29 /342
29
臺大管理論叢
2017/5
第
27
卷第
2S
期
29-62
DOI:10.6226/NTUMR.2017.MAR.F104-007
為頻繁單變量不確定樣式產生摘要
Generating Summaries for Frequent Univariate Uncertain Pattern
摘 要
在巨量資料的研究與應用裡，從中發現有用的知識是一個重要的主題。大部份關於這
個主題的研究著眼於發展於巨量資料中擷取知識的方法，然而如何呈現所發現的知識
仍是一個關鍵的議題。在文獻中，單變量不確定資料是一種巨量資料，並且從單變量
不確定資料裡取出的頻繁單變量不確定樣式是一有用的知識。可是，頻繁單變量不確
定樣式的數量通常都非常龐大而不易被人們利用，因此我們需要一個好的頻繁單變量
不確定樣式的呈現方式。我們提出為頻繁單變量不確定樣式產生摘要的研究，其使用
階層群集技術去產生摘要。人們只需檢視數十或數百個代表樣式，而不需處理大量的
頻繁樣式。實驗顯示我們方法的摘要品質高於最大頻繁單變量不確定樣式所提供的摘
要品質。
【關鍵字】
巨量資料、單變量不確定資料、頻繁單變量不確定樣式、摘要
Abstract
In big data related research and applications, discovery of useful knowledge from big data is
an important topic. While most studies concerning this topic focus on developing methods
for retrieving knowledge, presentation of the discovered knowledge remains a critical issue.
A good method of presentation allows ordinary people to quickly understand and utilize the
discovered knowledge. In the literature, univariate uncertain data is one kind of big data, and
frequent patterns retrieved from univariate uncertain data, i.e., frequent univariate uncertain
patterns, represent useful knowledge. However, the number of frequent univariate uncertain
patterns is often too large to be dealt with by ordinary user. In other words, a good method of
presenting the discovered frequent univariate uncertain patterns is required. To this end, we
propose a novel way of summarizing frequent univariate uncertain patterns. We use a
hierarchical clustering technique to generate a summary of a set of frequent univariate
uncertain patterns. Instead of examining a large number of frequent univariate uncertain
patterns, a user only needs to check tens, or perhaps hundreds, of representative frequent
univariate uncertain patterns. Experimental results show that the summarization quality of
our method is better than the summarization quality of maximum frequent univariate
uncertain patterns.
【
Keywords
】
big data, univariate uncertain data, frequent univariate uncertain patterns,
summary
劉英和
/
國立東華大學資訊管理系副教授
Ying-Ho Liu
, Associate Professor, Department of Information Management, National Dong Hwa
University
Received, 2015/6, Final revision received 2016/3