Distribution of User Behavior in String Type Modeling

• Main Authors: CHEN, YEN-CHIEH, 陳彥杰
• Other Authors: WEN, YEAN-FU
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/486pq6

碩士 === 國立臺北大學 === 資訊管理研究所 === 106 === Due to the flourish in the field of computer science, several types of machines and control systems are continuously launched. People usually bring a lot of behaviors on these machines and systems. However, when using different types of machines and systems, different groups have different behaviors. It is an important topic for discussion of appropriately modeling the distribution of behaviors, expressing the patterns of behaviors in a simpler model, and generating representative samples. From the perspective on the merge strategies, this study proposes two algorithms, ‘One-Step Comparison’ and ‘Several-Times Merge’, which are the improvement on ALERGIA. In the ‘One-Step Comparison’ algorithm, we only compare the frequency of stay and the frequency of transition between two states, without comparing the following paths and states. In the ‘Several-Times Merge’ algorithm, we will compare the total numbers of states every time the ALERGIA finished. Once the numbers of states after running the algorithm are less than the numbers of states before running the algorithm, which means that some states are merged, then we will run the ALERGIA again until no states are merged. This study compares the numbers of states, running time, average distance, RMSE to evaluate the performances on ALERGIA, ‘One-Step Comparison’, ‘Several-Times Merge’. This research shows that ‘One-Step Comparison’ algorithm has significant improvement on shortening running time; Though ‘Several-Times Merge’ algorithm slightly increases running time compared to ALERGIA, it can effectively reduce the numbers of states with almost the same performances on average distance and RMSE, to express the behavior distribution with a simpler model.