Methods and systems to identify collaborative communities from multiplex healthcare providers
Methods and systems to identify collaborative communities of individuals from graphs of multiple types of relationships amongst the individuals, including to mine data related to multiple types of relationships amongst individuals, construct graphs to represent the respective types of relationships amongst individuals, and perform a multiplex graph convolutional network (MGCN) artificial intelligence machine learning (AIML) analysis across the multiple graphs to identify the collaborative communities. A mathematical representation of the graphs may be learned/tuned to optimize clustering of the individuals. Multiple parameters (inter-graph weights, consensus regularization function) may be jointly tuned based on a joint optimization function. The collaborative communities may be displayed such that relative positions of the individuals represent measures of influence exerted by the respective individuals within the respective collaborative communities.
1 . A machine-implemented method, comprising:
mining data related to multiple types of relationships amongst individuals;
constructing multiple training graphs to represent the respective types of relationships amongst the individuals; and
performing a multiplex graph convolutional network (MGCN) artificial intelligence machine learning (AIML) analysis across the multiple training graphs to identify collaborative communities of the individuals based on the multiple types of relationships amongst the individuals, the performing comprising clustering the individuals into communities based on a mathematical representation of the training graphs, wherein the mathematical representation is machine tuned to optimize the clustering and to optimize a number of the collaborative communities relative to densities of the collaborative communities.
2 . The machine-implemented method of claim 1 , wherein the machine tuning comprises:
machine tuning weights applied to the training graphs to optimize the clustering.
3 . The machine-implemented method of claim 1 , wherein the performing further comprises:
displaying the collaborative communities of the individuals such that relative positions of the individuals within the collaborative communities represent measures of influence exerted by the respective individuals within the collaborative communities.
4 . The machine-implemented method of claim 1 , wherein the performing comprises:
translating the training graphs into a first mathematical function;
generating predicted graphs based on the first mathematical function;
translating the predicted graphs into a second mathematical function;
determining consensus regularization function based on the first and second mathematical functions;
clustering the individuals based on the first mathematical function;
computing a joint optimization function based on a measure of the clustering and the consensus regularization function; and
machine tuning multiple parameters to optimize the joint optimization function, wherein the multiple parameters include weights applied to the training and predicted multiplex graphs and a weight applied to the consensus regularization function.
5 . The machine-implemented method of claim 1 , wherein the constructing comprises:
constructing a peer sociometric training graph of healthcare professionals (HCPs) based on peer nomination surveys and link predictions generated from the peer nomination surveys;
constructing a patient centric training graph of the HCPs based on medical claims, prescription claims, and/or hospital claims of patients of the HCPs;
constructing an affiliation training graph of the HCPs based on organization affiliations of the HCPs;
constructing a scientific collaboration training graph based on medical conference co-presenter data and publication co-authorship data; and
constructing a social medial training graph of the HCPs based on social media contacts of the HCPs.
6 . The machine-implemented method of claim 1 , wherein the performing further comprises, for each collaborative community, determining a measure of the collaborative community, wherein the measure of the collaborative community includes density, volume, betweenness, connected value, and Rx size.
7 . The machine-implemented method of claim 1 , wherein the performing further comprises, for each collaborative community, determining a community leader score for each member of each collaborative community.
8 . The machine-implemented method of claim 1 , wherein the performing further comprises, for each collaborative community, determining a community leader score for each member of each collaborative community.
9 . An apparatus, comprising a processor and memory configured to:
mine data related to multiple types of relationships amongst individuals;
construct multiple training graphs to represent the respective types of relationships amongst individuals; and
perform a multiplex graph convolutional network (MGCN) artificial intelligence machine learning (AIML) analysis across the multiple training graphs to identify collaborative communities of the individuals based on the multiple types of relationships amongst the individuals, the performing comprising clustering the individuals into communities based on a mathematical representation of the training graphs, wherein the mathematical representation is machine tuned to optimize the clustering and to optimize a number of the collaborative communities relative to densities of the collaborative communities.
10 . The apparatus of claim 9 , wherein the processor and memory are further configured to:
machine tune weights applied to the training graphs to optimize the clustering.
11 . The apparatus of claim 9 , wherein the processor and memory are further configured to, for each collaborative community:
determine a community role for each collaborative community;
determine a community leader score for each member of each collaborative community;
determine a measure of centrality for each member of each collaborative community; and
determine a measure of the collaborative community, wherein the measure of the collaborative community includes density, volume, betweenness, connected value, and Rx size.
12 . The apparatus of claim 9 , wherein the processor and memory are further configured to:
display the collaborative communities of the individuals such that relative positions of the individuals within the collaborative communities represent measures of influence exerted by the respective individuals within the collaborative communities.
13 . The apparatus of claim 9 , wherein the processor and memory are further configured to:
translate the training graphs into a first mathematical function;
generate predicted graphs based on the first mathematical function;
translate the predicted graphs into a second mathematical function;
determine consensus regularization function based on the first and second mathematical functions;
cluster the individuals based on the first mathematical function;
compute a joint optimization function based on a measure of the clustering and the consensus regularization function; and
machine tune multiple parameters to optimize the joint optimization function, wherein the multiple parameters include weights applied to the training and predicted multiplex graphs and a weight applied to the consensus regularization function.
14 . The apparatus of claim 9 , wherein the processor and memory are further configured to:
construct a peer sociometric training graph of healthcare professionals (HCPs) based on peer nomination surveys and link predictions generated from the peer nomination surveys;
construct a patient centric training graph of the HCPs based on medical claims, prescription claims, and/or hospital claims of patients of the HCPs;
construct an affiliation training graph of the HCPs based on organization affiliations of the HCPs;
construct a scientific collaboration training graph based on medical conference co-presenter data and publication co-authorship data; and
construct a social medial training graph of the HCPs based on social media contacts of the HCPs.
15 . A non-transitory computer readable medium encoded with a computer program that comprises instructions to cause a processor to:
mine data related to multiple types of relationships amongst individuals;
construct multiple training graphs to represent the respective types of relationships amongst individuals; and
perform a multiplex graph convolutional network (MGCN) artificial intelligence machine learning (AIML) analysis across the multiple training graphs to identify collaborative communities of the individuals based on the multiple types of relationships amongst the individuals, the performing comprising clustering the individuals into communities based on a mathematical representation of the training graphs, wherein the mathematical representation is machine tuned to optimize the clustering and to optimize a number of the collaborative communities relative to densities of the collaborative communities.
16 . The non-transitory computer readable medium of claim 15 , further comprising instructions to cause the processor to:
machine tune weights applied to the training graphs to optimize the clustering.
17 . The non-transitory computer readable medium of claim 15 , further comprising instructions to cause the processor to, for each collaborative community:
determine a community role for each collaborative community;
determine a community leader score for each member of each collaborative community;
determine a measure of centrality for each member of each collaborative community; and/or
determine a measure of the collaborative community, wherein the measure of the collaborative community includes density, volume, betweenness, connected value, and Rx size.
18 . The non-transitory computer readable medium of claim 15 , further comprising instructions to cause the processor to:
display the collaborative communities of the individuals such that relative positions of the individuals within the collaborative communities represent measures of influence exerted by the respective individuals within the collaborative communities.
19 . The non-transitory computer readable medium of claim 15 , further comprising instructions to cause the processor to:
translate the training graphs into a first mathematical function;
generate predicted graphs based on the first mathematical function;
translate the predicted graphs into a second mathematical function;
determine consensus regularization function based on the first and second mathematical functions;
cluster the individuals based on the first mathematical function;
compute a joint optimization function based on a measure of the clustering and the consensus regularization function; and
machine tune multiple parameters to optimize the joint optimization function, wherein the multiple parameters include weights applied to the training and predicted multiplex graphs and a weight applied to the consensus regularization function.
20 . The non-transitory computer readable medium of claim 15 , further comprising instructions to cause the processor to:
construct a peer sociometric training graph of healthcare professionals (HCPs) based on peer nomination surveys and link predictions generated from the peer nomination surveys;
construct a patient centric training graph of the HCPs based on medical claims, prescription claims, and/or hospital claims of patients of the HCPs;
construct an affiliation training graph of the HCPs based on organization affiliations of the HCPs;
construct a scientific collaboration training graph based on medical conference co-presenter data and publication co-authorship data; and
construct a social medial training graph of the HCPs based on social media contacts of the HCPs.