This page was translated automatically from Turkish. Read the original

Bu sayfa, 26. Dünya Psikiyatri Kongresi'nde sunulan çalışmanın yapay zekâ desteğiyle hazırlanmış Türkçe özetidir. Çalışmanın sahibi sunan yazarlardır; atıf için orijinal kaynağa başvurun.

EV0452

Symbolic entropy and semantic network disorder: a computational approach to cognitive coherence in psychiatric models

SYMBOLIC ENTROPY AND SEMANTIC NETWORK DISORGANIZATION: A COMPUTATIONAL APPROACH TO COGNITIVE COHERENCE IN PSYCHIATRIC MODELS

While one-step local entropy misses changes in relational topology, multistep entropy and clustering capture them; entropy should therefore be interpreted alongside topology.

The aim was to test, in a word association network, whether entropy could distinguish changes in relational topology from changes in the strength of existing associations; psychiatric relevance remains hypothetical. SWOW-EN18 first-response (R1) strengths were filtered at a threshold of 0.06, the 500 most frequent words were selected, self-loops were excluded, and duplicate edges were merged by retaining the highest strength; the resulting directed network contained 467 words and 847 edges, with 448 words having outgoing associations and 19 sinks. As controlled perturbations, 0, 0.25, 1, and 5 accepted double-edge swaps per original edge were applied using 50 random seeds; the swaps preserved each node's in-degree and out-degree, as well as its multiset of outgoing weights. Rewiring increased three-step endpoint entropy (H3) and reduced clustering, while one-step local entropy (H1) remained constant by design: H1 was initially 0.684 bits and remained at 0.684 ± 0.000 bits at 5 swaps per edge, H3 changed from 1.692 to 2.112 ± 0.042 bits, and the directed clustering coefficient C changed from 0.0981 to 0.0042 ± 0.0015. In the final condition, H3 increased by 0.420 bits (24.8%), while C decreased by 95.7%; these effects describe the perturbed network, not a clinical population. In the weight control condition, flattening the outgoing distributions increased H1 from 0.684 to 0.758 bits, while C remained at 0.0981; this increase is a structural consequence of making the distributions uniform and does not indicate disorder. The authors note that local entropy may miss topological changes and that multistep distributions and clustering add structural characterization, but none of these measures alone establishes cognitive coherence or psychopathology. Limitations include the use of a single thresholded English-language network and a fixed three-step horizon, with sensitivity to threshold, language, and horizon left untested; clinical utility requires independent speech data and controls for network size, density, and speech length.

Tüm WCP 2026 özetleri