Social graph construction and abnormal behavior early warning method for old people

Abstract

The invention relates to the technical field of information processing, and particularly discloses a social graph construction and abnormal behavior early warning method for old people, and the method comprises the steps: S1, continuously collecting the position data of the old people; obtaining a pre-constructed place semantic knowledge graph; for every two old people, whether a position co-occurrence relation exists between the two old people is judged, and when the position co-occurrence relation exists, the position co-occurrence intensity is calculated; calculating a social interaction strength score between the old people according to the position co-occurrence strength and the place social function weight; according to the social interaction strength score and the place function information, constructing an old people community social graph; according to the social graph construction method for the old people, the social interaction information can be effectively mined from the position data, and then the social graph of the old people community is constructed, so that the real social interaction between the old people can be more accurately identified and quantified, and data support is provided for the old people institution to optimize the nursing service and activity arrangement.

Description

Technical Field

[0001] This application relates to the field of information processing technologies, and more particularly, to a method for constructing a social graph and warning of abnormal behaviors for the elderly. Background Art

[0002] Currently, in elderly care institutions, in order to ensure the safety of the elderly with limited mobility, indoor positioning systems based on low-power wearable tags are widely deployed. The main function of these systems is to record the location information of the elderly for quick positioning in case of emergencies. However, with the increasing demand for refined elderly care services, institutions need to understand the social interaction patterns of the elderly more deeply to optimize care services and activity arrangements, especially for predicting whether the elderly are at risk of social isolation to prevent them from making abnormal behaviors. The method of directly observing or manually recording social interactions has the disadvantages of time-consuming and low efficiency, and may also violate the privacy of the elderly. Using knowledge graph technology to mine social information from positioning data has become a potentially feasible solution.

[0003] Although knowledge graph technology has the ability to enhance semantics, inferring social interactions directly from basic positioning data still faces many challenges. Basic positioning data can only reflect the proximity of physical locations, and location proximity does not completely equal real social interactions. For example, the coincidence of locations in corridors or non-social places may not represent the occurrence of social behaviors. In addition, elderly care institutions usually lack professional technical personnel and have extremely high requirements for data privacy protection. How to economically and effectively mine information from existing low-power positioning data that can accurately reflect the real social interaction situation of the elderly and construct a community social graph without increasing additional hardware investment and without violating the privacy of the elderly has become a technical problem that needs to be solved urgently by current elderly care institutions.

[0004] For the above problems, there is currently no effective technical solution. Summary of the Invention

[0005] The purpose of this application is to provide a method for constructing a social graph and warning of abnormal behaviors for the elderly, so as to more deeply mine the social interaction relationships between the elderly based on their location data, and then construct a more accurate social graph.

[0006] In a first aspect, this application provides a method for constructing a social graph for the elderly, which is used to analyze the social relationships of the elderly population. The method includes the following steps: S1. Continuously collect the location data of the elderly; S2. Obtain a pre-constructed semantic knowledge graph of places, where the graph nodes include place area codes, and each place area code is preset with a place social function weight, a place type adjustment coefficient, and place function information; S3. For every two elderly people, analyze the co-occurrence frequency and duration of the same venue area codes corresponding to their location data in multiple consecutive time periods to determine whether there is a location co-occurrence relationship between them. When there is a location co-occurrence relationship, calculate the location co-occurrence intensity according to the co-occurrence frequency, the duration, and the venue type adjustment coefficient; S4. Calculate the social interaction intensity score between the elderly according to the location co-occurrence intensity and the venue social function weight; S5. Construct a social graph of the elderly community according to the social interaction intensity score and the venue function information.

[0007] The method for constructing a social graph of the elderly in this application uses a venue semantic knowledge graph to semantically enhance the location data of the elderly, overcomes the limitation that simply being close in location does not equal social interaction, more accurately mines the real social interaction information between the elderly, and constructs a social graph of the elderly community for analyzing the social relationships of the elderly population. Among them, this method uses the continuously collected location data of the elderly as the data basis, combines with a pre-constructed venue semantic knowledge graph that can endow the location data with venue semantic information, analyzes the location co-occurrence relationship between different elderly people, and calculates and obtains the social interaction intensity score, which can effectively mine the social interaction information from the location data of the elderly, and then combines with the venue function information to construct a social graph of the elderly community that can be used to analyze the social relationships of the elderly population, so as to more accurately identify and quantify the real social interaction between the elderly, providing data support for optimizing the care services and activity arrangements of nursing homes.

[0008] The described method for constructing a social graph of the elderly, wherein the step of analyzing the co-occurrence frequency and duration of the same venue area codes corresponding to the location data of every two elderly people in multiple consecutive time periods to determine whether there is a location co-occurrence relationship between them includes: S31. For every two elderly people, count the co-occurrence frequency and duration of the same venue area codes corresponding to their location data in multiple consecutive time periods, and preliminarily determine whether there is a location co-occurrence relationship between them according to the co-occurrence frequency and the duration; S32. When it is preliminarily determined that there is a location co-occurrence relationship between them, count the effective co-occurrence frequency and effective co-occurrence duration of the distance between the two elderly people being less than the interaction distance threshold in these multiple consecutive time periods according to the location data; S33. Finally determine whether there is a location co-occurrence relationship between them according to the effective co-occurrence frequency and the effective co-occurrence duration.

[0009] The above processing method can exclude the situation where the venue area codes are the same but the actual distance is far by introducing the interaction distance threshold, the effective co-occurrence frequency, and the effective co-occurrence duration, more accurately determine whether there is a location co-occurrence relationship between the elderly, and effectively avoid the problem of inaccurate judgment caused by simply judging based on the same venue area code. Because just having the same venue area code does not mean that there is actual social interaction between two elderly people. When judging whether there is a location co-occurrence relationship between the elderly, the above processing method not only considers the co-occurrence frequency and duration of the venue area codes, but also considers the actual distance between the elderly, making the judgment of the location co-occurrence relationship more accurate and more in line with the actual social interaction situation of the elderly. Among them, step S31 plays a role in preliminary screening, step S32 makes an accurate judgment through the interaction distance threshold, and step S33 gives the final judgment result. These three steps work together to improve the accuracy of the determination of the location co-occurrence relationship, making the subsequent constructed social graph more accurate and reliable.

[0010] The described method for constructing a social graph of the elderly, wherein the process of determining the interaction distance threshold includes: A1. According to the venue function information corresponding to the venue area code, obtain the interaction distance threshold from the mapping relationship table between the venue function information and the distance threshold.

[0011] In this way, the interaction distance threshold can be adaptively adjusted in different venues, so as to more accurately determine whether there is a location co-occurrence relationship between the elderly. Thus, the accuracy of the determination of the location co-occurrence relationship is improved, and the accuracy of constructing the social graph of the elderly community can be effectively enhanced.

[0012] The described method for constructing a social graph of the elderly, wherein each venue area code is also preset with venue area information; the process of determining the interaction distance threshold further includes: A2. Adjust the interaction distance threshold according to the venue area information corresponding to the venue area code.

[0013] The described method for constructing a social graph of the elderly, wherein the step of calculating the location co-occurrence intensity according to the co-occurrence frequency, the duration, and the venue type adjustment coefficient includes: S34. According to the duration, split and obtain the sub-co-occurrence duration within each time period for every two elderly people in multiple consecutive time periods; S35. According to the venue type adjustment coefficient, set a time decay factor sequence for time decay for each time period; S36. Based on the time decay factor sequence, weight the sub-co-occurrence duration within each time period to obtain the weighted co-occurrence duration; S37. Multiply the weighted co-occurrence duration by the co-occurrence frequency to obtain the location co-occurrence intensity.

[0014] The described method for constructing a social graph of the elderly, wherein step S4 includes: S41. Obtain the age score and physical ability score of the elderly; S42. Based on the location co-occurrence relationship, adjust the venue social function weight according to the age score and physical ability score; S43. Calculate the social interaction intensity score between the elderly according to the adjusted venue social function weight and the location co-occurrence intensity.

[0015] The described method for constructing a social graph of the elderly, wherein step S5 includes: S51. According to the social interaction intensity score, construct an initial social graph of the elderly community, where the graph nodes represent the elderly, and the weight of the edges is initially set to the social interaction intensity score; S52. Based on the guidance of the venue area coding, match the social interaction types corresponding to the venue function information for the edges of the initial social graph of the elderly community to form the social graph of the elderly community.

[0016] In the second aspect, the present application also provides a method for warning of abnormal behaviors of the elderly, and the method includes the following steps: B1. Analyze the social graph of the elderly community obtained by the method for constructing a social graph of the elderly provided in the first aspect based on the community discovery algorithm to obtain community structure information; B2. Obtain community parameter information according to the community structure information; B3. For the positions in the community structure information where the elderly are located, dynamically evaluate the social isolation risk level and its change trend of each elderly person in combination with the community parameter information; B4. Predict whether the social behaviors of each elderly person are abnormal according to the social isolation risk level and its change trend.

[0017] The method for warning of abnormal behaviors of the elderly in the present application analyzes the social graph of the elderly community obtained by the method for constructing a social graph of the elderly provided in the first aspect to identify the community structure information in the elderly population and extract the community parameter information, so that for each elderly person at the node in the community structure, and in combination with the community parameter information, dynamically evaluate the social isolation risk level and its change trend of each elderly person to determine whether there is a social isolation risk, thereby predicting whether the future social behaviors of the elderly are abnormal, realizing the warning of abnormal behaviors of the elderly, providing technical support for pension institutions, and helping to improve the quality of pension services.

[0018] The described method for warning of abnormal behaviors of the elderly, wherein step B1 includes: B11. Use the Louvain algorithm optimized by modularity to perform community detection on the social graph of the elderly community. Among them, the resolution parameter of the Louvain algorithm is adaptively optimized in the following way: B111. Construct a candidate parameter set for the resolution parameter. Using the grid search method, select different resolution parameter values in the candidate parameter set, and run the Louvain algorithm to perform community partitioning on the social graph of the elderly community; B112. For each community partitioning result, calculate the silhouette coefficient of the community partitioning, and select the resolution parameter corresponding to the maximum silhouette coefficient as the optimal resolution parameter; B113. Based on the community partitioning result under the optimal resolution parameter, extract the community structure information.

[0019] Through the above processing method, the resolution parameter of the Louvain algorithm can be adaptively adjusted to the optimal state, improving the accuracy and robustness of community detection.

[0020] In the abnormal behavior warning method for the elderly described above, the community parameter information includes community scale, community connection strength, and node centrality index. Among them, the community scale is characterized by the number of nodes in the community, the community connection strength is characterized by the average weight of the connection edges between the nodes in the community, and the node centrality index is characterized by the degree centrality index.

[0021] As can be seen from the above, the present application provides a method for constructing a social graph and warning of abnormal behavior for the elderly. Among them, the method for constructing a social graph of the elderly uses the place semantic knowledge graph to semantically enhance the location data of the elderly, overcomes the limitation that simple location proximity is not equivalent to social interaction, more accurately mines the real social interaction information between the elderly, and constructs a community social graph for analyzing the social relationships of the elderly population. Among them, this method uses the continuously collected location data of the elderly as the data basis, combines the pre-constructed place semantic knowledge graph that can endow the location data with place semantic information, analyzes the location co-occurrence relationship between different elderly people, and calculates and obtains the social interaction intensity score, which can effectively mine the social interaction information from the location data of the elderly, and then combines the place function information to construct an elderly community social graph that can be used to analyze the social relationships of the elderly population, so as to more accurately identify and quantify the real social interaction between the elderly, providing data support for optimizing the care services and activity arrangements of nursing homes. Brief Description of the Drawings

[0022] Figure 1 It is a flowchart of a method for constructing a social graph of the elderly provided by an embodiment of the present application.

[0023] Figure 2Flowchart of an abnormal behavior warning method for the elderly provided by an embodiment of the present application. Detailed implementation manners

[0024] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Components of the embodiments of the present application described and illustrated herein can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application to be protected, but merely represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative efforts fall within the scope of protection of the present application.

[0025] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present application, terms such as "first" and "second" are only used for differential description and cannot be understood as indicating or implying relative importance.

[0026] In a first aspect, please refer to Figure 1 , some embodiments of the present application provide a method for constructing a social graph of the elderly, which is used to analyze the social relationships of the elderly population. The method includes the following steps: S1. Continuously collect the location data of the elderly; S2. Obtain a pre-constructed semantic knowledge graph of places, where the graph nodes include place area codes, and each place area code is preset with a place social function weight, a place type adjustment coefficient, and place function information; S3. For every two elderly people, analyze the co-occurrence frequency and duration of the same place area codes corresponding to their location data in multiple consecutive time periods to determine whether there is a location co-occurrence relationship between them. When there is a location co-occurrence relationship, calculate the location co-occurrence intensity according to the co-occurrence frequency, duration, and place type adjustment coefficient; S4. Calculate the social interaction intensity score between the elderly according to the location co-occurrence intensity and the place social function weight; S5. Construct a social graph of the elderly community according to the social interaction intensity score and the place function information.

[0027] Specifically, in step S1, the location data can be continuously collected through low-power wearable tags and a positioning system, or through a wearable device with a positioning function carried by the elderly.

[0028] More specifically, in step S2, the location semantic knowledge graph is pre-constructed. The location area codes in the graph can be, for example, the room numbers or area identifiers of different areas within a pension institution.

[0029] More specifically, in step S2, the location semantic knowledge graph is pre-constructed. The location semantic knowledge graph includes multiple graph nodes and edges connecting the graph nodes, and it can be constructed by existing knowledge graph generation algorithms through collecting relevant location data; the location area codes in the graph can be, for example, the room numbers or area identifiers of different areas within a pension institution. The location social function weight can be preset according to the social attributes of the location. For example, the social function weight of a restaurant is higher than that of a corridor. The location type adjustment coefficient can be set according to the type of the location. For example, for non-social locations, the location type adjustment coefficient can be set to a lower value to reduce the impact of location co-occurrence in these locations on the strength of social relationships. The location function information can include a description of the use of the location, such as "restaurant", "activity room", etc.

[0030] More specifically, step S3 is used to determine the location co-occurrence relationship and calculate the location co-occurrence strength. Specifically, for every two elderly people, within multiple consecutive time periods, such as time periods in units of 10 minutes, half an hour, 1 hour, or a day, the co-occurrence frequency and duration of the same location area codes corresponding to their location data are counted. The co-occurrence frequency refers to the number of time periods in which the location area codes of the two elderly people are the same within these multiple consecutive time periods, and the duration can refer to the total length of time in which the location area codes of the two elderly people are the same within the statistical period. The determination of the location co-occurrence relationship is based on setting thresholds for the co-occurrence frequency and duration. When both meet the preset conditions, it is determined that there is a location co-occurrence relationship. For example, when the co-occurrence frequency exceeds a certain number of times and the duration exceeds a certain length, it is determined that there is a location co-occurrence relationship between the two. The calculation of the location co-occurrence strength takes into account the co-occurrence frequency, duration, and location type adjustment coefficient. The location type adjustment coefficient can be set according to the actual social function of the location. For example, for locations with a weaker social function, the location type adjustment coefficient can be set to a lower value, and vice versa to a higher value. The calculation formula for this location co-occurrence strength can be simply recorded as: Location co-occurrence strength = Co-occurrence frequency * Duration * Location type adjustment coefficient.

[0031] More specifically, step S4 is used to calculate the social interaction strength score, and its calculation process can be simply recorded as: Social interaction strength score = Location co-occurrence strength * Location social function weight. The location social function weight can have different preset values in different locations to reflect the differences in the social attributes of different locations.

[0032] More specifically, step S5 is used to construct a social graph of the elderly community. Among them, the graph nodes represent the elderly, and the edges can represent the social relationships or social types among the elderly, which can be determined by the venue function information, or the attributes of the edges can be directly configured as the venue function information. The weight of the edges can be the social interaction intensity score. Thus, the social graph of the elderly community is constructed for analyzing the social relationships of the elderly population.

[0033] The method for constructing the social graph of the elderly in the embodiments of the present application uses the venue semantic knowledge graph to semantically enhance the location data of the elderly, overcomes the limitation that simply being in close proximity does not necessarily mean social interaction, more accurately mines the real social interaction information among the elderly, and constructs a community social graph for analyzing the social relationships of the elderly population. Among them, the method uses the continuously collected location data of the elderly as the data basis, combines the pre-constructed venue semantic knowledge graph that can endow the location data with venue semantic information, analyzes the location co-occurrence relationship between different elderly people, and calculates and obtains the social interaction intensity score, which can effectively mine the social interaction information from the location data of the elderly. Furthermore, combined with the venue function information, it constructs a social graph of the elderly community that can be used to analyze the social relationships of the elderly population, so as to more accurately identify and quantify the real social interaction among the elderly, providing data support for optimizing the care services and activity arrangements in nursing homes.

[0034] In some preferred embodiments, for every two elderly people, the steps of analyzing the co-occurrence frequency and duration of the same venue area codes corresponding to the location data in multiple consecutive time periods to determine whether there is a location co-occurrence relationship between the two include: S31. For every two elderly people, count the co-occurrence frequency and duration of the same venue area codes corresponding to the location data in multiple consecutive time periods, and preliminarily determine whether there is a location co-occurrence relationship between the two based on the co-occurrence frequency and duration; S32. When it is preliminarily determined that there is a location co-occurrence relationship between the two, count the effective co-occurrence frequency and effective co-occurrence duration of the distance between the two elderly people being less than the interaction distance threshold in these multiple consecutive time periods according to the location data; S33. Finally determine whether there is a location co-occurrence relationship between the two based on the effective co-occurrence frequency and effective co-occurrence duration.

[0035] Specifically, step S31 is used to analyze the characteristics of the location data of every two elderly people in multiple consecutive time periods, mainly for counting the co-occurrence frequency and duration of the same venue area codes. The preliminary determination of whether there is a location co-occurrence relationship between the two can be completed based on a preset frequency threshold and duration threshold. For example, if the co-occurrence frequency exceeds the frequency threshold and the duration exceeds the duration threshold, it is preliminarily determined that there is a location co-occurrence relationship between the two.

[0036] More specifically, step S31 is equivalent to quickly screening out the elderly who may have a location co-occurrence relationship based on the consistency of the venue area codes. Therefore, it belongs to the preliminary judgment step. After screening out these elderly who may have a location co-occurrence relationship, steps S32 and S33 perform a detailed analysis. It calculates the effective co-occurrence frequency and effective co-occurrence duration in which the distances between two elderly people are less than the interaction distance threshold in multiple consecutive time periods, so as to accurately judge whether there is a location co-occurrence relationship between the two based on the distance relationship between the elderly. Among them, the interaction distance threshold represents the maximum distance at which social interaction may occur between the elderly; the effective co-occurrence frequency refers to the number of time periods in which the distances between two elderly people are less than the interaction distance threshold and the venue area codes are the same in multiple consecutive time periods; the effective co-occurrence duration refers to the total duration in which the distances between two elderly people are less than the interaction distance threshold and the venue area codes are the same. The interaction distance threshold can be determined according to the venue function information. The final determination in step S33 can be completed based on a preset effective frequency threshold and effective duration threshold. For example, if the effective co-occurrence frequency exceeds the effective frequency threshold and the effective co-occurrence duration exceeds the effective duration threshold, it is finally determined that there is a location co-occurrence relationship between the two. The effective frequency threshold and effective duration threshold can be the same as or different from the frequency threshold and duration threshold of the preliminary determination.

[0037] More specifically, by introducing the interaction distance threshold, effective co-occurrence frequency, and effective co-occurrence duration, the above processing method can exclude the situation where the venue area codes are the same but the actual distances are far apart, and more accurately determine whether there is a location co-occurrence relationship between the elderly, effectively avoiding the problem of inaccurate judgment caused by simply judging based on the same venue area code. Because just having the same venue area code does not mean that there is actual social interaction between two elderly people. When judging whether there is a location co-occurrence relationship between the elderly, the above processing method not only considers the co-occurrence frequency and duration of the venue area codes, but also considers the actual distance between the elderly, making the judgment of the location co-occurrence relationship more accurate and more in line with the actual social interaction situation of the elderly. Among them, step S31 plays a role of preliminary screening, step S32 makes an accurate judgment through the interaction distance threshold, and step S33 gives the final judgment result. These three steps work together to improve the accuracy of the determination of the location co-occurrence relationship, making the subsequent constructed social graph more accurate and reliable.

[0038] In some preferred embodiments, the process of determining the interaction distance threshold includes: A1. According to the venue function information corresponding to the venue area code, obtain the interaction distance threshold from the mapping relationship table between the venue function information and the distance threshold.

[0039] Specifically, the mapping table of venue function information and distance threshold is a pre-established mapping table that records the interaction distance thresholds corresponding to different venue function information. The venue function information can be extracted based on the venue area coding of the venue semantic knowledge graph. The interaction distance threshold is the maximum distance used to determine whether there is an effective co-occurrence relationship between the elderly.

[0040] More specifically, in step S32, the rationality of the interaction distance threshold affects the accuracy of the determination of the location co-occurrence relationship. The mapping table of venue function information and distance threshold stores the corresponding relationship between venue function information and interaction distance threshold. When it is necessary to determine the interaction distance threshold, first, the corresponding venue function information is obtained from the venue semantic knowledge graph according to the venue area coding. Then, the venue function information is used as an index to find the corresponding interaction distance threshold in the mapping table of venue function information and distance threshold. For example, the mapping table of venue function information and distance threshold can be set as follows: when the venue function information is "public activity area", the interaction distance threshold is set to 5 meters; when the venue function information is "rest area", the interaction distance threshold is set to 3 meters; when the venue function information is "corridor", the interaction distance threshold is set to 2 meters. In this way, the interaction distance threshold can be adaptively adjusted in different venues, so as to more accurately determine whether there is a location co-occurrence relationship between the elderly. Thus, the accuracy of the determination of the location co-occurrence relationship is improved, and the accuracy of constructing the social graph of the elderly community can be effectively enhanced.

[0041] In some preferred embodiments, each venue area coding also presets venue area information; the process of determining the interaction distance threshold further includes: A2. Adjust the interaction distance threshold according to the venue area information corresponding to the venue area coding.

[0042] Specifically, the venue area information is preset in the venue area code of the venue semantic knowledge graph. It can be the actual floor area of the venue or the area scale of the corresponding functional venue. After obtaining the initial interaction distance threshold from the mapping relationship table between venue function information and distance threshold according to the venue function information, step A2 will further adjust the initial interaction distance threshold according to the venue area information. Specifically, if the venue area is large, the interaction distance threshold can be appropriately increased; conversely, if the venue area is small, the interaction distance threshold can be appropriately decreased. The adjustment method can be achieved by presetting an area adjustment coefficient. For example, establish a correspondence table between the venue area range and the area adjustment coefficient, or use a functional relationship to calculate the area adjustment coefficient according to the venue area size, and then multiply the initial interaction distance threshold by the area adjustment coefficient to obtain the final interaction distance threshold. Step A2 adjusts the interaction distance threshold through the venue area information, so that the setting of the interaction distance threshold can better conform to the actual venue situation, and thus can more accurately determine whether there is a real social interaction relationship between the elderly.

[0043] In some preferred embodiments, the steps of calculating the position co-occurrence intensity according to the co-occurrence frequency, duration, and venue type adjustment coefficient include: S34. Split and obtain the sub-co-occurrence duration within each time period for every two elderly people in multiple consecutive time periods according to the duration; S35. Set a time decay factor sequence regarding time series decay for each time period according to the venue type adjustment coefficient; S36. Weight the sub-co-occurrence duration within each time period based on the time decay factor sequence to obtain the weighted co-occurrence duration; S37. Multiply the weighted co-occurrence duration by the co-occurrence frequency to obtain the position co-occurrence intensity.

[0044] Specifically, in step S34, the position data of every two elderly people in multiple consecutive time periods is further analyzed to obtain the duration during which the venue area codes of the two are the same within each time period, that is, the sub-co-occurrence duration.

[0045] It should be noted that in the embodiments using steps S31 - S33 for position co-occurrence relationship determination, since it has been determined that there is a position co-occurrence relationship between two elderly people, therefore, step S34 can use the complete duration that better reflects the co-occurrence of the two elderly people within the venue with the same venue area code to split and obtain the sub-co-occurrence duration for deeper analysis.

[0046] More specifically, in step S35, the setting of the time decay factor sequence is associated with the venue type adjustment coefficient. For social interactions in different venues, the time series decay characteristics may vary. Therefore, the setting of the time decay factor sequence needs to consider the venue type adjustment coefficient. The time decay factor can be a linear decay factor or an exponential decay factor, etc., which is used to indicate that the shorter the sub - co - occurrence duration is in time, the higher its weight. The introduction of the time decay factor sequence reduces the influence of early position co - occurrences and makes the scoring focus more on recent social interactions.

[0047] More specifically, in step S36, the time decay factor sequence is used to weight the sub - co - occurrence durations in each time period to obtain the weighted co - occurrence duration. The weighting process reflects that the sub - co - occurrence duration in the recent time period has a greater impact on the position co - occurrence intensity, while the sub - co - occurrence duration in the far - away time period has a smaller impact.

[0048] More specifically, in step S37, the weighted co - occurrence duration replaces the original duration, so that the calculated position co - occurrence intensity can more accurately reflect the social interaction intensity with time - series decay characteristics, thus reflecting the timeliness of social interactions and obtaining a position co - occurrence intensity that is more in line with the actual social interaction situation.

[0049] More specifically, the position co - occurrence intensity finally obtained by the above - mentioned processing method comprehensively considers the co - occurrence frequency, the time length considering time decay and the venue type adjustment coefficient, as well as the overall duration, so as to more comprehensively and finely quantify the strength of the position co - occurrence relationship between the elderly, enabling it to more accurately and meticulously reflect the social interaction intensity between the elderly based on location data, making the subsequent constructed social graph and abnormal behavior early warning analysis more reliable and effective.

[0050] In some preferred embodiments, step S4 includes: S41. Obtain the age score and physical ability score of the elderly; S42. Adjust the venue social function weight based on the position co - occurrence relationship according to the age score and physical ability score; S43. Calculate the social interaction intensity score between the elderly according to the adjusted venue social function weight and the position co - occurrence intensity.

[0051] Specifically, in step S41, the age score and physical ability score of the elderly quantify the individual characteristics of the elderly. The acquisition methods of the age score and physical ability score are not limited. For example, the pre - recorded score data can be retrieved from the personal files of the elderly, or obtained through real - time evaluation by a health assessment system.

[0052] More specifically, step S42 is used to adjust the weight of the social function of the venue based on the co-occurrence relationship of the positions among the elderly, as well as their age scores and physical ability scores. Specifically, when it is determined that there is a co-occurrence relationship between two elderly people, the method for constructing the social graph of the elderly in the embodiments of the present application will further consider the age and physical ability differences between these two elderly people. The adjustment of the weight of the social function of the venue can be achieved in various ways. For example, an adjustment model can be preset, which takes the age difference coefficient and the physical ability difference coefficient as inputs and outputs a comprehensive adjustment coefficient, and the weight of the social function of the venue is adjusted according to this comprehensive adjustment coefficient.

[0053] More specifically, in step S43, since the weight of the social function of the venue has been adjusted according to the individual characteristics of the elderly, the finally obtained social interaction intensity score can more accurately reflect the real social interaction level among the elderly.

[0054] More specifically, in order to more accurately evaluate the social interaction intensity among the elderly, the method for constructing the social graph of the elderly in the embodiments of the present application further introduces age scores and physical ability scores to adjust the weight of the social function of the venue, so that the influence of individual differences of the elderly on the social interaction intensity is taken into account to more accurately evaluate the social interaction intensity among the elderly; In some preferred embodiments, step S42 includes: S421. For each pair of elderly people with a co-occurrence relationship, calculate and obtain the age difference coefficient and the physical ability difference coefficient according to the age score and the physical ability score; S422. Calculate the comprehensive adjustment coefficient by using the weighted average method according to the age difference coefficient and the physical ability difference coefficient; S423. Adjust the weight of the social function of the venue according to the comprehensive adjustment coefficient.

[0055] Specifically, in step S421, as a way of implementation, the age difference coefficient is obtained by taking the absolute value of the age difference between two elderly people and then normalizing it. For example, the age difference coefficient can be calculated as the absolute value of the age difference divided by the maximum possible age difference. The physical ability difference coefficient can be obtained by quantifying the normalized difference in the physical ability scores of two elderly people based on a preset physical ability scoring standard. For example, the physical ability difference coefficient can be calculated as the absolute value of the difference in physical ability scores divided by the maximum possible difference in physical ability scores. The age difference coefficient and the physical ability difference coefficient calculated in step S421 can quantify the degree of difference in age and physical ability between the elderly, providing a quantitative basis for subsequent adjustment of the weight of the social function of the venue.

[0056] More specifically, in step S422, the comprehensive adjustment coefficient is the weighted average of the age difference coefficient and the physical ability difference coefficient. The weight coefficient used for the weighting process can be flexibly adjusted according to the actual application scenario. For example, in a scenario where more attention is paid to the physical ability difference, the weight of the physical ability difference coefficient can be set higher. This processing method can flexibly adjust the influence degree of age and physical ability differences on the comprehensive adjustment coefficient according to the actual situation, making the adjustment process more refined.

[0057] More specifically, step S423 uses the comprehensive adjustment coefficient to adjust the weight of the social function of the venue. It can be adjusted by multiplication or addition, preferably by multiplication. This step uses the comprehensive adjustment coefficient to adjust the weight of the social function of the venue, so that the weight of the social function of the venue can more accurately reflect the impact of the age and physical ability differences of the elderly on social interaction.

[0058] In some preferred embodiments, step S5 includes: S51. According to the social interaction intensity score, construct an initial social graph of the elderly community. The nodes of the graph represent the elderly, and the weight of the edge is initialized to the social interaction intensity score; S52. Based on the guidance of the venue area code, match the social interaction types corresponding to the venue function information for the edges of the initial social graph of the elderly community to form a social graph of the elderly community.

[0059] Specifically, in step S51, the nodes of the initial social graph of the elderly community are set to represent individual elderly people, and the weight of the edge connecting two elderly nodes is initialized to the social interaction intensity score calculated in step S4. Thus, the initial graph can preliminarily quantify the social interaction intensity between the elderly.

[0060] More specifically, in step S52, the place area code is used as a guide to process the edges of the initial social graph of the elderly community constructed in step S51. Specifically, for each edge in the graph, retrieve the place area code corresponding to the place where the two elderly people connected by the edge have social interactions. Subsequently, according to the place area code, search the pre-constructed place semantic knowledge graph to obtain the place function information corresponding to the place area code. The place function information reflects the type of social interaction in this place. For example, a park may correspond to the social function of "leisure and entertainment", and a restaurant may correspond to the social function of "dining and socializing". The type of social interaction corresponding to the place function information is matched to the edge of the initial social graph of the elderly community as the attribute information of this edge. Through the processing of step S52, in the finally formed social graph of the elderly community, the edge information is enriched, including not only the weight information of the social interaction intensity score but also the information of the type of social interaction corresponding to the place function information, which can more comprehensively and finely reflect the social interaction patterns among the elderly and provide a richer data basis for subsequent community analysis and abnormal behavior warning.

[0061] It should be noted that all the calculation processes involved in the method for constructing the social graph of the elderly in the embodiments of the present application are dimensionless calculations.

[0062] In a second aspect, please refer to Figure 2 , some embodiments of the present application also provide a method for warning of abnormal behavior of the elderly, which includes the following steps: B1. Analyze the social graph of the elderly community obtained by the method for constructing the social graph of the elderly provided in the first aspect based on the community discovery algorithm to obtain community structure information; B2. Obtain community parameter information according to the community structure information; B3. For the position in the community structure information where the elderly are located, dynamically evaluate the social isolation risk level and its change trend of each elderly person in combination with the community parameter information; B4. Predict whether the social behavior of each elderly person is abnormal according to the social isolation risk level and its change trend.

[0063] Specifically, in step B1, the community discovery algorithm is used to analyze the social graph of the elderly community to identify the community structure existing in the graph. The community structure information reflects the community division situation within the elderly group and the relationship between communities. The community discovery algorithm can adopt existing algorithms such as the Leiden algorithm and the modularity optimization algorithm. The community structure information can be embodied as the community division result. For example, the identifier of the community to which each elderly person belongs. As an implementation manner, the Louvain algorithm is adopted as the community discovery algorithm. The Louvain algorithm discovers the community structure in the graph by iteratively optimizing the modularity, and the modularity is used to evaluate the quality of the community division.

[0064] More specifically, in step B2, the community parameter information can quantitatively describe the characteristics of the community.

[0065] More specifically, in step B3, for the position of each elderly person in the community structure information, combined with the community parameter information, the social isolation risk level and its change trend of the elderly are dynamically evaluated. The social isolation risk level reflects the degree of connection between the elderly and their affiliated community, and the change trend reflects the change of the connection degree over time.

[0066] More specifically, step B4 is used to predict whether the social behavior of each elderly person is abnormal according to the social isolation risk level and its change trend. For example, if the social isolation risk level of an elderly person continues to rise, the prediction result is that the probability of the elderly person having abnormal social behavior in the future is relatively high. Through in-depth analysis of the social graph of the elderly community, the leap from the social relationship graph to the early warning of abnormal behavior is achieved, and the problem that simply constructing a social graph cannot be directly applied to the early warning of abnormal behavior is solved, providing an active and intelligent security guarantee means for pension institutions.

[0067] The abnormal behavior early warning method for the elderly in the embodiments of the present application analyzes the social graph of the elderly community obtained by the social graph construction method for the elderly provided in the first aspect to identify the community structure information in the elderly group and extract the community parameter information, so as to be able to target the nodes of each elderly person in the community structure, and combined with the community parameter information, dynamically evaluate the social isolation risk level and its change trend of each elderly person to determine whether there is a social isolation risk, so as to predict whether the future social behavior of the elderly is abnormal, realizing the early warning of the abnormal behavior of the elderly, providing technical support for pension institutions, and helping to improve the quality of pension services.

[0068] In some preferred embodiments, step B1 includes: B11. Use the modularity-optimized Louvain algorithm to perform community discovery on the social graph of the elderly community. Among them, the resolution parameter of the Louvain algorithm is adaptively optimized in the following way: B111. Construct a candidate parameter set for the resolution parameter, use the grid search method to select different resolution parameter values in the candidate parameter set, and run the Louvain algorithm to perform community division on the social graph of the elderly community; B112. For each community division result, calculate the silhouette coefficient of the community division, and select the resolution parameter corresponding to the maximum silhouette coefficient as the optimal resolution parameter; B113. Based on the community division result under the optimal resolution parameter, extract the community structure information.

[0069] Specifically, step B11 proposes an adaptive optimization method for the resolution parameter of the Louvain algorithm to perform community discovery on the social graph of the elderly community. First, step B111 creates a candidate parameter set for the resolution parameter, and then uses the grid search method to select different resolution parameter values from the candidate parameter set, so that the Louvain algorithm runs under each selected resolution parameter value, thereby obtaining multiple community partition results. Step B112 calculates the silhouette coefficient of the community partition for each community partition result. The silhouette coefficient is used as an indicator to evaluate the quality of the community partition, and the magnitude of its value reflects the quality of the community partition effect. The selection of the optimal resolution parameter is achieved by comparing the silhouette coefficients of the community partitions under different resolution parameters, and the resolution parameter corresponding to the maximum silhouette coefficient is determined as the optimal resolution parameter. After the optimal resolution parameter is determined, step B113 extracts the community partition result under this parameter as the community structure information for subsequent analysis.

[0070] More specifically, through the above processing method, the resolution parameter of the Louvain algorithm can be adaptively adjusted to the optimal state, improving the accuracy and robustness of community discovery.

[0071] In some preferred embodiments, the community parameter information includes community size, community connection strength, and node centrality index. Among them, the community size is characterized by the number of nodes in the community, the community connection strength is characterized by the average weight of the connection edges between the nodes in the community, and the node centrality index is characterized by the degree centrality index.

[0072] Specifically, in step B2, the community parameter information is obtained according to the community structure information obtained in step B1. The community parameter information is a quantitative description of the community structure characteristics, including community size, community connection strength, and node centrality index. Among them, the community size is determined by counting the number of elderly nodes included in the community; the community connection strength is quantified by calculating the average weight of all connection edges in the community, and the weight of the edge represents the social interaction strength between the elderly; the node centrality index uses the degree centrality index, which refers to the number of other elderly nodes directly connected to a specific elderly node in the community social graph. Thus, the community size can intuitively reflect the size of the community, the community connection strength can reflect the tightness of the connection among the internal members of the community, and the node centrality index can quantify the social activity of a specific elderly person in the community. Through these three pieces of community parameter information, the social graph of the elderly community can be quantitatively analyzed from different dimensions of the community. The pension institution can quantify the structure characteristics of the community and the position of individuals in the community, providing data support for subsequent social isolation risk assessment and abnormal behavior warning.

[0073] In some preferred embodiments, step B3 includes: B31. Analyze and obtain the node types of each elderly person in the social network graph of the elderly community according to their positions in the community structure information of the elderly. The node types are classified into: isolated nodes, marginal nodes, and core nodes. B32. Query the association relationship between the node type and community parameters according to the node type, and obtain the community scale threshold, community connection strength threshold, and node centrality index threshold corresponding to each elderly person. B33. Determine whether at least one of the following three conditions holds for each elderly person: the community scale of the elderly person is less than the community scale threshold, the community connection strength is less than the community connection strength threshold, and the node centrality index is less than the node centrality index threshold. B34. If the judgment result in step B33 is yes, it is determined that the corresponding elderly person has a risk of social isolation. B35. For the elderly with a risk of social isolation, based on the community structure information, extract the connection strength, connection quantity, and connection object type of the elderly person in the community, and use a pre-trained risk assessment model to fuse the connection strength, connection quantity, and connection object type to calculate the social isolation risk level of the elderly person. B36. Use the time series analysis method to analyze the social isolation risk level of the elderly to obtain the change trend of the social isolation risk level.

[0074] Specifically, in step B31, according to the positions of the elderly in the community structure information, the node types of the elderly in the community social network graph are divided into isolated nodes, marginal nodes, and core nodes. This division can reflect the different degrees of social participation of the elderly in the community. The node division process can use a variety of graph theory metrics to determine the node type. For example, it can be divided according to degree centrality. Nodes with extremely low degree centrality can be classified as isolated nodes, nodes between isolated nodes and core nodes can be classified as marginal nodes, and nodes with high degree centrality can be regarded as core nodes.

[0075] More specifically, the relevant thresholds set in step B32 are used as the benchmarks for evaluating the social isolation risk of the elderly. The setting of the thresholds can be based on the statistical distribution of community structure parameters. For example, the community scale, community connection strength, and node centrality index of all communities can be statistically analyzed, and then the corresponding thresholds can be set according to the statistical distribution of different node types. As a preferred implementation, these thresholds can be set manually according to experience or expert knowledge, combined with the actual community social network graph data.

[0076] More specifically, steps B33 and B34 judge whether there is at least one index lower than the threshold by comparing the community scale, community connection strength, and node centrality index of the community where the elderly are located with their respective corresponding thresholds. If so, it is initially determined that the elderly person has a risk of social isolation.

[0077] More specifically, step B35 conducts a refined assessment of the elderly who are initially judged to be at risk of social isolation. By inputting community structure information such as the connection strength, connection quantity (the number of other elderly people connected in the community structure information), and connection object type (the node type of other elderly people connected in the community structure information) of the elderly who are initially judged to be at risk of social isolation into a pre-trained risk assessment model, after the model fuses this information, the social isolation risk level of the elderly is output. The assessment model can use a machine learning classification model, such as a support vector machine or a random forest. The training data of the model can use historical community social data and the social isolation risk assessment results of the elderly. The input features of the model include community structure information such as the connection strength, connection quantity, and connection object type of the elderly in the community.

[0078] More specifically, step B36 uses time series analysis methods to perform time series analysis on the social isolation risk level of the elderly, so as to obtain the change trend of the risk level. The time series analysis method can adopt methods such as the autoregressive integrated moving average model to model and predict the time series data of the social isolation risk level of the elderly, so as to obtain the change trend of the risk level.

[0079] More specifically, the above processing method can effectively evaluate the social isolation risk level of the elderly and its change trend, provide early warning information for nursing homes, so as to take intervention measures in time to improve the social situation of the elderly. This method can conduct multi-level and dynamic risk assessments for the elderly of different social types, combining community parameters and individual connection characteristics, and provide data support for early warning of abnormal behaviors.

[0080] In some preferred embodiments, step B4 includes: B41. Based on a pre-constructed abnormal behavior prediction model, input the social isolation risk level and the change trend of the social isolation risk level of the elderly with social isolation risk, and predict the probability of the target elderly participating in social activities within a future time window. If the probability is lower than a preset threshold, it is determined that the social activities of the target elderly have abnormal behaviors.

[0081] Specifically, the abnormal behavior prediction model can adopt machine learning algorithms. For example, the model can be trained with historical data, and the historical data includes the social isolation risk level of the elderly and its change trend, as well as the corresponding social activity participation situation.

[0082] More specifically, the social isolation risk level and the trend of change in the social isolation risk level are used as inputs to an abnormal behavior prediction model that predicts the probability of a target elderly person participating in social activities within a future time window, which can be set according to actual needs. For example, it can be set to the next week or the next month. A preset threshold is set as the criterion for determining whether a social activity is abnormal. The threshold setting can be based on empirical data or expert knowledge. For example, if the average probability of an elderly person's normal social activity participation is 50%, the preset threshold can be set to 30%. When the probability predicted by the model is lower than the preset threshold, the system determines that the elderly person's social activity shows abnormal behavior, indicating that the willingness or ability of the elderly person to participate in social activities within the future time window has significantly decreased.

[0083] In this document, relational terms such as first and second are used solely to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0084] The above description is only for the embodiments of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for constructing a social graph of the elderly, used to analyze the social relationships of the elderly group, characterized in that: The method comprises the following steps: S1. Continuously collect location data of the elderly; S2. Obtain a pre-constructed place semantic knowledge graph, where the graph nodes include place area codes, and each place area code is preset with a place social function weight, a place type adjustment coefficient, and place function information; S3. For every two elderly people, the co-occurrence frequency and duration of the same place area code corresponding to the location data of the two in multiple consecutive time periods are analyzed to determine whether there is a location co-occurrence relationship between the two, and if there is a location co-occurrence relationship, the location co-occurrence intensity is calculated according to the co-occurrence frequency, the duration and the place type adjustment coefficient; S4, calculating the social interaction intensity score between the elderly according to the location co-occurrence intensity and the social function weight of the venue; S5. Construct a social graph of the elderly community based on the social interaction intensity score and the venue function information.

2. A method for constructing a social graph for the elderly according to claim 1, characterized in that: The step of analyzing the co-occurrence frequency and duration of the same location area codes corresponding to the location data of each two elderly people in multiple consecutive time periods to determine whether there is a location co-occurrence relationship between the two elderly people includes: S31, for every two elderly people, counting the co-occurrence frequency and duration of the same place area code corresponding to the location data of the two in multiple consecutive time periods, and preliminarily determining whether there is a location co-occurrence relationship between the two according to the co-occurrence frequency and the duration; S32, when it is preliminarily determined that there is a location co-occurrence relationship between the two elderly people, the effective co-occurrence frequency and effective co-occurrence duration in which the distance between the two elderly people is less than the interaction distance threshold in the multiple consecutive time periods are counted according to the location data; S33. Finally determine whether there is a position co-occurrence relationship between the two according to the effective co-occurrence frequency and the effective co-occurrence duration.

3. A method for constructing a social graph for the elderly according to claim 2, characterized in that: The process of determining the interaction distance threshold includes: A1. According to the venue function information corresponding to the venue area code, the interactive distance threshold is obtained from the venue function information and distance threshold mapping relationship table.

4. A method for constructing a social graph for the elderly according to claim 3, characterized in that: Each venue area code is also preset with venue area information; the process of determining the interaction distance threshold also includes: A2. Adjust the interaction distance threshold according to the venue area information corresponding to the venue area code.

5. The method for constructing a social graph for the elderly according to claim 1, characterized in that: The step of calculating the position co-occurrence intensity according to the co-occurrence frequency, the duration and the venue type adjustment coefficient comprises: S34, splitting according to the duration to obtain the sub-co-occurrence durations of each two elderly people in multiple consecutive time periods in each time period; S35. According to the venue type adjustment coefficient, a time decay factor sequence about time decay is set for each time period; S36, based on the time decay factor sequence, weighting the sub-co-occurrence durations in each time period to obtain a weighted co-occurrence duration; S37. Multiply the weighted co-occurrence duration by the co-occurrence frequency to obtain the position co-occurrence strength.

6. A method for constructing a social graph for the elderly according to claim 1, characterized in that: Step S4 includes: S41. Obtaining the age score and physical ability score of the elderly; S42, adjusting the weight of the social function of the venue according to the age score and physical ability score based on the location co-occurrence relationship; S43. Calculate the social interaction intensity score between the elderly based on the adjusted venue social function weight and the location co-occurrence intensity.

7. A method for constructing a social graph for the elderly according to claim 1, characterized in that: Step S5 includes: S51, constructing an initial social graph of the elderly community according to the social interaction intensity score, where the graph nodes represent the elderly and the edge weights are initialized to the social interaction intensity score; S52. Based on the guidance of the venue area coding, the social interaction type corresponding to the venue function information is edge-matched on the initial social graph of the elderly community to form the social graph of the elderly community.

8. A method for early warning of abnormal behavior of the elderly, characterized in that: The method comprises the following steps: B1. Analyzing the social graph of the elderly community obtained by the method for constructing a social graph of the elderly according to any one of claims 1 to 7 based on a community discovery algorithm to obtain community structure information; B2. Obtaining community parameter information according to the community structure information; B3. Dynamically evaluate the social isolation risk level and its changing trend of each elderly person based on their position in the community structure information and the community parameter information; B4. Predict whether the social behavior of each elderly person is abnormal based on the social isolation risk level and its changing trend.

9. The abnormal behavior early warning method for the elderly according to claim 8, characterized in that: Step B1 includes: B11. Use the modularity-optimized Louvain algorithm to perform community discovery on the elderly community social graph, wherein the resolution parameter of the Louvain algorithm is adaptively optimized in the following manner: B111, constructing a candidate parameter set of resolution parameters, using a grid search method to select different resolution parameter values ​​in the candidate parameter set, and running the Louvain algorithm to divide the elderly community social graph into communities; B112. Calculate the silhouette coefficient of each community division result, and select the resolution parameter corresponding to the maximum silhouette coefficient as the optimal resolution parameter; B113. Based on the community division results under the optimal resolution parameters, extract the community structure information.

10. The method for early warning of abnormal behavior of the elderly according to claim 8, characterized in that: The community parameter information includes community size, community connection strength and node centrality index, wherein the community size is represented by the number of nodes in the community, the community connection strength is represented by the average weight of the connection edges between the nodes in the community, and the node centrality index is represented by the degree centrality index.

Images