Furniture sales recommendation method and system based on artificial intelligence

Abstract

The application discloses a furniture sales recommendation method and system based on artificial intelligence, relates to the technical field of furniture recommendation, and deeply fuses multiple source heterogeneous data, constructs a high-dimensional user portrait, covers static attributes, behavior preferences and logistics tolerance and other implicit dimensions, relies on a composite recall strategy, realizes accurate insight into user intentions, simultaneously innovatively introduces space size matching detection, logistics path simulation and disassembly scheme optimization mechanism, completes the whole-process feasibility prediction of furniture from entering a house to placing in a recommended stage, effectively avoids the delivery failure risk caused by physical space limitation through output of the optimal carrying posture and bottleneck node information, and constructs end-to-end delivery guarantee capability.

Description

Technical Field

[0001] This invention relates to the field of furniture recommendation technology, specifically to a furniture sales recommendation method and system based on artificial intelligence. Background Technology

[0002] Against the backdrop of the rapid development of e-commerce and intelligent manufacturing, online furniture sales have become the mainstream consumption model. However, most existing recommendation systems are based on algorithms such as collaborative filtering and deep learning ranking, which focus on mining users' clicks, favorites and purchase preferences in their historical behavior to achieve interest matching. These methods have achieved remarkable results in the field of standardized products such as books and clothing, but they have fundamental limitations when directly transferred to the furniture sales scenario. As a typical large, non-standard commodity, furniture consumption decisions are strongly constrained by physical space and logistics conditions. When purchasing sofas, beds, wardrobes, and other products, users not only need to consider style, material, and price, but also need to assess whether the size matches the walls in their homes, whether it can pass through elevator doors and turn in the corridor, and whether it needs to be disassembled and reassembled to enter the home. Existing systems generally ignore the logistics feasibility dimension, resulting in a large number of orders being returned or canceled during the delivery process, which seriously damages user experience and platform operational efficiency. In addition, traditional methods lack the ability to perceive users' tolerance for logistics. Therefore, there is an urgent need for an intelligent recommendation method that integrates user preferences, environmental constraints, and logistics capabilities, and completes full-chain feasibility verification during the recommendation stage, thereby achieving a paradigm upgrade from interest matching to delivery assurance. Summary of the Invention

[0003] The purpose of this invention is to provide a furniture sales recommendation method and system based on artificial intelligence to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a furniture sales recommendation method based on artificial intelligence, comprising the following steps: S1. Collect raw data from the furniture manufacturer database and user-uploaded information. After aligning and standardizing the data, generate furniture standardized parameter vectors and user environment parameter vectors. Use the standardized parameter vectors to construct a knowledge graph. S2. After constructing a user profile vector based on the user environment parameter vector and historical behavior data, perform multi-way recall in parallel, filter candidate furniture from the furniture standardized parameter vector, and merge and deduplicate to form an initial candidate set; S3. Perform logistics feasibility verification on all furniture in the initial candidate set. Calculate the wall matching degree, passageway capacity, and obstacle obstruction through space size matching detection, and output the space matching score. S4. Simulate the furniture handling process in the initial candidate set, search for a collision-free path from outside the unit door to the placement position in the three-dimensional posture space, and output the path feasibility flag, the optimal handling posture and the set of bottleneck nodes. S5. For furniture where the path is not feasible but the user accepts disassembly and assembly, perform disassembly scheme optimization, find the component removal scheme with the lowest cost, and calculate the deformation difficulty index. S6. Calculate the estimated click-through rate score for each piece of furniture in the candidate set, and combine it with the space matching score, feasibility score and transformation difficulty index to obtain a comprehensive score. Based on the adjustment factor, analyze the fusion score of each piece of furniture and generate a recommendation list according to the score.

[0005] Preferably, step S1 specifically includes the following steps: S101. Collect the original furniture parameter set from the furniture manufacturer database, and obtain the original environmental data from the floor plan uploaded by the user, the property cooperation interface and historical logistics records. Analyze the floor plan, output the user's room size vector, and extract structured information from the furniture description in the original furniture parameter set. S102. After performing alignment and standardization on all data, generate a standardized parameter vector for each piece of furniture and an environmental parameter vector for all users. Store the standardized parameter vector and the environmental parameter vector in the furniture data window and the environmental data window, respectively. S103. Using the standardized parameter vector of each piece of furniture, construct furniture nodes and attribute nodes in the knowledge graph, aggregate all user historical behavior data, calculate the relationship weights between nodes, and thus generate a globally shared knowledge graph.

[0006] Preferably, step S2 specifically includes the following steps: S201. Construct a corresponding user profile vector based on the environmental data window and user historical behavior data. The user profile vector specifically includes the apartment area, assembly compatibility, preference scores for different styles, price sensitivity, decision cycle, acceptance of disassembly and assembly, and the longest waiting days. S202, Select the first For each user, the cosine similarity between the current user and other users is calculated based on the historical behavior data of all users. Users with similarity scores greater than a preset value are selected as similar users of the current user. Furniture that similar users have purchased but the current user has not yet interacted with is retrieved and added to the first candidate set. Indicates the serial number; S203. Extract the preference scores for different styles from the current user profile, filter the attribute nodes according to the preference scores, collect the corresponding furniture nodes using the knowledge graph and the current attribute nodes, average the embedding vector of the current attribute nodes to obtain the user preference center vector, calculate the correlation between this vector and the collected furniture node embedding vector, sort the furniture according to the correlation, select the furniture with a correlation higher than the preset value, and add it to the second candidate set. S204. After obtaining all candidate sets, the user's most recent 20 interactive furniture embedding vector sequences are used as input. A Transformer encoder is used to model the sequence and output a sequence representation. The sequence representation is then multiplied by the embedding vectors of the furniture in each candidate set to obtain the user's interest score for each piece of furniture. The furniture with the highest scores is then added to the third candidate set. S205. Aggregate the global furniture sales of the past 30 days from the order database, take the top-ranked furniture as the fourth candidate set, merge all candidate sets, remove duplicate furniture, and thus obtain the initial candidate set.

[0007] Preferably, step S3 specifically includes the following steps: S301. Extract the predefined type-room mapping table and the first [room type] from the candidate set. Furniture type, original dimension vector Packaging size vector ,in , , They represent the first The length, width, and height of the furniture pieces They represent the first The length, width, and height of the furniture package. Indicates the sequence number, retrieves the room size vector for the current user. and a list of obstacles According to the type-room mapping table and the first The type of furniture determines the corresponding wall length. ,use and Calculate the wall matching degree ,in ; S302, Setting a safety margin Then, using and Analyze the required width of the passageway ,in ,from Extract the narrowest aisle width ,like If the conflict type is recorded as "aisle width mismatch," the furniture placement will be adjusted. and Each obstacle in Occupied Area Perform an overlap test if If so, the conflict type is recorded as "obstacle exists"; S303. Utilization , , , as well as Calculate the spatial matching score ,in ,like Then the spatial matching score flag will be set. Set it to 1, otherwise set it to 0.

[0008] Preferably, step S4 specifically includes the following steps: S401. Extract the current user's logistics path node sequence. After reading the 3D mesh model of the furniture and the geometric model of the obstacles, set the first... The positional and rotational boundaries of a piece of furniture in three-dimensional space, where , Represents spatial coordinates, This represents the rotation angle around the three coordinate axes, and initializes the search tree parameters, sampling probability, expansion step size, maximum number of iterations, and starting attitude parameters. and endpoint attitude parameters ; S402. In each iteration, using the sampling probability Randomly sample a reference pose With probability Will As the target pose, find the node closest to the target pose from the current search tree. ,exist In the direction toward the target posture, with step length Expand to generate new nodes Preset endpoint judgment threshold ,like If a node does not collide with any obstacles, it is added to the search tree. If a node in the tree has the same pose parameters as the endpoint, then the node is added to the search tree. The Euclidean distance is less than And the node is in the same pose as the target. If there is no collision between the paths, return from arrive Valid paths between; S403. After extracting all nodes on the effective path, match the spatial coordinates of each node with the logistics path node sequence, count the logistics nodes to which all attitude parameters belong, record all attitude parameters belonging to the logistics node for each logistics node, and calculate the minimum gap between the furniture enclosure box and the obstacle of the logistics node under each attitude parameter. If there is a minimum gap less than the gap threshold, add the logistics node to the bottleneck node set. S404. After extracting the bottleneck node set, select the node with the smallest gap from the set, backtrack the path, and filter out the first attitude parameter that passes through the node and has the same gap value. Use this attitude parameter as the optimal handling attitude parameter. S405. If the number of iterations does not exceed the maximum value and the path is returned directly, set the feasibility flag to 1. If the number of iterations exceeds the maximum value and the path is still not returned, use the original size vector. Replace packaging size vector Rerun the simulation; if the path is still not returned, set the feasibility flag to 0. S406. After a series of logistics path simulations, output the feasibility flag, optimal handling posture parameters, and bottleneck node set. If the feasibility flag is set to 0 and the furniture cannot be disassembled, then directly record the conflict type of the furniture as logistics infeasibility. If the furniture can be disassembled and the disassembly acceptance rate in the current user profile is greater than 0.3, then optimize the disassembly scheme.

[0009] Preferably, step S5 specifically includes the following steps: S501, Obtain the furniture to be disassembled, for the first... Each piece of furniture is used to construct an assembly tree model. ,in Represents a component set. Represent the edge set and define dimensional decision variables , , Indicates the demolition of the first Each component Indicates that the first [number] is reserved. Each component This represents the total number of components. All dismantled components are removed from the component set. The remaining components are counted, and each vertex is selected... The maximum and minimum values ​​in three directions are used to calculate the overall dimensions of the outer enclosure of the remaining components; S502, Set the disassembly cost function ,in , This represents the fixed penalty value for removing a single component. This represents the weighting coefficient for the disassembly time per second. Indicates the demolition of the first Estimated time required for each component Indicates the first Decision variables for each component This indicates the sequence number, and the specific constraint is that the dimensions of the disassembled enclosure must be able to pass through all logistics bottleneck nodes. S503. Traverse all decomposition combinations, and for each combination, determine the decision variables. Determine whether the disassembled dimensions can pass through all logistics bottlenecks. If so, calculate the disassembly cost. Choose the decision variable with the lowest cost as the optimal decision variable. And record the corresponding dimensions after disassembly, using Calculate the deformation difficulty index ,in , This indicates the preset maximum disassembly time, and sets the feasibility flag to 1; S504. If all decision variables do not meet the constraints, it is determined that the furniture cannot be moved into the user's home even after being disassembled. The conflict type is recorded as logistics infeasibility, and the feasibility flag is set to 0.

[0010] Preferably, step S6 specifically includes the following steps: S601. After obtaining the space matching score, feasibility score and transformation difficulty index of each piece of furniture in the candidate set, the click-through rate prediction score of each candidate piece of furniture is calculated by analyzing user profiles, standardized parameter vectors of furniture and historical user behavior data. S602, Utilizing spatial matching scoring indicators for each piece of furniture Feasibility indicators Deformation difficulty index Calculate the overall feasibility score ,in ; S603, Assembly compatibility based on user profiles Acceptance of disassembly / reassembly markings and deformation difficulty index Calculate the adjustment factor ,in ; S604. Calculate the fusion score using the click-through rate prediction score, feasibility comprehensive score, and adjustment factor. Sort the candidate set in descending order and select the top 20 furniture items as the [number]. The final recommendation list for each user is generated and output through a visual interface. This process is repeated to determine and output the recommendation list for each user.

[0011] The artificial intelligence-based furniture sales recommendation system includes a data acquisition unit, a candidate set generation unit, a scoring label output unit, a path planning unit, an index calculation unit, and a recommendation determination unit. The data acquisition unit collects raw data from the furniture manufacturer database and user-uploaded information. After aligning and standardizing the data, it generates furniture standardized parameter vectors and user environment parameter vectors, and uses the standardized parameter vectors to construct a knowledge graph. The candidate set generation unit constructs a user profile vector based on the user environment parameter vector and historical behavior data, and then performs multi-way recall in parallel to filter candidate furniture from the furniture standardized parameter vector. After merging and deduplication, an initial candidate set is formed. The scoring flag output unit performs logistics feasibility verification on all furniture in the initial candidate set, calculates the wall matching degree, passageway capacity and obstacle obstruction through space size matching detection, and outputs space matching scoring flags. The path planning unit simulates the handling process of furniture in the initial candidate set, searches for a collision-free path from outside the unit door to the placement position in the three-dimensional posture space, and outputs the path feasibility flag, the optimal handling posture, and the set of bottleneck nodes. For furniture where the path is not feasible but the user accepts disassembly and assembly, the index calculation unit performs disassembly scheme optimization, finds the component removal scheme with the lowest cost, and calculates the deformation difficulty index. The recommendation determination unit calculates the estimated click-through rate score for each piece of furniture in the candidate set, and combines it with the spatial matching score, feasibility score, and deformation difficulty index to obtain a comprehensive score. Based on the adjustment factor, it analyzes the fusion score of each piece of furniture and generates a recommendation list according to the score.

[0012] Compared with the prior art, the beneficial effects of the present invention are: This invention deeply integrates multi-source heterogeneous data to construct a high-dimensional user profile, covering implicit dimensions such as static attributes, behavioral preferences, and logistics tolerance. Relying on a composite recall strategy, it achieves accurate insight into user intent. At the same time, it innovatively introduces spatial size matching detection, logistics path simulation, and disassembly scheme optimization mechanisms. It completes the feasibility prediction of the entire process of furniture from delivery to placement during the recommendation stage. By outputting the optimal handling posture and bottleneck node information, it effectively avoids the risk of delivery failure caused by physical space limitations and builds end-to-end delivery assurance capabilities. Attached Figure Description

[0013] Figure 1 An overall method flowchart is provided for embodiments of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Example 1: Please see Figure 1 This invention provides a technical solution: a furniture sales recommendation method based on artificial intelligence, comprising the following steps: S1. Collect raw data from the furniture manufacturer database and user-uploaded information. After aligning and standardizing the data, generate furniture standardized parameter vectors and user environment parameter vectors. Use the standardized parameter vectors to construct a knowledge graph. S2. After constructing a user profile vector based on the user environment parameter vector and historical behavior data, perform multi-way recall in parallel, filter candidate furniture from the furniture standardized parameter vector, and merge and deduplicate to form an initial candidate set; S3. Perform logistics feasibility verification on all furniture in the initial candidate set. Calculate the wall matching degree, passageway capacity, and obstacle obstruction through space size matching detection, and output the space matching score. S4. Simulate the furniture handling process in the initial candidate set, search for a collision-free path from outside the unit door to the placement position in the three-dimensional posture space, and output the path feasibility flag, the optimal handling posture and the set of bottleneck nodes. S5. For furniture where the path is not feasible but the user accepts disassembly and assembly, perform disassembly scheme optimization, find the component removal scheme with the lowest cost, and calculate the deformation difficulty index. S6. Calculate the estimated click-through rate score for each piece of furniture in the candidate set, and combine it with the space matching score, feasibility score and transformation difficulty index to obtain a comprehensive score. Based on the adjustment factor, analyze the fusion score of each piece of furniture and generate a recommendation list according to the score.

[0016] S1 specifically includes the following steps: S101. Collect the original furniture parameter set from the furniture manufacturer database, and obtain the original environmental data from the floor plan uploaded by the user, the property cooperation interface and historical logistics records. Analyze the floor plan, output the user's room size vector, and extract structured information from the furniture description in the original furniture parameter set. S102. After performing alignment and standardization on all data, generate a standardized parameter vector for each piece of furniture and an environmental parameter vector for all users. Store the standardized parameter vector and the environmental parameter vector in the furniture data window and the environmental data window, respectively. S103. Construct furniture nodes and attribute nodes in the knowledge graph using the standardized parameter vector of each piece of furniture, aggregate all user historical behavior data, calculate the relationship weights between nodes, and thus generate a globally shared knowledge graph. S2 specifically includes the following steps: S201. Construct a corresponding user profile vector based on the environmental data window and user historical behavior data. The user profile vector specifically includes: apartment size, assembly compatibility, preference scores for different styles, price sensitivity, decision-making cycle, acceptance of disassembly / reassembly, and maximum waiting days. Extract the first... Room size vector for each user ,use The system calculates the apartment area, and then analyzes the percentage of time a user clicks on installation videos and browses assembly-related products in their historical behavior data. It uses the Sigmoid function to map these data and obtain the assembly compatibility score. It iterates through each interaction event in the user's historical behavior data (including purchase, adding to cart, liking, and clicking), and weights furniture belonging to corresponding style tags according to preset weights to obtain the user's preference score for different styles. It also analyzes price sensitivity based on the ratio between historical purchase prices and the median price across the entire site. The system calculates the average number of days from the user's first click to their final purchase, using this as the current user's decision-making cycle. It directly collects the user's acceptance of disassembly / assembly through a questionnaire pop-up, setting the default value to 0.5. Finally, it calculates the longest waiting time using the median of historical order delivery intervals in the user's historical behavior data. S202. Select the i-th user, calculate the cosine similarity between the current user and other users based on the historical behavior data of all users, filter out other users whose similarity is greater than the preset value, and take them as similar users of the current user. Search for furniture that similar users have purchased but the current user has not yet interacted with, and include them in the first candidate set, where i represents the sequence number. S203. Extract the preference scores for different styles from the current user profile, filter the attribute nodes according to the preference scores, collect the corresponding furniture nodes using the knowledge graph and the current attribute nodes, average the embedding vector of the current attribute nodes to obtain the user preference center vector, calculate the correlation between this vector and the collected furniture node embedding vector, sort the furniture according to the correlation, select the furniture with a correlation higher than the preset value, and add it to the second candidate set. S204. After obtaining all candidate sets, the user's most recent 20 interactive furniture embedding vector sequences are used as input. A Transformer encoder is used to model the sequence and output a sequence representation. The sequence representation is then multiplied by the embedding vectors of the furniture in each candidate set to obtain the user's interest score for each piece of furniture. The furniture with the highest scores is then added to the third candidate set. S205. Aggregate the global furniture sales of the past 30 days from the order database, take the top-ranked furniture as the fourth candidate set, merge all candidate sets, remove duplicate furniture, and thus obtain the initial candidate set; S3 specifically includes the following steps: S301. Extract the predefined type-room mapping table and the first [room type] from the candidate set. Furniture type, original dimension vector Packaging size vector ,in , , They represent the first The length, width, and height of the furniture pieces They represent the first The length, width, and height of the furniture package. Indicates the sequence number, retrieves the room size vector for the current user. and a list of obstacles According to the type-room mapping table and the first The type of furniture determines the corresponding wall length. ,use and Calculate the wall matching degree ,in ; S302, Setting a safety margin Then, using and Analyze the required width of the passageway ,in ,from Extract the narrowest aisle width ,like If the conflict type is recorded as "aisle width mismatch," the furniture placement will be adjusted. and Each obstacle in Occupied Area Perform an overlap test if If so, the conflict type is recorded as "obstacle exists"; S303. Utilization , , , as well as Calculate the spatial matching score ,in ,like Then the spatial matching score flag will be set. Set to 1, otherwise set to 0; S4 specifically includes the following steps: S401. Extract the current user's logistics path node sequence. After reading the 3D mesh model of the furniture and the geometric model of the obstacles, set the first... The positional and rotational boundaries of a piece of furniture in three-dimensional space, where , Represents spatial coordinates, This represents the rotation angle around the three coordinate axes, and initializes the search tree parameters, sampling probability, expansion step size, maximum number of iterations, and starting attitude parameters. and endpoint attitude parameters ; S402. In each iteration, using the sampling probability Randomly sample a reference pose With probability Will As the target pose, find the node closest to the target pose from the current search tree. ,exist In the direction toward the target posture, with step length Expand to generate new nodes Preset endpoint judgment threshold ,like If a node does not collide with any obstacles, it is added to the search tree. If a node in the tree has the same pose parameters as the endpoint, then the node is added to the search tree. The Euclidean distance is less than And the node is in the same pose as the target. If there is no collision between the paths, return from arrive Valid paths between; S403. After extracting all nodes on the effective path, match the spatial coordinates of each node with the logistics path node sequence, count the logistics nodes to which all attitude parameters belong, record all attitude parameters belonging to the logistics node for each logistics node, and calculate the minimum gap between the furniture enclosure box and the obstacle of the logistics node under each attitude parameter. If there is a minimum gap less than the gap threshold, add the logistics node to the bottleneck node set. S404. After extracting the bottleneck node set, select the node with the smallest gap from the set, backtrack the path, and filter out the first attitude parameter that passes through the node and has the same gap value. Use this attitude parameter as the optimal handling attitude parameter. S405. If the number of iterations does not exceed the maximum value and the path is returned directly, set the feasibility flag to 1. If the number of iterations exceeds the maximum value and the path is still not returned, use the original size vector. Replace packaging size vector Rerun the simulation; if the path is still not returned, set the feasibility flag to 0. S406. After a series of logistics path simulations, output the feasibility flag, optimal handling posture parameters and bottleneck node set. If the feasibility flag is set to 0 and the furniture cannot be disassembled, then directly record the conflict type of the furniture as logistics infeasibility. If the furniture can be disassembled and the disassembly acceptance rate in the current user profile is greater than 0.3, then optimize the disassembly scheme. S5 specifically includes the following steps: S501, Obtain the furniture to be disassembled, for the first... Each piece of furniture is used to construct an assembly tree model. ,in Represents a component set. Represent the edge set and define dimensional decision variables , , Indicates the demolition of the first Each component Indicates that the first [number] is reserved. Each component This represents the total number of components. All dismantled components are removed from the component set. The remaining components are counted, and each vertex is selected... The maximum and minimum values ​​in three directions are used to calculate the overall size of the outer enclosure of the remaining parts. The assembly tree model structure is obtained by parsing the product instruction manual PDF file of the furniture. Specifically, the layout analysis model is used to perform area detection on the PDF page. After reading the name label of each part using optical character recognition, the tree structure is constructed according to preset rules. S502, Set the disassembly cost function ,in , This represents the fixed penalty value for removing a single component. This represents the weighting coefficient for the disassembly time per second. Indicates the demolition of the first Estimated time required for each component Indicates the first Decision variables for each component This indicates the sequence number, and the specific constraint is that the dimensions of the disassembled enclosure must be able to pass through all logistics bottleneck nodes. S503. Traverse all decomposition combinations, and for each combination, determine the decision variables. Determine whether the disassembled dimensions can pass through all logistics bottlenecks. If so, calculate the disassembly cost. Choose the decision variable with the lowest cost as the optimal decision variable. And record the corresponding dimensions after disassembly, using Calculate the deformation difficulty index ,in , This indicates the preset maximum disassembly time, and sets the feasibility flag to 1; S504. If all decision variables do not meet the constraints, it is determined that the furniture cannot be moved into the user's home even after being disassembled. The conflict type is recorded as logistics infeasibility, and the feasibility flag is set to 0. S6 specifically includes the following steps: S601. After obtaining the space matching score, feasibility score and transformation difficulty index of each piece of furniture in the candidate set, the click-through rate prediction score of each candidate piece of furniture is calculated by analyzing user profiles, standardized parameter vectors of furniture and historical user behavior data. S602, Utilizing spatial matching scoring indicators for each piece of furniture Feasibility indicators Deformation difficulty index Calculate the overall feasibility score ,in ; S603, Assembly compatibility based on user profiles Acceptance of disassembly / reassembly markings and deformation difficulty index Calculate the adjustment factor ,in ; S604. Calculate the fusion score using the click-through rate prediction score, feasibility comprehensive score, and adjustment factor. Sort the candidate set in descending order and select the top 20 furniture items as the [number]. The final recommendation list for each user is generated and output through a visual interface. The same operation is repeated to determine and output the recommendation list for each user. Example 2: The present invention also provides an artificial intelligence-based furniture sales recommendation system, including a data acquisition unit, a candidate set generation unit, a scoring label output unit, a path planning unit, an index calculation unit, and a recommendation determination unit; The data acquisition unit collects raw data from the furniture manufacturer database and user-uploaded information. After aligning and standardizing the data, it generates furniture standardized parameter vectors and user environment parameter vectors. The standardized parameter vectors are then used to construct a knowledge graph. The candidate set generation unit constructs a user profile vector based on the user environment parameter vector and historical behavior data, and then performs multi-way recall in parallel to select candidate furniture from the standardized furniture parameter vector. After merging and deduplication, the initial candidate set is formed. The scoring tag output unit performs logistics feasibility verification on all furniture in the initial candidate set, calculates the wall matching degree, aisle passage capacity and obstacle obstruction through space size matching detection, and outputs space matching scoring tags; The path planning unit simulates the handling process of furniture in the initial candidate set, searches for a collision-free path from outside the unit door to the placement position in the three-dimensional posture space, and outputs the path feasibility indicator, the optimal handling posture, and the set of bottleneck nodes. For furniture where the path is not feasible but the user accepts disassembly and assembly, the index calculation unit performs disassembly scheme optimization, finds the component removal scheme with the lowest cost, and calculates the deformation difficulty index. The recommended unit calculates the estimated click-through rate score for each piece of furniture in the candidate set, and combines it with the space matching score, feasibility score and transformation difficulty index to obtain a comprehensive score. Based on the adjustment factor, the integration score of each piece of furniture is analyzed, and a recommendation list is generated according to the score.

[0017] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0018] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A furniture sales recommendation method based on artificial intelligence, characterized in that, The method includes the following steps: S1. Collect raw data from the furniture manufacturer database and user-uploaded information. After aligning and standardizing the data, generate furniture standardized parameter vectors and user environment parameter vectors. Use the standardized parameter vectors to construct a knowledge graph. S2. After constructing a user profile vector based on the user environment parameter vector and historical behavior data, perform multi-way recall in parallel, filter candidate furniture from the furniture standardized parameter vector, and merge and deduplicate to form an initial candidate set; S3. Perform logistics feasibility verification on all furniture in the initial candidate set. Calculate the wall matching degree, passageway capacity, and obstacle obstruction through space size matching detection, and output the space matching score. S4. Simulate the furniture handling process in the initial candidate set, search for a collision-free path from outside the unit door to the placement position in the three-dimensional posture space, and output the path feasibility flag, the optimal handling posture and the set of bottleneck nodes. S5. For furniture where the path is not feasible but the user accepts disassembly and assembly, perform disassembly scheme optimization, find the component removal scheme with the lowest cost, and calculate the deformation difficulty index. S6. Calculate the estimated click-through rate score for each piece of furniture in the candidate set, and combine it with the space matching score, feasibility score and transformation difficulty index to obtain a comprehensive score. Based on the adjustment factor, analyze the fusion score of each piece of furniture and generate a recommendation list according to the score.

2. The furniture sales recommendation method based on artificial intelligence according to claim 1, characterized in that: S1 specifically includes the following steps: S101. Collect the original furniture parameter set from the furniture manufacturer database, and obtain the original environmental data from the floor plan uploaded by the user, the property cooperation interface and historical logistics records. Analyze the floor plan, output the user's room size vector, and extract structured information from the furniture description in the original furniture parameter set. S102. After performing alignment and standardization on all data, generate a standardized parameter vector for each piece of furniture and an environmental parameter vector for all users. Store the standardized parameter vector and the environmental parameter vector in the furniture data window and the environmental data window, respectively. S103. Using the standardized parameter vector of each piece of furniture, construct furniture nodes and attribute nodes in the knowledge graph, aggregate all user historical behavior data, calculate the relationship weights between nodes, and thus generate a globally shared knowledge graph.

3. The furniture sales recommendation method based on artificial intelligence according to claim 1, characterized in that: S2 specifically includes the following steps: S201. Construct a corresponding user profile vector based on the environmental data window and user historical behavior data. The user profile vector specifically includes the apartment area, assembly compatibility, preference scores for different styles, price sensitivity, decision cycle, acceptance of disassembly and assembly, and the longest waiting days. S202, Select the first For each user, the cosine similarity between the current user and other users is calculated based on the historical behavior data of all users. Users with similarity scores greater than a preset value are selected as similar users of the current user. Furniture that similar users have purchased but the current user has not yet interacted with is retrieved and added to the first candidate set. Indicates the serial number; S203. Extract the preference scores for different styles from the current user profile, filter the attribute nodes according to the preference scores, collect the corresponding furniture nodes using the knowledge graph and the current attribute nodes, average the embedding vector of the current attribute nodes to obtain the user preference center vector, calculate the correlation between this vector and the collected furniture node embedding vector, sort the furniture according to the correlation, select the furniture with a correlation higher than the preset value, and add it to the second candidate set. S204. After obtaining all candidate sets, the user's most recent 20 interactive furniture embedding vector sequences are used as input. A Transformer encoder is used to model the sequence and output a sequence representation. The sequence representation is then multiplied by the embedding vectors of the furniture in each candidate set to obtain the user's interest score for each piece of furniture. The furniture with the highest scores is then added to the third candidate set. S205. Aggregate the global furniture sales of the past 30 days from the order database, take the top-ranked furniture as the fourth candidate set, merge all candidate sets, remove duplicate furniture, and thus obtain the initial candidate set.

4. The furniture sales recommendation method based on artificial intelligence according to claim 1, characterized in that: S3 specifically includes the following steps: S301. Extract the predefined type-room mapping table and the first [room type] from the candidate set. Furniture type, original dimension vector Packaging size vector ,in , , They represent the first The length, width, and height of the furniture pieces They represent the first The length, width, and height of the furniture package. Indicates the sequence number, retrieves the room size vector for the current user. and a list of obstacles According to the type-room mapping table and the first The type of furniture determines the corresponding wall length. ,use and Calculate the wall matching degree ,in ; S302, Setting a safety margin Then, using and Analyze the required width of the passageway ,in ,from Extract the narrowest aisle width ,like If the conflict type is recorded as "aisle width mismatch," the furniture placement will be adjusted. and Each obstacle in Occupied Area Perform an overlap test if If so, the conflict type is recorded as "obstacle exists"; S303. Utilization , , , as well as Calculate the spatial matching score ,in ,like Then the spatial matching score flag will be set. Set it to 1, otherwise set it to 0.

5. The furniture sales recommendation method based on artificial intelligence according to claim 1, characterized in that: S4 specifically includes the following steps: S401. Extract the current user's logistics path node sequence. After reading the 3D mesh model of the furniture and the geometric model of the obstacles, set the first... The positional and rotational boundaries of a piece of furniture in three-dimensional space, where , Represents spatial coordinates, This represents the rotation angle around the three coordinate axes, and initializes the search tree parameters, sampling probability, expansion step size, maximum number of iterations, and starting attitude parameters. and endpoint attitude parameters ; S402. In each iteration, using the sampling probability Randomly sample a reference pose With probability Will As the target pose, find the node closest to the target pose from the current search tree. ,exist In the direction toward the target posture, with step length Expand to generate new nodes Preset endpoint judgment threshold ,like If a node does not collide with any obstacles, it is added to the search tree. If a node in the tree has the same pose parameters as the endpoint, then the node is added to the search tree. The Euclidean distance is less than And the node is in the same pose as the target. If there is no collision between the paths, return from arrive Valid paths between; S403. After extracting all nodes on the effective path, match the spatial coordinates of each node with the logistics path node sequence, count the logistics nodes to which all attitude parameters belong, record all attitude parameters belonging to the logistics node for each logistics node, and calculate the minimum gap between the furniture enclosure box and the obstacle of the logistics node under each attitude parameter. If there is a minimum gap less than the gap threshold, add the logistics node to the bottleneck node set. S404. After extracting the bottleneck node set, select the node with the smallest gap from the set, backtrack the path, and filter out the first attitude parameter that passes through the node and has the same gap value. Use this attitude parameter as the optimal handling attitude parameter. S405. If the number of iterations does not exceed the maximum value and the path is returned directly, set the feasibility flag to 1. If the number of iterations exceeds the maximum value and the path is still not returned, use the original size vector. Replace packaging size vector Rerun the simulation; if the path is still not returned, set the feasibility flag to 0. S406. After a series of logistics path simulations, output the feasibility flag, optimal handling posture parameters, and bottleneck node set. If the feasibility flag is set to 0 and the furniture cannot be disassembled, then directly record the conflict type of the furniture as logistics infeasibility. If the furniture can be disassembled and the disassembly acceptance rate in the current user profile is greater than 0.3, then optimize the disassembly scheme.

6. The furniture sales recommendation method based on artificial intelligence according to claim 1, characterized in that: S5 specifically includes the following steps: S501, Obtain the furniture to be disassembled, for the first... Each piece of furniture is used to construct an assembly tree model. ,in Represents a component set. Represent the edge set and define dimensional decision variables , , Indicates the demolition of the first Each component Indicates that the first [number] is reserved. Each component This represents the total number of components. All dismantled components are removed from the component set. The remaining components are counted, and each vertex is selected... The maximum and minimum values ​​in three directions are used to calculate the overall dimensions of the outer enclosure of the remaining components; S502, Set the disassembly cost function ,in , This represents the fixed penalty value for removing a single component. This represents the weighting coefficient for the disassembly time per second. Indicates the demolition of the first Estimated time required for each component Indicates the first Decision variables for each component This indicates the sequence number, and the specific constraint is that the dimensions of the disassembled enclosure must be able to pass through all logistics bottleneck nodes. S503. Traverse all decomposition combinations, and for each combination, determine the decision variables. Determine whether the disassembled dimensions can pass through all logistics bottlenecks. If so, calculate the disassembly cost. Choose the decision variable with the lowest cost as the optimal decision variable. And record the corresponding dimensions after disassembly, using Calculate the deformation difficulty index ,in , This indicates the preset maximum disassembly time, and sets the feasibility flag to 1; S504. If all decision variables do not meet the constraints, it is determined that the furniture cannot be moved into the user's home even after being disassembled. The conflict type is recorded as logistics infeasibility, and the feasibility flag is set to 0.

7. The furniture sales recommendation method based on artificial intelligence according to claim 1, characterized in that: S6 specifically includes the following steps: S601. After obtaining the space matching score, feasibility score and transformation difficulty index of each piece of furniture in the candidate set, the click-through rate prediction score of each candidate piece of furniture is calculated by analyzing user profiles, standardized parameter vectors of furniture and historical user behavior data. S602, Utilizing spatial matching scoring indicators for each piece of furniture Feasibility indicators Deformation difficulty index Calculate the overall feasibility score ,in ; S603, Assembly compatibility based on user profiles Acceptance of disassembly / reassembly markings and deformation difficulty index Calculate the adjustment factor ,in ; S604. Calculate the fusion score using the click-through rate prediction score, feasibility comprehensive score, and adjustment factor. Sort the candidate set in descending order and select the top 20 furniture items as the [number]. The final recommendation list for each user is generated and output through a visual interface. This process is repeated to determine and output the recommendation list for each user.

8. A furniture sales recommendation system based on artificial intelligence, characterized in that: The furniture sales recommendation system is applicable to the artificial intelligence-based furniture sales recommendation method according to any one of claims 1-7, and includes a data acquisition unit, a candidate set generation unit, a scoring flag output unit, a path planning unit, an index calculation unit, and a recommendation determination unit; The data acquisition unit collects raw data from the furniture manufacturer database and user-uploaded information. After aligning and standardizing the data, it generates furniture standardized parameter vectors and user environment parameter vectors, and uses the standardized parameter vectors to construct a knowledge graph. The candidate set generation unit constructs a user profile vector based on the user environment parameter vector and historical behavior data, and then performs multi-way recall in parallel to filter candidate furniture from the furniture standardized parameter vector. After merging and deduplication, an initial candidate set is formed. The scoring flag output unit performs logistics feasibility verification on all furniture in the initial candidate set, calculates the wall matching degree, passageway capacity and obstacle obstruction through space size matching detection, and outputs space matching scoring flags. The path planning unit simulates the handling process of furniture in the initial candidate set, searches for a collision-free path from outside the unit door to the placement position in the three-dimensional posture space, and outputs the path feasibility flag, the optimal handling posture, and the set of bottleneck nodes. For furniture where the path is not feasible but the user accepts disassembly and assembly, the index calculation unit performs disassembly scheme optimization, finds the component removal scheme with the lowest cost, and calculates the deformation difficulty index. The recommendation determination unit calculates the estimated click-through rate score for each piece of furniture in the candidate set, and combines it with the spatial matching score, feasibility score, and deformation difficulty index to obtain a comprehensive score. Based on the adjustment factor, it analyzes the fusion score of each piece of furniture and generates a recommendation list according to the score.

Images