The goal was to convert noisy, repetitive, or opinionated raw claims into a single clean canonical claim.
Deduplication
Raw claims
“The earliest actual poll ... showed that 62 percent of respondents believed Hillary had won the debate.”
“An early CNN/ORC poll showed that 62% of respondents thought Hillary Clinton won the debate.”
→
Canonical claim
According to a CNN/ORC poll, 62% of respondents believed Hillary Clinton won the debate.
Remove loaded language
Raw claims
“Trumpkins from both 4chan and Reddit were stuffing last night’s poll boxes...”
“Online polls often have skewed results due to manipulation by groups such as Trump supporters...”
→
Canonical claim
Trump supporters manipulated some online polls during the debate.
Aggregation
Raw claims
“60% of those favoring the right-wing coalition back ending Muslim migration.”
“40% of Labor Party supporters and 34% of Green Party supporters favor ending Muslim migration.”
→
Canonical claim
Among political affiliations, 60% of right-wing coalition supporters, 40% of Labor Party supporters, and 34% of Green Party supporters favor ending Muslim migration.