A Recognition Approach For Adversarial Planning Based On Complete Goal Graph

Since plan recognition as a new research problem has been proposed, the related research has been focused on the simple traditional areas, such as natural language understanding, intelligent help systems. In recent years, some scholars have turned the focus on the nature of the confrontation research fields, such as the game, military, and confrontation of network information and so on. These areas in possession of nature of confrontation are called adversary field. Planning in the field of adversary is called adversarial plan, and planning to deal with adversary is called confrontation planning.Based on classical plan graph and goal graph, complete goal graph (CGG) is constructed against the characteristics of adversarial domain. The complete goal graph makes the action relate with its goal directly, which is more efficient to recognize adversarial goals. The conception of complete degree of goals will be put forward to distinguish adversarial high level goals. Then, a deep research is done on adversarial plan recognition algorithm based on CGG, which not only can predict adversarial next step action, but also can recognize adversarial goals in different levels with complete degrees. The research is of great significance for the plan recognition problem of uncontrollable and nondeterministic domain such as contest robot, information security, business strategy, game role design and etc. At last, the paper presents role value and checking function for checking adversarial planning recognition, and a recognizer with recognition checking method would do better, which can correct its recognized message for the information that fed back by checking method. And then, the recognizer can recognize adversarial planning and goals correctly.C++ language has been used on the .NET platform in this paper, and Gobang as an example to achieve the algorithm involved in the paper comparing with the existing MaxMin search method. The algorithm proposed by the paper could correctly identify not only the adversarial actions and targets, but also identify a more timely and effective confrontation planning. At the same time, the effective verification of the algorithm in the field of Gobang indicates that the algorithm has also good prospects in the other fields such as games, war, network information confrontation.