Pet intelligent product supply chain matching method based on layered AI architecture

Abstract

The invention discloses a pet intelligent product supply chain matching method based on a hierarchical AI architecture. The method comprises the following steps: firstly, obtaining a pet intelligent product modularization scheme customized by a user; then semantic enhancement and compatibility check are carried out through a product-manufacturer knowledge graph driven by symbology AI, and an initial manufacturer matching scheme is generated; mining a group reputation index of the similar user group by using a collaborative filtering algorithm in the linking AI; inputting the scheme, the reputation index and a constraint condition set including hardware compatibility, production cost and productivity into a constraint satisfaction problem solver for optimization solution, and outputting a final manufacturer matching scheme conforming to all constraints; and finally, generating a personalized explanation text through the generative AI. Through hierarchical fusion of three AI technologies, full-process automation from user personalized customization to manufacturer intelligent matching is realized, the problems of configuration conflict, insufficient resource integration and opaque decision are effectively solved, and the precision and efficiency of supply chain matching are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical fields of artificial intelligence and e-commerce, in particular to a pet intelligent product supply chain matching method combining a knowledge graph based on symbolic AI, collaborative filtering and constraint satisfaction algorithm based on connectionist AI, and generative AI based on hierarchical AI architecture. BACKGROUND

[0002] With the rapid development of the pet economy, the pet intelligent product market shows a personalized and customized trend. Users want to customize products that meet the specific needs of their pets through modular combination, such as intelligent feeders, pet monitoring devices, etc.

[0003] However, traditional recommendation systems are mostly based on simple rules or historical purchase records, which cannot deeply understand the complex relationship between pet attributes, behavior habits and product modules, and lack comprehensive consideration of multi-dimensional information such as manufacturer production capacity, quality certification, production capacity saturation, cost interval, etc., resulting in insufficient matching accuracy.

[0004] At the same time, when users customize multiple functional modules, traditional methods have difficulty in automatically detecting compatibility problems in electrical interfaces, communication protocols, physical structures, etc., relying on manual judgment, which is inefficient and prone to errors.

[0005] In addition, traditional systems lack the ability to explain matching results, making it difficult for users to understand the reasons for the recommendations, affecting decision-making confidence.

[0006] Therefore, there is an urgent need in the art for an individualized recommendation solution that can overcome the above-mentioned deficiencies, provide accurate, comprehensive, interpretable and conflict-free configuration. SUMMARY

[0007] The present application overcomes the above-mentioned shortcomings of the prior art and provides a pet intelligent product supply chain matching method based on hierarchical AI architecture to solve the problems of low matching accuracy, compatibility detection difficulty, insufficient supply chain resource integration and non-transparent decision-making in the prior art.

[0008] To achieve the above-mentioned purpose, the present application proposes a pet intelligent product supply chain matching method based on hierarchical AI architecture, the core idea of which is to integrate the semantic reasoning ability of the knowledge graph based on symbolic AI, the group wisdom of collaborative filtering based on connectionist AI, and the precise solving ability of constraint satisfaction problems, the language understanding and generation ability of generative AI, forming a hybrid recommendation strategy. The method comprises the following steps: S1. Obtain the user-defined pet intelligent product modularization customization scheme; S2. Based on the pre-constructed product-manufacturer knowledge graph, the customization scheme is semantically enhanced and compatibility is checked, and an initial manufacturer matching scheme is generated according to the pet portrait, the knowledge graph including product module entities, manufacturer entities and their attributes, and the association relationship between them; S3. Using a collaborative filtering algorithm, find a similar user group to the current user, and obtain the group reputation index of the similar user group to the potential manufacturer; S4. Input the initial manufacturer matching scheme, the group reputation index, and the preset constraint condition set into the constraint satisfaction problem solver, the constraint condition set including hardware compatibility constraints, production cost constraints, and production capacity constraints; S5. The constraint satisfaction problem solver optimizes and resolves conflicts in the initial manufacturer matching scheme, and outputs one or more final manufacturer matching schemes that meet all constraint conditions; S6. Use a generative artificial intelligence component to generate a personalized explanation text for the final manufacturer matching scheme and present it to the user or enterprise.

[0009] Preferably, the step S2 of generating an initial recommended product set and an initial configuration scheme based on the product-manufacturer knowledge graph specifically includes: Matching the module entities in the customization scheme with the nodes in the knowledge graph; By traversing the edges connected to the matched nodes in the knowledge graph, check the compatibility between modules, and find associated modules according to the pet portrait; According to the weight and type of the edge, calculate the correlation degree of the manufacturer entity and the customization scheme, and generate an initial manufacturer matching scheme based on the correlation degree.

[0010] Preferably, the step S3 of using a collaborative filtering algorithm specifically uses a model-based collaborative filtering algorithm, which is trained by the following steps: Collect all anonymized user manufacturer cooperation data and satisfaction feedback data on the platform; Construct a user-manufacturer feature matrix; Use matrix decomposition or deep learning models to learn the latent feature vectors of users and manufacturers to predict the target user's preference score for uncooperative manufacturers and generate a group reputation index.

[0011] Preferably, the constraint condition set in step S4 includes: Hardware compatibility constraints ensure that each selected product function module is compatible with each other in terms of electrical interface, communication protocol and physical structure; Production cost constraints ensure that the total production cost of the final manufacturer matching scheme does not exceed the budget threshold; Production capacity constraints ensure that the production capacity saturation of the manufacturer allows timely completion of production.

[0012] Preferably, the optimization and conflict resolution in step S5 employs a backtracking search algorithm or a local search algorithm to maximize a comprehensive objective function composed of a knowledge graph-based matching degree score, a collaborative filtering-based group reputation score, a business factor score, and a constraint violation penalty term.

[0013] The present application improves matching accuracy by deeply understanding the complex correlation between pet attributes, product characteristics, and manufacturer capabilities through the semantic reasoning ability of the knowledge graph based on symbolic AI. It also automatically detects hardware compatibility issues to avoid production problems caused by module mismatch. It also integrates multi-objective optimization, considering factors such as technical compatibility, production cost, capacity constraints, and group preferences to achieve Pareto optimal solutions. Finally, it generates personalized explanation texts through generative AI to enhance user trust and decision-making efficiency. It realizes the full-process intelligentization from user customization to manufacturer matching, significantly improving supply chain collaboration efficiency.

[0014] The present application realizes full-process automation from user individual customization to manufacturer intelligent matching through the hierarchical integration of three AI technologies, effectively solving the problems of configuration conflict, insufficient resource integration, and opaque decision-making, significantly improving the accuracy and efficiency of supply chain matching.

[0015] The present application has the following advantages: The existing technology lacks semantic depth understanding, hardware compatibility detection is difficult, supply chain resource integration is insufficient, and the decision-making process is not transparent. The present application hierarchically integrates symbolic AI (knowledge graph), connectionist AI (collaborative filtering and constraint solving), and generative AI (natural language generation). Through the symbolic representation and logical reasoning ability of the knowledge graph, it realizes accurate understanding and processing of complex domain rules, laying the foundation for the technical rationality and safety of the solution. Through the collaborative filtering model based on connectionist AI, it learns potential preference patterns from massive user data and performs multi-objective optimization search through the constraint solver, making the recommendation results not only conform to the group trend but also optimal under multiple restrictions, realizing group wisdom and global optimization. Generative AI improves decision-making transparency and user experience: it converts complex structured data and reasoning results into natural language explanations of generative AI, greatly reducing the user's understanding threshold and enhancing the transparency and credibility of the system decision-making process. It realizes precise, automatic, and reliable supply chain matching. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is the overall flowchart of the method of the present application.

[0017] Figure 2 is a partial example graph of the knowledge graph in an embodiment of the present application.

[0018] Figure 3 is a module structure diagram of a personalized configuration recommendation system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0019] The present application will be described in detail below with reference to the accompanying drawings and embodiments.

[0020] Embodiment 1

[0021] The present embodiment is used to implement the aforementioned pet intelligent product supply chain matching method.

[0022] The user completes the modular configuration of the pet intelligent product through the interactive interface, and the system obtains a customized scheme containing each functional module and its parameter setting.

[0023] The system matches the customized scheme with a pre-constructed product-manufacturer knowledge graph. The knowledge graph (as shown in FIG. 1) contains product module entities, manufacturer entities and their attributes, and the association relationship therebetween. Figure 2

[0024] Through graph traversal and reasoning, the system checks the compatibility between the modules, and according to the portrait data such as pet breed, age, health condition, discovers and suggests adding modules with high correlation, and realizes semantic enhancement of the scheme.

[0025] A model-based collaborative filtering algorithm is used to map the current user and his customized scheme to a latent feature space, find a similar user group through cosine similarity, analyze the long-term cooperation tendency and satisfaction of the group to a specific manufacturer, and form a manufacturer group reputation index.

[0026] The problem is modeled as a constraint satisfaction problem, and the constraint conditions include hardware compatibility constraints (to ensure that the electrical interfaces, communication protocols and physical structures between the modules are compatible), production cost constraints (to ensure that the total cost does not exceed the budget threshold), and production capacity constraints (to ensure that the manufacturer's production capacity saturation allows timely delivery).

[0027] The solver uses backtracking search and local search algorithms to maximize the comprehensive objective function:

[0028] Wherein each component represents the knowledge graph matching degree, the collaborative filtering prediction score, the business factor score and the constraint violation penalty respectively.

[0029] The generative AI component receives the structured matching result, generates a natural language explanation text, and explains the recommendation reasons, manufacturer advantages and scheme characteristics, thereby enhancing the credibility and persuasiveness of the scheme.

[0030] Embodiment 2

[0031] Reference Figure 1 ​For example, the user customizes a smart litter box for his pet cat, the recommended method of the application is as follows: The user selects a basic litter box structure C, a weight sensor module D (used to record pet weight) and an automatic scale removal module E through the front-end interactive interface. The system obtains this structured scheme: {product: smart litter box; module: [basic structure C, weight sensor D, automatic scale removal module E]}.

[0032] The system matches and reasons this scheme with the pre-built product-manufacturer knowledge graph, locates the "automatic scale removal module E" node in the graph, traverses its "needs to be compatible" edge, and finds that the module clearly requires the "basic structure" to have "lateral slide rail" and "12V DC power interface". After checking, the power interface of "basic structure C" selected by the user is "5V USB", and the physical structure is "bottom rotation", and there is an "incompatible" edge connecting the two in the graph. Based on this explicit symbolic rule, the symbolic AI detects a hardware compatibility conflict. At the same time, the system traverses the "applicable-pet breed" edge according to the user's pet portrait (breed: British Shorthair, large weight), finds that "widened and reinforced basin body" has a very high correlation with this breed, and can provide the required "lateral slide rail" and "12V interface". Therefore, the system automatically corrects the user's scheme to use the compatible "widened and reinforced basin body C1" to replace "basic structure C". According to the "production" edge, the system finds a set of manufacturers that can produce these compatible modules, for example: manufacturer M can produce {C1, E}, and manufacturer N can produce {D}. Thus, the initial manufacturer matching scheme is generated.

[0033] The system inputs the current user (portrait: multi-pet family, health monitoring, low price sensitivity) and his customized scheme into the trained collaborative filtering model (trained using matrix decomposition method). The model maps users and manufacturers to the same latent feature space. The system finds the K user groups most similar to the current user in the feature space by calculating the cosine similarity. This group generally prefers to cooperate with manufacturers with "mature technology and strong durability". Analyzing the historical cooperation scores and repeat purchase rates of the similar group to candidate manufacturers M and N, the group reputation index is calculated: manufacturer M (0.88), manufacturer N (0.95).

[0034] The system formalizes the matching problem as a constraint satisfaction problem, requiring to meet: hardware compatibility; total production cost ≤ user budget (1500 yuan)'selected manufacturer's module production capacity saturation < 85%. The goal of the solver (using backtracking search algorithm) is to maximize the comprehensive score function:

[0035] Where: KGscore (Knowledge graph matching degree): According to the correlation edge weight calculation of "widened and reinforced basin body" and "doll cat" in the graph, this example is 0.9; CF score (Co-filter group reputation index): That is, the manufacturer's reputation index; Biz score (Business factor score) is calculated by the manufacturer's quotation and logistics cost, and the score after normalization. Weight coefficient α , β , γ Through training a logistic regression model on the platform's historical transaction data, the model coefficients are 0.5, 0.4, and 0.1, indicating that the current platform users value technical matching degree and group reputation the most. The penalty coefficient δ is set to 1000, which is much larger than the theoretical maximum value of the objective function reward item (theoretical upper limit <3), to ensure that any scheme that violates the hard constraint is automatically excluded. Penalty is defined as 1 if any hard constraint is not met, otherwise 0.

[0036] The solver traverses the possible manufacturer combinations. Finally, the scheme {manufacturer M: [basin C1, cleaning module E], manufacturer N: [weight sensor D]} is output as the final matching scheme because it satisfies all constraints, has high technical matching degree and good group reputation, and scores the highest in the objective function.

[0037] The generative AI component receives the final matching scheme and its key data: {scheme manufacturer list, core advantages: [hardware fully compatible, group evaluation excellent (95%), within budget], pet individualization points: [selected widened and reinforced basin body for your doll cat]}. Based on templates and data, the following natural language explanation text is generated: "The'manufacturer M + manufacturer N' combination scheme we recommend for you is a double guarantee of technology and reputation. First, we have selected a widened and reinforced basin body for your doll cat to ensure stable and comfortable use, and perfect compatibility with the automatic cleaning system, fundamentally avoiding hardware conflicts. The manufacturer N responsible for the core sensor has received 95% of the highest praise from similar pet owner groups, and its data monitoring is accurate and reliable. The total price of the whole set of scheme is 1420 yuan, strictly controlled within your budget, and all manufacturers have confirmed that they have sufficient capacity to deliver on time." The content described in the embodiments of the present specification is only a list of implementation forms of the inventive concept, and the protection scope of the present application should not be considered as limited to the specific forms described in the embodiments, and the protection scope of the present application also extends to equivalent technical means that can be thought of by those skilled in the art based on the inventive concept.

Claims

1. A pet intelligent product supply chain matching method based on a hierarchical AI architecture, characterized in that, The method comprises the following steps: S1. Obtain a user-defined pet intelligent product modular customization scheme; S2. Based on a pre-built product-manufacturer knowledge graph, perform semantic enhancement and compatibility checking on the customization scheme, and generate an initial manufacturer matching scheme according to a pet portrait, wherein the knowledge graph comprises product module entities, manufacturer entities and their attributes, and the association relationship therebetween; S3. Use a collaborative filtering algorithm to find a similar user group similar to the current user, and obtain a group reputation index of potential manufacturers of the similar user group; S4. Input the initial manufacturer matching scheme, the group reputation index, and a preset constraint condition set into a constraint satisfaction problem solver, wherein the constraint condition set comprises hardware compatibility constraints, production cost constraints, and production capacity constraints; S5. The constraint satisfaction problem solver optimizes and resolves conflicts of the initial manufacturer matching scheme, and outputs one or more final manufacturer matching schemes that meet all constraint conditions; S6. Use a generative artificial intelligence component to generate a personalized explanation text of the final manufacturer matching scheme, and present it to the user or enterprise.

2. The method of claim 1, wherein, Step S2 of generating an initial recommended product set and an initial configuration scheme based on the product-manufacturer knowledge graph comprises: Matching the module entities in the customization scheme with the nodes in the knowledge graph; By traversing the edges connected to the matched nodes in the knowledge graph, checking the compatibility between modules, and discovering associated modules according to the pet portrait; According to the weight and type of the edge, the association degree of the manufacturer entity and the customization scheme is calculated, and the initial manufacturer matching scheme is generated based on the association degree.

3. The method of claim 1, wherein, Step S3 uses a collaborative filtering algorithm, which is a model-based collaborative filtering algorithm. The algorithm model is trained by the following steps: Collect all anonymized user manufacturer cooperation data and satisfaction feedback data on the platform; Build a user-manufacturer feature matrix; Use matrix decomposition or deep learning models to learn the latent feature vectors of users and manufacturers to predict the target user's preference score for uncooperative manufacturers and generate a group reputation index.

4. The method of claim 1, wherein, The constraint condition set in step S4 comprises: Hardware compatibility constraints ensure that each selected product function module is compatible with each other in terms of electrical interface, communication protocol and physical structure; Production cost constraints ensure that the total production cost of the final manufacturer matching scheme does not exceed the budget threshold; Production capacity constraints ensure that the production capacity saturation of the manufacturer allows timely production.

5. The method of claim 1, wherein, The optimization and conflict resolution in step S5 use a backtracking search algorithm or a local search algorithm to maximize a comprehensive objective function, which is composed of a matching degree score based on a knowledge graph, a group reputation score based on collaborative filtering, a business factor score, and a constraint violation penalty term.

Images