Efficient Mining Methods for Latent Association Rules and their Application for Generating Latent Event Sequence Corpora

2016 Fiscal Year Final Research Report

• Project/Area Number: 25330256
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Intelligent informatics
• Research Institution: University of Yamanashi
• Principal Investigator: IWANUMA Koji 山梨大学, 総合研究部, 教授 (30176557)
• Project Period (FY): 2013-04-01 – 2017-03-31
• Keywords: データマイニング / 負の相関ルール / 極小生成子 / 飽和アイテム集合 / オンライン型アルゴリズム / データストリーム / 潜在因子 / 無損失圧縮

Outline of Final Research Achievements

In this research, we developed several efficient algorithms for negative association rule mining, which can be regraded as a concrete form of a latent association mining. We also studied some online approximation algorithms for a huge transaction stream, which is an essential tool for generating latent event sequence corpora from a very large sequential text data such as newspaper data. The details are as follows: First, we proposed a new efficient top-down search algorithm for valid negative association rules, which uses a suffix tree over frequent itemsets. Second, we studied lossless compression for a set of negative rules, and gave a novel lossless compression method based on minimal generators. Third, we developed two online approximation algorithms for mining a huge transaction stream: one achieves a resource-oriented computation, and another uses an incremental intersection computation for frequent closed itemsets, both of which can avoid the combinatorial explosion phenomena.

Free Research Field

人工知能基礎