Summarizing Text Conversations - Methods for Mining and Summarizing Text Conversations - page 94

In much of the work on meeting summarization, the summaries are meant to be a component
of a meeting browser, serving as an index into the audio-video meeting record. The summaries

meeting
browsers

may be time-aligned with other artefacts from a meeting such as notes, slides and visualizations of
speaker activity. In these cases, extractive summary sentences will almost always have a one-to-one
mapping with transcript sentences, while abstract sentences can have a many-to-many mapping with
the transcript. This highlights the fact that abstractive systems are identifying patterns, messages
or events that aggregate numerous sentences. Figure 4.3 shows an example of a meeting browser
that incorporates abstract summaries, with the summary linked to the transcript and the transcript
time-aligned with the audio and video.

Figure 4.3: Meeting browser incorporating abstractive summaries.

One alternative to summarizing meeting speech is simply to speed it up. Experiments have
shown users still have good comprehension of meeting discussions even when the discussion is
played several times faster than the original speed [ Tucker and Whittaker , 2006 ]. However, user
satisfaction is not high.

4.3.3 SUMMARIZING CHATS AND BLOGS
In this section we first present case studies of summarization applied to online chats and blogs, and
then discuss assumptions and inputs, measures of informativeness, along with outputs and interfaces
in these contexts.