A Framework for Rank Computation and
Aggregation in Fuzzy Environments
Asheq Khan :University at Buffalo (SUNY)
Chunming Qiao : University at Buffalo (SUNY)
Satish K. Tripathi: University at Buffalo (SUNY)
INFOCOM 2009
Outline
Introduction
Service Management Environment
Ranking Metrics
Resolution Efficiency
Information Centrality
Ranking Policies
Results
Conclusion
2
Introduction
When user submit loosely-defined requests to obtain answers,
we should capturing the importance, relevance or rank of an
object according to a set of metrics or rules.
Information retrieval (IR)
Database
World wide web (WWW)
Network domain
IT service management system
3
Service Management Environment
Comprises of SMEs
(a SME may serve in multi-groups)
Capability : {c(d1),…, c(dd)}
Probability : {p(d1),…, p(dd)}
4
Service Management Environment
Routing to possible resolver
Which SME should we choose?
Re-route
My email is not working!
Assign signature
5
Outline
Introduction
Service Management Environment
Ranking Metrics
Resolution Efficiency
Information Centrality
Ranking Policies
Results
Conclusion
6
Ranking Metrics
We propose a ranking system to evaluate the performance of
each SME in resolving tickets originating from different
domains.
Acquires knowledge from both the information flow and the
organizational structure.
Rank of SME
Resolution
efficiency
7
Information
centrality
Resolution Efficiency(1)
The resolution efficiency of SME Sk, RE(Sk) represents the
effectiveness of Sk to resolve each distinct type of problems.
How quickly Sk processes a
problem ticket?
Current capability
Resolution
efficiency
Past performance
Workload
Δ is a parameter tuning constant that can be adjusted to set the range of the processing time.
8
Resolution Efficiency(2)
Resolution proximity(RP)
The additional time when the ticket passed SME Sk.
How fastResolution
can Sk route a ticket
efficiency
to a potential
resolver?
Current capability
Past performance
Workload
Sr is the resolver for each ticket with PDSm, that has passed through SME Sk
Nm is the total number of such tickets handled by Sk
Tl(Sk, Sr) represents the latency to reach Sr from Sk
9
Resolution Efficiency(3)
A SME Sk with low processing time and resolution proximity
might be repeatedly chosen to handle a large portion of tickets.
Resolution
efficiency
Current capability
Past performance
How many tickets
waiting in the queue?
Workload
ti ∈ Nq represents each of the Nq tickets waiting in the queue for service by Sk.
10
Information Centrality(1)
The resolution efficiency does not capture the interactions
between SMEs as they resolve a problem ticket.
Information
centrality
degree
in-degree
out-degree
M is the total number of SMEs in the system.
Global centrality of Sk
11
betweenness
Cb(Sk) = SMEs (Si, Sj) uses Sk as a
medium to reach each other
closeness
Outline
Introduction
Service Management Environment
Ranking Metrics
Resolution Efficiency
Information Centrality
Ranking Policies
Results
Conclusion
12
Ranking Policies(1)
SME Selection
The system analyzes historical PDSs that are similar to the PDS
of the pending ticket and selects the SMEs that have worked on
those.
K-Nearest Neighbors (KNN):
choose the k most closely matched PDS
Global Selection (G)
(d is the total number of domains )
13
Ranking Policies(2)
SME Ranking
Determine the performance score (PS) of the SMEs
Resolution efficiency (RE)
Information centrality (C)
Both
14
Outline
Introduction
Service Management Environment
Ranking Metrics
Resolution Efficiency
Information Centrality
Ranking Policies
Results
Conclusion
15
Performance Evaluation(1)
simulated IT service management model
500 SMEs and 25 work-groups with each SME working in at most 3
groups.
5 different service domains, each supported by 6 different work-groups.
Each SME is assigned a capability vector
{c(d1), .., c(d5)} based on the domain he/she services.
Assign each SME a random capability between 0.4 and 1 to resolve a
ticket in each domain.
We generate synthetic problem tickets and randomly assigned a PDS
{p(d1),…,p(d5)} (where,
5
p(d ) 1)
i
i 1
16
Performance Evaluation(2)
Two approach
Probabilistic
determine the likelihood of each SME to resolve the ticket
Likelihood score (LS) of a SME Sk to resolve the ticket
Iterative Domain Probing
Inquire each supported domain in an orderly manner to find the resolver
Do not take into consideration the queuing delay (wq = 0)
17
Performance Evaluation(3)
SME selection
K-nearest neighbors (KNN)
Global selection (G)
SME ranking policy
Centrality (C)
Resolution efficiency (RE)
Both
18
Probabilistic
Results(1)
routing algorithm
Avg resolution
time
Avg number of
hops
19
IDP
routing algorithm
Results(2)
The k-nearest neighbors (KNN) approach performs better
than the global selection.
C(G)
RE(G)
C(KNN)
Both(G)
Both(KNN)
RE(KNN)
20
(set k=15)
Results(3)
Using RE and Both policies perform better than C policy in
the ticket traces generated using the probabilistic algorithm.
RE(G)
Both(G)
RE(KNN)
Both(KNN)
21
Results(4)
Using information centrality (C) policy performs the best in
the ticket traces generated using the IDP algorithm.
It’s due to the adaptability of the routing
algorithm to the fuzziness of the environment.
C(G)
C(KNN)
22
Results(5)
A probabilistic approach seems to adapt much better.
8.3
10.3
It’s because IDP always starts by probing the most
likely troubled domain and moves down the list.
Only 1/3 tickets belong to the domain with the highest probability
and 1/5 of the tickets are from the two least likely support domains
23
Conclusion
This paper provided a framework for computing the rank of
Specialists in an IT service management system.
The results suggest the importance of considering both
routing efficiency and social connectivity to minimize the
resolution time of a ticket.
24
© Copyright 2026 Paperzz