levels to obtain the deconstruction results.

[0123] In this embodiment, the structural hierarchy involved is divided into four deconstruction levels, from coarse to fine and from shallow to deep: components, modules, attributes, and levels. The deconstruction method is as follows:

[0124] 1) Based on the principle of minimum functionality, the product is first deconstructed into "components" that achieve different functions. The "components" can be large or small, but each "component" is required to achieve a minimum functional unit facing the consumer, such as the handlebar assembly of a bicycle.

[0125] 2) Deconstruct the "component" into different "modules". The characteristic of a "module" is that it is physically indivisible to the consumer, such as the handlebar grip module of a handlebar assembly.

[0126] 3) Deconstruct the “module” into different “attributes”. “Attributes” are not concrete substances, but rather a certain external characteristic of a module, such as the hardness of a sleeve module.

[0127] 4) Deconstruct the "attribute" into different "levels". Treat a certain "attribute" as a variable, then the "level" refers to the different values ​​of that variable. For example, the hardness of a sleeve includes multiple "levels" such as hard, medium, and soft. It should be noted that the "level" does not need to exhaust all possibilities, but only needs to include the common "levels" involved in this category of products on the market.

[0128] The deconstruction of design elements for a product category should include multiple deconstruction paths. A complete deconstruction path is "component" - "module" - "attribute" - "level".

[0129] The results obtained from deconstruction at any of the deconstruction levels—"component," "module," "attribute," and "level"—can be collectively referred to as design elements. The deconstruction results include design elements deconstructed at multiple levels. In practice, each deconstruction path can stop at any of the "component," "module," "attribute," and "level" levels, without necessarily deconstructing to the "level."

[0130] Figure 2 This is a structural schematic diagram of a design element space provided in an embodiment of this application. For example... Figure 2 As shown, following the preset four deconstruction levels of "components," "modules," "attributes," and "levels," the product design elements of the target product category are structured. The deconstruction at the "component" level yields the following results: Figure 2 The shown components 1 through 3 comprise three design elements. Further, components 1 and 3 are deconstructed at the "module" level, yielding modules 1 through 3, and modules 4 through 5. Further, modules 2, 4, and 5 are deconstructed at the "attribute" level, yielding attributes 1 through 2, 3 through 4, and 5. Further, attributes 1 and 2 are deconstructed at the "level" level, yielding levels 1 through 3, and 4 through 5.

[0131] Combination Figure 2It can be seen that component 2 is deconstructed to the "component" level and then deconstructs; components 1 and 3 are deconstructed to the "module" level; modules 1 and 3 are deconstructed to the "module" level and then deconstructs; modules 2, 4, and 5 are deconstructed to the "attribute" level. Attributes 3 through 7 are deconstructed to the "attribute" level and then deconstructed; attributes 1 and 2 are further deconstructed to the "horizontal" level. The above example only uses four deconstruction levels. In practical applications, depending on the diverse settings of the deconstruction levels, various deconstruction effects can be achieved, resulting in other possible deconstruction results. This embodiment does not limit this. Furthermore, in Figure 2 The naming of Component 1 to Component 3, Module 1 to Module 5, Attribute 1 to Attribute 7, and Level 1 to Level 5 in this document is only an example. In actual applications, the naming of design elements at each level in the deconstruction result will conform to the specific scenario and situation.

[0132] S103. Based on the design elements in the deconstruction results and the relationship between different design elements in the deconstruction results at the deconstruction level, construct the design element space of the Kansei Engineering knowledge graph.

[0133] In practical applications, a design element space for a sensory engineering knowledge graph can be built on the Neo4j platform. Neo4j is an open-source NoSQL graph database with its proprietary query language, Cypher, which is intuitive and efficient. First, design element entities are constructed in the Neo4j platform. As illustrated in the previous example, the design elements are deconstructed using four levels: components, modules, attributes, and levels. The design element space contains four entity types: "components," "modules," "attributes," and "levels."

[0134] (1) The name of the “component” design element that needs to be constructed into the corresponding entity in the graph is denoted as component, and the corresponding entity code is denoted as n;

[0135] (2) The name of the “module” design element that needs to be constructed into the corresponding entity in the graph is denoted as module, and the corresponding entity code is denoted as m;

[0136] (3) The name of the “attribute” design element that needs to be constructed into the corresponding entity in the map is denoted as attribute, and the corresponding entity code is denoted as p;

[0137] (4) The name of the “horizontal” design element that needs to be constructed in the map is denoted as level, and the corresponding entity code is denoted as q.

[0138] In the Neo4j platform, the Cypher language is used to perform specific operations. This step uses four types of entity construction statements for design elements, listed from top to bottom: statements for constructing "component" entities, statements for constructing "module" entities, statements for constructing "attribute" entities, and statements for constructing "horizontal" entities.

[0139] CREATE (n:component{name: 'component'});

[0140] CREATE (m:module{name: 'module'});

[0141] CREATE (p:attribute{name: 'attribute'});

[0142] CREATE (q:level{name: 'level'}).

[0143] Next, relationships between design element entities are constructed in the Neo4j platform. The relationships between "component" and "module," "module" and "attribute," and "attribute" and "level" entities are all "containment." This is because dividing design elements into four levels is merely for a more detailed and comprehensive deconstruction of product categories; however, in terms of their connection to the sensory imagery space, these four levels are equally important and functional. The specific relationship construction statements are as follows:

[0144] (1) The statement describing the containment relationship between the "component" entity and the "module" entity is:

[0145] MATCH (n: component {name: 'component'}), (m: module {name: 'module'}), CREATE(n) - [r: contain] -> (m)

[0146] (2) The statement describing the containment relationship between the "module" entity and the "attribute" entity is:

[0147] MATCH (m: module {name: 'module'}), (p: attribute {name: 'attribute'}), CREATE(m) - [r: contain] -> (p)

[0148] (3) The statement describing the containment relationship between the "attribute" entity and the "level" entity is:

[0149] MATCH (p:attribute{name: 'attribute'}), (q:level{name: 'level'}), CREATE(p) - [r:contains] -> (q)

[0150] By executing steps S102 and S103 above, the design element deconstruction and design element space construction are completed sequentially. In practical applications, step S101 can be executed first, followed by steps S102 and S103, or steps S102 and S103 (deconstruction and design element space construction) can be executed first, followed by step S101 (obtaining the core sensory vocabulary). The order of execution for steps S101 and S102 and S103 is not limited here.

[0151] Based on the core sensory vocabulary database obtained in step S101 above, steps S104-S105 below will specifically introduce the implementation process of constructing the sensory imagery space.

[0152] S104. Based on the ability of the product images of the target product category to express the multiple core emotional terms, construct a first set of emotional terms and a second set of emotional terms based on the multiple core emotional terms.

[0153] In step S101, a core emotional vocabulary database for the target product category was obtained, containing multiple core emotional terms for the target product category. In step S104, the core emotional terms are categorized based on the expressive power of product images of the target product category, thereby constructing a first emotional vocabulary set and a second emotional vocabulary set. The emotional terms in the first emotional vocabulary set are representative of one or more emotional terms in the second emotional vocabulary set.

[0154] The following is an example:

[0155] In the technical solution of this application, in order to construct the first set of sensory vocabulary and the second set of sensory vocabulary, the following steps S1041~S1044 can be performed (Note: S1041~S1044 are not shown in the figure):

[0156] S1041. Construct a first semantic difference scale based on the product images of the target product category and the multiple core perceptual terms.

[0157] Product images for the target product category can be collected from e-commerce platforms. Generally, there are no strict requirements on the number of images collected at this stage, and a further screening process can be performed after collection. The goal is to minimize the number of images after screening, while still covering all design elements derived from the S102 deconstruction. Screening as few images as possible reduces the workload of distributing semantic difference scales. Covering all design elements ensures that all design elements are considered during the subsequent relationship establishment process. Otherwise, the influence of some design elements on sensory vocabulary may not be reflected in the regression equation. Both image collection and screening processes can be performed by design experts. Alternatively, intelligent technologies can be used to complete the collection and screening steps.

[0158] In this embodiment, a semantic difference scale can be compiled based on the product images of the selected target product categories and the core emotional vocabulary in the core emotional vocabulary database obtained in S101. For example, an odd-point Likert scale can be used. This semantic difference scale is referred to herein as the first semantic difference scale. The first semantic difference scale can be distributed in the form of an electronic questionnaire, thereby facilitating the collection of first evaluation results regarding the performance ability of the product images of the target product categories to the multiple core emotional vocabulary. Recipients of the electronic questionnaire can use the first semantic difference scale to score the performance ability of the product images to the core emotional vocabulary, and the scoring results serve as the first evaluation results. For example, product image A has a strong performance ability to express core emotional vocabulary a, product image A cannot express core emotional vocabulary b, and product image B has the ability to express core emotional vocabulary b, but the performance ability is weak. All of these can be evaluated using the first semantic difference scale.

[0159] S1042. Based on multiple first evaluation results, perform cluster analysis to obtain N groups of emotional vocabulary.

[0160] In practical applications, SPSS software can be used to perform cluster analysis on multiple first-evaluation results collected from the first semantic difference scale questionnaire. For example, cluster analysis can yield 4-5 groups of sensory vocabulary. Here, N is an integer greater than 1; N=4 or N=5 are merely examples. The purpose of cluster analysis is to categorize core sensory vocabulary in the core sensory vocabulary corpus based on similarities and distinguish between those based on differences. SPSS software can be replaced by any existing or future platform or tool available for cluster analysis.

[0161] S1043. For each group of sensory words, determine a representative sensory word, or determine a representative sensory word and at least one secondary sensory word.

[0162] There may exist groups of sensory words containing only one sensory word, meaning that during clustering, a certain category contains only one sensory word, and in this case, there are no secondary sensory words within that group. For groups of sensory words with multiple sensory words, a representative sensory word and at least one secondary sensory word can be identified. The relationship between the representative sensory word and the secondary sensory words identified for the same group of sensory words is called a representative relationship. The representative sensory word should be able to encompass the semantics of most of the sensory words in that group, and the representative sensory words from different groups should have clear distinguishability. Other words in a group of sensory words that are not used as representative sensory words are considered secondary sensory words.

[0163] S1043 has two possible ways of determining the representative emotional vocabulary: one is to select a representative emotional vocabulary from within the group of emotional vocabulary; the other is to use a new emotional vocabulary obtained through summarization and induction based on the group of emotional vocabulary as the representative emotional vocabulary of that group. The following describes the two methods of determining the representative emotional vocabulary, using the target group of emotional vocabulary as the object of description.

[0164] (1) In the target emotional vocabulary group, a core emotional vocabulary that covers the semantics of the core emotional vocabulary in the target emotional vocabulary group at a predetermined proportion or more is determined as the representative emotional vocabulary corresponding to the target emotional vocabulary group, and the other words in the target emotional vocabulary group other than the representative emotional vocabulary are taken as the secondary emotional vocabulary corresponding to the target emotional vocabulary group.

[0165] (2) Based on the semantics of each core emotional word in the target emotional word group, determine an emotional word that does not belong to the target emotional word group as the representative emotional word corresponding to the target emotional word group, and take each core emotional word in the target emotional word group as the secondary emotional word corresponding to the target emotional word group.

[0166] Figure 3 This is a schematic diagram of determining representative sensory words and secondary sensory words according to an embodiment of this application, corresponding to the implementation of the above-described (1) scheme. Figure 4 This diagram illustrates another method for determining representative and secondary sensory words in an embodiment of this application, corresponding to the implementation of the above-described (2) scheme. It should be noted that in the above-described (2) implementation scheme, although the determined representative sensory words do not belong to the sensory word group, they are required to cover the semantics of the core sensory words in the sensory word group at a predetermined proportion or higher. The predetermined proportion can be customized; for example, it can be 70%.

[0167] S1044. Construct the first set of sensory words using the representative sensory words determined for each of the N sets of sensory words, and construct the second set of sensory words using the secondary sensory words determined for each of the N sets of sensory words.

[0168] In practical applications, the representative sensory words identified by each of the N sensory word groups can be added to a single set, called the first sensory word set, which contains the representative sensory words identified by each sensory word group. Conversely, the secondary sensory words from each of the N sensory word groups can be added to another set, called the second sensory word set, which contains the secondary sensory words identified by each sensory word group.

[0169] S105. Based on the first set of sensory vocabulary, the second set of sensory vocabulary, and the association between the sensory vocabulary in the first set of sensory vocabulary and the sensory vocabulary in the second set of sensory vocabulary, construct the sensory image space of the sensory engineering knowledge graph.

[0170] In practical applications, a sensory imagery space can be constructed within the Neo4j platform. First, sensory imagery entities are constructed within Neo4j. This space must contain at least two entity types: "representative sensory words" and "secondary sensory words." The name of the "representative sensory word" for which an entity needs to be constructed in the graph is denoted as `representative`, and its corresponding entity code is 'a'. The name of the "secondary sensory word" for which an entity needs to be constructed in the graph is denoted as `subordinate`, and its corresponding entity code is denoted as `b`. The statement for constructing entities in the sensory imagery space is as follows:

[0171] CREATE (a: represents a subjective term {name: 'representative'})

[0172] CREATE (b:subordinate)

[0173] Next, relationships between the sensory image entities are constructed in the Neo4j platform. The relationship between the "representative sensory vocabulary" and the "secondary sensory vocabulary" entities is "representative". The specific relationship construction statement is as follows:

[0174] MATCH (a:representative) {name: 'representative'}, (b:subordinate) {name: 'subordinate'}, CREATE (a) - [r:representative] -> (b)

[0175] After completing steps S104-S105 above, the construction of the sensory imagery space is finished. Since the design element space and sensory imagery space have been completed, we are only "one step away" from building a complete sensory engineering knowledge graph for this target product category. This involves establishing the connection between the design element space and the sensory imagery space. By building a bridge between these two different spaces, we can then achieve effective representation, management, and retrieval of sensory engineering knowledge based on the knowledge graph.

[0176] S106. Based on the correlation between the design elements in the design element space and the sensory vocabulary in the sensory image space, construct the connection between the design element space and the sensory image space, and based on the design element space, the sensory image space, and the connection between the design element space and the sensory image space, obtain the sensory engineering knowledge graph of the target product category.

[0177] As previously introduced, the various design elements in the design element space are derived from the deconstruction of design elements of the target product category. Similarly, the various sensory terms in the sensory imagery space are obtained by evaluating and analyzing the expressive power of different core sensory terms through product images of the target product category. Therefore, both the design elements in the design element space and the sensory terms in the sensory imagery space are associated with products or product images of the target product category. If the correlation between the design elements in the design element space and the sensory terms in the sensory imagery space is analyzed, the connection between the design element space and the sensory imagery space can be naturally constructed. Based on the design element space, the sensory imagery space, and the connection between the design element space and the sensory imagery space, a sensory engineering knowledge graph of the target product category can be obtained.

[0178] This application proposes establishing a design element space and a sensory image space, and provides specific implementation methods for establishing both. Based on the connection between the design element space and the sensory image space, the two are organically combined to construct a sensory ergonomics knowledge graph. Sensory ergonomics knowledge is described using triples of entities and relationships between entities. Compared to related technologies, this application's technical solution can represent sensory ergonomics knowledge in a simpler and easier-to-implement way, thereby achieving effective management of sensory ergonomics knowledge. The technical solution provided in this application, by constructing a sensory ergonomics knowledge graph, helps to make sensory ergonomics knowledge easier to apply. Furthermore, knowledge graphs are a machine- and human-understandable means of knowledge representation. By utilizing the scalability and interpretable reasoning capabilities of knowledge graphs, it can better respond to the retrieval needs of designers or consumers and meet the needs for the continuous enrichment and improvement of sensory ergonomics knowledge.

[0179] Regarding the method for constructing a knowledge graph of sensible engineering described in the above embodiments, this application also provides an implementation method for analyzing the correlation between design elements in the design element space and sensory words in the sensory image space, thereby facilitating the execution of step S106. In the technical solutions described above, before constructing the connection between the design element space and the sensory image space based on the correlation between design elements in the design element space and sensory words in the sensory image space, the method may further include S1~S4 (Note: S1~S4 are not shown in the figures):

[0180] S1. Encode the product images of the target product category based on the deconstruction results to obtain an x-dimensional encoding vector.

[0181] For example, the deconstruction result includes x design elements, and the x dimensions correspond one-to-one with the x design elements, with each dimension corresponding to a different design element. If the value of the x-dimensional encoding vector in the i-th dimension is a first value, it means that the encoded product image reflects the design element corresponding to the i-th dimension; if the value of the x-dimensional encoding vector in the i-th dimension is a second value, it means that the encoded product image does not reflect the design element corresponding to the i-th dimension. The first value and the second value are different values, used to indicate that the product image reflects the design elements in completely opposite ways. For example, the first value is 1, and the second value is 0. In other words, the x-dimensional encoding result of the product image reflects whether the product image reflects each design element in the deconstruction result or not; this x-dimensional encoding result is equivalent to directly mapping the image to the design element space.

[0182] S2. Based on the coded product images and the emotional words in the first set of emotional words, construct a second semantic difference scale.

[0183] The second semantic difference scale can also be represented using an odd-point Likert scale and distributed to different individuals in the form of an electronic questionnaire. This second semantic difference scale is used to collect secondary evaluation results regarding the ability of product images to represent the sensory words in the first sensory vocabulary set. By collecting these secondary evaluation results, it is essentially equivalent to mapping the images onto a sensory imagery space, clarifying the connection between the images and the sensory words representing those words in that space.

[0184] S3. Based on the second evaluation results and the x-dimensional coding vector, a multiple linear regression analysis is performed to obtain a regression equation with the emotional words in the first set of emotional words as the dependent variable and x design elements as independent variables.

[0185] Here, the multiple linear regression analysis can be performed using SPSS software, which can be replaced by any existing or future platform or tool available for multiple linear regression analysis.

[0186] S4. Extract the regression coefficients corresponding to the x design elements that are independent variables in the regression equation; the magnitude of the regression coefficients reflects the correlation between the corresponding design elements and the emotional vocabulary.

[0187] The magnitude of the regression coefficients in the regression equation reflects the strength of the correlation between design elements and sensory terms. Understandably, in a regression equation where a particular sensory term from the first set of sensory terms is the dependent variable, negative regression coefficients are first eliminated. Among the remaining positive coefficients, higher coefficients correspond to a stronger correlation between the design element's independent variable and that sensory term; conversely, lower coefficients correspond to a weaker correlation.

[0188] Next, taking a target sensory word (which is any representative sensory word) from the first set of sensory words as an example, the construction of the connection between the design element space and the sensory image space based on the correlation between the design elements in the design element space and the sensory words in the sensory image space, as described in S106, may specifically include the following steps:

[0189] For each target emotional word in the first set of emotional words, the design elements corresponding to regression coefficients greater than or equal to a preset relevance threshold in the regression equation with the target emotional word as the dependent variable and x design elements as independent variables are determined as the relevant design elements of the target emotional word. Then, based on the determined relevant design elements of each emotional word in the first set of emotional words, a connection is constructed between the design element space and the emotional image space. The preset relevance threshold is a positive number.

[0190] In other words, in this embodiment, a preset relevance threshold is used as the threshold for the relevance reflected by the regression coefficient. If the regression coefficient is greater than or equal to the preset relevance threshold, it is determined that the design element corresponding to the regression coefficient has a strong relevance to the target emotional vocabulary, and the design element can be regarded as a relevant design element of the target emotional vocabulary. By analogy, the relevant design elements of each emotional vocabulary in the first emotional vocabulary can be obtained. The determination of relevant design elements means that the relevant design elements belonging to the design element space and the target emotional vocabulary in the emotional image space can be connected.

[0191] In the embodiments described above, the constructed sensory imagery space includes entities corresponding to representative sensory words belonging to a first set of sensory words, and entities corresponding to secondary sensory words belonging to a second set of sensory words. The sensory words introduced in the sensory engineering knowledge base constructed using related technologies are very limited. Therefore, the sensory engineering knowledge base constructed based on related technologies may have unsatisfactory retrieval results. To address this problem, this application proposes to introduce synonyms of core sensory words (representative and secondary sensory words) from existing corpora, thereby expanding the searchable scope of the knowledge base to better respond to searches by designers or consumers. This fully utilizes the scalability of knowledge graphs.

[0192] Specifically, in this application's technical solution, for each group of emotional vocabulary, representative and secondary emotional vocabulary words, synonyms not included in the core emotional vocabulary corpus can be collected from a corpus as supplementary emotional vocabulary. The relationship between the synonyms collected based on the representative emotional vocabulary and the representative emotional vocabulary is a similarity relationship, and the relationship between the synonyms collected based on the secondary emotional vocabulary and the secondary emotional vocabulary is also a similarity relationship. Here, the corpus used for supplementary vocabulary can be the Hownet sentiment dictionary, or it can be replaced by any existing or future relatively complete Chinese corpus that can assist in collecting synonyms.

[0193] Next, a third set of sensory vocabulary is constructed based on the collected supplementary sensory vocabulary. Unlike the first set of sensory vocabulary, which contains representative sensory vocabulary, and the second set of sensory vocabulary, which contains secondary sensory vocabulary, in this application, the third set of sensory vocabulary contains supplementary sensory vocabulary.

[0194] Based on this, the construction of the sensory image space of the sensory engineering knowledge graph, based on the first sensory vocabulary set, the second sensory vocabulary set, and the association between sensory words in the first sensory vocabulary set and sensory words in the second sensory vocabulary set, includes:

[0195] Based on the names of representative sensory words in the first set of sensory words, construct representative sensory word entities one by one; based on the names of secondary sensory words in the second set of sensory words, construct secondary sensory word entities one by one; based on the names of supplementary sensory words in the third set of sensory words, construct supplementary sensory word entities one by one. That is, construct entities of three types of sensory words.

[0196] Let's denote the name of the "representative lexicon" that needs to be used to construct the corresponding entity in the graph as "representative" and the corresponding entity code as "a". Let's denote the name of the "secondary lexicon" that needs to be used to construct the corresponding entity in the graph as "subordinate" and the corresponding entity code as "b". Let's denote the name of the "supplementary lexicon" that needs to be used to construct the corresponding entity in the graph as "supplementary" and the corresponding entity code as "c". Then, there are three specific entity construction statements:

[0197] CREATE (a:representative keyword{name: 'representative'});

[0198] CREATE (b:subordinate_vocabulary{name: 'subordinate'});

[0199] CREATE (c:supplementary vocabulary{name: 'supplementary'}).

[0200] Based on the representational relationship between representative and secondary sensory words, construct the corresponding representational relationship between the representative sensory word entity and the secondary sensory word entity; based on the similarity relationship between the representative sensory word and the supplementary sensory word, construct the corresponding similarity relationship between the representative sensory word entity and the supplementary sensory word entity; based on the similarity relationship between the secondary sensory word and the supplementary sensory word, construct the corresponding similarity relationship between the secondary sensory word entity and the supplementary sensory word entity. The specific relationship construction statements are as follows:

[0201] MATCH (a:representative term {name: 'representative'}), (b:subordinate term {name: 'subordinate'}), CREATE (a) - [r:representative] -> (b);

[0202] MATCH (a:representative) (c:supplementary) (a:supplementary) (c:supplementary) (a) - [r:supplementary] -> (c);

[0203] MATCH (b: secondary sensory vocabulary {name: 'subordinate'}), (c: supplementary sensory vocabulary {name: 'supplementary'}), CREATE (b) - [r: similar] -> (c).

[0204] Figure 5 A schematic diagram of the structure of a sensory image space provided in an embodiment of this application, such as... Figure 5 As shown, vocabulary supplementation can be performed on representative and secondary emotional vocabulary to obtain supplementary emotional vocabulary, and based on this, a more comprehensive emotional imagery space can be formed.

[0205] It's important to note that in the Neo4j platform, when constructing relationships between sensory image entities and design element entities, only entities representing sensory terms in the sensory image space can establish direct relationships with the four types of entities in the design element space: "components," "modules," "attributes," and "levels," and these relationships are all "related." "Secondary sensory terms" and "supplementary sensory terms" entities can only form indirect relationships with design element entities through entities representing sensory terms. This is to ensure the reliability and accuracy of the connections. The standard for a "related" relationship between a design element entity and a "representative sensory term" entity is that the design element entity is a related design element of that "representative sensory term" entity. The specific relationship construction statements are as follows:

[0206] MATCH (m: component {name: 'component'}), (a: representative {name: 'representative'}), CREATE (m) - [r: related] -> (a);

[0207] MATCH (n: module {name: 'module'}), (a: representative term {name: 'representative'}), CREATE (n) - [r: related] -> (a);

[0208] MATCH (p: attribute {name: 'attribute'}), (a: representative term {name: 'representative'}), CREATE (p) - [r: related] -> (a);

[0209] MATCH (q: level {name: 'level'}), (a: representative word {name: 'representative'}), CREATE (q) - [r: related] -> (a).

[0210] from Figure 5As can be seen from the illustrated structure of the sensory imagery space, the three types of entities have different statuses: "representative sensory words" are the most important, "secondary sensory words" are next, and "supplementary sensory words" are the least important. This importance is reflected in the process of retrieving the sensory engineering knowledge graph. When a designer or consumer inputs a sensory word to search for corresponding design elements, it first matches sensory word entities labeled "representative sensory words." If no results are found, it then matches sensory word entities labeled "secondary sensory words," and finally matches sensory word entities labeled "supplementary sensory words." This retrieval strategy, relying on the structure of the sensory imagery space described in this invention, can greatly improve retrieval efficiency.

[0211] When a designer or consumer enters a specific emotional term, once the search results are retrieved, the system will return the design element entities that are connected to the retrieved emotional image entity in the knowledge graph, regardless of the distance between the two entities. If we denote the emotional term the designer or consumer wants to search for as kansei, then there are 12 possible query statements:

[0212] MATCH (m: component) --> ( : represents the emotional word {name: 'kansei'}), RETURN m;

[0213] MATCH (n: module) --> ( : represents the emotional word {name: 'kansei'}), RETURN n;

[0214] MATCH (p:attribute) --> ( :represents the emotional word {name: 'kansei'}), RETURN p;

[0215] MATCH (q: level) --> ( : represents emotional vocabulary {name: 'kansei'}), RETURN q;

[0216] MATCH (m: component) - -> ( : secondary sensory vocabulary {name: 'kansei'}), RETURN m;

[0217] MATCH (n: module) --> ( : secondary sensory vocabulary {name: 'kansei'}), RETURN n;

[0218] MATCH (p:attribute) --> ( :secondary sensual vocabulary {name: 'kansei'}), RETURN p;

[0219] MATCH (q: level) - -> ( : secondary sensory vocabulary {name: 'kansei'}), RETURN q;

[0220] MATCH (m: component) - -> ( : supplementary emotional vocabulary {name: 'kansei'}), RETURN m;

[0221] MATCH (n: module) --> ( : supplementary emotional vocabulary {name: 'kansei'}), RETURN n;

[0222] MATCH (p:attribute) --> ( :supplementary emotional vocabulary{name: 'kansei'}), RETURN p;

[0223] MATCH (q: level) --> ( : supplementary emotional vocabulary {name: 'kansei'}), RETURN q.

[0224] While related technologies have conducted in-depth deconstruction of the internal structure of the design element space based on the FBS model, they have not provided sufficient explanation of the internal structure of the sensory image space and the connection between the two spaces. This issue is addressed in this invention. Based on the implementation process of the sensual engineering method, this invention divides the entities in the sensory image space into "representative sensory words," "secondary sensory words," and "supplementary sensory words," and establishes two relationships within them: "representative" and "similar," making the structure of the entire sensory image space and the status and role of each sensory word clearer.

[0225] Related technologies only mention using the Python programming language to build a Kansei Engineering knowledge base, but do not provide any specific code or execution methods. This invention, however, builds a Kansei Engineering knowledge base based on knowledge graph tools and provides specific execution methods and Cypher statements for building the knowledge base on the Neo4j platform.

[0226] Existing technologies have not provided a reasonable solution to the retrieval failures that may result from the ambiguity in the expressions of designers and consumers. This invention proposes to expand the searchable scope of the knowledge base by introducing synonyms of core emotional terms (representing emotional terms and secondary emotional terms) from the existing corpus, so as to better respond to the searches of designers or consumers. This fully utilizes the scalability of knowledge graphs.

[0227] Based on the method for constructing a knowledge graph of kanji science described in the above embodiments, this application further provides a retrieval method based on a knowledge graph of kanji science. Figure 6This is a flowchart of a retrieval method based on a knowledge graph of kanji engineering. (For example...) Figure 6 As shown, the method includes:

[0228] S601. Receive the search keywords and search purpose information provided by the user.

[0229] This application provides an information retrieval function that allows users to obtain retrieval results based on a Kansei Engineering knowledge graph by entering search keywords and search objectives. As an example, the search terms provided by the user can be Kansei words, which may be representative Kansei words, secondary Kansei words, or supplementary Kansei words mentioned above, or words with similar meanings or types to the representative Kansei words, secondary Kansei words, or supplementary Kansei words mentioned above.

[0230] The search objective clearly specifies the type of content to be retrieved. For example, the search objective could be: design elements or product images. The search objective depends on the user's search needs, and combined with the established Kansei Engineering knowledge graph, search results are provided to meet the user's actual needs.

[0231] S602. Based on the search keywords, match sensory words in the sensory image space of the sensory engineering knowledge graph.

[0232] As mentioned above, the sensibility engineering knowledge graph used in this embodiment is constructed through any of the implementation methods described in the previous embodiments. When matching sensibility words in the sensibility intention space based on search terms, priority is given to matching sensibility words belonging to the first sensibility word set. If no match is found, then matching sensibility words belonging to the second sensibility word set is considered.

[0233] If the sensory image space is constructed based on the names and relationships between representative sensory words in the first sensory vocabulary set, secondary sensory words in the second sensory vocabulary set, and supplementary sensory words in the third sensory vocabulary set, then the step of matching sensory words in the sensory image space of the sensory engineering knowledge graph based on the search keywords in this step includes:

[0234] Based on the search keywords, representative emotional words are first matched in the emotional image space of the emotional engineering knowledge graph. If a representative emotional word is successfully matched, the matching of emotional words for the search keywords in the emotional image space is stopped.

[0235] If a match for a representative emotional word fails, the matching of secondary emotional words continues in the emotional image space of the emotional engineering knowledge graph. If a secondary emotional word is successfully matched, the matching of emotional words for the search keyword in the emotional image space is stopped.

[0236] If a match for a secondary sensory term fails, the matching process continues in the sensory imagery space of the sensory engineering knowledge graph to find supplementary sensory terms. If a supplementary sensory term is successfully matched, the matching process stops in the sensory imagery space for the search keywords.

[0237] Such a retrieval strategy relies on the structure of the sensory image space pointed out in this invention, which can greatly improve retrieval efficiency.

[0238] S603. In the aforementioned sensibility engineering knowledge graph, identify the design elements associated with the matched sensibility terms.

[0239] Since the knowledge graph of sensual engineering is constructed based on the correlation between sensual vocabulary in the sensual imagery space and design elements in the design element space, it is possible to easily identify the design elements associated with sensual vocabulary once the sensual vocabulary has been matched against the search keywords, based on the connections between the two spaces. This path is specifically manifested in the following ways:

[0240] (1) S602 matched words representing emotions.

[0241] The path to determine design elements is: representative emotional terms → related design elements.

[0242] (2) S602 matched secondary emotional words.

[0243] The path to determine design elements is: secondary sensory terms → representative sensory terms → related design elements.

[0244] (3) S602 matches the supplementary emotional vocabulary, which is obtained by supplementing the representative emotional vocabulary.

[0245] The path to determine design elements is: supplementing sensory vocabulary → representing sensory vocabulary → related design elements.

[0246] (4) S602 matches a supplementary emotional vocabulary, which is obtained by supplementing the secondary emotional vocabulary.

[0247] The path to determine design elements is: supplementing sensory vocabulary → secondary sensory vocabulary → representative sensory vocabulary → related design elements.

[0248] S604. Based on the determined design elements, provide the user with search results that match the search objective information.

[0249] Based on the different information to be retrieved, the following describes several possible implementations of S604:

[0250] (1) If the search objective indicated by the search objective information is a design element, the step of providing the user with search results matching the search objective information based on the determined design element includes:

[0251] The identified design elements will be provided to users as search results.

[0252] (2) If the search objective indicated by the search objective information is a product image, the step of providing the user with search results matching the search objective information based on the determined design elements includes:

[0253] The query can deconstruct individual products containing the identified design elements;

[0254] The product images of the individual products retrieved will be provided to the user as search results.

[0255] Based on the above embodiments, it is easy to see that this application provides a variety of search functions based on the Kansei Engineering knowledge graph, which can search for both design elements and product images. This provides users with diverse search services. Furthermore, since the Kansei Engineering knowledge graph used for retrieval constructs a design element space and a Kansei image space, where the Kansei image space specifically involves several different Kansei vocabulary entities, including representative Kansei words, secondary Kansei words, and supplementary Kansei words, there is a clear definition of Kansei words. The overall feasibility of the solution is improved.

[0256] Based on the method for constructing a knowledge graph of kanji science and the retrieval method based on the knowledge graph of kanji science described in the above embodiments, this application also provides a device for constructing a knowledge graph of kanji science and a retrieval device based on the knowledge graph of kanji science. The following description is in conjunction with the accompanying drawings.

[0257] Figure 7 This is a schematic diagram of the structure of a device for constructing a sensory engineering knowledge graph, provided in an embodiment of this application. Figure 8 This is a schematic diagram of the structure of a retrieval device based on a knowledge graph of kinetic engineering, provided in an embodiment of this application.

[0258] like Figure 7 As shown, the apparatus for constructing a knowledge graph of sensible engineering includes:

[0259] The vocabulary acquisition module 701 is used to acquire the core emotional vocabulary of the target product category. The core emotional vocabulary includes multiple core emotional words, which are emotional words that have a positive impact on consumers' behavior of purchasing products of the target product category.

[0260] The design element deconstruction module 702 is used to deconstruct the design elements of the target product category according to multiple preset deconstruction levels to obtain deconstruction results; the deconstruction results include the design elements deconstructed at multiple deconstruction levels respectively.

[0261] Design element space construction module 703 is used to construct the design element space of the Kansei Engineering knowledge graph based on the design elements in the deconstruction result and the relationship between different design elements in the deconstruction result at the deconstruction level.

[0262] The emotional vocabulary evaluation module 704 is used to construct a first emotional vocabulary set and a second emotional vocabulary set based on the multiple core emotional vocabulary and the product images of the target product category; wherein, the emotional vocabulary in the first emotional vocabulary set has representative ability for one or more emotional vocabulary in the second emotional vocabulary set.

[0263] The sensory imagery space construction module 705 is used to construct the sensory imagery space of the sensory engineering knowledge graph based on the first sensory vocabulary set, the second sensory vocabulary set, and the association between the sensory vocabulary in the first sensory vocabulary set and the sensory vocabulary in the second sensory vocabulary set.

[0264] The spatial association module 706 is used to construct a connection between the design element space and the sensory image space based on the correlation between the design elements in the design element space and the sensory vocabulary in the sensory image space, and to obtain a sensory engineering knowledge graph of the target product category based on the design element space, the sensory image space, and the connection between the design element space and the sensory image space.

[0265] In one alternative implementation, the sensory vocabulary evaluation module 704 includes:

[0266] The scale construction unit is used to construct a first semantic difference scale based on the product images of the target product category and the multiple core emotional words; the first semantic difference scale is used to collect first evaluation results on the performance ability of the product images of the target product category to the multiple core emotional words;

[0267] The clustering analysis unit is used to perform clustering analysis based on multiple first evaluation results to obtain N groups of sensory vocabulary groups; where N is an integer greater than 1.

[0268] The vocabulary type determination unit is used to determine a representative emotional word for each group of emotional vocabulary, or to determine a representative emotional word and at least one secondary emotional word; wherein, the relationship between the representative emotional word and the secondary emotional word determined for the same group of emotional vocabulary is a representative relationship.

[0269] The vocabulary set construction unit is used to construct the first emotional vocabulary set using representative emotional words determined for each of the N emotional vocabulary groups, and to construct the second emotional vocabulary set using secondary emotional words determined for each of the N emotional vocabulary groups.

[0270] In one optional implementation, for the target lexical group among the N groups of sensual lexical groups, the lexical type determination unit is specifically used for:

[0271] In the target emotional vocabulary group, a core emotional vocabulary word that encompasses the semantics of a predetermined proportion or more of the core emotional vocabulary words in the target emotional vocabulary group is determined as the representative emotional vocabulary word corresponding to the target emotional vocabulary group, and the other words in the target emotional vocabulary group other than the representative emotional vocabulary word are designated as the secondary emotional vocabulary words corresponding to the target emotional vocabulary group; or,

[0272] Based on the semantics of each core emotional word in the target emotional vocabulary group, an emotional word that does not belong to the target emotional vocabulary group is determined as the representative emotional word corresponding to the target emotional vocabulary group, and each core emotional word in the target emotional vocabulary group is taken as the secondary emotional word corresponding to the target emotional vocabulary group.

[0273] In one alternative implementation, the apparatus for constructing the sensible engineering knowledge graph also includes:

[0274] The vocabulary supplementation module is used to collect synonyms not included in the core vocabulary database from the corpus for the representative and secondary emotional vocabulary of each emotional vocabulary group, as supplementary emotional vocabulary; wherein, the relationship between the synonyms collected based on the representative emotional vocabulary and the representative emotional vocabulary is a similarity relationship, and the relationship between the synonyms collected based on the secondary emotional vocabulary and the secondary emotional vocabulary is a similarity relationship.

[0275] The vocabulary set construction module is used to construct a third set of sensory vocabulary based on the collected supplementary sensory vocabulary;

[0276] The sensory imagery space construction module 705 includes:

[0277] The vocabulary entity construction unit is used to construct representative emotional vocabulary entities one by one according to the names of representative emotional vocabulary in the first set of emotional vocabulary, construct secondary emotional vocabulary entities one by one according to the names of secondary emotional vocabulary in the second set of emotional vocabulary, and construct supplementary emotional vocabulary entities one by one according to the names of supplementary emotional vocabulary in the third set of emotional vocabulary.

[0278] The entity relation construction unit is used to construct the representation relationship between the representative emotional word entity and the secondary emotional word entity based on the representation relationship between the representative emotional word and the secondary emotional word; to construct the similarity relationship between the representative emotional word and the supplementary emotional word entity based on the similarity relationship between the representative emotional word and the supplementary emotional word; and to construct the similarity relationship between the secondary emotional word and the supplementary emotional word entity based on the similarity relationship between the secondary emotional word and the supplementary emotional word.

[0279] In one alternative implementation, the apparatus for constructing the sensible engineering knowledge graph also includes:

[0280] The image encoding module is used to encode product images of the target product category based on the deconstruction result, obtaining an x-dimensional encoding vector. The deconstruction result includes x design elements, and the x-dimensional vector corresponds one-to-one with the x design elements. If the value of the x-dimensional encoding vector in the i-th dimension is a first value, it means that the encoded product image reflects the design element corresponding to the i-th dimension. If the value of the x-dimensional encoding vector in the i-th dimension is a second value, it means that the encoded product image does not reflect the design element corresponding to the i-th dimension.

[0281] The scale construction module is used to construct a second semantic difference scale based on the encoded product image and the sensory words in the first sensory word set; the second semantic difference scale is used to collect a second evaluation result on the ability of the product image to express the sensory words in the first sensory word set;

[0282] The regression analysis module is used to perform multiple linear regression analysis based on the second evaluation result and the x-dimensional encoding vector to obtain a regression equation with the emotional words in the first set of emotional words as the dependent variable and x design elements as independent variables.

[0283] The regression coefficient extraction module is used to extract the regression coefficients corresponding to the x design elements that are independent variables in the regression equation; the magnitude of the regression coefficients reflects the correlation between the corresponding design elements and the emotional vocabulary.

[0284] In one optional implementation, the spatial association module 706 includes:

[0285] The relevant design element determination unit is used to determine, for the target emotional word in the first set of emotional words, the design element corresponding to the regression coefficient that is greater than or equal to a preset correlation threshold in the regression equation with the target emotional word as the dependent variable and x design elements as independent variables, as the relevant design element of the target emotional word; the target emotional word is any emotional word in the first set of emotional words;

[0286] The spatial connection construction unit is used to construct the connection between the design element space and the sensory image space based on the relevant design elements of each sensory word in the determined first set of sensory words.

[0287] like Figure 8 As shown, the retrieval device based on the Kansei Engineering knowledge graph includes:

[0288] The information receiving module 801 is used to receive search keywords and search purpose information provided by the user.

[0289] The vocabulary matching module 802 is used to match sensory words in the sensory image space of the sensory engineering knowledge graph based on the search keywords; the sensory engineering knowledge graph is constructed according to the construction method of the sensory engineering knowledge graph.

[0290] The design element determination module 803 is used to determine the design elements associated with the matched sensory vocabulary in the sensory engineering knowledge graph.

[0291] The search results providing module 804 is used to provide the user with search results that match the search objective information based on the determined design elements.

[0292] In one optional implementation, if the search objective indicated by the search objective information is a design element, the search result providing module 804 is specifically used for:

[0293] The identified design elements will be provided to users as search results.

[0294] In one optional implementation, if the search purpose indicated by the search purpose information is product images, the search result providing module 804 is specifically used for:

[0295] The query can deconstruct individual products containing the identified design elements;

[0296] The product images of the individual products retrieved will be provided to the user as search results.

[0297] In one optional implementation, the sensory image space is constructed based on the names and relationships between the representative sensory words in the first sensory vocabulary set, the secondary sensory words in the second sensory vocabulary set, and the supplementary sensory words in the third sensory vocabulary set; wherein, the supplementary sensory words in the third sensory vocabulary set are obtained by searching for synonyms based on the representative sensory words in the first sensory vocabulary set and the secondary sensory words in the second sensory vocabulary set.

[0298] The vocabulary matching module 802 includes: a first matching unit, a second matching unit, and a third matching unit;

[0299] The first matching unit is used to first match representative emotional words in the emotional image space of the emotional engineering knowledge graph based on the search keywords. If the first matching unit successfully matches a representative emotional word, the word matching module 802 stops matching emotional words for the search keywords in the emotional image space.

[0300] The second matching unit is used to continue matching secondary sensory words in the sensory image space of the sensory engineering knowledge graph if the matching of representative sensory words fails. If the second matching unit successfully matches a secondary sensory word, the word matching module 802 stops matching sensory words for the search keyword in the sensory image space.

[0301] The third matching unit is used to continue matching supplementary sensory words in the sensory image space of the sensory engineering knowledge graph if the matching of secondary sensory words fails. If the third matching unit successfully matches supplementary sensory words, the word matching module 802 stops matching sensory words for the search keywords in the sensory image space.

[0302] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. The components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment solution according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0303] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A search method based on an emotional engineering knowledge graph, characterized by, The method comprises the following steps: receiving a search keyword and search purpose information provided by a user; matching a sensibility vocabulary in a sensibility image space of a sensibility engineering knowledge graph based on the search keyword; determining a design element associated with the matched sensibility vocabulary in the sensibility engineering knowledge graph; providing a search result matched with the search purpose information to the user according to the determined design element; The method for constructing the sensibility engineering knowledge graph comprises the following steps: obtaining a core sensibility vocabulary library of a target product category, wherein the core sensibility vocabulary library comprises a plurality of core sensibility vocabularies, and the core sensibility vocabulary is a sensibility vocabulary that has a positive impact on the behavior of consumers purchasing products of the target product category; performing design element decomposition on products of the target product category according to a plurality of preset decomposition levels to obtain a decomposition result, wherein the decomposition result comprises design elements decomposed at the plurality of decomposition levels, respectively; constructing a design element space of the sensibility engineering knowledge graph based on the design elements in the decomposition result and the relationship between different design elements in the decomposition result at the decomposition levels; constructing a first sensibility vocabulary set and a second sensibility vocabulary set based on the plurality of core sensibility vocabularies according to the expression ability of the plurality of core sensibility vocabularies to product pictures of the target product category, wherein a sensibility vocabulary in the first sensibility vocabulary set has a representative ability to one or more sensibility vocabularies in the second sensibility vocabulary set; constructing a sensibility image space of the sensibility engineering knowledge graph based on the first sensibility vocabulary set, the second sensibility vocabulary set, and the association between the sensibility vocabulary in the first sensibility vocabulary set and the sensibility vocabulary in the second sensibility vocabulary set; constructing the connection between the design element space and the sensibility image space according to the correlation between the design elements in the design element space and the sensibility vocabularies in the sensibility image space, and obtaining the sensibility engineering knowledge graph of the target product category based on the design element space, the sensibility image space, and the connection between the design element space and the sensibility image space.

2. The method of claim 1, wherein, The method for constructing the first sensibility vocabulary set and the second sensibility vocabulary set based on the plurality of core sensibility vocabularies according to the expression ability of the plurality of core sensibility vocabularies to product pictures of the target product category comprises the following steps: constructing a first semantic difference scale according to the product pictures of the target product category and the plurality of core sensibility vocabularies, wherein the first semantic difference scale is used to collect first evaluation results about the expression ability of the plurality of core sensibility vocabularies to the product pictures of the target product category; performing clustering analysis based on a plurality of the first evaluation results to obtain N groups of sensibility vocabulary groups, wherein N is an integer greater than 1; determining a representative sensibility vocabulary for each group of sensibility vocabulary groups, or determining a representative sensibility vocabulary and at least one secondary sensibility vocabulary for each group of sensibility vocabulary groups, wherein the relationship between the representative sensibility vocabulary and the secondary sensibility vocabulary determined for the same group of sensibility vocabulary groups is a representative relationship. The first set of emotional words is constructed by using the representative emotional words respectively determined for the N groups of emotional words, and the second set of emotional words is constructed by using the secondary emotional words respectively determined for the N groups of emotional words.

3. The method of claim 2, wherein, For a target group of emotional words in the N groups of emotional words, one representative emotional word is determined for the target group of emotional words, or one representative emotional word and at least one secondary emotional word are determined, including: In the target group of emotional words, a core emotional word covering the semantics of more than a preset proportion of the core emotional words in the target group of emotional words is determined as the representative emotional word corresponding to the target group of emotional words, and the other words in the target group of emotional words except the representative emotional word are determined as the secondary emotional words corresponding to the target group of emotional words; or, Based on the semantics of each core emotional word in the target group of emotional words, an emotional word not belonging to the target group of emotional words is determined as the representative emotional word corresponding to the target group of emotional words, and each core emotional word in the target group of emotional words is determined as the secondary emotional word corresponding to the target group of emotional words.

4. The method according to claim 2 or 3, characterized in that, The method further includes: For the representative emotional words and the secondary emotional words of each group of emotional words, near-synonyms not included in the core emotional word library are collected from a corpus as supplementary emotional words; wherein the relationship between the near-synonyms collected based on the representative emotional words and the representative emotional words is a similar relationship, and the relationship between the near-synonyms collected based on the secondary emotional words and the secondary emotional words is a similar relationship; A third set of emotional words is constructed according to the collected supplementary emotional words; The construction of the emotional image space of the emotional engineering knowledge graph based on the first set of emotional words, the second set of emotional words, and the association between the emotional words in the first set of emotional words and the emotional words in the second set of emotional words includes: Representative emotional word entities are constructed according to the names of the representative emotional words in the first set of emotional words, secondary emotional word entities are constructed according to the names of the secondary emotional words in the second set of emotional words, and supplementary emotional word entities are constructed according to the names of the supplementary emotional words in the third set of emotional words; Representative relationships between the corresponding representative emotional word entities and the secondary emotional word entities are constructed according to the representative relationships between the representative emotional words and the secondary emotional words, similar relationships between the corresponding representative emotional word entities and the supplementary emotional word entities are constructed according to the similar relationships between the representative emotional words and the supplementary emotional words, and similar relationships between the corresponding secondary emotional word entities and the supplementary emotional word entities are constructed according to the similar relationships between the secondary emotional words and the supplementary emotional words.

5. The method according to any one of claims 1-3, characterized in that, Before the construction of the connection between the design element space and the emotional image space according to the relevance of the design elements in the design element space and the emotional words in the emotional image space, the method further includes: According to the deconstruction result, product pictures of the target product category are encoded to obtain an x-dimensional encoding vector; the deconstruction result includes x design elements, and the x dimension corresponds to the x design elements one by one; if a value of the x-dimensional encoding vector on the i-th dimension is a first value, it represents that the encoded product picture reflects the design element corresponding to the i-th dimension; if the value of the x-dimensional encoding vector on the i-th dimension is a second value, it represents that the encoded product picture does not reflect the design element corresponding to the i-th dimension; According to the encoded product picture and the perceptual vocabulary in the first perceptual vocabulary set, a second semantic difference scale is constructed; the second semantic difference scale is used to collect a second evaluation result about the expression ability of the product picture to the perceptual vocabulary in the first perceptual vocabulary set; Based on the second evaluation result and the x-dimensional encoding vector, multiple linear regression analysis is performed to obtain a regression equation with the perceptual vocabulary in the first perceptual vocabulary set as the dependent variable and the x design elements as the independent variable; The regression coefficients corresponding to the x design elements in the regression equation are extracted; the size of the regression coefficient reflects the correlation between the corresponding design element and the perceptual vocabulary.

6. The method of claim 5, wherein, The correlation between the design elements in the design element space and the perceptual vocabulary in the perceptual image space is constructed according to the correlation between the design elements in the design element space and the perceptual vocabulary in the perceptual image space, comprising: For a target perceptual vocabulary in the first perceptual vocabulary set, the design elements corresponding to the regression coefficients greater than or equal to a preset correlation threshold in the regression equation with the target perceptual vocabulary as the dependent variable and the x design elements as the independent variable are determined as the relevant design elements of the target perceptual vocabulary; the target perceptual vocabulary is any perceptual vocabulary in the first perceptual vocabulary set; According to the determined relevant design elements of each perceptual vocabulary in the first perceptual vocabulary set, the correlation between the design element space and the perceptual image space is constructed.

7. The method of claim 1, wherein, If the search purpose information indicates that the search purpose is a design element, the search result matching the search purpose information is provided to the user according to the determined design element, comprising: The determined design element is provided to the user as the search result.

8. The method of claim 1, wherein, If the search purpose information indicates that the search purpose is a product picture, the search result matching the search purpose information is provided to the user according to the determined design element, comprising: Querying product individuals capable of deconstructing the determined design elements; The product picture of the queried product individual is provided to the user as the search result.

9. The method of claim 1, 7 or 8, wherein, The perceptual image space is constructed based on the names of the representative perceptual vocabulary in the first perceptual vocabulary set, the secondary perceptual vocabulary in the second perceptual vocabulary set, and the supplementary perceptual vocabulary in the third perceptual vocabulary set, and the relationship between them; wherein the supplementary perceptual vocabulary in the third perceptual vocabulary set is supplemented by searching for synonyms based on the representative perceptual vocabulary in the first perceptual vocabulary set and the secondary perceptual vocabulary in the second perceptual vocabulary set; The matching of the emotional vocabulary in the emotional image space of the kansei engineering knowledge graph based on the search keyword comprises: Firstly, matching the representative emotional vocabulary in the emotional image space of the kansei engineering knowledge graph based on the search keyword, and if the representative emotional vocabulary is successfully matched, then stop matching the emotional vocabulary in the emotional image space for the search keyword; If the matching of the representative emotional vocabulary fails, then continue matching the secondary emotional vocabulary in the emotional image space of the kansei engineering knowledge graph, and if the secondary emotional vocabulary is successfully matched, then stop matching the emotional vocabulary in the emotional image space for the search keyword; If the matching of the secondary emotional vocabulary fails, then continue matching the supplementary emotional vocabulary in the emotional image space of the kansei engineering knowledge graph, and if the supplementary emotional vocabulary is successfully matched, then stop matching the emotional vocabulary in the emotional image space for the search keyword. 10.A retrieval device based on an emotional engineering knowledge graph, characterized in that, Comprise: An information receiving module configured to receive a search keyword and search purpose information provided by a user; A vocabulary matching module configured to match emotional vocabulary in an emotional image space of a kansei engineering knowledge graph based on the search keyword; A design element determining module configured to determine design elements associated with the matched emotional vocabulary in the kansei engineering knowledge graph; A search result providing module configured to provide a search result matched with the search purpose information to the user according to the determined design elements; The construction device of the kansei engineering knowledge graph comprises: A vocabulary library obtaining module configured to obtain a core emotional vocabulary library of a target product category, wherein the core emotional vocabulary library comprises a plurality of core emotional vocabulary, and the core emotional vocabulary is an emotional vocabulary having a positive impact on the consumer's behavior of purchasing a product of the target product category; A design element deconstruction module configured to perform design element deconstruction on the product of the target product category according to a plurality of preset deconstruction levels to obtain a deconstruction result, wherein the deconstruction result comprises design elements deconstructed at the plurality of deconstruction levels respectively; A design element space construction module configured to construct a design element space of a kansei engineering knowledge graph based on the design elements in the deconstruction result and the relationship between different design elements in the deconstruction result at the deconstruction levels; A kansei vocabulary evaluation module configured to evaluate the performance of the plurality of core emotional vocabulary according to product pictures of the product of the target product category, and construct a first emotional vocabulary set and a second emotional vocabulary set based on the plurality of core emotional vocabulary, wherein the emotional vocabulary in the first emotional vocabulary set has a representative ability for one or more emotional vocabulary in the second emotional vocabulary set; A kansei image space construction module configured to construct an emotional image space of a kansei engineering knowledge graph based on the first emotional vocabulary set, the second emotional vocabulary set, and the association between the emotional vocabulary in the first emotional vocabulary set and the emotional vocabulary in the second emotional vocabulary set. The spatial correlation module is configured to: construct a correlation between the design element space and the Kansei image space according to a correlation between a design element in the design element space and a Kansei vocabulary in the Kansei image space; and obtain a Kansei ergonomics knowledge graph of the target product category based on the design element space, the Kansei image space, and the correlation between the design element space and the Kansei image space.

11. The apparatus of claim 10, wherein, The Kansei vocabulary evaluation module comprises: A scale construction unit configured to: construct a first semantic difference scale according to product pictures of the target product category and the plurality of core Kansei vocabularies; and collect first evaluation results of the product pictures of the target product category on expression abilities of the plurality of core Kansei vocabularies by using the first semantic difference scale; A cluster analysis unit configured to: perform cluster analysis based on the plurality of first evaluation results to obtain N groups of Kansei vocabulary groups; and N is an integer greater than 1; A vocabulary type determination unit configured to: determine a representative Kansei vocabulary for each group of Kansei vocabulary groups, or determine a representative Kansei vocabulary and at least one secondary Kansei vocabulary for each group of Kansei vocabulary groups; and a relationship between the representative Kansei vocabulary and the secondary Kansei vocabulary determined for the same group of Kansei vocabulary groups is a representative relationship; A vocabulary set construction unit configured to: construct the first Kansei vocabulary set by using the representative Kansei vocabularies determined for the N groups of Kansei vocabulary groups, and construct the second Kansei vocabulary set by using the secondary Kansei vocabularies determined for the N groups of Kansei vocabulary groups.

12. The apparatus of claim 11, wherein, For a target Kansei vocabulary group in the N groups of Kansei vocabulary groups, the vocabulary type determination unit is specifically configured to: determine a core Kansei vocabulary covering semantics of more than a preset proportion of core Kansei vocabularies in the target Kansei vocabulary group as a representative Kansei vocabulary corresponding to the target Kansei vocabulary group, and determine other vocabularies in the target Kansei vocabulary group other than the representative Kansei vocabulary as secondary Kansei vocabularies corresponding to the target Kansei vocabulary group; Or, determine a Kansei vocabulary not belonging to the target Kansei vocabulary group as a representative Kansei vocabulary corresponding to the target Kansei vocabulary group based on semantics of each core Kansei vocabulary in the target Kansei vocabulary group, and determine each core Kansei vocabulary in the target Kansei vocabulary group as a secondary Kansei vocabulary corresponding to the target Kansei vocabulary group.

13. The apparatus of claim 11 or 12, wherein, The device further comprises: A vocabulary supplementing module configured to: collect synonyms not included in the core Kansei vocabulary library as supplementary Kansei vocabularies from a corpus for the representative Kansei vocabulary and the secondary Kansei vocabulary of each group of Kansei vocabulary groups; and a relationship between the synonyms collected based on the representative Kansei vocabulary and the representative Kansei vocabulary is a similarity relationship, and a relationship between the synonyms collected based on the secondary Kansei vocabulary and the secondary Kansei vocabulary is a similarity relationship; A vocabulary set construction module configured to: construct a third Kansei vocabulary set according to the supplementary Kansei vocabularies collected; The Kansei image space construction module comprises: a vocabulary entity constructing unit configured to construct a representative emotional vocabulary entity according to a name of a representative emotional vocabulary in the first emotional vocabulary set, construct a secondary emotional vocabulary entity according to a name of a secondary emotional vocabulary in the second emotional vocabulary set, and construct a supplementary emotional vocabulary entity according to a name of a supplementary emotional vocabulary in the third emotional vocabulary set; an entity relationship constructing unit configured to construct a representative relationship between a representative emotional vocabulary entity and a secondary emotional vocabulary entity according to a representative relationship between the representative emotional vocabulary and the secondary emotional vocabulary, construct a similarity relationship between a representative emotional vocabulary entity and a supplementary emotional vocabulary entity according to a similarity relationship between the representative emotional vocabulary and the supplementary emotional vocabulary, and construct a similarity relationship between a secondary emotional vocabulary entity and a supplementary emotional vocabulary entity according to a similarity relationship between the secondary emotional vocabulary and the supplementary emotional vocabulary.

14. The apparatus of any one of claims 10-12, wherein, The apparatus further includes: a picture encoding module configured to encode a product picture of the target product category according to the deconstruction result to obtain an x-dimensional encoding vector; the deconstruction result includes x design elements, and the x dimension corresponds to the x design elements one by one; if a value of the x-dimensional encoding vector on an i-th dimension is a first value, it indicates that the encoded product picture reflects a design element corresponding to the i-th dimension; if the value of the x-dimensional encoding vector on the i-th dimension is a second value, it indicates that the encoded product picture does not reflect the design element corresponding to the i-th dimension; a scale constructing module configured to construct a second semantic difference scale according to the encoded product picture and the emotional vocabulary in the first emotional vocabulary set; the second semantic difference scale is used to collect a second evaluation result about an expression ability of the product picture on the emotional vocabulary in the first emotional vocabulary set; a regression analysis module configured to perform multiple linear regression analysis based on the second evaluation result and the x-dimensional encoding vector to obtain a regression equation with the emotional vocabulary in the first emotional vocabulary set as a dependent variable and with the x design elements as independent variables; a regression coefficient extracting module configured to extract regression coefficients corresponding to the x design elements as independent variables in the regression equation; the regression coefficients reflect correlations between the corresponding design elements and the emotional vocabulary.

15. The apparatus of claim 14, wherein, The spatial correlation module includes: a relevant design element determining unit configured to, for a target emotional vocabulary in the first emotional vocabulary set, determine, as a relevant design element of the target emotional vocabulary, a design element corresponding to a regression coefficient greater than or equal to a preset correlation threshold in a regression equation with the target emotional vocabulary as a dependent variable and with the x design elements as independent variables; the target emotional vocabulary is any emotional vocabulary in the first emotional vocabulary set; a spatial relationship constructing unit configured to construct a relationship between the design element space and the emotional image space according to the determined relevant design elements of the emotional vocabularies in the first emotional vocabulary set.

16. The apparatus of claim 10, wherein, If the search purpose information indicates that the search purpose is a design element, the search result providing module is specifically configured to: The determined design elements are provided to the user as search results.

17. The apparatus of claim 10, wherein, If the search purpose information indicates that the search purpose is a product picture, the search result providing module is specifically configured to: query a product individual capable of deconstructing the determined design elements; provide the product picture of the queried product individual to the user as the search result.

18. The apparatus of claim 10, 16, or 17, wherein, The sensibility image space is constructed based on the names of the representative sensibility words in the first sensibility word set, the secondary sensibility words in the second sensibility word set, and the supplementary sensibility words in the third sensibility word set, and the relationships therebetween; the supplementary sensibility words in the third sensibility word set are obtained by searching for synonymous words based on the representative sensibility words in the first sensibility word set and the secondary sensibility words in the second sensibility word set; The word matching module comprises a first matching unit, a second matching unit, and a third matching unit. The first matching unit is configured to match the representative sensibility words in the sensibility image space of the sensibility engineering knowledge graph based on the search keyword; if the first matching unit successfully matches the representative sensibility words, the word matching module stops matching the sensibility words in the sensibility image space for the search keyword; The second matching unit is configured to continue matching the secondary sensibility words in the sensibility image space of the sensibility engineering knowledge graph if the matching of the representative sensibility words fails; if the second matching unit successfully matches the secondary sensibility words, the word matching module stops matching the sensibility words in the sensibility image space for the search keyword; The third matching unit is configured to continue matching the supplementary sensibility words in the sensibility image space of the sensibility engineering knowledge graph if the matching of the secondary sensibility words fails; if the third matching unit successfully matches the supplementary sensibility words, the word matching module stops matching the sensibility words in the sensibility image space for the search keyword.

Images