Individual Belief Updating Model And Verification In Stochastic Decision Making

This study comes of a practical problem. There are so many different database and researching products provided by lots of webs for Sci-tech periodicals, such as CNKI. However, only little customers use the products or even little ones know them. It's why? What brings the difference between the expectation of operator and the behavior of customers? Therefore, it must be focused on the users' information behavior. Compared with the factors which form stochastic decision problem, it is appropriate that the analytical method be adopted in the study on users' information behavior.The subject of this study is the two sides of one problem: (1) the learning rule in stochastic decision-making, especially the belief updating rule; (2) the path of the cognition evolution. The empirical method used in the field of behavioral game theory is adopted here to make cleat whether the agents' behavioral rules are some kinds of non-conscious learning model, such as reinforcement learning rule, or belief learning model.Because the model is the bridge between reality situation and the form of the experiment, one point here is to select a model or even modeling the experimental context. It is simple to prick off an appropriate one from existing models. Taking the decision situation with the lowest information constraints, only reinforcement learning model is chosen. On the other hand, reinforcement learning is some of non-conscious learning process, so it cannot research agent's belief. So one model (BBAM) based on the process of belief updating is founded to simulate this kind of decision problem. By the way, this method can be refered to solve the same kind problems.Parameter test is adopted in reinforcement model and non-parameter test in BBAM. The former uses the EWA (Experience weighted attraction learning model) estimation for reference, and the latter tests three learning hypotheses with two samples.The conclusions are drawed as following: (1) both reinforcement learning moedel and BBAM can fit the information searching behavior on web in given context; (2) faced with a general restriction, which means a strong uncertainty, the decision making behavior of bounded rationality agent should involve completely rationally calculation and routinism without any conscious reflection on the situation; (3) agent's cognition evolution should be directed by consciousness, the convergence is affected by three factors: the initial belief system, the response to the feedback, the times of attempt.