Cognitive mechanism of creative thinking: Integrating the semantic network and spreading activation model

Computational modeling offers a principled way to link the structure of semantic networks to the processes that support creative problem solving. We combined item-level semantic network analyses (Experiment 1, behavioral method) with the Spreading Activation Model simulation (SAM; Experiment 2, computational modeling) for the Chinese Word Remote Associates Test (CWRAT). In Experiment 1, we built customized item-semantic networks based on the CWRAT-free association task and quantified network properties for each item. Results showed that greater network efficiency and higher modularity were associated with higher item pass rates, whereas larger network size was associated with lower item pass rates. In Experiment 2, we used SAM to simulate activation dynamics on each item's semantic network across a grid of retention (r, the proportion of activation retained at each time step) and decay (d, the proportion of activation lost at each time step) parameters. At the item level, we extracted target-activation metrics (mean maximum activation, mean activation, and mean activation retention duration), and the results showed that these metrics correlated positively with the empirical item pass rate and negatively with item response time. At the individual level, we estimated best-fitting parameter pairs (r, d) for each participant. Results showed that these parameters correlated negatively with individuals' performance in both full and high-difficulty item sets. Particularly on the high-difficulty subset, results showed that parameters correlated positively with individual ratings of insightfulness and interest. Together, these results describe how bottom-up spreading activation may interact with top-down control to shape and support the creative thinking process.