2003 Fiscal Year Final Research Report Summary
STUDIES ON KNOWLEDGE DISCOVERY FROM TEMPORAL DATABASES
| Project/Area Number |
14580392
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
計算機科学
|
|---|
| Research Institution | Hosei University |
|---|
Principal Investigator |
MIURA Takao Hosei University, Dept. of Elect., Elect. & Camp. Engr., Prof., 工学部, 教授 (00219586)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
SHIOYA Isamu SANNO University, Dept. of Mgmt & Informatics, Prof., 経営情報学部, 教授 (70170850)
|
|---|
| Project Period (FY) |
2002 – 2003
|
| Keywords | Temporal database / Data Mining / Knowledge Extraction / Decision Trees / Self Organizing Map / Stream Clustering / Color Petrinet / OLAP |
|---|
| Research Abstract |
In the first year of this investigation, our main goals have been targeted for applying concept learning from databases to temporal domain to extract behavior knowledge of temporal databases and to establish modeling methodologies. Also we have examined fundamental features of meta modeling approach as well as abstraction/aggregation to obtain language framework. In the second year, we have paid much attention on data stream. In this work, we have proposed a new design methodology to extract temporal class from temporal logs. At the same time we have proposed incremental clustering technique by using regression analysis, thus temporal database schemes and data instances could be reconfigured automatically.
Of course we have to examine whether database scheme works well as classification criteria. We have extended Decision Trees for class hierarchy and pre-pruning. The former approach involves disjunctive classes to simplify schemes and path entropy to measure interestingness. A new pre-pruning technique has been also proposed. We have examined self-Organizing Map (SON) for classification as well as clustering. SOM is inherently different from decision trees because of unsupervised property which can be expected useful to data stream. In this work, we have proposed k-propagated property and applied to TaxSOM.
Two major applications have been discussed. One is classification based on Bayes theory for news-text by using polysemy and synonyms. Another one is information system design where several descriptions have been made in different era. Here we have discussed how to translate DFD description into UML.
|
