UNSUPERVISED PRIORITIZATION AND VISUALIZATION OF CLUSTERS
Techniques are disclosed that automatically identify and order the most differentiated clusters from a given collection of clusters within a dataset. A measure of dissimilarity is computed for each cluster from a defined reference cluster, and the clusters are ordered according to the chosen dissimilarity. At least N clusters are selected as the most differentiated clusters relative to the defined reference. Within each cluster, the top-M most distinguishing cluster attributes can be automatically identified by an analogous process that computes the dissimilarity of each cluster attribute to its corresponding attribute in the reference cluster, and orders the attributes by dissimilarity. This then allows for automatic surfacing of what it is about a cluster that differentiates its members relative to the population as a whole, and to provide insight on what action or treatment might be made to address that specific segment of the underlying population.
1 - 20 . (canceled)
21 . A network device, comprising:
a transceiver to send and receive data over a network; and
a processor that is operative to perform actions, including:
receiving data having values for a plurality of attributes for a plurality of entities, wherein each entity is a user of a cell phone and has an associated set of attribute values that include values for one or more types of use of communication capabilities of the cell phone; and
performing, via unsupervised learning, a prioritized visualization for multiple clusters generated by machine analysis of multiple attributes from the received data, wherein each of the multiple clusters includes a distinct group of multiple entities that are grouped together based at least in part on the one or more types of use of the communication capabilities of the cell phone, the performing including:
computing, for a first cluster of the multiple clusters and for each of the multiple attributes, an aggregate value of the attribute for the first cluster by combining values of the attribute for the multiple entities in the first cluster;
determining, for a selected entity from the multiple entities of the first cluster, attributes that are most differentiating between the selected entity and the first cluster, based on computed attribute dissimilarities between values of the multiple attributes for the selected entity and the aggregate values of the multiple attributes for the first cluster; and
displaying, on a display device, information about the determined attributes to indicate how the selected entity is differentiated from the multiple entities in the first cluster.
22 . The network device of claim 21 wherein the performing of the prioritized visualization further includes:
determining, for the selected entity, the values of the multiple attributes; and
computing the attribute dissimilarities between the values of the multiple attributes for the selected entity and the aggregate values of the multiple attributes for the first cluster,
and wherein the displaying of the information includes displaying indications of one or more of the computed attribute dissimilarities.
23 . The network device of claim 22 wherein the performing of the prioritized visualization further includes:
determining attributes for each of the multiple entities in the first cluster that are most differentiating from the aggregate values of the multiple attributes for the first cluster; and
identifying the entity to select based at least in part on the determined attributes for each of the multiple entities.
24 . The network device of claim 22 wherein the performing of the prioritized visualization further includes determining attributes for each of the multiple entities in the first cluster that are most differentiating from the aggregate values of the multiple attributes for the first cluster, and wherein the displaying of the information includes displaying an ordered ranking of the multiple entities based on the determined attributes for each of the multiple entities.
25 . The network device of claim 21 wherein the associated set of attribute values for each of the entities further includes one or more types of demographic information, and wherein the performing of the prioritized visualization further includes generating, via unsupervised learning, the multiple clusters based at least in part on the one or more types of demographic information and on the one or more types of use of the communication capabilities of the cell phone.
26 . The network device of claim 25 wherein the determining of the attributes that are most differentiating between the selected entity and the first cluster includes determining at least one attribute separate from the multiple attributes used in generating the multiple clusters.
27 . The network device of claim 21 wherein the performing of the prioritized visualization further includes determining one or more attributes that are most differentiating between the selected entity and one or more additional clusters of the multiple clusters, and wherein the displaying of the information further includes displaying information about the determined one or more attributes and the determined one or more additional clusters.
28 . The network device of claim 21 wherein the determined attributes include a set of numerical vectors interpreted as at least one of probability density functions or probability mass functions over one or more characteristics, and wherein the determining of the attributes that are most differentiating between the selected entity and the first cluster includes comparing distributions of the one or more characteristics for the first cluster and the selected entity.
29 . The network device of claim 21 wherein the determining of the attributes that are most differentiating between the selected entity and the first cluster is performed at least in part using a Kullback-Leibler divergence.
30 . The network device of claim 21 wherein the determined attributes include two or more attributes, and wherein the displaying of the information further includes displaying an ordering of the two or more attributes that is based on amounts of differentiation for the two or more attributes.
31 . A non-transitory computer-readable medium having computer-executable instructions stored thereon that, in response to execution by a computing device, cause the computing device to perform operations including:
receiving data having values for a plurality of attributes for a plurality of entities, wherein each entity is a user of a cell phone and has an associated set of attribute values that include values for one or more types of use of communication capabilities of the cell phone; and
performing unsupervised prioritization for multiple clusters generated by machine analysis of multiple attributes from the received data, wherein each of the multiple clusters includes a distinct group of multiple entities that are grouped together based at least in part on the one or more types of use of the communication capabilities of the cell phone, the performing including:
computing, for a first cluster of the multiple clusters and for each of the multiple attributes, an aggregate value of the attribute for the first cluster by combining values of the attribute for the multiple entities in the first cluster;
determining, for a selected entity from the multiple entities of the first cluster, one or more attributes of the multiple attributes that are most differentiating between the selected entity and the first cluster, based on computed attribute dissimilarities between values of the one or more attributes for the selected entity and the aggregate values of the one or more attributes for the first cluster; and
displaying, on a display device, information about the determined one or more attributes to indicate how the selected entity is differentiated from the multiple entities in the first cluster.
32 . The non-transitory computer-readable medium of claim 31 wherein the performing of the unsupervised prioritization further includes:
determining, for the selected entity, the values of the one or more attributes; and
computing the attribute dissimilarities between the values of the one or more attributes for the selected entity and the aggregate values of the one or more attributes for the first cluster,
and wherein the displaying of the information includes displaying indications of one or more of the computed attribute dissimilarities.
33 . The non-transitory computer-readable medium of claim 32 wherein the performing of the unsupervised prioritization further includes:
determining attributes for each of the multiple entities in the first cluster that are most differentiating from aggregate values of the multiple attributes for the first cluster; and
identifying the entity to select based at least in part on the determined attributes for each of the multiple entities.
34 . The non-transitory computer-readable medium of claim 32 wherein the performing of the unsupervised prioritization further includes determining attributes for each of the multiple entities in the first cluster that are most differentiating from aggregate values of the multiple attributes for the first cluster, and wherein the displaying of the information includes displaying an ordered ranking of the multiple entities based on the determined attributes for each of the multiple entities.
35 . The non-transitory computer-readable medium of claim 31 wherein the determined one or more attributes include at least two attributes, and wherein the displaying of the information further includes displaying an ordering of the at least two attributes that is based on amounts of differentiation for the at least two attributes.
36 . The non-transitory computer-readable medium of claim 31 wherein the associated set of attribute values for each of the entities further includes one or more types of demographic information, and wherein the performing of the unsupervised prioritization further includes generating, via unsupervised learning, the multiple clusters based at least in part on the one or more types of demographic information and on the one or more types of use of the communication capabilities of the cell phone.
37 . The non-transitory computer-readable medium of claim 31 wherein the determining of the attributes that are most differentiating between the selected entity and the first cluster includes determining at least one attribute separate from the multiple attributes used in generating the multiple clusters.
38 . The non-transitory computer-readable medium of claim 31 wherein the determined attributes include a set of numerical vectors interpreted as at least one of probability density functions or probability mass functions over one or more characteristics, and wherein the determining of the one or more attributes that are most differentiating between the selected entity and the first cluster includes comparing distributions of the one or more characteristics for the first cluster and the selected entity.
39 . The non-transitory computer readable medium of claim 31 wherein the determining of the one or more attributes that are most differentiating between the selected entity and the first cluster is performed at least in part using a Kullback-Leibler divergence, a Battacharrya distance, an mean-squared error, an Lp-norm, or an Euclidean distance.
40 . The non-transitory computer readable medium of claim 31 wherein the determined one or more attributes are each a vector-valued attribute.