Method of distributed aggregation in a call center
View Patent ↗A method for partitioning a call center having N agents associated with M agent groups, for computation by a plurality of computational nodes, has steps for (a) assigning each agent as a vertex in a hypergraph; (b) assigning each agent group as a hyper-edge in the hypergraph; and (c) applying a hypergraph partitioning algorithm to partition the agents and groups relative to the nodes with the hypergraph cost function awarding equal load distribution to nodes and minimizing inter-node traffic.
1. A method for partitioning a call center having N agents associated with M agent groups, for computation by K computational nodes, the method comprising:
(a) assigning each agent of the N agents as a vertex in a hypergraph;
(b) assigning each agent group of the M agent groups as a hyper-edge in the hypergraph;
(c) applying a hypergraph partitioning algorithm configured to partition the N agents and M groups among the K nodes with a hypergraph cost function configured to balance load distribution among the nodes and minimize traffic between the nodes;
(d) mapping each agent and group as an object O to one of the K nodes by a Hash function H:O→[1,K], such that each node H(O) is an owner for the object O;
(e) for each agent A, computing agent aggregates AA — 1, AA_P locally by the associated owner node H(A); and
(f) for each group G, computing group aggregates AG — 1, . . . , AG_S using respective agent aggregates AA — 1, AA_P for each agent belonging to group G, by:
i) creating a partial state record (PSR) of each of the group aggregates AG_j by node i by aggregating the agent aggregates AA — 1, . . . , AA_P for each of the agents owned by node i and belonging to group G; and
ii) sending the PSRs to the associated owner node H(G), which combines all the PSRs into corresponding total state records of the group aggregates.
2. A method for partitioning a contact center having N agents associated with M agent-groups, for computation by K computational nodes, the method comprising:
generating, by a processor, a hypergraph with N vertices corresponding to the N agents and M hyperedges corresponding to the M agent-groups;
partitioning, by the processor, the hypergraph into K partitions corresponding to the K nodes based on a function configured to balance load distribution among the nodes and minimize traffic between the nodes; and
using, by the processor, the partitioning to assign each of the agents and agent-groups to a corresponding one of the nodes,
wherein each node of the computational nodes is configured to:
compute agent aggregates for each of the agents assigned to the node;
compute group aggregates for each of the agent-groups assigned to the node;
compute partial group aggregates corresponding to the group aggregates of each agent-group of the agent-groups by aggregating the agent aggregates for each of the agents assigned to the node and belonging to the agent-group; and
send each partial group aggregate of the partial group aggregates to a respective one of the nodes corresponding to the agent group.
3. The method of claim 2 further comprising assigning, by the processor, a weight to each vertex of the N vertices, the vertex corresponding to an agent of the N agents, the weight being based on a computational complexity of the agent aggregates of the agent.
4. A system for partitioning a contact center having N agents associated with M agent-groups, the system comprising:
a processor; and
a non-transitory physical medium, wherein the medium has instructions stored thereon that, when executed by the processor, causes the processor to:
generate a hypergraph with N vertices corresponding to the N agents and M hyperedges corresponding to the M agent-groups;
partition the hypergraph into K partitions corresponding to K computational nodes based on a function configured to balance load distribution among the nodes and minimize traffic between the nodes; and
use the partitioning to assign each of the agents and agent-groups to a corresponding one of the nodes,
wherein each node of the computational nodes is configured to:
compute agent aggregates for each of the agents assigned to the node;
compute group aggregates for each of the agent-groups assigned to the node;
compute partial group aggregates corresponding to the group aggregates of each agent-group of the agent-groups by aggregating the agent aggregates for each of the agents assigned to the node and belonging to the agent-group; and
send each partial group aggregate of the partial group aggregates to a respective one of the nodes corresponding to the agent group.
5. The system of claim 4 , wherein the instructions, when executed by the processor, further cause the processor to assign a weight to each vertex of the N vertices, the vertex corresponding to an agent of the N agents, the weight being based on a computational complexity of the agent aggregates of the corresponding agent.