Ji-Gui Sun

We propose a new visualization method WDM to classify documents by adding in the position-factors of words such as the title-factor and the first-sentence-factor based on a SOM neural network. We also discuss the selection of the function which is used to calculate the belong-to-probability in neurons' reflecting process. The experimental results indicate that WDM makes the boundaries of different documents greatly more clear, and thus it can produce more accurate and intuitive classification compared to the visualization methods which do not have position-factors considered

Rigor is one of the most important factors required by agent-based technology. Based on SLABS, a formal specification language of multi-agent system (MAS) providing language facilities that support modularity, advocates a description logic method to represent the static knowledge of a MAS's specification, which depicts the basic modules: caste and agent, the inheritance relationship between castes (or agents) and the instance relationship between agents and castes. This paper provides a basis for formalizing a MAS in description logics (DLs), whose powerful expressiveness and the decidable reasoning mechanisms help to proof and analyze the properties of agent-based systems

In recent years there has been plenty of interest in random constraint satisfaction problem, both from an experimental and a theoretical point of view. In this paper we study and analyze the four popular problem instance generating models, and present the extended model B+ based on the most used model B, which has the different domains and constraint tightness meeting some probability distribution function. In the subsequent section we give the relation matrix version of backtracking integrated forward checking algorithms, and introduce the implementation of instances generator and solver based on the new model. Finally we show the experiment results and conclude the paper, point that our extended model B+ has the common phase transition region with the transitional models and it has the advantage of being suited to the testing of heuristic based constraint solving algorithms, such as variables selection heuristic algorithms

Fail First Principle (FFP) is a general heuristic algorithm (for solving constraint satisfaction problems) considering the order of variables. However, FFP neglects the order of values in domains. In this paper, we describe two algorithms, DH_FFP and Rank_FFP based on FFP. DH_FFP partitions a domain recursively with the middle value after choosing the variable that has the smallest domain. Rank_FFP makes use of the heuristic information between variables and constraints, evaluating and ranking every partition of a domain. We implemented these two algorithms and embedded into the constraint solving platform "Ming Yue SOLVER 1.0" designed by us. The experimental results show that both DH_FFP and Rank_FFP enhance FFP to much extent on both solving efficiency and size