In some our work for the IARPA program, we looked at the relationship between the structure of topic chains and group cohesiveness (as judged by external evaluators). We found a nice correlation that works like this
a.) identify the noun phrases that have the most references (pronominal, synonyms, repetitions) and call them the meso-topics of the discourse.
b.) count the number of "gaps" in the chain of references to the meso-topic, where a gap is a turn without a mention.
c.) Plot the rank of the meso-topic chain against the gappiness of the topic.
We found a neat correlation like the following:
a.) identify the noun phrases that have the most references (pronominal, synonyms, repetitions) and call them the meso-topics of the discourse.
b.) count the number of "gaps" in the chain of references to the meso-topic, where a gap is a turn without a mention.
c.) Plot the rank of the meso-topic chain against the gappiness of the topic.
We found a neat correlation like the following:
No comments:
Post a Comment