abstract |
The invention discloses an action knowledge extraction method combined with a Markov decision process, comprising: training a random forest model H; defining an action knowledge extraction problem AKE: for the random forest model H, segmenting attributes, defining attribute changes and actions, On this basis, define the action knowledge extraction problem AKE; use the Markov decision process to solve the AKE optimization problem: for any input data, define the Markov decision process MDP, and define the strategy, update the strategy through strategy iteration, and finally solve to get an optimal Optimal strategy; the action defined by the action knowledge extraction in the present invention can change multiple attribute values of the state, and will give accurate and feasible suggestions in practical applications. |