#1013: Quantifying Knowledge Convergence through Change Point Detection of Word Abstractness in Collaborative Discussions

The quantification of knowledge convergence often relies on qualitative coding, limiting insights into how the dispersion of abstractness evolves during collaborative discussion. This study proposes a quantitative method that models each discussion as a time series of abstractness and applies change point detection to identify distributional shifts in its mean and dispersion. Six groups engaged in ten-minute discussions. Nouns in their utterances were extracted and assigned abstractness scores, after which change-point detection was applied to each group’s abstractness sequence with a minimum segment size of 20%, and the number of change points was determined by BIC. Results showed that the mean followed a U-shaped trajectory while the standard deviation showed the inverse in four groups, reflecting a mid-discussion phase in which abstract reasoning was anchored in concrete examples. These results demonstrate that knowledge convergence appears as interpretable shifts thereby bridging qualitative analysis of knowledge convergence with quantitative segmentation.