Mining Text Conversations - Methods for Mining and Summarizing Text Conversations - page 46

Document

Sliding
Windows

Gap 1
Gap 2
…..

Gap 11
…..

Gap 25

…..

y-axis:
Cohesion Score

High
Cohesion

…..

Low
Cohesion

Gap 1 ….. Gap 11 …..

Gap 25
x-axis:
Gaps between sliding
windows, from top to
bottom of document

Figure 3.1: TextTiling topic modeling algorithm. A cohesion score is measured at each sentence gap
as two sliding windows are moved down on a target document. Bottoms of deep valleys in the cohesion
score function indicate segment boundaries.