User care-based home context graph construction method and system

Abstract

The application provides a user care-based home context graph construction method and system, which comprises the following steps: acquiring environment perception data flow of a target house, wherein the environment perception data flow comprises at least one of a wireless beamforming feedback sequence, a radar occupation sequence, household appliance use information and environment sensor readings; determining the topological relationship between each room in the target house, the room where each device in the target house is located and at least one life rule of a first user in the target house based on the environment perception data flow, wherein the device is a perceiver or a household appliance perceived by the perceiver that provides the environment perception data flow; and constructing a home context graph for caring for the first user based on the topological relationship between each room, the room where each device is located and the at least one life rule of the first user. The application can generate a home context graph that adapts to the natural drift of the family's work and rest rules without manual configuration or predefined rules.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, smart home, user care, and environmental intelligence and privacy computing, and particularly to a method and system for constructing a family context graph based on user care. Background Technology

[0002] With the accelerating aging of the global population, home-based elderly care has become the mainstream choice. How to achieve continuous and reliable behavioral monitoring and care inference using smart home environments without infringing on the privacy of the elderly or increasing the burden of wearing smart devices is a current research hotspot in the field of environmental intelligence. Existing smart home systems typically employ multiple technical approaches to achieve care inference for the elderly, but each has significant shortcomings.

[0003] In terms of context modeling, many commercial smart home systems require users or installers to manually assign room names, define device locations, and set daily routines, or rely on expert-preset rule bases to build a home model. These methods are costly to deploy, difficult to adapt to dynamic changes in family life (such as elderly people's routines shifting with the seasons or health status), and cannot be extended to undefined abnormal patterns. In other words, context modeling methods based on manual configuration or predefined rules are costly to deploy and difficult to adapt to the natural shifts in family routines.

[0004] Therefore, an effective solution is urgently needed to address the above problems. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a method and system for constructing a family context graph based on user care.

[0006] This invention provides a method for constructing a family context graph based on user care, including: Acquire environmental sensing data stream of the target residence, wherein the environmental sensing data stream includes at least one of wireless beamforming feedback sequence, radar occupancy sequence, home appliance usage information and environmental sensor readings; Based on the environmental perception data stream, the topological relationship between rooms in the target residence, the room where each device in the target residence is located, and at least one living pattern of the first user in the target residence are determined. The device is a sensor that provides the environmental perception data stream or a home appliance sensed by the sensor. Based on the topological relationships between the rooms, the rooms where the devices are located, and at least one of the living patterns of the first user, a family context graph for caring for the first user is constructed.

[0007] According to the present invention, a method for constructing a family context graph based on user care, wherein determining the topological relationships between rooms in a target residence, the rooms where each device in the target residence is located, and at least one daily routine of a first user in the target residence based on the environmentally aware data stream, includes: Based on the spatial co-occurrence and temporal correlation of the environmental perception data stream, the topological relationship between the rooms is determined; For each device, the room where the device is located is determined by overlapping analysis of the active time period of the device and the occupancy time period of each room in the environmental perception data stream. At least one repetitive data group is extracted from the environmental perception data stream based on the time clustering algorithm. For each repetitive data group, the corresponding life pattern is identified. The repetitive data group contains multiple data of the same device being active in the same location within the same time period.

[0008] A method for constructing a family context graph based on user care provided by the present invention further includes: Set confidence levels for each node and edge in the family context graph, and / or add source metadata for each node and edge in the family context graph, wherein the source metadata includes at least one of observation source, first observation timestamp, latest observation timestamp, and number of observations.

[0009] According to a method for constructing a family context graph based on user care provided by the present invention, after constructing a family context graph for caring for the first user based on the topological relationships between the rooms, the rooms where the devices are located, and at least one daily routine of the first user, the method further includes: Based on the current environmental perception data stream, determine the current activity data of the first user; Based on the family context graph, the current activity data is detected to obtain detection data; When the detection data indicates a change in the first user's daily routine, or when the detection data indicates a conflict between the current activity data and the edge and / or node corresponding to the current activity data in the family context graph, a behavioral description text corresponding to the first user is generated. The behavioral description text includes at least one of the following: time offset of the daily routine, change in duration, disappearance of the daily routine, decrease in frequency of the daily routine, and newly emerging daily routine. The behavioral description text is fed back to the second user to inform the second user.

[0010] According to a method for constructing a family context graph based on user care provided by the present invention, after feeding back the behavioral description text to the second user to inform the second user, the method further includes: Receive feedback from the second user regarding the behavior description text; If the feedback information confirms the assessment, increase the confidence of the nodes and / or edges corresponding to the behavioral description text in the family context graph. If the feedback information is rejected, reduce the confidence of the nodes and / or edges corresponding to the behavioral description text in the family context graph, or delete the nodes and / or edges corresponding to the behavioral description text in the family context graph.

[0011] According to a method for constructing a family context graph based on user care provided by the present invention, after receiving feedback information from the second user on the behavioral description text, the method further includes: If the feedback information includes the remarks input by the second user, the remarks are appended as additional information to the node and / or edge corresponding to the behavior description text. And / or, update the source metadata of the nodes and / or edges corresponding to the behavior description text based on the feedback information; And / or, record the feedback source and feedback time of the feedback information, as well as the operation type for updating the family context graph.

[0012] According to a method for constructing a family context graph based on user care provided by the present invention, after constructing a family context graph for caring for the first user based on the topological relationships between the rooms, the rooms where the devices are located, and at least one daily routine of the first user, the method further includes: Based on the current environmental awareness data stream, user event data is determined, which includes at least one of the following: event occurrence time, event location, event-associated device type, and event duration. Based on the life patterns corresponding to the user event data in the family context graph and the user event data, determine the degree of deviation corresponding to the user event data; Based on the degree of deviation and the relevant data of the life patterns corresponding to the user event data, an explanation of the care status of the first user is generated. The relevant data includes at least one of confidence level, latest update time, source metadata and additional information. The explanation of the care status includes at least one of deviation description text, deviation reason prompt and suggested operation information. The care status is explained and fed back to the second user to inform the second user.

[0013] A method for constructing a family context graph based on user care provided by the present invention further includes: Generate a bounded inference context package, wherein the bounded inference context package contains a subgraph corresponding to the user event data in the family context graph; The bounded inference context packet is anonymized and then sent to the authorization system.

[0014] A method for constructing a family context graph based on user care provided by the present invention further includes: Received data export instruction; When the data export instruction indicates that specified data should be output, the specified data is output. The specified data is a care status interpretation and / or comprehensive care data, and the comprehensive care data represents the activity of the first user within a set time period. If the data export instruction indicates that non-specified data should be output, the output of the non-specified data shall be rejected. The non-specified data is data other than the specified data.

[0015] A method for constructing a family context graph based on user care provided by the present invention further includes: For a specified element in the family context graph, the confidence of the specified element is reduced according to a preset decay rate. The specified element is a node or edge that has not been confirmed by relevant information within a set time period. The relevant information is newly acquired environmental perception data stream and / or feedback information. If the confidence level of the specified element is lower than the lower confidence limit, the specified element is deleted from the family context graph.

[0016] A method for constructing a family context graph based on user care provided by the present invention further includes: Receive adjustment instructions from the second user; Based on the adjustment instruction, the nodes and / or edges corresponding to the adjustment instruction in the family context graph are adjusted.

[0017] A method for constructing a family context graph based on user care provided by the present invention further includes: Perform the following steps for each graph element in the family context graph: Record the historical versions of the aforementioned atlas elements; If a rollback operation is detected for the graph element, the graph element in the family context graph is rolled back to the historical version specified by the rollback operation.

[0018] The present invention also provides a family context graph construction system based on user care, comprising: An environmental data acquisition module is configured to acquire an environmental sensing data stream of a target residence, the environmental sensing data stream including at least one of wireless beamforming feedback sequence, radar occupancy sequence, home appliance usage information and environmental sensor readings; The graph construction module is configured to determine, based on the environmental perception data stream, the topological relationships between rooms in the target residence, the rooms where each device is located in the target residence, and at least one living pattern of the first user in the target residence, wherein the device is a sensor that provides the environmental perception data stream or a home appliance sensed by the sensor; and to construct a family context graph for caring for the first user based on the topological relationships between rooms, the rooms where each device is located, and at least one living pattern of the first user.

[0019] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the user care-based family context graph construction method as described above.

[0020] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the family context graph construction method based on user care as described above.

[0021] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the family context graph construction method based on user care as described above.

[0022] The present invention provides a method and system for constructing a family context graph based on user care. This method acquires an environmental perception data stream of a target residence, including at least one of wireless beamforming feedback sequences, radar occupancy sequences, appliance usage information, and environmental sensor readings. Based on this data stream, it determines the topological relationships between rooms in the target residence, the rooms where each device is located, and at least one daily routine of a first user. The devices are sensors providing the environmental perception data stream or appliances sensed by the sensors. Based on the topological relationships between rooms, the rooms where each device is located, and at least one daily routine of the first user, a family context graph for caring for the first user is constructed. This invention eliminates the need for manual configuration or predefined rules, enabling the construction of a structured context graph containing room topological relationships and daily routines based on existing passive observation activities. It also adapts to the natural drift of family routines and reduces deployment costs compared to context modeling methods that require manual configuration or predefined rules. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is one of the flowcharts illustrating the family context graph construction method based on user care provided by the present invention.

[0025] Figure 2 This is the second flowchart illustrating the family context graph construction method based on user care provided by the present invention.

[0026] Figure 3 This is a schematic diagram of the process for generating a family context graph using the passive startup engine provided by the present invention.

[0027] Figure 4 This is a schematic diagram of the structure of the family context graph provided by the present invention.

[0028] Figure 5 This is a schematic diagram of the caregiver interaction interface provided by the present invention.

[0029] Figure 6 This is a schematic diagram of the process for updating the family context graph provided by the present invention.

[0030] Figure 7 This is a schematic diagram of the process for generating care status interpretations provided by the present invention.

[0031] Figure 8 This is a schematic diagram of the process for generating bounded inference context packages provided by the present invention.

[0032] Figure 9 This is a schematic diagram of the interactions between the modules in the family context graph construction system based on user care provided by the present invention.

[0033] Figure 10 This is a schematic diagram of the structure of the family context graph construction system based on user care provided by the present invention.

[0034] Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0036] First, a brief description of the relevant content involved in this invention will be given.

[0037] In passive observation activity recognition, existing technologies derive data-action units from raw data by passively observing sensor data streams and aggregate them into activity models through motif learning. The advantage of this approach is that it requires no pre-configuration or labeling, enabling automatic system startup. However, these methods construct activity classification models (e.g., identifying atomic activities like "cooking" and "sleeping"), rather than a structured "family context map" that includes family topology, daily routines, caregiver roles, and behavioral baselines. Furthermore, these methods rely solely on passive observation for a one-time startup, lacking mechanisms for subsequent proactive optimization, correction, and confidence management through caregiver interaction, making it difficult to correlate long-term observed behavioral drift with actual care needs.

[0038] In the application of knowledge graphs to home intelligence, existing technologies propose a semantic framework that utilizes smart home sensor data to detect problems in daily life activities. However, its graph structure and detection rules rely on pre-defined expert knowledge and cannot be passively initiated from environmental observation. Other studies combine knowledge graphs with attention-guided learning for activity recognition, but the construction of such knowledge graphs requires pre-labeled data, essentially remaining a supervised or semi-supervised paradigm, unsuitable for passive deployment in unlabeled home environments. Existing home safety monitoring methods based on Artificial Intelligence (AI) behavioral analysis also rely on predefined risk rules to construct a three-dimensional "person-environment-risk" knowledge graph, rather than being passively initiated from scratch.

[0039] Regarding active learning and user feedback, existing technologies utilize resident feedback to optimize sensor-activity associations extracted from the ontology. Researchers have designed active learning interfaces for multi-resident scenarios. These methods share the common premise of a pre-built ontology or initial model; feedback is used only to fine-tune existing association rules, rather than passively initiating the graph from a blank state, and it fails to utilize feedback for creating, pruning, or upgrading the confidence of graph nodes (such as daily routines).

[0040] In the field of multimodal non-contact sensing technology, existing technologies have made progress in monitoring human activity using various non-contact sensing technologies such as WiFi signals (e.g., Channel State Information (CSI), Beamforming Feedback Information (BFI)), radar, and thermal imaging. While these technologies utilize sensing data, they do not address how to automatically construct interpretable, privacy-constrained family context graphs using these passive sensing streams, nor do they address how to integrate caregiver interaction feedback into the continuous optimization of these graphs.

[0041] In summary, existing technologies suffer from the following unresolved shortcomings: a lack of ability to passively initiate family context graphs from environmental perception streams (current passive initiation methods are limited to activity classification models); a lack of active learning optimization mechanisms involving caregivers (current active learning methods rely on initial ontologies or predefined rules); insufficient privacy protection (most systems upload family models or behavioral data to the cloud); and a lack of interpretability in reasoning (current anomaly detection systems typically output anomaly scores rather than natural language explanations based on specific family contexts). Therefore, there is an urgent need for a passively initiated, caregiver-participatory, privacy-preserving, and interpretable method for constructing and reasoning family context graphs to address the aforementioned deficiencies of existing technologies in home-based elderly care scenarios.

[0042] To address the following technical shortcomings of existing home-based elderly care intelligent systems: Context modeling methods based on manual configuration or predefined rules have high deployment costs and struggle to adapt to the natural shifts in family routines; existing passive observation activity recognition methods can only construct atomic activity classification models and cannot automatically generate structured context graphs containing family topology, daily routines, caregiver roles, and behavioral baselines; existing knowledge graph-based smart home systems rely on expert-defined rules or pre-labeled data and cannot be passively initiated from environmental perception flows; existing active learning methods rely on initial ontologies or preset models, lacking mechanisms for caregivers to participate in graph optimization by confirming or rejecting behavioral hypotheses, and cannot directly use feedback for graph node creation, pruning, or confidence upgrades; most existing systems upload family models or behavioral data to the cloud, failing to achieve localized storage and non-exportable constraints for family-specific context data; and sixth, existing anomaly detection systems typically output general anomaly scores, lacking interpretable caregiving reasoning based on family-specific contexts. Therefore, this invention provides a method and system for constructing a family context graph based on user care to overcome these deficiencies.

[0043] The following description, in conjunction with the accompanying drawings, describes the method and system for constructing a family context graph based on user care according to the present invention.

[0044] Figure 1 This is one of the flowcharts illustrating the family context graph construction method based on user care provided by the present invention, such as... Figure 1 As shown, the method includes the following steps 101 to 103.

[0045] Step 101: Obtain the environmental perception data stream of the target residence. The environmental perception data stream includes at least one of the following: wireless beamforming feedback sequence, radar occupancy sequence, home appliance usage information, and environmental sensor readings.

[0046] It should be noted that the execution entity of the user care-based family context graph construction method in this invention can be a user care-based family context graph construction system, hereinafter referred to as the system.

[0047] Specifically, the target residence can be a target house, i.e., a place where user care is required. The target residence may include at least one of the following: bedroom, living room, balcony, kitchen, bathroom, and dining room.

[0048] Environmental perception data streams can be acquired through sensors (smart devices) deployed in the target residence, such as wireless routers that support wireless beamforming feedback acquisition (used to acquire wireless beamforming feedback sequences, such as WiFi beamforming feedback sequences (BFI)), millimeter-wave radar sensors (used to acquire radar occupancy sequences), smart sockets (used to acquire home appliance usage information, such as usage status), and door magnetic sensors (used to acquire environmental sensor readings), etc.

[0049] Step 102: Based on the environmental perception data stream, determine the topological relationship between rooms in the target residence, the room where each device in the target residence is located, and at least one living pattern of the first user in the target residence. The device is a sensor that provides the environmental perception data stream or a home appliance sensed by the sensor.

[0050] Specifically, the target residence includes at least one room, such as a bedroom, living room, balcony, kitchen, bathroom, and dining room. The first user can be a person being cared for, such as an elderly person living alone, a minor, or an animal being cared for, such as a pet cat or dog.

[0051] In practical applications, the system passively starts up after acquiring the environmental awareness data stream. Passive startup means that the system automatically executes the steps of generating a family context graph simply by accumulating observations of the environmental awareness data stream, without any pre-configured room topology, device locations, daily routines, or caregiver role definitions in its initial state. The caregiver role definition refers to the role of the primary user, such as an elderly person living alone or a minor.

[0052] Specifically, when the system passively starts, it can infer the topological relationships between rooms in the target residence based on the environmental perception data stream. For example, it can infer the adjacency relationship from the bedroom to the corridor, then to the bathroom and kitchen by detecting the sequence of changes in the wireless beamforming feedback anchor point. It can also infer the room where each device in the target residence is located based on the environmental perception data stream. For example, if appliance A is in use based on appliance usage information, and room B is occupied based on environmental sensor readings, then appliance A is located in room B. Furthermore, it can infer at least one daily routine of the first user in the target residence based on the environmental perception data stream. For example, if the kettle in the kitchen is in use at 7:00 AM every day, then it can be inferred that the first user's morning routine (daily routine) is drinking water around 7:00 AM. Here, the anchor point refers to the wireless router that supports wireless beamforming feedback acquisition.

[0053] It should be noted that the passive startup step supports inferring device status through device co-occurrence analysis and automatically identifying typical work and rest periods through time clustering algorithms.

[0054] Step 103: Based on the topological relationships between the rooms, the rooms where the devices are located, and at least one of the living patterns of the first user, construct a family context graph for caring for the first user.

[0055] Specifically, after obtaining the topological relationship between each room, the room where each device is located, and at least one daily routine of the first user, for each device, the positional relationship between the device and the room where the device is located can be determined based on the room where the device is located (i.e., the device is located in that room). For each daily routine, the triggering relationship between the daily routine and the corresponding device can be determined based on the device corresponding to the daily routine (e.g., the use of the kettle triggers the morning routine of drinking water around seven o'clock in the morning).

[0056] Furthermore, a family context graph can be constructed using each room, each device, and each daily routine as nodes, and each topological relationship, each location relationship, and each usage relationship as edges. This family context graph is then used to provide care for the primary user.

[0057] The present invention provides a method for constructing a family context graph based on user care. This method acquires an environmental perception data stream of a target residence, including at least one of wireless beamforming feedback sequences, radar occupancy sequences, appliance usage information, and environmental sensor readings. Based on this data stream, it determines the topological relationships between rooms in the target residence, the rooms where various devices are located, and at least one daily routine of a first user. The devices are sensors providing the environmental perception data stream or appliances detected by the sensors. Based on the topological relationships between rooms, the rooms where various devices are located, and at least one daily routine of the first user, a family context graph for caring for the first user is constructed. This invention eliminates the need for manual configuration or predefined rules, enabling the construction of a structured context graph containing room topological relationships and daily routines based on existing passive observation activities. It also adapts to the natural drift of family routines and reduces deployment costs compared to context modeling methods that require manual configuration or predefined rules.

[0058] Optionally, determining the topological relationships between rooms in the target residence, the rooms where each device in the target residence is located, and at least one daily routine of the first user in the target residence based on the environmentally perceived data stream includes: Based on the spatial co-occurrence and temporal correlation of the environmental perception data stream, the topological relationship between the rooms is determined; For each device, the room where the device is located is determined by overlapping analysis of the active time period of the device and the occupancy time period of each room in the environmental perception data stream. At least one repetitive data group is extracted from the environmental perception data stream based on the time clustering algorithm. For each repetitive data group, the corresponding life pattern is identified. The repetitive data group contains multiple data of the same device being active in the same location within the same time period.

[0059] Specifically, spatial co-occurrence characterizes data within the same room in the environmental perception data stream, while temporal correlation characterizes data at the same time in the environmental perception data stream.

[0060] In practical applications, the system can determine the topological relationships between rooms based on the spatial co-occurrence and temporal correlation of the environmental perception data stream. The specific reasoning process for inferring the topological relationships is as follows: The system abstracts each wireless beamforming feedback anchor point, millimeter-wave radar detection grid unit, door magnet, and other sensing sources into independent spatial sensing units. First, it extracts the signal strength or occupancy state sequence of each spatial sensing unit within a continuous time window, calculates the Pearson correlation coefficient between any two spatial sensing units, and constructs a spatial co-occurrence matrix. If two spatial sensing units exhibit a high positive correlation in their time series (e.g., the bedroom radar and bedroom wireless anchor signal fluctuate synchronously when a person is active in the bedroom), they are determined to belong to the same physical space area (the same room). Simultaneously, for spatial sensing units that may be located in different rooms, spatial adjacency relationships are inferred by analyzing the sequence and time delay of their signal changes. For example, the system records the first user movement event sequence: when the occupancy signal of "spatial sensing unit A" (bedroom radar) changes from "occupied" to "unoccupied," after a 2-5 second delay, the occupancy signal of "spatial sensing unit B" (corridor radar) changes from "unoccupied" to "occupied," and after another 3-8 second delay, "spatial sensing unit C" (bathroom radar) changes to "occupied," a directed transfer chain A→B→C is established. After accumulating a large number of transfer events (e.g., more than 200), the system constructs a family spatial topology using a directed graph: nodes in the directed graph represent candidate spatial regions (candidate rooms) obtained through co-occurrence clustering, and edges represent high-frequency movement paths with time delays consistent with walking characteristics. Combining transfer probabilities and delay distributions, the system infers that "bedroom-corridor-bathroom" are adjacent connected regions.

[0061] The system can determine the room where each device is located by overlapping analysis of the active periods of each device and the occupancy periods of each room in the environmental perception data stream. The active period refers to the time when the device is in use, and the occupancy period refers to the time when the first user is in the room.

[0062] Specifically, the inference process for the spatial area (room) where the device is located is as follows: For each device with detectable usage rhythm (such as a smart socket or smart light controller), the system constructs its "active time period feature vector". For example, with a 24-hour day as the horizontal axis and each 15-minute time slot as a time slot, the system counts the number of times the device power exceeds a threshold (such as an electric kettle > 800 watts (W)) within that time slot, forming a 96-dimensional activity distribution vector. Simultaneously, for each inferred spatial area (such as a kitchen or living room), a "regional occupancy distribution vector" of the same granularity is generated using radar or wireless occupancy data, representing the probability or duration of occupancy in that area within that time slot. Then, the Spearman rank correlation coefficient between the device activity vector and the occupancy vectors of each spatial area is calculated, because the usage rhythm of electrical appliances in different areas may be monotonically correlated with the human activity rhythm, and strict linearity is not required. The system selects the spatial area with the highest correlation and a p-value below 0.01 as the candidate location. In addition, a "transient event co-occurrence" verification is introduced: the timestamp of each device's on / off event is extracted, and it is checked whether there are corresponding personnel appearance / disappearance events (triggered by radar or door sensor) in the target space area within 30 seconds before and after the event; if the accuracy of such transient matching exceeds a certain threshold (e.g., 70%), the inference confidence is further strengthened. Combining the long-term rhythm correlation score and the transient event matching rate, the system determines the spatial affiliation of the device (e.g., "Smart socket A is located in the kitchen").

[0063] The system can use a time clustering algorithm to process the environmental perception data stream, identify recurring time-location-patterns (at least one recurring data group), and then identify candidate daily routines based on the recurring data groups (for example, moving from the bedroom area to the bathroom area from 6:20 to 6:40 every day, staying for nine to thirteen minutes, and then moving to the kitchen area).

[0064] In this embodiment of the invention, by analyzing the spatial co-occurrence and temporal correlation of the environmental sensing data stream, the overlap between the active periods of devices and the occupancy periods of each room in the environmental sensing data stream, and the repeated data groups in the environmental sensing data stream, the completeness, accuracy, and reliability of the topological relationships of rooms, the rooms where devices are located, and at least one living pattern can be improved.

[0065] Optionally, the method further includes: Set confidence levels for each node and edge in the family context graph, and / or add source metadata for each node and edge in the family context graph, wherein the source metadata includes at least one of observation source, first observation timestamp, latest observation timestamp, and number of observations.

[0066] In practical applications, after obtaining the family context graph, an initial confidence score can be generated for each node (including spatial regions (rooms), devices, and daily routines) and edge (including topological relationships, locational relationships, and triggering relationships) in the family context graph. This confidence score can be a default confidence score or a calculated one.

[0067] For example, the initial confidence level is obtained by weighting and summing the long-term rhythm correlation score and the instantaneous event matching rate.

[0068] After obtaining the family context graph, source metadata can be attached to each node (including spatial regions (rooms), devices, and living patterns) and edge (including topological relationships, location relationships, and triggering relationships) in the family context graph. The source metadata includes the observation source, the first observation timestamp, the most recent observation timestamp, and the number of observations.

[0069] In this embodiment of the invention, by setting confidence levels and / or adding source information to nodes and edges in the family context graph, the family context graph is upgraded from a "static snapshot" to a "dynamic evidence system." This allows for precise resolution of information conflicts and weighted reasoning; convenient source tracing and error correction, and rapid location of contamination sources; enhanced interpretability and support for transparent citation; support for permission isolation and data lifecycle management; realization of knowledge temporal evolution and automatic elimination; and quantification of multi-source data quality. Ultimately, this endows AI with the ability to rationally question and the confidence to provide factual answers.

[0070] Optionally, after constructing the family context graph for caring for the first user based on the topological relationships between the rooms, the rooms where the devices are located, and at least one of the first user's daily routines, the process further includes: Based on the current environmental perception data stream, determine the current activity data of the first user; Based on the family context graph, the current activity data is detected to obtain detection data; When the detection data indicates a change in the first user's daily routine, or when the detection data indicates a conflict between the current activity data and the edge and / or node corresponding to the current activity data in the family context graph, a behavioral description text corresponding to the first user is generated. The behavioral description text includes at least one of the following: time offset of the daily routine, change in duration, disappearance of the daily routine, decrease in frequency of the daily routine, and newly emerging daily routine. The behavioral description text is fed back to the second user to inform the second user.

[0071] Specifically, the behavioral description text is a behavioral hypothesis, that is, text that makes assumptions about the behavior of the first user. The second user can be the person caring for the first user, such as the adult children of an elderly person living alone.

[0072] In practical applications, the system can generate behavioral description text based on cumulative observations (the environmental perception data stream continuously acquired after building the family context graph, i.e., the current environmental perception data stream) and present it to a second user.

[0073] Specifically, the system continuously monitors real-time environmental awareness data streams, comparing newly observed patterns (current activity data) with the constructed family context graph to obtain detection data. This includes detecting time shifts in recorded daily routines, changes in duration, new occurrences of routines, or the disappearance or decrease in frequency of existing routines. When these changes are detected, indicating a change in the first user's daily routine, the system automatically generates a behavioral hypothesis in natural language (behavioral description text) and presents it to the second user through an interactive interface (e.g., mobile app notifications, web dashboards, or voice assistants). This behavioral description text may include a description of the change, possible reasons, and options for confirmation or rejection.

[0074] In addition, the system also has a conflict detection function: when a newly observed pattern (current activity data) conflicts with a high-confidence node in the constructed family context graph, the aforementioned step of generating behavioral description text is triggered. Here, high confidence is defined as a confidence level greater than the first confidence threshold.

[0075] In this embodiment of the invention, behavioral hypotheses combined with natural language interaction serve as a lightweight interaction vehicle. Real-time sensing data is continuously compared with the baseline of the spectral map to identify behavioral drift phenomena such as shifts in sleep patterns, changes in duration, the generation of new patterns, and the decay of existing patterns. A conflict detection mechanism is superimposed to capture conflicts between observed data and a high-confidence baseline, generating behavioral hypotheses and feeding them back to the second user. This allows the second user to receive early warnings and prevent accidents, transforming passive rescue into proactive intervention; it breaks down the information black box, reducing the second user's anxiety; it assists the second user in prioritizing their energy, improving collaboration efficiency; it promotes emotional interaction and detects early signs of depression. This forms a monitoring-response closed loop, extending the first user's independent life and significantly improving care satisfaction.

[0076] Optionally, after feeding back the behavior description text to the second user to inform the second user, the method further includes: Receive feedback from the second user regarding the behavior description text; The family context graph is updated based on the feedback information.

[0077] The step of updating the family context graph based on the feedback information can be: If the feedback information confirms the assessment, increase the confidence of the nodes and / or edges corresponding to the behavioral description text in the family context graph. If the feedback information is rejected, reduce the confidence of the nodes and / or edges corresponding to the behavioral description text in the family context graph, or delete the nodes and / or edges corresponding to the behavioral description text in the family context graph.

[0078] In practical applications, the system can receive feedback information from a second user and use a multi-factor mechanism to update the family context graph based on this feedback. This multi-factor mechanism can update the family context graph based on feedback information, as well as other factors (such as time decay, user commands, etc.).

[0079] Specifically, the system receives feedback from the second user on the behavior description text, and the feedback information may include the second user's confirmation or rejection of the behavior description text.

[0080] Based on the feedback information, the system performs at least one of the following map update operations: For the confirmed behavior description text, i.e. feedback information characterization confirmation, the node and / or edge corresponding to the behavior description text in the family context graph is upgraded to a high confidence baseline node, i.e. the confidence of the node and / or edge corresponding to the behavior description text is increased, such as increasing the confidence to a preset high threshold. For rejected behavior description text, i.e. feedback information representing rejection, the nodes and / or edges corresponding to the behavior description text in the family context graph are pruned from the family context graph, or the confidence of the nodes and / or edges corresponding to the behavior description text in the family context graph is reduced and marked as to be monitored, such as increasing the confidence to a preset low threshold.

[0081] In addition, selection criteria can be set to choose whether to perform pruning operations on nodes and / or edges or assume confidence levels.

[0082] For example, for rejected behavioral description text, if the confidence of the node and / or edge corresponding to the behavioral description text in the family context graph is greater than the second confidence threshold, the confidence of the node and / or edge corresponding to the behavioral description text in the family context graph is reduced and marked as to be monitored; if the confidence of the node and / or edge corresponding to the behavioral description text in the family context graph is less than or equal to the second confidence threshold, the node and / or edge corresponding to the behavioral description text in the family context graph is pruned from the family context graph.

[0083] In this embodiment of the invention, relying on the feedback information of the second user, the family context graph is optimized by linking confidence level adjustment and node pruning, forming a closed-loop active learning system of "perception-hypothesis-feedback-update", which realizes the long-term dynamic evolution of the graph.

[0084] Optionally, after receiving feedback from the second user on the behavior description text, the method further includes: If the feedback information includes the remarks input by the second user, the remarks are appended as additional information to the node and / or edge corresponding to the behavior description text. And / or, update the source metadata of the nodes and / or edges corresponding to the behavior description text based on the feedback information; And / or, record the feedback source and feedback time of the feedback information, as well as the operation type for updating the family context graph.

[0085] Specifically, the feedback information may also include notes input by the second user, which are contextual explanation text, i.e., notes on the text describing the behavior.

[0086] In practical applications, for confirmed behavioral description text, the notes in the feedback information can be appended as annotations to the nodes and / or edges corresponding to the behavioral description text in the family context graph.

[0087] For behavior description text that is confirmed or rejected, the source metadata of the node and / or edge corresponding to the behavior description text in the family context graph can be updated based on the feedback information.

[0088] When or after updating the family context graph based on feedback information, the source of feedback, the time of feedback, and the type of operation for updating the family context graph can be recorded based on the feedback information.

[0089] In this embodiment of the invention, relying on the feedback information from the second user, the confidence level is adjusted, node pruning is performed, and attribute updates are completed to optimize the graph, forming a closed-loop active learning system of "perception-hypothesis-feedback-update" to achieve long-term dynamic evolution of the graph.

[0090] Optionally, after constructing the family context graph for caring for the first user based on the topological relationships between the rooms, the rooms where the devices are located, and at least one of the first user's daily routines, the process further includes: Based on the current environmental awareness data stream, user event data is determined, which includes at least one of the following: event occurrence time, event location, event-associated device type, and event duration. Based on the life patterns corresponding to the user event data in the family context graph and the user event data, determine the degree of deviation corresponding to the user event data; Based on the degree of deviation and the relevant data of the life patterns corresponding to the user event data, an explanation of the care status of the first user is generated. The relevant data includes at least one of confidence level, latest update time, source metadata and additional information. The explanation of the care status includes at least one of deviation description text, deviation reason prompt and suggested operation information. The care status is explained and fed back to the second user to inform the second user.

[0091] Specifically, the care status interpretation can be a descriptive text of the first user's status.

[0092] In practical applications, the system uses the family context graph to perform local care reasoning and generate care status explanations.

[0093] Specifically, the system receives environmental awareness data streams in real time, determines user events from the current environmental awareness data streams, and extracts user event data, such as event occurrence time, location of occurrence, associated device type, and event duration. The associated device type refers to the type of device associated with (involved in) the user event.

[0094] Furthermore, query the activity patterns corresponding to the user event in the family context graph, that is, the baseline of the life pattern of the time window and the associated spatial area (room) of the activity pattern, the status of the caregiver and the historical behavior pattern, and calculate the degree of deviation between the user event and the activity pattern.

[0095] Based on the degree of deviation, and combined with metadata such as the confidence level of the activity pattern in the family context graph, the most recent update time, and whether it has been confirmed by the caregiver, a care status explanation is generated. This explanation includes a deviation description (e.g., "This morning's bathroom visit was delayed by 45 minutes from the baseline"), a reason for the deviation (e.g., "Possibly due to changes in sleep quality"), and suggested actions (e.g., "It is recommended to pay attention to the elderly person's condition"). The care status explanation is output in natural language text format and does not contain the original sensor data or specific node identifiers of the graph.

[0096] In this embodiment of the invention, differentiated anomaly detection and interpretable inference are performed. The local inference engine compares real-time events with family-specific baselines (life patterns) in the graph, enabling it to distinguish between normal individual behavioral changes and genuine care-related anomalies. The output care status explanation is based on the specific family context (e.g., "delayed morning bathroom visit") rather than a generic anomaly score, making it easier for the first user to understand the cause of the alarm.

[0097] Optionally, the method further includes: Generate a bounded inference context package, wherein the bounded inference context package contains a subgraph corresponding to the user event data in the family context graph; The bounded inference context packet is anonymized and then sent to the authorization system.

[0098] Specifically, a bounded reasoning context package refers to a graph subgraph that generates explanations of care status through reasoning.

[0099] In practical applications, while the care status interpretation is generated, the system can generate a bounded inference context package. This bounded inference context package contains a graph subgraph related to user events, that is, a subgraph corresponding to user event data in the family context graph. Before being sent, it undergoes privacy filtering to remove attributes that directly identify the first user.

[0100] In this embodiment of the invention, a bounded inference context package highly correlated with user events and with defined boundaries is generated from the full family context graph. This eliminates noise and reduces analysis costs for subsequent analysis of user time. Furthermore, the bounded inference context package undergoes privacy filtering before export, effectively preventing data leakage to the cloud or third parties.

[0101] Optionally, the method further includes: Received data export instruction; When the data export instruction indicates that specified data should be output, the specified data is output. The specified data is a care status interpretation and / or comprehensive care data, and the comprehensive care data represents the activity of the first user within a set time period. If the data export instruction indicates that non-specified data should be output, the output of the non-specified data shall be rejected. The non-specified data is data other than the specified data.

[0102] Specifically, comprehensive care data can be aggregated-level summaries.

[0103] In practical applications, the system enforces localized privacy constraints on the family context graph. The entire family context graph is stored in an encrypted database on the home gateway or local server, and any operation that exports the raw data of the family context graph (including nodes, edges, confidence scores, metadata, and first-user annotations) to external networks is prohibited. Only care status interpretations and / or comprehensive care data (e.g., "Overall activity is normal today") are allowed to be output to external systems; the internal structure of the family context graph or specific parameters that identify individual behavior are not permitted.

[0104] The technical implementation of prohibiting the export of raw data from the home context graph relies on multi-layered protection. For example: (1) At the network layer, the home gateway is configured with strict (iptables / nftables) firewall rules, allowing only whitelisted outbound connections (such as Message Queuing Telemetry Transport (MQTT) topics for push services to applications used by second users), and discarding Transmission Control Protocol (TCP) / User Datagram Protocol (UDP) traffic to all other destination ports by default. The graph database process is bound to the local loopback address, and any remote connection requests are rejected by the kernel. (2) At the transport layer, even internal communication is forced to use encrypted channels based on Transport Layer Security 1.3 (TLS 1.3), and client certificates are issued by the home hardware security module, distrusting external Certificate Authorities (CAs). (3) At the application layer, a dedicated data export agent is deployed. This agent performs deep packet inspection on all outgoing data packet payloads, filtering out regular expression patterns that match features such as the graph node identifier (ID), original sensor (sensor) values, and specific timestamps; only the generated care status interpretation and / or comprehensive care data are allowed. (4) In terms of key management, the graph encryption key is generated and sealed by the device's built-in Trusted Platform Module (TPM) / security chip. The decryption key cannot be exported under any circumstances, so even if the local database file is physically stolen, it cannot be decrypted. Through the above combined technical means, the privacy constraint of the graph's original data not being exportable is fully realized from the link, network, application to cryptographic level.

[0105] In this embodiment of the invention, through strict local privacy protection, the family context graph never leaves the family's local storage and only outputs aggregated summaries or explanatory text to the outside world, thereby improving data security.

[0106] Optionally, the method further includes: For a specified element in the family context graph, the confidence of the specified element is reduced according to a preset decay rate. The specified element is a node or edge that has not been confirmed by relevant information within a set time period. The relevant information is newly acquired environmental perception data stream and / or feedback information. If the confidence level of the specified element is lower than the lower confidence limit, the specified element is deleted from the family context graph.

[0107] In practical applications, the system also sets up a periodic application time decay rule independent of the second user feedback: for map elements that have not been corroborated by new observations (newly acquired environmental perception data streams) or confirmed by the second user (representing confirmed feedback information), i.e. designated elements, the confidence of the designated elements is reduced according to a preset decay rate. When the confidence of the designated element is lower than the preset elimination threshold (confidence lower limit), the designated element is automatically deleted.

[0108] In this embodiment of the invention, by automatically eliminating outdated spectral elements through a matching periodic time decay mechanism, the long-term dynamic evolution of the spectrum can be achieved. This ensures that the system can automatically process outdated information, reduce noise interference, and maintain the reliability and self-cleaning ability of the spectrum during long-term operation.

[0109] Optionally, the method further includes: Receive adjustment instructions from the second user; Based on the adjustment instruction, the nodes and / or edges corresponding to the adjustment instruction in the family context graph are adjusted.

[0110] In practical applications, the system also supports second users to actively initiate manual correction requests (adjustment commands) to the family context graph. The second user can edit any node and / or edge in the family context graph through the interactive interface, including adding new rules, deleting erroneous nodes, modifying confidence or adding annotations. The system adjusts the nodes and / or edges in the family context graph corresponding to the adjustment commands based on the adjustment commands.

[0111] In this invention, by supporting manual correction, the second user can passively respond to hypotheses or actively edit the graph, providing flexible human-machine collaborative governance capabilities.

[0112] Optionally, the method further includes: Perform the following steps for each graph element in the family context graph: Record the historical versions of the aforementioned atlas elements; If a rollback operation is detected for the graph element, the graph element in the family context graph is rolled back to the historical version specified by the rollback operation.

[0113] Specifically, graph elements refer to nodes and / or edges in a family context graph.

[0114] In practical applications, when updating or after updating the family context graph, the system also maintains the version history of each graph element, supports rollback to any historical state of the graph element, and provides a complete graph update log, that is, records all historical versions of each graph element. When a rollback operation is detected for a graph element, the graph element is rolled back to the historical version specified by the rollback operation.

[0115] In this embodiment of the invention, the complete version history and update log of the graph elements are preserved, providing flexible human-machine collaborative governance capabilities.

[0116] The following section provides a further explanation of the family context graph construction method based on user care provided by this invention, using specific scenarios as examples.

[0117] This invention takes a single elderly person's home as an example. The home (target residence) is equipped with a wireless router supporting WiFi beamforming feedback acquisition, a millimeter-wave radar sensor (installed on the living room ceiling), several smart sockets (for detecting the usage status of kitchen appliances, televisions, etc.), and door magnetic sensors (installed on the bedroom and bathroom doors). All sensors are connected through a home gateway. Initially, the system has no pre-configured room topology, device locations, daily routines, or caregiver roles.

[0118] (1) Passive start-up phase See Figure 2 and Figure 3 , Figure 2 This is the second flowchart illustrating the family context graph construction method based on user care provided by this invention. Figure 3 This is a schematic diagram of the process for generating a family context graph using the passive startup engine provided by the present invention.

[0119] The first 72 hours after system startup constitute a passive startup phase, during which a home context graph is constructed. The system continuously receives multimodal environmental sensing data streams (BFI, radar, appliance usage, and environmental sensors) through the environmental sensing interface (environmental data acquisition module): the router collects BFI information from associated WiFi clients every 30 seconds; the radar outputs room occupancy status (occupied / unoccupied and approximate location coordinates) every 10 seconds; smart sockets record the on / off times and power changes of each appliance; and door sensors record on / off events. This includes WiFi beamforming feedback (BFI), radar occupancy sequences, and appliance usage information.

[0120] The graph construction module passively starts generating an initial family context graph (nodes: spatial regions, devices, and daily routines; edges: topological and temporal relationships; with accompanying confidence and source metadata) according to the following steps.

[0121] First, spatial topology (topological relationship) inference. The system analyzes the spatial co-occurrence of BFI anchor point signal strength (spatial co-occurrence analysis) and finds a temporal correlation between the signal changes of the "bedroom router BFI anchor point" and the "WiFi client near the bathroom": whenever the bedroom client's signal decreases, the bathroom client's signal increases, and there is a fixed delay between the two (approximately 2-3 seconds). Combining this with radar-detected human movement trajectories (moving from the bedroom area to the bathroom area), the system infers a connected path between the bedroom and the bathroom. Similarly, the system continues to infer the adjacency relationship of bathroom → corridor → kitchen. After cumulative observation, the system automatically generates a topology map containing four spatial nodes: "bedroom," "bathroom," "corridor," and "kitchen." Edges between nodes are labeled "adjacent," and source metadata is attached to each inference result (e.g., "based on BFI anchor point changes, number of observations = 217, first observation time = day 1, 06:32"), with an initial confidence level set to 0.60.

[0122] Secondly, device location inference, i.e., device activity analysis. The system detected that smart socket A (connected to an electric kettle) regularly exhibited power peaks during three time periods each day: 06:30-07:00, 12:00-12:30, and 17:30-18:00. These time periods highly overlapped with the "occupants in the kitchen area" detected by radar, thus inferring that smart socket A and the electric kettle were located in the kitchen area. Similarly, smart socket B (connected to a TV) had its active period concentrated between 19:00-21:00, and this overlapped with the radar occupancy signal in the living room, inferring that it was located in the living room. The system added these device nodes to the map and added "located" edges connecting them to the corresponding spatial nodes. For example, the router is located in the living room, the smart speaker is located in the living room, the range hood is located in the kitchen, and the water heater is located in the bathroom.

[0123] Secondly, the system infers daily routine patterns. It uses a temporal clustering algorithm (based on Density-Based Spatial Clustering of Applications with Noise (DBSCAN), with a 30-minute time window and a minimum sample size of 5) to perform temporal correlation analysis on sensor events over three consecutive days. The algorithm identifies the following recurring pattern: every day from 06:20 to 06:40, radar detects human movement in the bedroom area → door sensor (bedroom door) opens → approximately 5-10 seconds later, bathroom radar is occupied → door sensor (bathroom door) closes → 9-13 minutes later, bathroom door opens → approximately 10-20 seconds later, kitchen radar is occupied. The system abstracts these event sequences into a "morning routine" (daily routine) node, containing the starting time window (06:20-06:40), movement path (bedroom → bathroom → kitchen), and bathroom stay duration (9-13 minutes). The initial confidence level is set to 0.65 (based on 10 observations), and the source metadata is attached as "BFI + radar + door sensor joint pattern, number of observations = 10". Simultaneously, the system also identified patterns in daily life such as "midday nap patterns" (the bedroom radar remains active and stationary from approximately 13:00-14:00, lasting 40-70 minutes) and "evening living room activity patterns" (the living room radar is active from 19:00-21:00, along with the smart TV socket), with initial confidence levels of 0.55 and 0.70, respectively. Figure 3 The times shown are "06:20-06:40 moving", "06:40-06:52 bathroom", and "06:52-07:10 kitchen".

[0124] See Figure 4 , Figure 4 This is a schematic diagram of the family context graph provided by the present invention: After 72 hours, the system constructed a family context graph containing space (house) nodes (5: bathroom, kitchen, living room, corridor, and bedroom), device nodes (5: water heater, kettle, television, corridor light, and air conditioner), life pattern nodes (5 life patterns), and several associated edges. All nodes have confidence scores and source metadata. This family context graph is stored in the encrypted database of the home gateway. Specifically, the water heater is located in the bathroom and can trigger life pattern 1; the kettle is located in the kitchen and can trigger life pattern 2; the television is located in the living room and can trigger life pattern 3; the corridor light is located in the corridor and can trigger life pattern 4; and the air conditioner is located in the bedroom and can trigger life pattern 5. The kitchen, bathroom, and living room are adjacent to each other, and the corridor is adjacent to both the living room and the bedroom.

[0125] (2) Behavioral hypothesis generation and caregiver interaction See Figure 2 and Figure 5 , Figure 5 This is a schematic diagram of the caregiver interaction interface provided by the present invention: The system continuously runs and monitors the real-time environmental perception data stream, and compares it with the baseline (life routine) in the family context graph.

[0126] On day 30, the hypothesis (behavioral description text) generation module detected the following changes: (1) the time spent in the bathroom during the morning routine gradually increased from 9-13 minutes to 20-25 minutes; (2) the morning start time window was delayed from 06:20-06:40 to 07:00-07:20; (3) the above changes have occurred for 7 consecutive days without any signs of recovery. Based on the preset deviation threshold (duration extension exceeding 40%, time deviation exceeding 30 minutes and lasting for more than 3 days), the system determined that the change was a candidate behavior drift. That is, behavior hypotheses are generated based on cumulative observations (detecting time slice discounts, duration changes, new patterns, or disappearing patterns).

[0127] The system automatically generates a natural language description of the behavior (behavior plus device): "Morning bathroom visit time has been delayed from approximately 06:30 to approximately 07:15, and bathroom stay time has increased from 10-12 minutes to 22 minutes. Is this an expected change? (e.g., due to sleep schedule adjustment, health condition, or medication effects)." This description is presented to the elderly person's children (caregivers) through a second user interaction module (mobile app push notification), providing two buttons: "Confirm (expected change)" and "Reject (abnormal situation)," as well as an optional text box for inputting contextual explanations. In other words, the assumption is presented to the caregiver (natural language prompts and confirm / reject options).

[0128] Upon seeing the notification, the caregiver recalls that the elderly person recently started taking a new medication for blood pressure issues, which has caused them to wake up slower and move more slowly in the morning. The caregiver then clicks the "Confirm (Expected Change)" button and adds a note in the text box: "New medication causes slower waking up," as in the example. They can also add a note like, "The doctor said this is a normal adaptation period and is expected to improve in two weeks." The caregiver can also click the "Reject (Abnormal Circumstances)" button.

[0129] also, Figure 4 The interactive interface also displays historical hypotheses, namely behavioral hypotheses generated in the past, such as the caregiver's confirmed "3 days ago: the midday nap routine appeared" and the pending "5 days ago: the evening jogging routine appeared".

[0130] (3) Map update based on feedback information See Figure 2 and Figure 6 , Figure 6This is a flowchart illustrating the process of updating the family context graph provided by the present invention: After the system receives the caregiver's confirmation feedback and explanatory text (feedback information), the graph update engine performs the following operations: receiving the caregiver's feedback and updating the graph (confirmation → upgrade to high confidence limit; rejection → value or demotion; attachment notes).

[0131] First, the initial confidence level of the "Morning Routine" node is increased from 0.65 to a preset high threshold of 0.92 (upgrading it to a high-confidence baseline node confidence level: 0.92), and it is marked as a "high-confidence baseline confirmed by caregivers." This process is passively initiated. Simultaneously, the text entered by the caregiver is appended to this node as a comment, including the confirmation time, caregiver identity (anonymized), and an explanation.

[0132] Secondly, the system updated the "Toilet Stay Duration" attribute for this node, adjusting its typical range from 9-13 minutes to 20-25 minutes, and recording the reason for the adjustment as "caregiver-confirmed behavior drift". The start time window was also adjusted from 06:20-06:40 to 07:00-07:20.

[0133] In addition, a new record has been added to the source metadata: "Feedback Source = Caregiver, Feedback Type = Confirmation, Feedback Time = Day 30, 14:23, with accompanying comments".

[0134] If the caregiver clicks "Reject" (e.g., the elderly person has no actual medication changes, but their children consider the changes abnormal), and the caregiver rejects the request, the system performs pruning (node ​​deletion): the candidate pattern is marked as pending monitoring, its confidence level is reduced to 0.20, and its sensitivity is increased for the next 7 days (any similar deviations are immediately escalated to an alarm). If there are no repeat observations for 14 consecutive days, i.e., no content related to the node is detected, the node is automatically deleted.

[0135] The system also implements a time decay rule: Each week, the confidence level of all nodes that have not received new observational evidence and have not been confirmed by caregivers is decayed at a rate of 5%, meaning a 5% decay per week. For example, a "midday nap pattern" with an initial confidence level of 0.55 will decay to 0.55 × (0.95)^4 ≈ 0.45 if it is not observed for four consecutive weeks (e.g., the elderly person changes their routine). When the confidence level falls below the culling threshold of 0.10, the node is automatically pruned.

[0136] A contradiction detection mechanism also operates in real time. Suppose that one night at 10:00 PM, the system detects that the kitchen radar is continuously occupied and the smart socket (electric kettle) is active (i.e., new observation: kitchen activity at 10:00 PM), while the high-confidence node in the graph (confirmed by the caregiver) shows "no one in the kitchen after 10:00 PM." The system detects this contradiction and automatically triggers a new hypothesis generation: "Nighttime kitchen activity detected, is this an abnormal situation (e.g., insomnia or sleepwalking)?" and pushes it to the caregiver. If the caregiver confirms that the elderly person got up to boil water that day because they were thirsty, which is an isolated incident, the system will mark this event as a "single anomaly" without modifying the baseline.

[0137] (4) Local care reasoning See Figure 2 and Figure 7 , Figure 7 This is a schematic diagram of the process for generating care status interpretations provided by this invention. Example of real-time reasoning: At 2:00 PM one afternoon, the radar sensor reports that the bedroom area has been continuously occupied for 60 minutes, during which there is no obvious body movement (only micro-breathing signals are present). That is, the bedroom radar remained occupied for 60 minutes at 2:00 PM without movement. Smart socket data shows that the bedside lamp in the bedroom was turned on at 1:50 PM, and the television was not turned on. The door sensor shows that the bedroom door remained closed.

[0138] The local inference engine receives the aforementioned real-time events and performs local care inference: real-time events are matched against the graph baseline to generate an interpretation of the care status. Specifically, the following steps are executed.

[0139] The event analysis module extracted the following information: Time = 14:00 (afternoon), Location = Bedroom, Type = Long-term Occupancy + Low Mobility, Duration = 60 minutes. This includes the event analysis module's time, location, type, and duration.

[0140] The graph query module retrieved the following from the family context graph: (1) the baseline of daily routines during the afternoon (13:00-15:00) – the “midday nap routine” node in the graph showed that the typical nap time was 13:00-14:00, lasting 40-70 minutes, in the bedroom, with a confidence level of 0.55 (not confirmed by the caregiver, only passively observed); (2) other baselines during the same time period – there was no positive record of “long-term bed rest in the afternoon” in the graph, but there was a high-confidence baseline of “usual living room activities from 14:00-16:00 in the afternoon” (confirmed by the caregiver, with a confidence level of 0.92). That is, the graph query module found “usual living room activities in the afternoon, with the bedroom baseline being short rest <30 people” from the family context graph.

[0141] The deviation calculation module calculates that the current event matches the "midday nap pattern" to a moderate degree (time window matches, but naps usually end before 2 PM, and it's already 2 PM), and deviates significantly from the "afternoon living room activity" baseline (expected in the living room, actually in the bedroom). In other words, the deviation calculation module calculates the degree of deviation between the current time and the limit (60 minutes vs. 30 minutes, deviation 100%). Based on confidence assessment and metadata: Since the baseline confidence of "afternoon living room activities" is 0.92 and has been confirmed by the caregiver, while the confidence of "midday nap routine" is low (0.55) and has not been confirmed, the system determines the current event as "abnormal behavior".

[0142] The generated care status explanation is: "Staying in the bedroom for more than 60 minutes at 2:00 PM with little activity is inconsistent with the usual afternoon living room pattern. The current time is an atypical nap time, and it is recommended to pay attention to the elderly person's condition" or "Staying in the bedroom for 60 minutes in the afternoon is abnormal compared to the usual afternoon living room pattern, and it is recommended to pay attention." This explanation is output in the form of natural language text and does not contain any raw sensor data or map node identifiers.

[0143] The system pushes the care status interpretation to the caregiver, who can then decide whether to contact the elderly person further or view the live feed (if the home has an optional camera, but this is not a mandatory requirement in this embodiment of the invention).

[0144] (5) Privacy protection mechanism See Figure 2 and Figure 8 , Figure 8 This is a flowchart illustrating the process of generating bounded inference context packages provided by this invention: Throughout the entire operation, privacy-preserving results are output (only care status explanations or aggregate summaries are provided; the original graph data cannot be exported). That is, the privacy control module enforces the following rules.

[0145] First, the home context graph (nodes, edges, confidence levels, source metadata, notes, historical versions, etc.) is stored only in an encrypted SQLite database on the home gateway (local home gateway / server). The database file is encrypted using Advanced Encryption Standard (AES-256) with a 256-bit key; that is, the database is encrypted, and the key is derived from the home hardware security module and does not leave the local device. Any attempt to send the raw graph data over the network to an external Internet Protocol (IP) address is explicitly prohibited by firewall rules.

[0146] Secondly, external output is limited to two forms: (1) plain text explaining the care status (as shown in the example above), which has been previously tagged and does not contain the elderly person's real name, specific address, or unique sensor identifier; (2) aggregated summary, such as "Today's overall activity is normal, morning routine is in line with baseline, no major abnormalities", which does not contain any specific values ​​for time or location. That is, output is allowed to be: aggregated summary (today's overall activity is normal) and / or care explanation (text).

[0147] For situations requiring more information to be provided to the emergency response agent, the privacy control module generates a "bounded inference context packet." This bounded inference context packet is generated through the following steps: Extracting a subgraph directly related to the current event from the graph, such as containing only spatial nodes (e.g., bedroom), corresponding pattern nodes (e.g., "midday nap pattern") within 30 minutes before and after the event, and the edges between them (nodes and edges involved within the 30 minutes before and after). Then, the bounded inference context packet generator removes all personally identifiable attributes through a privacy filter (e.g., replacing "bedroom" with "region A," and replacing specific time offsets with "medium offset"). This security context packet is allowed to be sent to authorized external systems, but the packet size is strictly limited to 2 kilobytes (KB) and does not contain the original confidence score (only discretized levels, such as "high / medium / low"). For example, the security context packet only contains the "bedroom → bathroom → kitchen" topology and "morning pattern," etc.

[0148] All outgoing operations are logged in an audit log, including the output time, output type (interpretation / summary / bounded packet), recipient identifier, and data size. Caregivers can review the audit log at any time to confirm that no raw data has been leaked.

[0149] (6) Manual correction and version management This embodiment also supports caregivers manually editing the atlas through the interactive interface. For example, if a caregiver finds that the system fails to automatically recognize the pattern that "the elderly do stretching exercises on the balcony every morning from 10:00 to 10:30," they can directly add a new pattern node through the "Manual Correction" entry in the interactive interface, specifying the time window, location (balcony), and activity description, and assigning a confidence level of 1.0 (caregiver forced confirmation). After accepting this manual input, the system adds it to the atlas as a high-confidence baseline and labels the source as "caregiver manually added."

[0150] If a caregiver discovers that an automatically generated pattern node is incorrect (for example, the system misidentifies "refrigerator door opening at 2:00 AM" as "nighttime eating pattern" when it is actually the elderly person getting up at night to drink water), the caregiver can choose to delete the node or mark it as "disabled". Deleted nodes are moved to the "Deleted Nodes" history table and can be restored within 30 days.

[0151] The graph update engine maintains a complete version history for each node and edge: each modification (including confidence changes, attribute adjustments, annotation additions or deletions) generates a new version number and records the modification time, operator (automatic system or caregiver identifier), old value, and new value. Caregivers can roll back to any historical version through the interface, for example, immediately restoring a regular node after accidental deletion.

[0152] (7) System module interaction See Figure 9 , Figure 9 This is a schematic diagram illustrating the interaction of various modules within the family context graph construction system based on user care provided by the present invention: the modules in the above embodiment work collaboratively. The environment perception interface continuously receives multimodal environment perception data streams and sends them to the graph initialization module (only during the startup phase) and the local inference engine (in real time). After constructing the initial family context graph, the graph initialization module stores it in the family context graph storage in the storage layer. The hypothesis generation module periodically scans the differences between the stored graph and the real-time data stream. When a candidate drift is detected, a hypothesis is generated and sent to the caregiver through the caregiver interaction module. The caregiver's feedback information is passed to the graph update engine via the caregiver interaction module, and the graph update engine modifies the corresponding nodes in the graph storage. When each real-time event arrives, the local inference engine retrieves relevant baselines from the family context graph storage, calculates the deviation, generates an explanation, and finally outputs a safe result through the privacy control module. The privacy control module and the caregiver interaction module are located in the interaction layer. The entire closed-loop system runs locally in the home and can complete continuous graph optimization and care inference without connecting to the cloud.

[0153] Furthermore, those skilled in the art should understand that the above embodiments only use WiFi BFI, radar, smart sockets, and door sensors as sensing data sources, but the present invention is not limited to these. In other embodiments, one or more of these sensors can be used alone, or they can be replaced with other environmental sensors (such as millimeter-wave radar, thermal imaging sensors, light sensors, and humidity sensors). The active learning interface is not limited to mobile applications, but can also be a voice assistant interaction (e.g., asking "I've noticed I've been getting up later in the mornings lately, is this normal?" through a smart speaker and receiving a voice answer). The map storage location is not limited to a home gateway, but can also be Network Attached Storage (NAS) or edge computing devices such as Raspberry Pi.

[0154] The embodiments of the present invention have the following advantages: First, passive startup and zero-configuration deployment. The system requires no manual room allocation, device location settings, or pattern definition. It can automatically infer the home topology, daily routines, and device locations simply by accumulating environmental perception data streams, significantly reducing the deployment threshold for smart care systems.

[0155] Second, caregiver-participatory active learning optimization. By presenting caregivers with specific behavioral hypotheses rather than requiring them to label activity categories, the system can acquire domain knowledge feedback through natural interaction. Confirmed hypotheses are upgraded to high-confidence baselines, while rejected hypotheses are pruned or downweighted, enabling the map to continuously adapt to dynamic changes within the family, without requiring caregivers to possess specialized skills in sensor configuration or data labeling.

[0156] Third, differentiated anomaly detection and interpretable inference. The local inference engine compares real-time events with family-specific baselines in the graph, distinguishing between normal individual behavioral changes and genuine care-related anomalies. The output care status interpretation is based on specific family contexts (e.g., "delayed morning bathroom visit") rather than generic anomaly scores, making it easier for caregivers to understand the reasons for alerts.

[0157] Fourth, strict local privacy protection. The family context graph never leaves the family's local storage, only outputting aggregated summaries or explanatory text. Bounded inference context packages undergo privacy filtering before export, effectively preventing the leakage of specific family behavioral patterns to the cloud or third parties.

[0158] Fifth, confidence management and robustness. Through confidence scoring, source metadata, time decay rules, and contradiction detection mechanisms, the system can automatically handle outdated information, reduce noise interference, and maintain the reliability and self-cleaning ability of the map during long-term operation. Caregiver feedback, as a high-weighted input, significantly improves the accuracy of key baselines.

[0159] Sixth, it supports manual correction and version management. Caregivers can either passively respond to hypotheses or actively edit the graph. The system retains a complete version history and audit logs, providing flexible human-machine collaborative governance capabilities.

[0160] In summary, this invention realizes a closed-loop family context graph construction and reasoning method that is passively initiated from environmental perception and actively optimized by caregiver interaction. It is superior to existing technologies in terms of reducing deployment costs, improving interpretability, and protecting privacy, and is particularly suitable for home-based elderly care scenarios.

[0161] The following describes the family context graph construction system based on user care provided by the present invention. The family context graph construction system based on user care described below can be referred to in correspondence with the family context graph construction method based on user care described above.

[0162] Figure 10 This is a schematic diagram of the structure of the family context graph construction system based on user care provided by the present invention, such as... Figure 10 As shown, the system includes: An environmental data acquisition module is configured to acquire an environmental sensing data stream of a target residence, the environmental sensing data stream including at least one of wireless beamforming feedback sequence, radar occupancy sequence, home appliance usage information and environmental sensor readings; The graph construction module is configured to determine, based on the environmental perception data stream, the topological relationships between rooms in the target residence, the rooms where each device is located in the target residence, and at least one living pattern of the first user in the target residence, wherein the device is a sensor that provides the environmental perception data stream or a home appliance sensed by the sensor; and to construct a family context graph for caring for the first user based on the topological relationships between rooms, the rooms where each device is located, and at least one living pattern of the first user.

[0163] The environmental data acquisition module serves as the environmental perception interface. The map construction module also functions as the map initialization module.

[0164] The family context graph construction system based on user care provided by this invention can construct a structured context graph containing room topology and daily routines based on existing passive observation activities without manual configuration or predefined rules. It can also adapt to the natural drift of family routines and reduces deployment costs compared to context modeling methods that require manual configuration or predefined rules.

[0165] In addition, the family context graph construction system based on user care also includes: The hypothesis generation module is used to generate behavioral hypotheses based on cumulative observations and present them to caregivers. The caregiver interaction module is used to receive feedback from caregivers regarding their confirmation or rejection of behavioral assumptions, and provides a manual editing interface; The graph update engine is used to update the graph based on feedback, confidence scores, and time decay rules, including confidence adjustment, node pruning, time decay, and contradiction detection. A local inference engine is used to combine real-time sensor events with the graph to generate an interpretation of care status; The privacy management module is used to enforce the policy that graph data cannot be exported, as well as to filter bounded inference context packets.

[0166] The map initialization module includes: a spatial topology inference unit, which infers the spatial region connectivity based on signal co-occurrence and temporal correlation; a device location inference unit, which determines the device location based on the overlap analysis between the device's active period and the space occupancy period; and a pattern recognition unit, which identifies recurring time-location-activity patterns based on a time clustering algorithm.

[0167] The hypothesis generation module includes: a monitoring unit for continuously comparing real-time sensing data with the map baseline; a change detection unit for identifying time shifts, duration changes, the appearance of new patterns, or the disappearance of patterns; and a natural language generation unit for converting the detected changes into behavioral hypothesis text.

[0168] The graph update engine includes: a confidence adjustment unit for increasing or decreasing node confidence based on feedback; a pruning unit for deleting rejected or expired nodes and edges; a time decay unit for periodically decreasing the confidence of unsupported graph elements; and a conflict detection unit for detecting conflicts between new observations and high-confidence baselines and triggering hypothesis generation.

[0169] The privacy management module includes: an encrypted storage unit for storing the graph in a local encrypted database; an export firewall unit for blocking the network export of the original graph data; a bounded packet generation unit for extracting relevant subgraphs and performing privacy filtering; and an audit log unit for recording all outgoing operations.

[0170] Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 11As shown, the electronic device may include: a processor 1110, a communications interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communications interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. The processor 1110 can call logical instructions in the memory 1130 to execute a family context graph construction method based on user care. The method includes: acquiring an environmental perception data stream of a target residence, the environmental perception data stream including at least one of a wireless beamforming feedback sequence, a radar occupancy sequence, appliance usage information, and environmental sensor readings; based on the environmental perception data stream, determining the topological relationships between rooms in the target residence, the rooms where each device is located in the target residence, and at least one living pattern of a first user in the target residence, wherein the device is a sensor providing the environmental perception data stream or an appliance sensed by the sensor; and constructing a family context graph for caring for the first user based on the topological relationships between rooms, the rooms where each device is located, and at least one living pattern of the first user.

[0171] Furthermore, the logical instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0172] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the family context graph construction method based on user care provided by the above methods. The method includes: acquiring an environmental perception data stream of a target residence, the environmental perception data stream including at least one of a wireless beamforming feedback sequence, a radar occupancy sequence, appliance usage information, and environmental sensor readings; determining, based on the environmental perception data stream, the topological relationships between rooms in the target residence, the rooms where each device is located in the target residence, and at least one living pattern of a first user in the target residence, wherein the device is a sensor that provides the environmental perception data stream or an appliance sensed by the sensor; and constructing a family context graph for caring for the first user based on the topological relationships between rooms, the rooms where each device is located, and at least one living pattern of the first user.

[0173] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for constructing a family context graph based on user care provided by the methods described above. This method includes: acquiring an environmental sensing data stream of a target residence, the environmental sensing data stream including at least one of a wireless beamforming feedback sequence, a radar occupancy sequence, appliance usage information, and environmental sensor readings; determining, based on the environmental sensing data stream, the topological relationships between rooms in the target residence, the rooms where each device is located in the target residence, and at least one living pattern of a first user in the target residence, wherein the device is a sensor providing the environmental sensing data stream or an appliance sensed by the sensor; and constructing a family context graph for caring for the first user based on the topological relationships between rooms, the rooms where each device is located, and at least one living pattern of the first user.

[0174] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0175] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a family context graph based on user care, characterized in that, include: Acquire environmental sensing data stream of the target residence, wherein the environmental sensing data stream includes at least one of wireless beamforming feedback sequence, radar occupancy sequence, home appliance usage information and environmental sensor readings; Based on the environmental perception data stream, the topological relationship between rooms in the target residence, the room where each device in the target residence is located, and at least one living pattern of the first user in the target residence are determined. The device is a sensor that provides the environmental perception data stream or a home appliance sensed by the sensor. Based on the topological relationships between the rooms, the rooms where the devices are located, and at least one of the living patterns of the first user, a family context graph for caring for the first user is constructed.

2. The method for constructing a family context graph based on user care according to claim 1, characterized in that, The step of determining the topological relationships between rooms in the target residence, the rooms where each device in the target residence is located, and at least one living pattern of the first user in the target residence based on the environmentally perceived data stream includes: Based on the spatial co-occurrence and temporal correlation of the environmental perception data stream, the topological relationship between the rooms is determined; For each device, the room where the device is located is determined by overlapping analysis of the active time period of the device and the occupancy time period of each room in the environmental perception data stream. At least one repetitive data group is extracted from the environmental perception data stream based on the time clustering algorithm. For each repetitive data group, the corresponding life pattern is identified. The repetitive data group contains multiple data of the same device being active in the same location within the same time period.

3. The method for constructing a family context graph based on user care according to claim 1, characterized in that, The method further includes: Set confidence levels for each node and edge in the family context graph, and / or add source metadata for each node and edge in the family context graph, wherein the source metadata includes at least one of observation source, first observation timestamp, latest observation timestamp, and number of observations.

4. The method for constructing a family context graph based on user care according to claim 1, characterized in that, After constructing a family context graph for caring for the first user based on the topological relationships between the rooms, the rooms where the devices are located, and at least one of the first user's daily routines, the process further includes: Based on the current environmental perception data stream, determine the current activity data of the first user; Based on the family context graph, the current activity data is detected to obtain detection data; When the detection data indicates a change in the first user's daily routine, or when the detection data indicates a conflict between the current activity data and the edge and / or node corresponding to the current activity data in the family context graph, a behavioral description text corresponding to the first user is generated. The behavioral description text includes at least one of the following: time offset of the daily routine, change in duration, disappearance of the daily routine, decrease in frequency of the daily routine, and newly emerging daily routine. The behavioral description text is fed back to the second user to inform the second user.

5. The method for constructing a family context graph based on user care according to claim 4, characterized in that, After feeding back the behavior description text to the second user to inform the second user, the method further includes: Receive feedback from the second user regarding the behavior description text; If the feedback information confirms the assessment, increase the confidence of the nodes and / or edges corresponding to the behavioral description text in the family context graph. If the feedback information is rejected, reduce the confidence of the nodes and / or edges corresponding to the behavioral description text in the family context graph, or delete the nodes and / or edges corresponding to the behavioral description text in the family context graph.

6. The method for constructing a family context graph based on user care according to claim 5, characterized in that, After receiving the feedback information from the second user regarding the behavior description text, the method further includes: If the feedback information includes the remarks input by the second user, the remarks are appended as additional information to the node and / or edge corresponding to the behavior description text. And / or, update the source metadata of the nodes and / or edges corresponding to the behavior description text based on the feedback information; And / or, record the feedback source and feedback time of the feedback information, as well as the operation type for updating the family context graph.

7. The method for constructing a family context graph based on user care according to any one of claims 1-6, characterized in that, After constructing a family context graph for caring for the first user based on the topological relationships between the rooms, the rooms where the devices are located, and at least one of the first user's daily routines, the process further includes: Based on the current environmental awareness data stream, user event data is determined, which includes at least one of the following: event occurrence time, event location, event-associated device type, and event duration. Based on the life patterns corresponding to the user event data in the family context graph and the user event data, determine the degree of deviation corresponding to the user event data; Based on the degree of deviation and the relevant data of the life patterns corresponding to the user event data, an explanation of the care status of the first user is generated. The relevant data includes at least one of confidence level, latest update time, source metadata and additional information. The explanation of the care status includes at least one of deviation description text, deviation reason prompt and suggested operation information. The care status is explained and fed back to the second user to inform the second user.

8. The method for constructing a family context graph based on user care according to claim 7, characterized in that, The method further includes: Generate a bounded inference context package, wherein the bounded inference context package contains a subgraph corresponding to the user event data in the family context graph; The bounded inference context packet is anonymized and then sent to the authorization system.

9. The method for constructing a family context graph based on user care according to any one of claims 1-6, characterized in that, The method further includes: Received data export instruction; When the data export instruction indicates that specified data should be output, the specified data is output. The specified data is a care status interpretation and / or comprehensive care data, and the comprehensive care data represents the activity of the first user within a set time period. If the data export instruction indicates that non-specified data should be output, the output of the non-specified data should be rejected. The non-specified data is data other than the specified data. And / or, the method further includes: For a specified element in the family context graph, the confidence of the specified element is reduced according to a preset decay rate. The specified element is a node or edge that has not been confirmed by relevant information within a set time period. The relevant information is newly acquired environmental perception data stream and / or feedback information. If the confidence level of the specified element is lower than the lower confidence limit, the specified element is deleted from the family context graph. And / or, the method further includes: Perform the following steps for each graph element in the family context graph: Record the historical versions of the aforementioned atlas elements; If a rollback operation is detected for the graph element, the graph element in the family context graph is rolled back to the historical version specified by the rollback operation.

10. A family context graph construction system based on user care, characterized in that, include: An environmental data acquisition module is configured to acquire an environmental sensing data stream of a target residence, the environmental sensing data stream including at least one of wireless beamforming feedback sequence, radar occupancy sequence, home appliance usage information and environmental sensor readings; The graph construction module is configured to determine, based on the environmental perception data stream, the topological relationships between rooms in the target residence, the rooms where each device is located in the target residence, and at least one living pattern of the first user in the target residence, wherein the device is a sensor that provides the environmental perception data stream or a home appliance sensed by the sensor; and to construct a family context graph for caring for the first user based on the topological relationships between rooms, the rooms where each device is located, and at least one living pattern of the first user.

Images