Furniture intelligent adaptation recommendation method based on family behavior data

Abstract

The invention belongs to the technical field of furniture recommendation, and discloses an intelligent furniture adaptation recommendation method based on family behavior data. Behavior sensor data, member interaction data and furniture use data in a family environment are collected, a family periodic activity mode model is constructed according to the behavior sensor data and the member interaction data, a behavior conflict index is calculated, and based on the family periodic activity mode model and the furniture use data, the behavior conflict index is calculated. A dynamic interaction model of behaviors and furniture is constructed, adaptive adjustment parameters are generated, and the dynamic interaction model integrates the periodic activity mode and furniture features; according to the behavior conflict index and the adaptive adjustment parameter, generating a furniture intelligent recommendation strategy, the furniture intelligent recommendation strategy including priority ranking and layout adjustment; and based on the intelligent furniture recommendation strategy, outputting an adaptive furniture list, and adjusting furniture layout parameters, thereby realizing intelligent furniture adaptive recommendation based on family behavior data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of furniture recommendation, more particularly, the present application relates to a furniture intelligent adaptation recommendation method based on family behavior data. BACKGROUND

[0002] In the field of furniture recommendation, existing systems mostly make recommendations based on static user preference data, which ignores the dynamic changes of family behavior, leading to mismatch between the recommended results and actual family life. Specifically, static data such as furniture size, color or style cannot capture the daily activity patterns of family members, such as space occupation when multiple people interact, safety needs when children play, or comfort requirements when the elderly rest. This neglect leads to conflicts in the actual layout of recommended furniture, such as sofas blocking children's activity paths or bed heights not suitable for the elderly. At the same time, existing recommendation algorithms lack temporal analysis of periodic activity patterns in the family, and cannot handle periodic fluctuations in behavior data, such as activity differences between weekdays and weekends, leading to recommendations that ignore time sensitivity, such as the need for multi-functional furniture during weekend gatherings. In addition, existing systems have difficulty integrating multi-source behavior data, such as sensor-captured movement trajectories and voice interaction records, resulting in incomplete input to the recommendation model, especially in multi-member families, where behavior conflicts are not identified, such as the overlap between parents' work areas and children's play areas. These problems make existing furniture recommendations unable to adapt to dynamic family environments, limiting user satisfaction and actual application effectiveness.

[0003] In view of this, the present application proposes a furniture intelligent adaptation recommendation method based on family behavior data to solve the above problems. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a furniture intelligent adaptation recommendation method based on family behavior data, comprising: Step S1: collecting behavior sensor data, member interaction data and furniture usage data in the family environment, wherein the behavior sensor data includes movement trajectory and activity frequency, the member interaction data includes voice and location interaction, and the furniture usage data includes occupation time and interaction frequency; Step S2: constructing a family periodic activity pattern model according to the behavior sensor data and the member interaction data, and calculating a behavior conflict index; Step S3: based on the family periodic activity pattern model and the furniture usage data, constructing a dynamic interaction model between behavior and furniture, and generating adaptation adjustment parameters, wherein the dynamic interaction model integrates periodic activity patterns and furniture characteristics; Step S4: generating a furniture intelligent recommendation strategy according to the behavior conflict index and the adaptive adjustment parameter, wherein the furniture intelligent recommendation strategy comprises priority ranking and layout adjustment; Step S5: outputting an adaptive furniture list and adjusting furniture layout parameters based on the furniture intelligent recommendation strategy, wherein the adjustment takes into account space coordinates and behavior matching.

[0005] Preferably, the behavior sensor data, member interaction data and furniture usage data in the home environment are collected, including: Mobile trajectory and activity frequency data are collected by sensors installed in the home space as behavior sensor data; Voice and location interactions between family members are recorded as member interaction data, wherein the voice is captured by a microphone and the location interaction is obtained by a positioning module; The occupancy duration and interaction frequency data of existing furniture are monitored as furniture usage data, wherein the monitoring is achieved by embedded sensors.

[0006] Preferably, the family periodic activity pattern model is constructed according to the behavior sensor data and the member interaction data, and the behavior conflict index is calculated, including: The behavior sensor data is time-series segmented to extract periodic activity patterns, wherein the time-series segmentation refers to dividing into weekday data and weekend data based on a preset time window; The member interaction data is fused to construct a family periodic activity pattern model, wherein the family periodic activity pattern model takes activity patterns as nodes, interaction frequency as edge weights, and embeds member role classification; The space occupancy conflict value and the time overlap value are calculated according to the family periodic activity pattern model, wherein the space occupancy conflict value is obtained by vector intersection, and the time overlap value is obtained by matrix operation; The space occupancy conflict value and the time overlap value are weighted and summed to generate a behavior conflict index, wherein the weighted sum is calculated by a self-defined conflict aggregation function, which embeds family member role weights and adjusts the function parameters in combination with activity frequency.

[0007] Preferably, the behavior and furniture dynamic interaction model is constructed based on the family periodic activity pattern model and the furniture usage data, and the adaptive adjustment parameter is generated, including: Furniture attributes are extracted from furniture usage data to obtain furniture size and functional characteristics, and then size, material and multi-functional attributes are represented by feature vectors; constructing a dynamic interaction model based on the household periodic activity pattern model and the functional characteristics of the furniture, wherein the dynamic interaction model takes the periodic activity pattern as input, the feature vector as a regulating variable, and defines a coupling matrix; iteratively optimizing the dynamic interaction model to calculate an adaptation score, wherein the iterative optimization includes updating parameters of the dynamic interaction model to minimize behavior-furniture mismatch; generating an adaptation adjustment parameter according to the adaptation score.

[0008] Preferably, the generating a furniture intelligent recommendation strategy according to the behavior conflict index and the adaptation adjustment parameter comprises: fusing the behavior conflict index and the adaptation adjustment parameter to calculate a recommendation priority score, wherein the fusion is achieved through weighted average operation and considers the nonlinear influence of the conflict index; filtering candidate furniture from a furniture database based on the recommendation priority score, wherein the filtering applies threshold filtering and sorting mechanisms; generating a recommendation strategy containing layout adjustments, wherein the layout adjustments map the conflict index to spatial coordinates through a self-defined transformation function, which embeds the timing constraints of the periodic activity pattern and defines a coordinate transformation matrix; integrating the geometric data of the household space during the generation process.

[0009] Preferably, the outputting an adaptation furniture list and adjusting furniture layout parameters based on the furniture intelligent recommendation strategy comprises: sorting candidate furniture according to the furniture intelligent recommendation strategy to output an adaptation furniture list, wherein the sorting is based on descending order of priority scores; calculating position parameters of furniture in the household space, wherein the calculation considers room boundaries and movement trajectories; simulating the interaction of movement trajectories with furniture boundaries through a self-defined layout simulation algorithm and iteratively optimizing the position coordinates of furniture to adjust the position parameters to match the periodic activity pattern; outputting adjusted furniture layout parameters, including furniture rotation angles and spacing values.

[0010] Preferably, the behavior sensor data, member interaction data, and furniture usage data are aligned using timestamps based on global clock synchronization; After alignment, data normalization processing is performed to form a unified behavior dataset.

[0011] Preferably, the calculating spatial occupancy conflict values and time overlap values according to the household periodic activity pattern model comprises: extracting spatial occupancy vectors from the household periodic activity pattern model through node projection; calculating an intersection value of the space occupation vectors by set operation, as a space occupation conflict value; extracting a time overlap matrix from the family periodic activity pattern model, and calculating a matrix trace as a time overlap value, wherein the matrix trace calculation is performed by a self-defined time convolution function which embeds the periodic pattern of member interaction and convolves a time window sequence.

[0012] Preferably, the dynamic interaction model is iteratively optimized to calculate an adaptation score, including: initializing parameters of the dynamic interaction model, wherein the initialization sets initial weights using a random distribution; calculating a loss increment of each iteration based on a feedback mechanism of periodic activity patterns by a self-defined gradient calculation, updating the calculation of the adaptation score; in the iteration, a learning rate decay strategy is applied to adjust the update step; terminating the iteration when the behavior-furniture mismatch degree converges to a preset threshold, and outputting a final adaptation score.

[0013] Preferably, the candidate furniture is screened from the furniture database based on the recommendation priority score, including: threshold filtering the recommendation priority score based on the median of the score distribution; extracting matching items from the furniture database by querying the index, and filtering furniture with unmatched functional features as candidate furniture; outputting the screened candidate furniture and composing a list.

[0014] The technical effects and advantages of the furniture intelligent adaptation recommendation method based on family behavior data of the present application are: By collecting multi-dimensional data and constructing a family periodic activity pattern model and calculating a behavior conflict index, real-time capture and quantitative analysis of daily activities of family members (such as multi-person interaction, children playing or the elderly resting) are realized. Potential conflicts (such as the overlap of parents' work area and children's play area) are identified, so as to avoid problems such as blocking the activity path or unsuitable height in actual layout of furniture recommendation.

[0015] A dynamic interaction model is constructed based on the periodic activity pattern model and furniture usage data, and adaptation adjustment parameters are generated. Deep integration of behavior data and furniture attributes is realized, providing high-precision adaptation parameters, so that furniture recommendation is more time-sensitive and functionally matched.

[0016] The fusion behavior conflict index and the adaptive adjustment parameter are used to calculate a recommended priority score, the whole-link optimization from recommendation to layout is realized, not only a priority-ordered furniture list is output, but also the parameter is adjusted through simulation of behavior interaction, the furniture is ensured to be conflict-free in actual environment, the practicability and executability of the recommendation are improved, and accurate processing of complex home scenes (such as multi-member interaction) is realized. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A furniture intelligent adaptive recommendation method based on home behavior data. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0019] The present application adaptively adjusts the closing time and the air volume switching strategy by introducing the modulation or demodulation link of the LED infrared active light source to perform multi-channel time sequence joint determination on the residual heat attenuation of the cooking surface and the oil smoke scattering attenuation.

[0020] Please refer to Figure 1 In the embodiments of the present application, a furniture intelligent adaptive recommendation method based on home behavior data comprises: In view of the problems in the background art that the static data cannot capture the daily activity patterns of family members, leading to conflict of recommended furniture in actual layout, the existing recommendation algorithm lacks time sequence analysis on periodic activity patterns of the family, cannot handle periodic fluctuations in behavior data, leading to time sensitivity being ignored in the recommendation, and the existing system cannot integrate multi-source behavior data, resulting in incomplete input of the recommendation model, especially in multi-member families, behavior conflicts are not identified, the design of the present system is as follows: Step S1: collecting behavior sensor data, member interaction data and furniture use data in the home environment, wherein the behavior sensor data comprises movement trajectory and activity frequency, the member interaction data comprises voice and position interaction, and the furniture use data comprises occupation time length and interaction frequency; Specifically: the movement trajectory (which refers to the spatial coordinate sequence of movement of the family members) and the activity frequency data (the number of events (here, the events refer to movement or stay events) occurring per unit time) are collected by infrared sensors (the working principle is to detect heat source movement) or pressure sensors (the working principle is to record ground pressure changes) installed in the home space as behavior sensor data. record the voice and location interaction between family members as member interaction data, wherein the voice is captured by a microphone array (i.e. the audio waveform generated by the voice exchange between family members needs to be converted into a digital audio signal), and the location interaction is obtained by a positioning module (such as GPS or indoor beacon), i.e. the change in spatial coordinates during the interaction between family members (i.e. calculate the relative coordinate difference between family member A and family member B); monitor the occupancy duration (continuous use time) and interaction frequency (touch or proximity times) through embedded sensors (such as contact sensors built into furniture) as furniture usage data (used to represent the utilization rate of furniture).

[0021] These data are collected through a high-precision sensor network. When the user determines that furniture intelligent adaptation recommendation is needed, the system extracts historical data for a certain period of time, for example, historical data for the past 30 days, or can start from the current time and collect data for the next 30 days as the main input for the subsequent recommendation, to ensure real-time and accuracy. During the collection process, a sampling rate of 10 times per second is used for continuous monitoring.

[0022] The behavior sensor data, member interaction data and furniture usage data are aligned using timestamps based on global clock synchronization. After alignment, data normalization is performed to form a unified behavior data set, wherein normalization is calculated by z-score method.

[0023] Multi-dimensional data fusion improves the completeness of model input and solves the problem of unrecognized behavior conflicts; time series alignment and normalization ensure data consistency, which is convenient for subsequent analysis; users can choose historical or future data, which is suitable for different family scenarios, and significantly improves the matching degree of recommendation results and actual life.

[0024] In the prior art, the dynamic changes of family behavior are often ignored, resulting in mismatch between the recommendation results and the actual family life; there is a lack of time series analysis of periodic activity patterns of the family, which cannot handle periodic fluctuations in behavior data, such as activity differences between weekdays and weekends; it is difficult to integrate multi-source behavior data, especially in multi-member families, behavior conflicts are not identified, such as the overlap between parent work area and child play area, etc. Based on the above, the system is designed as follows: Step S2: constructing a family periodic activity pattern model based on the behavior sensor data and the member interaction data, and calculating a behavior conflict index; the family periodic activity pattern model is a network structure based on time series data; the behavior conflict index is an index for quantifying incompatible activities (i.e. activity patterns that interfere with each other in space or time between family members, for example, children playing in the same area while parents working in the living room, resulting in a spatial occupancy conflict), to ensure that the model reflects the family rhythm.

[0025] Specifically: time series segmentation is performed on the behavior sensor data, frequency components in the time series data are analyzed by Fourier transform or ARIMA model, repeated patterns are identified, and similar sequences are classified using a clustering algorithm (such as K-means) to extract periodic activity patterns (referring to a sequence of family activities that repeatedly occurs within a specific time window, such as breakfast preparation in the morning on weekdays or family gatherings in the afternoon on weekends), and the data is grouped into daily sequences according to the global timestamp; then, the window is divided according to the preset rules (such as weekdays from Monday to Friday and weekends from Saturday to Sunday); statistical indicators (such as activity frequency mean and variance) are calculated within each window, and a sliding window (e.g., 1 hour step) is applied to smooth the data to extract patterns, thereby realizing the division of data into weekday data and weekend data based on the preset time window.

[0026] The member interaction data (voice and location interaction) is superimposed as additional features into the time series segmented activity patterns by a graph embedding algorithm (such as GraphSAGE), for example, the voice interaction frequency is used as a weight to adjust the intensity of the activity sequence, to realize the fusion of member interaction data and construct a family periodic activity pattern model, wherein the family periodic activity pattern model takes activity patterns as nodes, connects each node with edges, and the interaction frequency is the edge weight, and the member role classification is embedded, which classifies family members into categories such as adults (parents), children and the elderly; each node (activity pattern) is assigned a role label, for example, a "living room gathering" node is labeled as "adults + children"; the edge weight is calculated based on the interaction frequency, such as the higher the voice interaction frequency, the greater the edge weight, thereby forming a weighted role embedded graph network; From the family periodic activity pattern model, relevant subgraphs are extracted, for example, a subset of nodes involving the same space is selected; thereafter, for the space occupation conflict value, the nodes are projected to the space vector, and the vector intersection is calculated; at the same time, for the time overlap value, a time matrix is constructed from the edge weight and node attribute, and matrix operation is performed; The conflict in the family periodic activity pattern model is quantified using a graph algorithm (such as an adjacency matrix representation), for example, by node metric calculation of space vector intersection or by edge path analysis of time matrix trace, wherein the space occupation conflict value is obtained by vector intersection, and the time overlap value is obtained by matrix operation; The space occupancy conflict value and the time overlap value are weighted and summed to generate a behavioral conflict index. The weighted summation is calculated using a custom conflict aggregation function (that is, the weighted summation is not a simple linear addition, but is implemented through a custom function (conflict aggregation function). This function can be non-linear, for example, f(x,y)=αx+βy+γxy, where γ represents the embedded interaction item, x represents the space occupancy conflict value, y represents the time overlap value, and α and β are coefficients based on the member role weights, for example, α=0.6 (emphasizing space conflict) and β=0.4 (emphasizing time conflict). The coefficients can be dynamically adjusted by the activity frequency. This function embeds the family member role weights and adjusts the parameters of the function in combination with the activity frequency.

[0027] For example, suppose a family has parents and children, with a space conflict value of 0.8 (the living room is occupied by parents working and children playing, occupying 80% of the area); and a time overlap value of 0.5 (50% of the time overlaps on weekday evenings). Using the conflict aggregation function f = 0.6 × 0.8 + 0.4 × 0.5 + 0.1 × (0.8 × 0.5) (embedded interaction items), the calculated behavioral conflict index is approximately 0.72 (high conflict, indicating that furniture needs to be adjusted to resolve it). Parameters can also be adjusted; if the child's role has a high weight (e.g., 0.7), then α in the function increases to 0.7 to emphasize space conflict.

[0028] The calculation of space occupancy conflict value and time overlap value based on the family periodic activity pattern model includes: Space occupancy vectors are extracted from the family periodic activity pattern model through node projection (node ​​projection maps nodes in the family periodic activity pattern model to a low-dimensional vector space, for example, using GCN technology to convert node attributes into vector representations). The intersection value of the space occupancy vectors is calculated through set operations (using Jaccard similarity). The intersection value is a measure of the degree of overlap between vectors, for example, Jaccard index = |A∩B|÷|A∪B|, the higher the value, the greater the conflict, where A and B represent different spaces. This intersection value is used as the space occupancy conflict value.

[0029] For example: Assuming a family of four (parents and two children), the family's periodic activity pattern model is a graph network: The nodes include: "Working in the living room on weekdays" (parents' role, space occupation: 80% of the central area of ​​the living room, represented as a vector [living room coordinates x1-y1, area 0.8]), "Children playing on weekends" (children's role, space occupation: the entire living room, represented as a vector [living room coordinates x1-y2, area 1.0]), and "Elderly rest" (elderly role, space occupation: 20% of the corner area of ​​the living room, represented as a vector [living room coordinates x2-y2, area 0.2]), where x1 and x2 represent the x-coordinates and y1 and y2 represent the y-coordinates; Edges connect these nodes, with weights based on interaction frequency (e.g., the edge weight between "working in the living room on weekdays" and "kids playing on weekends" is 0.7, indicating high interaction).

[0030] Extracting space occupancy vectors: Using node projection (GCN algorithm traverses the network and aggregates node attributes), a vector is generated for each node. For example, extract vector V1=[0.8, central area] from the "Weekday living room work" node; extract V2=[1.0, whole area] from the "Weekend children play" node.

[0031] Calculate the intersection value: Apply Jaccard similarity (set operation) to V1 and V2: Assuming V1's occupancy set A1 = {central 80% of pixels} and V2's B1 = {all living room pixels}, then Jaccard = |A∩B| / |A∪B| = 0.8 / 1.0 = 0.8 (representing 80% overlap). If we also consider the "elderly resting" vector V3, calculate the average intersection of the multiple vectors, and the conflict value is 0.75.

[0032] Result: Space occupancy conflict value = 0.75, indicating high conflict (e.g., a 75% overlap between the parents' work area and the children's play area, potentially causing the sofa to block the children's path). This value is used to generate a behavioral conflict index to help recommend adjustments to furniture layout (such as moving the sofa).

[0033] A subset of relevant nodes (e.g., nodes involving the same member or space) is selected from a family periodic activity pattern model. Then, a matrix is ​​constructed from the node's temporal attributes (embedded periodic patterns, such as weekday / weekend labels) and edge weights. Specifically, the network is represented using an adjacency matrix, with edge weights as matrix elements, and a temporal dimension (e.g., each element multiplied by a time window overlap factor) is added to generate an N×N matrix. Rows / columns correspond to nodes, and element values ​​represent the time overlap ratio of two activities (based on timestamp data), serving as the time overlap matrix. The extracted temporal overlap matrix is ​​processed using a trace operation, and the final output trace value serves as the temporal overlap value (a higher value indicates a greater temporal conflict). The matrix trace calculation is performed using a custom temporal convolution function, which smooths and extracts patterns from the time series. Specifically, f(t) = ∑w_k × s_{tk}, where w_k is the convolution kernel weight (a custom parameter), and s_{tk} is the time window sequence. Before matrix trace calculation, this function convolves the matrix elements to capture periodic fluctuations (e.g., kernel size = 7 days, capturing periods). The convolution kernel weights w_k are dynamically adjusted to include member interaction data; for example, interaction frequency (extracted from model edge weights) is embedded as a multiplier in the kernel: w_k = base_w × (interaction frequency factor + periodic pattern embedding). For example, if children's interaction periodic patterns show peaks on weekends, the function amplifies the weights of the weekend window to ensure the convolution output reflects these patterns. The cyclical pattern of member interaction refers to the repetitive pattern of interactions (such as voice or location interaction) between family members over time. For example, a recurring pattern every week (such as brief interactions between parents and children on weekday evenings and intensive interactions throughout the weekend).

[0034] Following the example above, the explanation continues as follows: Nodes: "Living room work on weekdays" (time mode: weekdays 18:00-20:00), "Children's play on weekends" (time mode: weekends 10:00-18:00), "Elderly rest" (time mode: daily 14:00-16:00).

[0035] Extract the time overlap matrix: Construct a 3×3 matrix M, where M_{ij} = the time overlap ratio (e.g., calculated based on timestamps). For example, M[1,2]=0.3 (30% overlap between "working in the living room on weekdays" and "children playing on weekends" on weekend evenings); M[1,1]=1.0 (self-stability); Complete matrix M= [1.0, 0.3, 0.2] 0.3, 1.0, 0.4 [0.2, 0.4, 1.0]; To obtain the matrix trace, first, apply a temporal convolution function f to convolve the matrix rows (embedding the member interaction cycle pattern). Assume the cycle pattern is "children's interaction peaks on weekends" (embedding with kernel weights w=[0.2,0.5,0.3], emphasizing weekends).

[0036] After convolution, the matrix M' = [0.9, 0.4, 0.25 0.4, 0.95, 0.45 [0.25, 0.45, 0.9] Then, calculate the trace: trace(M') = 0.9 + 0.95 + 0.9 = 2.75. After normalization, the time overlap value is 2.75 / 3 ≈ 0.92 (a high value, indicating severe time conflicts, such as 92% overlap between work and playtime, which may require multifunctional furniture).

[0037] By constructing a model of cyclical family activity patterns and calculating a behavioral conflict index, the system achieves precise capture and quantification of dynamic family behavior. Specific effects include capturing temporal fluctuations (such as weekday / weekend differences), identifying spatial or temporal conflicts (such as overlapping activities), and generating quantitative indicators (conflict indices) as input for subsequent steps, ensuring that furniture recommendations adapt to real-life situations. This addresses the problem of missing temporal analysis, improving the time sensitivity of recommendations; identifying multi-member conflicts, enhancing model completeness; computational efficiency and easy scalability; and enhancing user satisfaction by avoiding layout mismatches.

[0038] Static data fails to capture the daily activity patterns of family members, leading to conflicts between recommended furniture and actual layout; the lack of dynamic integration of behavioral data and furniture characteristics makes it unable to handle time-sensitive factors, such as periodic fluctuations; and it ignores the dynamic interaction between behavior and furniture. Based on these issues, this design is as follows: Step S3: Based on the family periodic activity pattern model and the furniture usage data, construct a dynamic interaction model between behavior and furniture, and generate adaptation adjustment parameters, wherein the dynamic interaction model integrates periodic activity patterns and furniture features; Specifically, through statistical analysis and feature engineering methods, such as applying clustering algorithms (e.g., DBSCAN) or regression models to identify patterns in data on occupancy time and interaction frequency, furniture attributes are extracted from furniture usage data to obtain furniture size and functional characteristics. Furniture size refers to physical specifications, such as length, width, and height; functional characteristics refer to practical attributes, such as adjustability, durability, or specific uses (e.g., whether a sofa can be converted into a bed). Then, feature vectors are used to represent size, material, and multi-functional attributes, such as multi-functional attribute = 0.7 (the score represents the degree of multi-purpose use, such as 0-1 normalization). Based on the family's periodic activity pattern model and the functional characteristics of the furniture, a dynamic interaction model is constructed. This dynamic interaction model takes the periodic activity pattern as input, the feature vector as adjustment variables, and defines a coupling matrix. The dynamic interaction model is a mathematical framework, similar to a coupled neural network or matrix model. For example, it can be a function D=f(A,F,C), where A is the periodic activity pattern; F is the furniture feature vector; and C is the coupling matrix (an M×N matrix whose elements define the interaction strength between the activity pattern and the furniture features). The dynamic interaction model is iteratively optimized to calculate the fit score, wherein the iterative optimization includes updating the parameters of the dynamic interaction model to minimize the behavior-furniture mismatch. The adaptation score is used to generate adaptation adjustment parameters. That is, the adaptation score is converted into parameters through a mapping function. For example, a linear transformation P=k×S+b is used, where S is the adaptation score and k and b are preset coefficients; or a threshold rule is applied. If S>0.8, a positive adjustment parameter is generated (such as increasing the furniture size by 10%). The parameters include specific values, such as layout offset or functional weight adjustment.

[0039] The iterative optimization of the dynamic interaction model and the calculation of the adaptation score include: Initialize the parameters of the dynamic interaction model, wherein the initialization uses a random distribution to set the initial weights; The loss increment for each iteration is calculated using a custom gradient calculation and a feedback mechanism based on a periodic activity pattern, and the fitness score is updated accordingly; (What exactly does this process look like?) During the iteration, a learning rate decay strategy is applied to adjust the update step size (the step size refers to the magnitude of the weight update). The iteration terminates when the behavior-furniture mismatch converges to a preset threshold, and the final fit score is output.

[0040] Suppose a family of three (parents and child) with recurring activity patterns including "working in the living room on weekdays" (input vector A = [frequency = 0.8, duration = 2 hours]) and "children playing on weekends" (A = [frequency = 0.6, duration = 4 hours]). Existing furniture: sofa. Use the data to extract the feature vector F = [size = 180x80x90, material = fabric (encoded 2), multi-functional = 0.5 (foldable)].

[0041] process: Construct a dynamic interaction model D=A×C×F, where C is a 2x3 coupling matrix. The initial weights are randomly set, such as C=[[0.1,-0.05,0.2],[0.15,0.1,-0.1]] (weights represent the interaction intensity).

[0042] Iterative optimization: Iteration 1: Input A and F, calculate predicted matching degree (e.g., 0.4); actual matching degree (from the used data) = 0.6; loss L = (0.4 - 0.6)^2 = 0.04. Feedback mechanism: Calculate gradient ∂L / ∂w ≈ 0.1 based on activity pattern; update weights (e.g., C[1,1] = 0.1 - 0.01 × 0.1 = 0.099); fitness score S = 1 - L = 0.96. Loss increment ΔL = 0.04 (initial).

[0043] Iterate 2-5 times: Continue updating, applying learning rate decay (initial η=0.01, decaying to 0.009); gradually reduce the loss to 0.02, 0.015, etc. The feedback mechanism is adjusted using activity modes (e.g., weekend mode amplifies the weights related to children).

[0044] Iteration 6: Loss L = 0.009 < preset threshold 0.01, terminate.

[0045] Output: Final fit score S = 0.991 (high score, indicating that the sofa matches the behavior pattern 99.1%, such as the multi-functional attribute being suitable for weekend play). This score is used to generate adjustment parameters (such as "increase sofa height by 10%" to suit the elderly's rest).

[0046] By integrating multi-source data through feature vectors and coupling matrices, real-time interactions (such as periodic patterns affecting furniture functionality) are captured; iterative optimization (gradient calculation and learning rate decay) ensures efficient convergence of model parameters, and adaptation scores are calculated to quantify the matching degree; adaptation adjustment parameters are generated to support subsequent layout optimization. Overall, this improves the responsiveness of furniture recommendations to dynamic home environments, reduces conflicts, and enhances user satisfaction.

[0047] Step S4: Generate a furniture intelligent recommendation strategy based on the behavioral conflict index and the adaptation adjustment parameters, wherein the furniture intelligent recommendation strategy includes priority ranking and layout adjustment; The behavioral conflict index and the adaptation adjustment parameter are fused to calculate the recommendation priority score. The fusion is achieved through a weighted average operation, taking into account the nonlinear effect of the conflict index. For example, the formula P2=(w1×(1-conflict index)+w2×adjustment parameter) / (w1+w2), where w1 / w2 are the weights, and a nonlinear function (such as sigmoid) is introduced to handle the influence of the conflict index (e.g., if the conflict is high, its negative weight is nonlinearly amplified) to calculate the recommendation priority score for each potential furniture (the higher the score, the higher the priority).

[0048] Based on the recommendation priority score, candidate furniture is filtered from the furniture database, wherein the filtering applies a threshold filtering and sorting mechanism; A recommended strategy incorporating layout adjustments is generated. These adjustments map the conflict index to spatial coordinates using a custom transformation function that embeds temporal constraints of periodic activity patterns and defines a coordinate transformation matrix. The custom transformation function, such as f(conflict index = coordinate offset = a × tanh(conflict index) + b), maps the conflict index to spatial coordinates (e.g., adjusting furniture positions to avoid conflict). Temporal constraints include higher weighting for weekday patterns. The coordinate transformation matrix is ​​a 2D rotation / translation matrix, such as [[cosθ, -sinθ], [sinθ, cosθ]]. The entire process integrates geometric data of the home space (e.g., room dimensions, boundary coordinates).

[0049] During the generation process, geometric data of the home space is integrated.

[0050] Assume a family of four (parents, child, and elderly) living in a 20m² living room (geometric data: rectangular boundary coordinates [0,0] to [10,10] meters, with a central pillar obstructing the view). The preceding steps have already calculated: Behavioral Conflict Index = 0.75 (high conflict, e.g., 75% spatial / temporal overlap between the parents' work area and the child's play area); Adaptation Adjustment Parameters = [Size Adjustment + 10%, Functional Weight 0.8] (calculated from the recommendation priority score in step S3). Input: Behavioral conflict index = 0.75 (high value, indicating that conflict needs to be resolved first); Adaptation adjustment parameter = 0.8 (high value, indicating good adaptation potential).

[0051] The weighted average P2 = w1 × (1 - conflict index) + w2 × adjustment parameter is used, where w1 = 0.6 (emphasizing conflict resolution) and w2 = 0.4 (emphasizing adaptation). Considering nonlinear effects: the sigmoid function is applied to adjust the conflict component, sigmoid(0.75) = 0.68 (nonlinear amplification of negative impacts), thus the adjusted (1 - conflict index) = 0.25 × (1 - 0.68) = 0.08.

[0052] Calculation: P=(0.6×0.08)+(0.4×0.8)=0.048+0.32=0.368 (The initial score is low, and there is a conflict due to the high score).

[0053] Repeat the calculation for each piece of furniture in the database (e.g., for "Multifunctional Sofa Ar": adjusted P2=0.75; for "Standard Table and Chair Br": P2=0.45; for "Adjustable Bed Cr": P2=0.82). Result: Generate a list of scores.

[0054] Database example: Sofa Ar (score 0.75, size 180x80, multi-functional 0.7); Table and chair Br (score 0.45, size 100x60, multi-functional 0.4); Bed Cr (score 0.82, size 200x100, multi-functional 0.9); Desk Dr (score 0.65, size 120x70, multi-functional 0.6).

[0055] Set the threshold to 0.6, and only keep scores > 0.6 (Sofa Ar = 0.75, Bed Cr = 0.82, Desk Dr = 0.65; filter out Table and Chair B = 0.45).

[0056] Candidate list sorted in descending order: [Bed Cr (0.82), Sofa Ar (0.75), Desk Dr (0.65)].

[0057] Integrate the geometric data of the family space (living room boundary [0,0]-[10,10], avoid the central column [4,4]-[5,5]).

[0058] Use a custom transformation function f(conflict index) = offset coordinate = 2 × tanh(0.75) + 1 ≈ 2 × 0.96 + 1 = 2.92 (calculate the offset); embed time constraints (e.g., weekday mode weight 0.7, weekend mode weight 0.3, adjustment function is f_t = 0.7 × f_work + 0.3 × f_weekend).

[0059] Define a coordinate transformation matrix: rotation matrix Me=[[cos(30°),-sin(30°)],[sin(30°),cos(30°)]]≈[[0.866,-0.5],[0.5,0.866]], used to adjust the furniture orientation to resolve conflicts.

[0060] For the bed Cr (high score 0.82), the original position is [3,3]. Apply the transformation: new coordinates = Me × [3,3] + [2.92,0] = [approximately 5.6, 4.1] (move to the edge of the living room to avoid conflicts with children's play paths); consider the geometric data to ensure that the new coordinates are within [0,0]-[10,10] and do not touch the pillars.

[0061] Generation strategy: including priority sorting (1. Bed Cr, 2. Sofa Ar, 3. Desk Dr) and layout adjustment (e.g., "Bed Cr: Rotate 30° and move to [5.6, 4.1] to match the weekend party pattern").

[0062] Step S5: Based on the intelligent furniture recommendation strategy, output a list of suitable furniture and adjust the furniture layout parameters, wherein the adjustment takes into account spatial coordinates and behavior matching.

[0063] Specifically: Candidate furniture is sorted according to the intelligent furniture recommendation strategy, and a list of suitable furniture is output, wherein the sorting is based on descending priority scores; Calculate the positional parameters of furniture within the home space, where the calculation considers room boundaries and movement trajectories. For example, the initial position is assigned via a center point (bed Cr initial [3,3]); boundary checks (ensuring position coordinates ∈ [0,10], not exceeding the room boundaries); trajectory integration (calculating the distance between the trajectory and the furniture boundary, ensuring a minimum distance > 1 meter to avoid obstruction, for example, if the trajectory passes through [4,4], the furniture position is offset to maintain clearance). A custom layout simulation algorithm (an iterative algorithm based on particle swarm optimization (PSO)) simulates the interaction between movement trajectories and furniture boundaries, and iteratively optimizes the furniture's position coordinates, adjusting the position parameters to match periodic activity patterns, i.e., embedding periodic activity patterns (weekday weight 0.7, weekend weight 0.3, for example, optimization function = 0.7 × weekday matching + 0.3 × weekend matching); minimizing mismatches (if overlap > 10%, iterate the offset coordinates until < 5%). Output the adjusted furniture layout parameters, including furniture rotation angle and spacing values.

[0064] The step of filtering candidate furniture from the furniture database based on the recommendation priority score includes: Based on the median of the score distribution, a threshold filter is applied to the recommendation priority score; The system retrieves matching items from the furniture database by querying the index and filters out furniture with mismatched functional features, using them as candidate furniture. Output the filtered candidate furniture and compile it into a list.

[0065] By considering dynamic behaviors and periodic patterns (such as differences between weekdays and weekends), recommended furniture better matches the actual needs of families (e.g., multifunctional furniture for gatherings), avoiding conflicts arising from static recommendations and improving user satisfaction. The output of layout adjustment parameters further simplifies the implementation process, allowing users to directly apply the recommendations.

[0066] Suitable for multi-member families, it embeds member role weights and interaction frequencies to identify and resolve behavioral conflicts (such as children's safety needs versus the comfort needs of the elderly), making recommendations more personalized.

[0067] Time-series analysis and dynamic interaction models capture behavioral fluctuations and generate time-sensitive recommendations (such as multifunctional furniture suitable for weekends), enhancing the system's flexibility and adapting to changes in family life at different stages.

[0068] This invention fills the gap in dynamic behavior analysis in the field of furniture recommendation and can be applied to smart home systems, furniture e-commerce platforms and other fields to improve practical application effects and commercial value.

[0069] In summary, this invention achieves precise adaptation to dynamic home environments through a behavior-driven intelligent recommendation method, overcoming the limitations of existing technologies and demonstrating significant technical effectiveness and competitive advantages.

[0070] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0071] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0072] Although embodiments of the invention have been shown and described, those skilled in the art will understand 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 claims and their equivalents.

Claims

1. A furniture intelligent adaptation recommendation method based on family behavior data, characterized in that, The method comprises the following steps: Step S1: collecting behavior sensor data, member interaction data and furniture usage data in a home environment, wherein the behavior sensor data comprises movement trajectory and activity frequency, the member interaction data comprises voice and location interaction, and the furniture usage data comprises occupancy duration and interaction frequency; Step S2: constructing a family periodic activity pattern model according to the behavior sensor data and the member interaction data, and calculating a behavior conflict index; Step S3: constructing a dynamic interaction model of behavior and furniture based on the family periodic activity pattern model and the furniture usage data, and generating an adaptive adjustment parameter, wherein the dynamic interaction model integrates the periodic activity pattern and the furniture characteristics; Step S4: generating a furniture intelligent recommendation strategy according to the behavior conflict index and the adaptive adjustment parameter, wherein the furniture intelligent recommendation strategy comprises priority sorting and layout adjustment; Step S5: outputting an adaptive furniture list based on the furniture intelligent recommendation strategy, and adjusting furniture layout parameters, wherein the adjustment considers space coordinates and behavior matching. 2.The furniture smart adaptation recommendation method based on home behavior data according to claim 1, characterized in that, The collection of behavior sensor data, member interaction data and furniture usage data in a home environment comprises: Collecting movement trajectory and activity frequency data through sensors installed in the home space as behavior sensor data; Recording voice and location interaction between family members as member interaction data, wherein the voice is captured by a microphone and the location interaction is obtained by a positioning module; Monitoring occupancy duration and interaction frequency data of existing furniture as furniture usage data, wherein the monitoring is realized through embedded sensors. 3.The furniture smart adaptation recommendation method based on home behavior data according to claim 2, characterized in that, The construction of a family periodic activity pattern model according to the behavior sensor data and the member interaction data, and the calculation of a behavior conflict index, comprises: Time series segmentation of the behavior sensor data to extract periodic activity patterns, wherein the time series segmentation refers to division into weekday data and weekend data based on a preset time window; Fusion of the member interaction data to construct a family periodic activity pattern model, wherein the family periodic activity pattern model takes activity patterns as nodes, interaction frequency as edge weights, and embeds member role classification; Calculation of space occupancy conflict value and time overlap value according to the family periodic activity pattern model, wherein the space occupancy conflict value is obtained by vector intersection, and the time overlap value is obtained by matrix operation; Weighted summation of the space occupancy conflict value and the time overlap value to generate a behavior conflict index, wherein the weighted summation is calculated by a self-defined conflict aggregation function, which embeds family member role weight and combines activity frequency to adjust the parameters of the function. 4.The furniture smart adaptation recommendation method based on home behavior data according to claim 3, characterized in that, The construction of a dynamic interaction model of behavior and furniture based on the family periodic activity pattern model and the furniture usage data, and the generation of an adaptive adjustment parameter, comprises: Extracting furniture attributes from furniture usage data to obtain furniture size and functional characteristics, and then representing size, material and multi-functional properties by feature vectors; constructing a dynamic interaction model based on the household periodic activity pattern model and the functional characteristics of the furniture, wherein the dynamic interaction model takes the periodic activity pattern as input, the feature vector as a regulating variable, and defines a coupling matrix; iteratively optimizing the dynamic interaction model to calculate an adaptation score, wherein the iterative optimization includes updating parameters of the dynamic interaction model to minimize behavior-furniture mismatch; generating an adaptation adjustment parameter according to the adaptation score. 5.The furniture smart adaptation recommendation method based on home behavior data according to claim 4, characterized in that, The generation of the furniture intelligent recommendation strategy according to the behavior conflict index and the adaptation adjustment parameter includes: fusing the behavior conflict index and the adaptation adjustment parameter to calculate a recommendation priority score, wherein the fusion is realized through weighted average operation and considers the nonlinear influence of the conflict index; based on the recommendation priority score, filtering candidate furniture from a furniture database, wherein the filtering applies threshold filtering and sorting mechanism; generating a recommendation strategy containing layout adjustment, wherein the layout adjustment maps the conflict index to spatial coordinates through a self-defined transformation function, which embeds the timing constraints of the periodic activity pattern and defines a coordinate transformation matrix; In the generation process, integrate the geometric data of the family space. 6.The furniture smart adaptation recommendation method based on home behavior data according to claim 5, characterized in that, The output of the adaptation furniture list and the adjustment of the furniture layout parameter based on the furniture intelligent recommendation strategy includes: sorting the candidate furniture according to the furniture intelligent recommendation strategy to output the adaptation furniture list, wherein the sorting is based on the descending order of the priority score; calculating the position parameter of the furniture in the family space, wherein the calculation considers the room boundary and the moving trajectory; simulate the interaction between the moving trajectory and the furniture boundary through a self-defined layout simulation algorithm, and iteratively optimize the position coordinates of the furniture to adjust the position parameter to match the periodic activity pattern; output the adjusted furniture layout parameter, including the furniture rotation angle and the spacing value. 7.The furniture smart adaptation recommendation method based on home behavior data according to claim 6, characterized in that, The behavior sensor data, member interaction data, and furniture usage data are timestamped and synchronized based on a global clock; After alignment, data normalization is performed to form a unified behavior dataset. 8.The furniture smart adaptation recommendation method based on home behavior data according to claim 7, characterized in that, The calculation of the space occupation conflict value and the time overlap value according to the household periodic activity pattern model includes: extracting a space occupation vector from the household periodic activity pattern model through node projection; calculate the intersection value of the space occupation vector as the space occupation conflict value through set operation; extract the time overlap matrix from the household periodic activity pattern model and calculate the matrix trace as the time overlap value, wherein the matrix trace calculation is performed through a self-defined time convolution function, which embeds the periodic pattern of member interaction and convolves the time window sequence. 9.The furniture smart adaptation recommendation method based on home behavior data according to claim 8, characterized in that, The iterative optimization of the dynamic interaction model to calculate the adaptation score includes: initializing the parameters of the dynamic interaction model, wherein the initialization uses a random distribution to set the initial weight; calculating the loss increment of each iteration based on the feedback mechanism of the periodic activity pattern through a self-defined gradient calculation to update the calculation of the adaptation score; In the iteration, a learning rate decay strategy is applied to adjust the update step; terminate the iteration when the behavior-furniture mismatch converges to a preset threshold, and output the final adaptation score. 10.The furniture smart adaptation recommendation method based on home behavior data according to claim 9, characterized in that, The candidate furniture is filtered from the furniture database based on the recommendation priority score, comprising: Threshold filtering of the recommendation priority score based on the median of the score distribution; Extracting matching items from the furniture database by querying the index, and filtering out furniture that does not match the functional features as candidate furniture; Output the filtered candidate furniture and form a list.

Images