The invention relates to the technical field of
social network analysis and generative
artificial intelligence, and discloses a multi-agent-based complex
social behavior simulation and public opinion deduction method, which comprises the following steps of: dividing social groups, and dividing users in a
social network into key opinion leader communities and common user communities; respectively constructing a key opinion leader agent and a common
user agent which are driven by the large
language model; constructing a dynamic weighted directed network, taking each key opinion leader agent model as a node, calculating a network node neighborhood through opinion index and node influence weighting, calculating an emotion
score based on a text emotion index, dynamically adjusting the network node neighborhood based on the emotion
score, and performing loop iteration for several times to complete deduction and
simulation of network public opinions; the problem that public opinion propagation modeling is difficult under the cross-domain dynamic background is solved, the difference of public opinion individuals and the complex social interaction relation can be simulated more accurately, and the method is feasible.