Wonders of Seabed: Difficulty Evaluation of Management Games Using Neural Network - Technologies and Applications of Artificial Intelligence - page 173

2. From Table 1(A) and Table 1(B), we can see that the effect of T cele and
T human is not so obvious for improving the total income. However, it can
increase slightly the time that the player needs to clear the stage.
3. Consider the results in Table 1(B) and Table 1(C) as examples. The larger
the range between T tax and T maintain is, the earlier the average total in-
come becomes positive. So, the difference between T tax and T maintain affects
greatly the amount of money earned.
6 Conclusions
We have developed a management game and proposed to use a neural network
to evaluate the diculty level for players. Our management game is a real time
3D game. Players can control a government unit to develop a city. Our game has
four major areas: a residential area, a commercial area, agricultural area, and an
administrative center. The players need to maximize the income within a game
time limit. We have applied the neural network to learn how to play the game.
Our system computes an expected income within a game time duration. We can
estimate the amount of time that a player needs to take to clear a stage. Thus,
we can evaluate the game diculty level with respect to a player while the game
is running.
Our approach has limitations. The complexity of the game rules is simple
in our game. There are only four parameters to control. If the complexity of
the game grows higher, our approach may not be adequate to evolve the neural
network to successfully play the game. This is due to that the output may be
highly non-linear with respect to the input data. The probability functions are
not smooth in our game and they affect the learning speed of the neural network.
Furthermore, the major parameters would be changed over time while the player
develops the city. The weights of the neural network are not useful anymore after
the major parameters are changed. The neural network must be trained again.
In the future, we would like to tackle these limitations.
Acknowledgements. We would like to thank the anonymous reviewers for
their constructive comments. This work was partially supported by the National
Science Council of ROC (Taiwan) under the grant no. NSC 102-2221-E-009-103-
MY2 and the Ministry of Science and Technology of ROC (Taiwan) under the
grant no. MOST 103-2221-E-009-122-MY3 and 103-2815-C-009-031-E.
References
1. Real-time game adaptation for optimizing player satisfaction. IEEE Transactions
on Computational Intelligence and AI in Games 1(2), 121-133 (2009)
2. Game object model version ii: a theoretical framework for educational game devel-
opment. Educational Technology Research and Development 55(1) (2007)
3. Andrade, G., Ramalho, G., Gomes, S., Corruble, V.: Dynamic game balancing: An
evaluation of user satisfaction. In: AAAI Conference on Artificial Intelligence and
Interactive Digital Entertainment, pp. 3-8 (2006)