Mining System-User Interaction Traces for Use Case Models

1
Mining System-User Interaction Traces for Use
Case Models Mohammed El-Ramly Eleni
Stroulia Paul Sorenson (presented by Hsiao-Ming
Tsou)?
2
"In previous works, we presented the CelLEST
method for legacy user interface (UI)
reengineering and integration. This method uses
recorded traces of the interaction between the
legacy system and its users as its input, and
does not require examination of the legacy
code." "In this paper, we discuss how the
CelLEST method can be further automated by
discovering the use case models, needed for the
GUI construction, from the interaction traces. We
use knowledge discovery and pattern searching
algorithms to search for frequently occurring
patterns in the traces."
3
Agenda

• Background
• Previous Works
• Current Works
• Summary
• Likes/Dislikes

4
Background

• Often times, new systems under development
  require integration with other systems.
• Such systems include
• Off-the-shelf components (COTS)?
• Usually comes with documentation on how to
  integrate.
• Legacy systems
• Usually lacks documentation.
• Source code, even if available, is often poorly
  structured in terms of application design.
• How to integrate a new system with an underlying
  legacy system, given a lack of internal
  specifications?
• Solution data mine user-interface (UI)
  specifications collect input/output traces (use
  cases) of real users using the legacy system, and
  build a behavioral model of the system out of it.

5
Previous Works

• CelLEST was developed as a method for legacy user
  interface reengineering and integration.
• "The CelLEST UI reengineering method is
  applicable when the reengineering objective is to
  provide a more usable GUI to current legacy
  services, or to make them available through the
  Web."
• The CelLEST method consists of two phases
• Phase 1 Reverse Engineering
• Produce a state-transition diagram out of use
  cases - recorded traces of the interaction
  between the legacy systems and its users. The
  traces are a series of screen snapshots of the
  user interface.
• Phase 2 Forward Engineering
• Construct an abstract specification of a GUI for
  the modeled task, such that it can be implemented.

6
Previous Works (2)?
7
Previous Works (3)?
8
Current Works

• "In this paper, we describe how patterns of
  frequent user tasks can be discovered
  automatically by applying data mining techniques
  to interaction traces, and hence eliminating the
  need to collect task-specific traces."
• There are some issues in automatically
  discovering frequent user tasks from interaction
  traces
• Not all frequent user tasks are interesting - we
  want only the interesting ones.
• How to define the criteria that makes a task
  interesting
• How to extract the tasks that are interesting
• Verifying that those extracted tasks are what we
  are looking for
• Building use case models for the extracted
  patterns

9
Current Works (2)?

• Finding common sequences
• The "interesting"-ness is scored in terms of
  three parameters
• pattF (pattern frequency
• pattF (pattern frequency)?
• density ("accuracy" of the pattern)?

10
Current Works (3)?
11
Likes/Dislikes

• Likes
• The diagrams were intuitive.
• Dislikes
• Intuitively (and logically), this isn't a very
  sound approach to engineering interfaces on the
  Web. Automation isn't the answer to everything,
  and this is an arguably inherently insecure way
  to build interfaces.
• Some of the equations in the paper aren't
  explained in terms of reason.
• Why not just ask the expert user?