Individualized dietary recommendation system for children obesity based on knowledge graph

Abstract

The application relates to the technical field of intelligent health management, and discloses a children obesity individualized meal recommendation system fused with a knowledge graph, which comprises the following modules: a graph construction module used for establishing food nodes with initial half-lives and volumes and directed edges with time sequence shielding coefficients; a constraint solving module used for extracting a candidate subgraph meeting a heat standard and arranging to generate multiple eating topological directed paths; a state machine deduction module used for stepping along the paths and calculating actual half-lives based on the time sequence shielding coefficients; a time sequence integral module used for obtaining satiety decay integrals by discretely accumulating residual volumes based on the actual half-lives; and a recommendation output module used for converting the path with the largest integral into a food material list and outputting a recommended eating instruction. The application quantifies the intervention of an eating order on gastric emptying, outputs a scheme with the longest satiety through optimization, and improves the scientificity and compliance of the children obesity individualized meal recommendation fused with the knowledge graph.

Description

Technical Field

[0001] This invention relates to the field of intelligent health management technology, specifically to a personalized dietary recommendation system for childhood obesity that integrates knowledge graphs. Background Technology

[0002] Childhood obesity is mainly characterized by excessive fat accumulation due to energy imbalance in the body. To control weight, individualized dietary recommendations are widely used. This technology aims to develop personalized energy intake plans based on the physiological indicators of the target population. Meanwhile, knowledge graphs, as a structured semantic network, are beginning to be introduced into dietary management to store and link complex food composition and nutritional data.

[0003] In existing dietary intervention technologies, the conventional approach involves setting specific calorie restrictions for each meal based on a human basal metabolic model. These technologies typically use a pre-defined calorie threshold to select food combinations from a food database that do not exceed the total calorie limit, and then provide the user with a weight-loss diet plan. This traditional approach achieves basic energy balance management and can effectively limit a child's overall calorie intake in a single meal.

[0004] However, existing technologies have significant shortcomings in practical applications. Traditional recommended plans only focus on achieving the total calorie target through static combinations, completely ignoring the dynamic changes that occur when different foods mix and empty from the digestive tract. In reality, there are complex physicochemical interactions between different foods, and a specific eating order can alter the physical matrix of the chyme in the stomach, thus significantly interfering with the digestion rate of subsequent foods. Current conventional plans lack quantitative assessment methods for the intervention effect of eating order, making it difficult to analyze the impact of different eating orders on overall emptying time. This results in the provided guidelines failing to effectively prolong postprandial satiety, making children prone to eating extra meals prematurely due to hunger, ultimately reducing adherence to the overall weight control process.

[0005] Therefore, this invention proposes a personalized dietary recommendation system for childhood obesity that integrates knowledge graphs to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a personalized dietary recommendation system for childhood obesity that integrates knowledge graphs. This system solves the problem that existing technologies neglect the dynamic intervention of the order of eating on digestion and emptying, making it difficult for dietary plans to effectively prolong children's post-meal satiety.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a personalized dietary recommendation system for childhood obesity integrating knowledge graphs, comprising: The graph construction module is used to establish food nodes and directed edges in the graph database, assign initial half-life and initial volume to the food nodes, and assign time-series masking coefficients to the directed edges. The constraint solving module is used to extract candidate subgraphs where the total calories are less than the single meal calorie threshold, and to arrange the food nodes in the candidate subgraphs to generate multiple eating topological directed paths. The state machine deduction module is used to perform state machine steps along the directed path of the feeding topology and calculate the actual half-life of the food node based on the temporal masking coefficient of the directed edge. The time-series integration module is used to calculate the residual volume within a preset time window based on the actual half-life and to perform discrete accumulation to obtain the satiety decay integral; The recommended output module is used to convert the directed path of the eating topology with the largest satiety decay integral into a food list and suggested eating instructions and output them to the user terminal.

[0008] Preferably, the map construction module receives structured data from a standardized food composition database to instantiate the food node, reads a preset food physiological constant database to extract the isolated gastric emptying time constant and physical space occupancy, defines the isolated gastric emptying time constant as the initial half-life, and defines the physical space occupancy as the initial volume.

[0009] Preferably, the graph construction module constructs bidirectional asymmetric directed edges for two food nodes that have the possibility of being mixed in the same meal, extracts a preset food physicochemical interaction parameter matrix, extracts the digestive retardation rate from the food physicochemical interaction parameter matrix according to the food category to which the food node belongs, converts the digestive retardation rate into a dimensionless scalar and defines it as the temporal masking coefficient.

[0010] Preferably, the constraint solving module obtains age parameters, height parameters, weight parameters, gender parameters, and daily exercise frequency parameters, calculates the basal metabolic rate based on the age parameters, height parameters, weight parameters, and gender parameters, and multiplies the basal metabolic rate by the physical activity coefficient mapped from the daily exercise frequency parameters and the meal distribution ratio to calculate the single meal calorie threshold.

[0011] Preferably, the constraint solving module uses a breadth-first search algorithm to perform a topological traversal of the graph database with the food nodes of the staple food category as the search starting point, extracts and accumulates the calorie attribute values ​​recorded by the food nodes, and rejects merging the next adjacent food node when the sum of the accumulated calorie attribute values ​​is greater than or equal to the single-meal calorie threshold. Connected food nodes whose sum of accumulated calorie attribute values ​​is strictly less than the single-meal calorie threshold are extracted as an independent data set as the candidate subgraph.

[0012] Preferably, the constraint solving module extracts all the food nodes contained in the candidate subgraph, performs a mathematical permutation operation without replacement on the food nodes in the data processing space to generate a linear sequence, and sequentially connects the adjacent directed edges according to the order of the food nodes in the linear sequence to generate the feeding topology directed path.

[0013] Preferably, the state machine deduction module instantiates a finite state machine cursor in the data processing space, moves the finite state machine cursor along the directed path of the feeding topology according to the unidirectional topological pointing of the directed edge for each food node, traces back to extract the temporal masking coefficient recorded in the directed edge pointing from the previous food node to the current target food node, and performs algebraic superposition operation in combination with the initial half-life of the current target food node to obtain the actual half-life of the current target food node.

[0014] Preferably, the time-series integration module sets the preset time window and divides it into multiple time slices according to the discrete time step. Based on the first-order dynamic exponential decay model, it uses the initial volume of the food node and the actual half-life to estimate the volume of the chyme in the residence state to obtain the residual volume.

[0015] Preferably, the time-series integration module synchronizes the emptying time and discrete slice time of all the food nodes, calculates the scalar sum of the residual volume of all the food nodes in each time slice within the preset time window, and discretely accumulates the scalar sum along the time axis to obtain the satiety decay integral.

[0016] Preferably, the recommendation output module uses a numerical sorting algorithm to sort the satiety decay integrals corresponding to multiple eating topology directed paths in descending order, locks the eating topology directed path with the largest satiety decay integral, extracts the pre-loaded food name text inside the eating topology directed path and encapsulates it with the benchmark intake quality value to generate the food list, determines the sequential position of the food nodes according to the unidirectional topology pointing, and concatenates the food name text into strings to generate the suggested eating instruction and sends it to the user terminal.

[0017] This invention provides a personalized dietary recommendation system for childhood obesity that integrates knowledge graphs. It has the following beneficial effects: 1. This invention establishes food nodes and directed edges in a graph database and assigns temporal shielding coefficients representing physical obstruction, transforming digestive interventions when different foods are consumed together into computable network parameters. This data architecture design based on an underlying knowledge graph overcomes the limitations of traditional methods that only count total calories, providing a dietary recommendation basis that includes the order of food intake for improving childhood obesity, and enhancing the objectivity of guidance plans from the perspective of digestive physiological mechanisms.

[0018] 2. This invention achieves the dynamic simulation of gastric emptying by executing state machine steps along the directed path of the eating topology and calculating the actual half-life by combining the masking weights of preceding food nodes. This algorithm effectively integrates the original digestive characteristics of a single food item with the dynamic intervention effects of combined eating, enabling the generation of individualized eating instructions with clear time-series constraints for users at risk of childhood obesity, thereby scientifically guiding their daily meal order.

[0019] 3. This invention quantifies the overall hunger resistance efficacy under different eating sequences by calculating the residual volume within a time window based on the actual half-life and obtaining the satiety decay integral through discrete accumulation. This evaluation mechanism can identify the food arrangement that maintains satiety for the longest time from a massive knowledge graph node, providing an individualized management strategy for intervening in childhood obesity without excessively restricting food intake, and effectively reducing the dropout rate caused by frequent hunger. Attached Figure Description

[0020] Figure 1 This is an architecture diagram of the child obesity individualized dietary recommendation system that integrates knowledge graphs according to the present invention; Figure 2 This is a flowchart of the individualized dietary recommendation method for childhood obesity that integrates knowledge graphs, as described in this invention. Figure 3 This is a schematic diagram illustrating the ingredient list and suggested eating instructions output according to the present invention; Figure 4 This is a flowchart of the feeding topology directed path of the present invention; Figure 5 This is a line graph showing the integral of satiety decay under different eating sequences according to the present invention.

[0021] Among them, 10 is the graph construction module; 20 is the constraint solving module; 30 is the state machine deduction module; 40 is the time series integration module; and 50 is the recommendation output module. Detailed Implementation

[0022] The technical solutions in 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.

[0023] Please see the appendix Figure 1 This invention provides a personalized dietary recommendation system for childhood obesity that integrates knowledge graphs, including: Graph construction module 10 is used to build food nodes and directed edges in the graph database, assign initial half-life and initial volume to food nodes, and assign temporal masking coefficients to directed edges. The constraint solving module 20 is used to extract candidate subgraphs whose total calories are less than the single meal calorie threshold, and to arrange the food nodes in the candidate subgraphs to generate multiple eating topological directed paths. The state machine deduction module 30 is used to perform state machine stepping along the directed path of the feeding topology and calculate the actual half-life of the food node based on the temporal masking coefficient of the directed edge. The time-series integration module 40 is used to calculate the residual volume within a preset time window based on the actual half-life and to perform discrete accumulation to obtain the satiety decay integral. The recommended output module 50 is used to convert the directed path of the eating topology with the largest satiety decay integral into a food list and suggested eating instructions and output them to the user terminal.

[0024] Please see the appendix Figure 2 This invention provides a method for individualized dietary recommendations for childhood obesity that integrates knowledge graphs, including the following steps: Step S1: Establish food nodes and directed edges in the graph database, assign initial half-life and initial volume to food nodes, and assign time-series masking coefficients to directed edges. Step S2: Extract candidate subgraphs whose total calories are less than the single meal calorie threshold, and arrange the food nodes in the candidate subgraphs to generate multiple eating topology directed paths; Step S3: Proceed through the state machine along the directed path of the feeding topology and calculate the actual half-life of the food node based on the temporal masking coefficient of the directed edge. Step S4: Calculate the residual volume within the preset time window based on the actual half-life and perform discrete accumulation to obtain the satiety decay integral; Step S5: The directed path of the eating topology with the largest satiety decay integral is converted into a food list and suggested eating instructions and output to the user terminal.

[0025] The method for individualized dietary recommendations for childhood obesity based on knowledge graphs and the system for individualized dietary recommendations for childhood obesity based on knowledge graphs of the present invention belong to the same inventive concept. Each logical module in the recommendation system is configured to execute the corresponding steps in the method for individualized dietary recommendations for childhood obesity based on knowledge graphs. Specifically, the graph construction module 10 in the recommendation system is used to execute step S1; the constraint solving module 20 is used to execute step S2; the state machine deduction module 30 is used to execute step S3; the temporal integration module 40 is used to execute step S4; and the recommendation output module 50 is used to execute step S5. The system for individualized dietary recommendations for childhood obesity based on knowledge graphs 100 relies on the sequential interaction of each module to achieve a computational closed loop from the construction of underlying graph data and the unfolding of the directed path of eating topology to the optimization of the temporal integral of satiety.

[0026] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0027] See attached document Figure 1 In this embodiment, the map construction module 10 is executed through the following sub-steps: In step S101, the graph construction module 10 establishes the underlying entities of food nodes and directed edges in the graph database, and assigns initial half-life and initial volume to the food nodes. As a preferred approach, the graph construction module 10 receives structured data from a standardized food composition database. The standardized food composition database refers to a reference data source recording objective nutritional values ​​such as macronutrients, micronutrients, and calories of various foods, while the graph database is a specialized storage medium for storing topological network data composed of nodes and directed connections; the two are independent data carriers. The graph construction module 10 sends node creation instructions using the application programming interface provided by the graph database. This instruction instantiates several discrete entity objects within the graph database, and the system defines these entity objects as food nodes, representing independent raw ingredients or processed foods in daily diets.

[0028] After food nodes are created, to achieve static data association of physical and biochemical attributes, the atlas construction module 10 reads a pre-set food physiological constants database. This database is a structured data table pre-stored in a storage medium, which compiles standard digestibility indicators (such as gastric emptying half-life and physical residence volume at standard intake) of various foods measured and published in the medical field. The atlas construction module 10 extracts the isolated gastric emptying time constant and the physical space occupancy at standard intake mass for each food from this data through a table lookup operation. The system defines the isolated gastric emptying time constant as the initial half-life. The unit is set to minutes. This parameter, from a physiological perspective, characterizes the baseline time required for the residual amount of food to decay to half of its initial amount when it exists alone in the stomach environment, due to digestive enzymes and physical peristalsis. The system defines the physical space occupied as the initial volume. Typically measured in milliliters. The atlas construction module 10, through attribute appending and updating operations, extracts food name text, category, calorie value, and baseline intake mass value from the standardized food composition database and the food physiological constant database, along with the aforementioned initial half-life. With initial volume The data is persistently written to the corresponding food node in key-value pair format. For the instantiation mechanism of the basic nodes within the graph database, those skilled in the art can use a general graph query language to perform conventional write operations. The basic database operation logic is well-known in the field and will not be elaborated upon here.

[0029] To further define the dynamic impact of the eating order, in step S102, the graph construction module 10 traverses the set of food nodes in the graph database. For any two food nodes that have the possibility of being eaten together in the same meal, a data connection representing the topological sequence of eating order is constructed. This data connection has absolute topological directionality, and the system defines it as a directed edge. Specifically, assuming there are food nodes... With food nodes The graph construction module 10 is used in the food node. With food nodes Establish a line between them point to directed edges This represents the node where food is first ingested during a single meal. The corresponding food intake node This corresponds to a specific order in which foods are eaten. Simultaneously, the map construction module 10 establishes a framework from... point to directed edges This forms a bidirectional asymmetric topological framework.

[0030] After constructing the graph topology framework, the system needs to quantify the physiological blocking effect caused by this sequence. In step S103, the graph construction module 10 assigns temporal shielding coefficients to the directed edges. Because different foods form different physical matrix states in gastric juice, the matrix of the food ingested first will encapsulate or adhere to the digestive enzymes of the food ingested later, hindering their contact area. Based on general physical principles, for example, the viscous network formed by high-fiber foods absorbing water and swelling in the gastric acid environment will significantly delay the emptying process of subsequently ingested carbohydrates. The graph construction module 10 extracts a pre-set food physicochemical interaction parameter matrix. This parameter matrix is ​​a two-dimensional data lookup table pre-loaded in memory, with its row and column dimensions corresponding to different basic food categories. The matrix elements in this parameter matrix record the digestive blocking rate between different food categories based on viscosity and cellulose content. The graph construction module 10 then uses directed edges... Corresponding food nodes With food nodes The corresponding food category is determined by performing row and column matching queries within the food physicochemical interaction parameter matrix to extract the corresponding digestive retardation rate.

[0031] The graph construction module 10 converts the obtained digestive retardation rate into a specific dimensionless scalar, which is defined as the time-series masking coefficient. In this embodiment, the timing masking coefficient... The value range is defined as a real number greater than or equal to zero, and its magnitude is positively correlated with the intensity of digestive repression. (Time-series shielding coefficient) Representing nodes that have previously ingested food Under the corresponding food conditions, the physicochemical properties of the preceding food matrix affect the subsequent food nodes. The asymmetric intervention weights are generated corresponding to the rate of food digestion and decomposition. The map construction module 10 extracts the temporal masking coefficients. Write the corresponding directed edge The edge attribute field is used for this purpose. Based on the above steps, the graph construction module 10 completes the construction of directed graph data that integrates temporal intervention factors, which provides underlying data structure support for the system to subsequently identify differences in satiety caused by different eating orders.

[0032] See attached document Figure 1 In this embodiment, the constraint solving module 20 is executed through the following sub-steps: In step S201, the constraint solving module 20 receives terminal input data from the target child to establish physiological calorie constraint boundaries. The constraint solving module 20 acquires the target child's age, height, weight, gender, and daily exercise frequency parameters. As a preferred method, the constraint solving module 20 calculates the basal metabolic rate (BMR) based on these physiological parameters using linear weighting, and multiplies the BMR by the physical activity coefficient mapped from the daily exercise frequency parameters and the meal distribution ratio to calculate the single-meal calorie threshold. To finely control energy intake at each meal, the constraint solving module 20 adjusts the daily total energy consumption limit according to a fixed meal distribution ratio. The specific calculation logic is expressed by the following formula: ; in, This represents a weight parameter, and its unit of measurement is kilograms. This represents height, and its unit of measurement is centimeters. This represents an age parameter, and its unit of measurement is years. , , These are the scaling constants for the corresponding physiological parameters; This represents the sex compensation constant. To ensure the completeness and feasibility of the formula, parameters are set based on the conventional basal metabolic rate assessment principles in the medical field. The value is 10. The value is 6.25. The value is 5. For male children, the gender compensation constant is 5. The value is 5; for female children, the gender compensation constant is... The value is -161. The physical activity coefficient ranges from 1.2 to 1.9 and is determined by the target child's daily activity frequency. This represents the proportion of meals allocated to different meals. As a preferred method, based on normal dietary habits, this parameter ranges from 0.2 to 0.4. The values ​​obtained through the above calculations... Defined as the single-meal calorie threshold. The single-meal calorie threshold represents the maximum total calorie scalar that a target child is allowed to consume in a single target meal, and it constitutes a hard cutoff condition for subsequent map retrieval.

[0033] In step S202, after establishing the energy constraint boundary, the constraint solving module 20 performs a retrieval and extraction operation in the graph database. Specifically, the constraint solving module 20 is used to extract candidate subgraphs where the total calories are less than the single-meal calorie threshold. The constraint solving module 20 employs a breadth-first search algorithm, performing a topological traversal of the graph database starting from food nodes of any staple food category. While traversing connected food nodes along directed edges, the constraint solving module 20 simultaneously extracts and accumulates the calorie attribute values ​​recorded within the traversed food nodes. To avoid algorithmic dead zones and ensure the rigor of the truncation condition, attempting to merge the next adjacent food node on the current traversal branch will result in the accumulated total calories being greater than or equal to the single-meal calorie threshold. When this happens, the constraint solving module 20 refuses to merge the next adjacent food node, thus terminating the continued downward traversal of this branch, and assigns the currently traversed nodes with a total calorie count strictly less than the single-meal calorie threshold. This batch of connected food nodes and their inherent directed edges are extracted as a separate data set.

[0034] The extracted data set is defined as a candidate subgraph. Essentially, a candidate subgraph represents a compliant food combination scheme that does not exceed an individual's caloric load; however, within this combination, the order in which each food node is consumed is not uniquely determined. For the node traversal and hierarchical backtracking execution logic of the breadth-first search algorithm at the underlying graph database, those skilled in the art can use existing graph computing frameworks to call relevant graph traversal function libraries to implement it. The specific pathfinding and optimization process is well-known in the field and will not be elaborated upon here.

[0035] Step S203: After obtaining the food combination scheme that meets the calorie limit, it is necessary to further analyze the temporal intervention possibilities brought about by different eating orders. The constraint solving module 20 arranges the food nodes in the candidate subgraph to generate multiple eating topological directed paths. For a single extracted candidate subgraph, the constraint solving module 20 extracts all the independent food nodes it contains and performs a mathematical permutation operation on these food nodes in the data processing space allocated by the system. Through the permutation operation without replacement, the node set that originally presented as a network and lacked temporal constraints is linearly expanded and transformed into multiple linear sequences with a clear order of arrangement.

[0036] Based on this, the constraint solving module 20 sequentially connects the directed edges between adjacent nodes according to the order of nodes in each linear sequence. The unidirectional graph theory path formed by connecting specific food combinations in a definite and unambiguous order of intake is defined as a topological directed path for eating. The topological directed path for eating represents a precise sequence of eating operations strictly followed by the target child during actual meals. This processing step exhaustively expands the static food combinations into a dynamic temporal eating sequence, thus providing a complete topological data space to be verified for subsequent modules to evaluate the substantial impact of different eating orders on the digestive emptying rate.

[0037] See attached document Figure 1 In this embodiment, the state machine deduction module 30 is executed through the following sub-steps: In step S301, the state machine deduction module 30 performs state machine stepping along the directed path of the eating topology. As a preferred method, the state machine deduction module 30 instantiates a finite state machine cursor with memory function in the data processing space. State machine stepping refers to the operation process where the finite state machine cursor moves along a specific directed path of the eating topology, starting from the initial food node of the path, and moves node by node according to the unidirectional topological pointing of the directed edges, while synchronously updating the current computing environment context. Each cursor movement between nodes represents the completion of a food intake sequence during the simulated meal. Through this directional transfer operation along the topological connection from front to back, the state machine deduction module 30 can accurately simulate and reproduce the discrete event sequence of the target child eating in a specific order in memory. For the underlying execution logic of the finite state machine based on pointer movement and state transition in the data structure, those skilled in the art can implement it using existing program state control statements; the software control method is well-known in the field and will not be elaborated here.

[0038] In step S302, when the state machine cursor reaches a target food node in the directed path of the feeding topology, the state machine deduction module 30 performs backtracking extraction of the preceding masking parameters. Since the directed path of the feeding topology has a strict sequential characteristic, nodes preceding the current state machine cursor position are collectively referred to as preceding food nodes. The state machine deduction module 30 identifies the sequence position index of the current target food node in the directed path of the feeding topology and sequentially traces back all preceding food nodes existing before that sequence position index. Based on the directed edge attributes pre-loaded in the graph database, the state machine deduction module 30 extracts the temporal masking coefficients recorded in the directed edges pointing from each preceding food node to the current target food node. Through this backtracking operation, the system summarizes the physicochemical intervention weights of all previously ingested food matrices on the current target food.

[0039] Step S303: After completing the parameter summary, the state machine deduction module 30 calculates the actual half-life of the food node based on the temporal shielding coefficients of the directed edges. The actual half-life refers to the dynamic time constant that, under the intervention of a specific eating sequence, the residual volume of the current target food is reduced to half its initial volume after its original emptying characteristics are prolonged due to the encapsulation and adhesion of preceding food. From the perspective of digestive physiology, gastric emptying is a complex process influenced by the mixing of various chyme matrices. High-viscosity or high-fat foods ingested earlier form a physical barrier on the gastric mucosa, thereby reducing the contact area between subsequent food and digestive enzymes in the gastric juice to a certain extent, thus prolonging the emptying half-life of subsequent food compared to its independent existence. The state machine deduction module 30 calls the pre-stored initial half-life within the current target food node and, combined with all temporal shielding coefficients extracted from backtracking, obtains the actual half-life through algebraic superposition. The specific mathematical calculation model is shown in the following formula: ; in, This represents the sequence position index of the current target food node in the directed path of the feeding topology. It is a positive integer greater than or equal to 2; The position index of the sequence is The actual half-life of the current target food node is calculated and is expressed in minutes. The initial half-life assigned to the target food node during the graph construction phase; Represents the sequence position index in Previous preceding food node index; Represented by the sequence position index The temporal masking coefficient carried by the directed edge from the preceding food node to the current target food node.

[0040] To avoid logical dead zones in the graph retrieval process and to constrain the completeness of the system computation, when the graph database does not contain a sequence position index... If there is a direct directed edge between the preceding food node and the current target food node, or if the corresponding directed edge does not contain a temporal masking coefficient, the state machine deduction module 30 determines that the food corresponding to the preceding food node does not cause temporal interference to the food corresponding to the current target food node, and automatically sets the corresponding... The value is set to 0 for algebraic summation calculation. Specifically, when the state machine cursor is at the starting node of the directed path in the feeding topology... When the value equals 1, since there is no digestive intervention from preceding food, the actual half-life of this starting node is directly equal to its initial half-life. The above calculation process transforms discrete topological network parameters into continuous dynamic time variables, establishing a crucial dynamic foundation for subsequent accurate assessment of the overall satiety decay process.

[0041] See attached document Figure 1 In this embodiment, the timing integration module 40 is executed through the following sub-steps: In step S401, the time-series integration module 40 establishes the time parameters for discretization calculation. From the perspective of human digestive physiology, the changes in gastric wall tension caused by gastric emptying after a single meal typically last for several hours. To fully assess the evolution of this postprandial physiological state, the system extracts a fixed duration starting from the occurrence of eating and defines it as a preset time window. As a preferred approach, the length of this preset time window is 240 minutes, covering the main interval from a regular meal to the next meal. Simultaneously, to enable numerical evaluation of the continuous digestive process via computer programs, the time-series integration module 40 divides this preset time window into multiple uniform time slices according to a set discrete time step. To ensure a balance between the accuracy of discrete integration calculations and the computer's processing load, the discrete time step is typically set to 1 minute.

[0042] In step S402, after defining the time assessment range, the time-series integration module 40 calculates the residual volume within a preset time window based on the actual half-life. Satiety, at a physical level, primarily stems from the tension stimulation received by gastric wall distension receptors, and this physiological tension directly depends on the total volume of chyme remaining in the stomach. This volume of chyme in a resident state is defined as the residual volume, which characterizes the amount of physical space occupied by food remaining in the stomach before entering the intestines after a period of digestion, enzymatic breakdown, and physical emptying. At any discrete time slice, for any food node in the directed path of the eating topology, the time-series integration module 40, based on a first-order kinetic exponential decay model, uses the initial volume and actual half-life of the food node to estimate its current residual volume. The specific formula for calculating the residual volume is as follows: ; in, The index of the sequence position in the directed path of the feeding topology is Food nodes undergo emptying time The remaining volume after processing, in milliliters; This represents the inherent initial volume of the food node; This represents the actual half-life of the food node; This represents the cumulative digestion time, expressed in minutes, since the food was ingested. Because the timing masking factor was defined as non-negative in the preceding steps, and the initial half-life is strictly greater than zero, therefore... It must be a positive real number greater than zero. This boundary condition, from a fundamental logical perspective, avoids the risk of dead zones caused by division by zero overflow during computer operation.

[0043] In step S403, after obtaining the volume data at discrete time points, the time-series integration module 40 performs discrete accumulation to obtain the satiety decay integral. To ensure a consistent benchmark for the discrete accumulation process, the time-series integration module 40 adopts macroscopic physical approximation logic, ignoring the total time of a single meal compared to a preset time window of several hours; that is, it sets the emptying time for all food nodes. All are synchronized with the global discrete slice time. The physiological intervention effects of different food intake sequences have been fully mapped and solidified in the independent actual half-life of each food node through the pre-processing steps. internal.

[0044] Based on the aforementioned time reference, the time-series integration module 40 calculates the scalar sum of the residual volumes of all food nodes at each discrete time point within the time window, and then discretely accumulates these volume sums along the time axis to obtain the satiety decay integral under a specific eating sequence. The satiety decay integral is a comprehensive quantitative indicator reflecting the overall satiety experience after a meal. Since the anti-hunger efficacy felt by the human body is not only related to the absolute amount of stomach contents at a given moment, but also depends on the duration of this fullness over time, it is defined using the area under the curve integral form. Its discrete accumulation calculation model is expressed as follows: ; in, The value represents the calculated satiety decay integral, with the dimension of milliliters per minute. The larger the value, the stronger the overall effect of maintaining satiety after a meal under a specific eating order. This represents the discrete time step established above; The step sequence index representing the discrete-time slice; This represents the total number of outliers within the preset time window, which is the preset time window length divided by [the number of outliers]. The business; This represents the total number of food nodes contained in the directed path of this feeding topology.

[0045] Through the aforementioned discrete accumulation operation, the temporal integration module 40 aggregates the dynamic decay process of micro-nodes into a single evaluation value characterizing the overall dietary hunger resistance. This computational method, which integrates discrete graph theory parameters with continuous dynamic equations, provides a comparable quantitative benchmark for subsequently selecting the optimal solution that provides the most lasting feeling of satiety from multiple feeding topological directed paths. The basic statements and memory iteration processes for executing multiple loop accumulation summations in computer programs can be implemented using conventional nested loop logic by those skilled in the art; the software control mechanism is well-known in the field and will not be elaborated upon here.

[0046] See attached document Figure 1 and Figure 3 In this embodiment, the recommended output module 50 is executed through the following sub-steps: Step S501: To select the dietary strategy with the best hunger-suppressing efficacy from numerous compliant solutions that meet calorie constraints, the recommendation output module 50 acquires and aggregates multiple sets of evaluation data calculated by the preceding module. This data includes multiple directed paths of the eating topology derived from the candidate subgraph and their corresponding satiety decay integrals. The recommendation output module 50 uses a numerical sorting algorithm to sort all satiety decay integrals in descending order and perform an extreme value search. Its optimization calculation logic can be expressed by the following formula: ; in, This represents the topological directed path of eating with the largest satiety decay integral selected after comparison. It represents the set of all directed paths of feeding topology derived from the full permutation of the candidate subgraph; Representative set The first in A specific feeding topological directed path; This represents the specific path. The satiety decay integral. To avoid algorithmic selection dead zones during the extreme value search process, when the set When multiple feeding topological directed paths have the same and largest satiety decay integral, as a preferred approach, it is recommended that output module 50 select the feeding topological directed path with the smallest sequence index in the traversed set as the optimal path. Through the above comparison and retrieval operations, the recommended output module 50 accurately identifies the graph theory connected path to which the peak value of the integral belongs.

[0047] In step S502, after determining the directed path of the eating topology with the largest satiety decay integral, the recommendation output module 50 deconstructs and extracts the static node attributes contained within it, thereby transforming the directed path of the eating topology with the largest satiety decay integral into a food list. The food list refers to structured data details recording the names, categories, and standard intake weights of various foods required for a specific meal. In the specific transformation process, the recommendation output module 50 traverses all independent food nodes contained within the directed path of the eating topology. For each food node, it reads the node labels and associated fields pre-persistently loaded in the underlying graph database, extracting the corresponding food name text and the baseline intake quality value matching the physiological caloric constraint. The recommendation output module 50 extracts and concatenates these multi-dimensional attribute data, encapsulating them into a standard structured data package in JSON or XML format.

[0048] In step S503, in addition to providing static food composition guidance, to accurately implement the physical temporal intervention effect contained in the underlying map into actual eating behavior, the recommendation output module 50 also transforms the directed eating topology path with the largest satiety decay integral into suggested eating instructions. Suggested eating instructions are prompts generated based on the principle of food physicochemical interaction inhibition, used to guide target children to strictly follow a specific food intake order during meals. Based on the locked directed eating topology path object, the recommendation output module 50 establishes the sequential position of each food node according to the unidirectional topological pointing of the directed edges in the path. According to this sequence position, the recommendation output module 50 concatenates the extracted food name texts in absolute order from first to last, and automatically inserts assigned numerical sequence fields or guiding characters representing the sequence between adjacent food name texts for string concatenation, thereby generating text prompt information with clear temporal order characteristics, thus constituting suggested eating instructions.

[0049] In step S504, after completing data encapsulation and guidance text generation, the recommendation output module 50 interacts and distributes the converted food list and suggested eating instructions to the user terminal through an external interface. By integrating the attribute retrieval of the underlying medical knowledge graph with the aforementioned mathematical deduction of dynamic digestion and emptying, a personalized dietary recommendation for childhood obesity based on the integrated knowledge graph is completed. The user terminal refers to an electronic communication device with information interaction and network access functions that is commonly used by the target child or their guardian, such as mainstream smartphones, tablets, or smart wearable watches. The recommendation output module 50 calls the underlying network communication components and uses wireless network protocols to send the data stream containing the food list and suggested eating instructions to the specific application receiving end of the target user terminal. After receiving the complete data, the user terminal triggers the rendering update mechanism of the front-end interface, displaying the weight loss diet plan to the user in multimedia form of text, images, or voice. For the data communication parsing and application interface rendering technology on the mobile device, those skilled in the art can use existing network transmission protocols and front-end UI frameworks to implement it. The underlying network communication and interface display methods are well-known technologies in the field and will not be described in detail here.

[0050] To verify the effectiveness and feasibility of the knowledge graph-integrated personalized dietary recommendation system for childhood obesity of the present invention, a complete application example is given below with specific data.

[0051] The target child is a 10-year-old male, 140 cm tall and weighing 50 kg. His daily exercise frequency corresponds to a physical activity coefficient of 1.3, and the set meal distribution ratio is 0.3.

[0052] For the target children, the graph construction module 10 establishes underlying entities of food nodes and directed edges in the graph database. The graph construction module 10 receives structured data and instantiates food nodes, creating three food nodes: "oatmeal," "apple," and "milk." The graph construction module 10 assigns initial half-lives and initial volumes to the food nodes. The initial half-life of "oatmeal" is set to 60 minutes, and its initial volume to 200 ml; the initial half-life of "apple" is set to 40 minutes, and its initial volume to 150 ml; the initial half-life of "milk" is set to 30 minutes, and its initial volume to 250 ml. Subsequently, the graph construction module 10 establishes directed edges between the food nodes and assigns temporal masking coefficients to these edges. Based on the food physicochemical interaction parameter matrix, the temporal masking coefficient carried by the directed edge from "oatmeal" to "milk" is set to 0.5, the temporal masking coefficient from "oatmeal" to "apple" is set to 0.2, and the temporal masking coefficient from "apple" to "milk" is set to 0.1.

[0053] The constraint solving module 20 receives the terminal input data from the target child and calculates the single-meal calorie threshold. Based on the aforementioned physiological parameters and substituting the male baseline physiological scaling constant, the specific calculation formula is as follows: ; The target child's single-meal calorie threshold was determined to be 518.7 kcal. The constraint solving module 20 extracted candidate subgraphs whose total calories were less than the single-meal calorie threshold. The system searched the graph database and found that the total calories of the combination of "oatmeal," "apple," and "milk" was 420 kcal, strictly less than 518.7 kcal, and this combination was extracted as a candidate subgraph. The constraint solving module 20 arranged the food nodes within the candidate subgraphs to generate multiple directed paths for eating topology. Linear sequences containing a specific order of oatmeal, apple, and milk were generated, specifically path A (oatmeal -> apple -> milk) and path B (milk -> apple -> oatmeal). Please refer to the appendix. Figure 4 ,Should Figure 4 The topological connections and temporal intervention weights among the three food nodes extracted above are shown. Figure 4 Each food node is represented by a light gray circle with a Chinese label, and the solid black lines with arrows between nodes are directed edges. The directed edge from the starting node "Oatmeal" to "Apple" is labeled with the exact value 0.2, the directed edge from "Oatmeal" directly to "Milk" is labeled with the value 0.5, and the directed edge from the intermediate node "Apple" to "Milk" is labeled with the value 0.1. This directed topology network forms the underlying basic structure for the state machine steps of subsequent modules.

[0054] The state machine derivation module 30 moves the state machine along the directed path of the feeding topology. For path A, the state machine cursor starts from "oat" and moves towards "apple," then towards "milk." The state machine derivation module 30 calculates the actual half-life of the food nodes based on the temporal masking coefficients of the directed edges. In path A, "oat" is the starting food node, and its actual half-life is equal to its initial half-life of 60 minutes; "apple" is affected by the temporal masking coefficient of 0.2 of the preceding "oat," which is substituted into the calculation formula: ; The actual half-life is extended to 48 minutes; the "milk" is affected by the combined time-series shielding coefficient of 0.5 from the preceding "oat" and 0.1 from the preceding "apple," which is then substituted into the calculation formula: ; The actual half-life is extended to 48 minutes. Similarly, the state machine deduction module 30 calculates the actual half-life of each food node in path B. Since the starting node of path B is "milk", it lacks the blocking effect of the preceding high-fiber food, and its overall actual half-life value is relatively short.

[0055] The time-series integration module 40 sets a preset time window of 240 minutes and a discrete time step of 1 minute. Based on the actual half-life, the time-series integration module 40 calculates the residual volume within the preset time window and performs discrete accumulation to obtain the satiety decay integral. The time-series integration module 40 calculates the residual volumes of "oatmeal," "apple," and "milk" in path A at each digestion time t, substituting them into the calculation formula respectively: ; ; ; The time-series integration module 40 calculates the scalar sum of the residual volumes of all food nodes at each discrete time point (i.e., + The total volume was then discretely accumulated along the time axis, yielding a satiety decay integral value of 43372 ml / min for path A. Using the same discrete accumulation method, the satiety decay integral value for path B was calculated to be 35833 ml / min. Please refer to the appendix. Figure 5 , Figure 5 The solid black line represents the volume decay process of the feeding topology directed path A, and the dashed black line represents the volume decay process of the feeding topology directed path B. At the starting point of digestion time of 0 minutes, the two curves overlap and indicate an accurate value of 600 ml; by the 120th minute, the solid line of path A shows that the total residual volume remains at 120.7 ml, while the dashed line of path B shows that the total residual volume has decreased to 84.4 ml.

[0056] In a geometric and physical sense, the region enclosed below the broken line and the horizontal axis corresponds to the satiety decay integral calculated by the computer. Figure 5 As can be seen from the chart's visual representation, the graphic space enclosed by the solid line is significantly larger than that enclosed by the dashed line. Combined with the precise values ​​calculated by the time-series integration module 40, the space corresponding to the solid line covers a total integration volume of 43372 ml / min, while the space corresponding to the dashed line only covers a total integration volume of 35833 ml / min. Figure 5 The table clearly demonstrates that Path A has a superior effect in maintaining satiety.

[0057] The recommendation output module 50 sorts the satiety decay integrals corresponding to multiple feeding topology directed paths numerically, and locks the feeding topology directed path with the largest satiety decay integral (i.e., path A). The recommendation output module 50 converts the feeding topology directed path with the largest satiety decay integral into a food list and suggested eating instructions, which are then output to the user terminal. The converted food list includes standard intake mass values ​​for oats, apples, and milk; the converted suggested eating instructions are generated as text according to the sequence of path A, with the content "Step 1: Eat oats; Step 2: Eat apples; Step 3: Drink milk". This data is then packaged and sent to the user terminal interface of the target child or their guardian for display.

[0058] 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 personalized dietary recommendation system for childhood obesity integrating knowledge graphs, characterized in that, include: The graph construction module is used to establish food nodes and directed edges in the graph database, assign initial half-life and initial volume to the food nodes, and assign time-series masking coefficients to the directed edges. The constraint solving module is used to extract candidate subgraphs where the total calories are less than the single meal calorie threshold, and to arrange the food nodes in the candidate subgraphs to generate multiple eating topological directed paths. The state machine deduction module is used to perform state machine steps along the directed path of the feeding topology and calculate the actual half-life of the food node based on the temporal masking coefficient of the directed edge. The time-series integration module is used to calculate the residual volume within a preset time window based on the actual half-life and to perform discrete accumulation to obtain the satiety decay integral; The recommended output module is used to convert the directed path of the eating topology with the largest satiety decay integral into a food list and suggested eating instructions and output them to the user terminal.

2. The personalized dietary recommendation system for childhood obesity based on knowledge graphs according to claim 1, characterized in that, The map construction module receives structured data from a standardized food composition database to instantiate the food nodes, reads a preset food physiological constant database to extract the isolated gastric emptying time constant and physical space occupancy, defines the isolated gastric emptying time constant as the initial half-life, and defines the physical space occupancy as the initial volume.

3. The personalized dietary recommendation system for childhood obesity based on knowledge graphs according to claim 2, characterized in that, The graph construction module constructs bidirectional asymmetric directed edges for two food nodes that have the possibility of being mixed in the same meal, extracts a preset food physicochemical interaction parameter matrix, extracts the digestive retardation rate from the food physicochemical interaction parameter matrix according to the food category to which the food node belongs, converts the digestive retardation rate into a dimensionless scalar and defines it as the temporal masking coefficient.

4. The personalized dietary recommendation system for childhood obesity based on knowledge graphs according to claim 1, characterized in that, The constraint solving module obtains age parameters, height parameters, weight parameters, gender parameters, and daily exercise frequency parameters. Based on the age parameters, height parameters, weight parameters, and gender parameters, it calculates the basal metabolic rate. The basal metabolic rate is then multiplied by the physical activity coefficient mapped from the daily exercise frequency parameters and the meal distribution ratio to calculate the single meal calorie threshold.

5. The personalized dietary recommendation system for childhood obesity based on knowledge graphs according to claim 4, characterized in that, The constraint solving module uses a breadth-first search algorithm to perform a topological traversal of the graph database, starting with the food nodes of the staple food category. It extracts and accumulates the calorie attribute values ​​recorded by the food nodes. When merging into the next adjacent food node would cause the sum of the accumulated calorie attribute values ​​to be greater than or equal to the single-meal calorie threshold, it rejects merging into the next adjacent food node. Connected food nodes whose sum of accumulated calorie attribute values ​​is strictly less than the single-meal calorie threshold are extracted as an independent data set as the candidate subgraph.

6. The personalized dietary recommendation system for childhood obesity based on knowledge graphs according to claim 5, characterized in that, The constraint solving module extracts all the food nodes contained in the candidate subgraph, performs a mathematical permutation operation without replacement on the food nodes in the data processing space to generate a linear sequence, and sequentially connects the adjacent directed edges according to the order of the food nodes in the linear sequence to generate the feeding topology directed path.

7. The personalized dietary recommendation system for childhood obesity based on knowledge graphs according to claim 1, characterized in that, The state machine deduction module instantiates a finite state machine cursor in the data processing space, moves the finite state machine cursor along the directed path of the feeding topology according to the unidirectional topological pointing of the directed edge for each food node, traces back to extract the temporal masking coefficient recorded in the directed edge pointing from the previous food node to the current target food node, and performs algebraic superposition operation in combination with the initial half-life of the current target food node to obtain the actual half-life of the current target food node.

8. The personalized dietary recommendation system for childhood obesity fused with knowledge graphs according to claim 1, characterized in that, The time-series integration module sets the preset time window and divides it into multiple time slices according to the discrete time step. Based on the first-order dynamic exponential decay model, it uses the initial volume of the food node and the actual half-life to estimate the volume of the chyme in the residence state and obtain the residual volume.

9. The personalized dietary recommendation system for childhood obesity fused with knowledge graphs according to claim 8, characterized in that, The time-series integration module synchronizes the emptying time and discrete slice time of all the food nodes, calculates the scalar sum of the residual volume of all the food nodes in each time slice within the preset time window, and discretely accumulates the scalar sum along the time axis to obtain the satiety decay integral.

10. The personalized dietary recommendation system for childhood obesity fused with knowledge graphs according to claim 1, characterized in that, The recommendation output module uses a numerical sorting algorithm to sort the satiety decay integrals corresponding to multiple eating topology directed paths in descending order, locks the eating topology directed path with the largest satiety decay integral, extracts the pre-loaded food name text and the baseline intake quality value inside the eating topology directed path to generate the food list, determines the sequential position of the food nodes according to the unidirectional topology pointing, and concatenates the food name text into strings to generate the suggested eating instruction and sends it to the user terminal.

Images