This paper proposes an artificial intelligence-driven personalized learning path optimization framework for the problems of high dropout rate and low course passing rate in online learning. A fuzzy cognitive diagnostic model (Fuzzy-CDF) is introduced to replace the traditional binary diagnosis, and through fuzzy intersection and merger operation and 4-Logistic parameter correction, the continuous cognitive level value is output, so as to realize the fine-grained quantification of the mastery degree of knowledge points. A two-dimensional learning state model of “basic knowledge + pattern knowledge” is constructed, in which the pattern knowledge dynamically portrays the cognitive structure from four attributes: overall level, feature point level, coverage set level, and coverage level. We also design a knowledge graph-based failure rate update mechanism to locate the weak points through the initial failure rate matrix, and dynamically correct the assessment results using the contribution value of the centrality of the knowledge points, so as to realize the accurate push of personalized resources. Experimental validation shows the effectiveness of Fuzzy-CDF diagnosis, in the test of 237 students, the model accurately identifies the weak points of the group, the mastery rate of curve integral A3 is 31.71%, the reintegration application A8 is only 27.42%, and the mastery rate of the strong knowledge point of the infinite number of steps A4 reaches 90.55%. Oriented to the four differentiated learning state users, the satisfaction of this method for planning paths reaches up to 4.823, which significantly exceeds the genetic algorithm GA and the ant colony algorithm ACO, with an average improvement of 14.7%, and the matching degree reaches 0.79-0.91.
