Topic Segmentation

Date:

Determining the positions at which topics change in a stream of text or speech.

Segmentation:

determining the positions at which topics change in a stream of text or speech.

Task:

  • Segment text into non-hierarchical, non-overlapping zones which contain the same subtopic
  • Equivalent definition: Detect subtopic shifts (changes of subtopic)
  • Reasons for not simply using paragraph or section boundaries:
    • Stark (1988) found not all paragraph boundaries reflect topic shifts
    • Paragraph conventions genre-dependent
    • Sections often too large

More information here