Intra-Speaker Topic Modeling for
Improved Multi-Party Meeting Summarization with Integrated Random Walk
Yun-Nung Chen and Florian Metze
• Important utterances are topically/lexically similar to each other
• Utterances from the same speaker usually focus on similar topics
Construct a graph to represent utterances in the document (node: utterance, edge: weighted by topical/lexical similarity) Use the graph to re-compute the importance
Topic Model
(ex. PLSA, LDA)
U2 97 U3 83
:
Multi-Party Corpus
ASR Topical Distribution
Manual
Integrated Random Walk
Summary
Rerank
U5 93 U2 88 Summary :
LTE
(baseline)