異質序列資料之共識探勘模式

Mining Consensus Patterns Across Heterogeneous Sequence Databases

朱煥霖(Huan-Lin Chu)

指導教授 ： 莊詠婷 ； 黃正魁

國立中正大學/管理學院/資訊管理學系/碩士(2016年)

摘要

現今許多企業常蒐集大量的資料，如：顧客對商品的偏好或是購買商品的順序等，進而利用資料探勘的方式來找到消費者的需求，擬定策略以增加企業的競爭優勢。但是在大量資料中必定會出現意見衝突的情況，所以如何解決這樣意見衝突的問題，找出具有代表性的共識，便成為傳統的推薦系統與決策支援系統最重要的議題。
以現實生活最常接觸的兩個序列為例，1.偏好，u_1 : (A ≥ B ≻ C)，表示u_1認為(1)A的偏好有時比B高有時卻差不多；(2)B的偏好比C高。2.時間序列，u_1 : (A < B = C)，表示u_1認為(1)A的發生的順序比B早；(2)B和C同時發生。人們經常以單一面向思考決定偏好或時間順序，但在現實生活中，人們在決策時必須多面向的考量，才能避免策略失敗，如：股票的投資必須考量個人偏好、風險、甚至整個供應鏈的狀態，這些都可能影響到收益。基於以上幾點，本研究定義一個創新的模型結合排名序列與時間序列找出具有代表性的共識。
本研究收集兩種資料，包含期刊推薦與股票購買，以全國資訊管理學系所的老師及博士生與熟悉半導體類股票者為目標族群，發送983份問券，回收期刊推薦有效問卷為55份，股票購買有效問卷為51份，並利用本研究之時間偏好序列共識探勘模式演算法進行資料探勘。研究結果顯示，在不需要承擔風險的選擇上，人們會感性的擬定策略；而在需要承擔風險的選擇上，人們會理性的擬定策略，多面向的考量，以至於擬定出相似的策略，這也表現出人們在決策時風險規避的心態。

關鍵字

資料探勘 ； 異質資料 ； 共識序列 ； 時間序列

並列摘要

Many modern enterprise collect large quantities of data, such as customer preference or their temporal purchasing behaviors, and utilize data mining as their competitive advantages. However, some conflicts in the data may exist, and determining how to aggregate many different opinions into a consensus is a traditional core problem in recommendation systems and decision support systems. Taking the preference ranking problem as an example, u_1 : (A ≥ B ≻ C) indicates that for the user u_1, (1)A is at least favorable than B; (2)B is more favorable compared to C. Another temporal ranking problem is to discover the possible temporal relationships among items, which refer to the temporal ordering of items. For example, u_1 : (A < B = C) indicates that the user considers that item A should occur before B and item B can occur simultaneously with C. However, in the real world, user may have many different aspects of consideration in regards to the same itemset at the same time. The real-life application is that when investors purchase stocks, they consider not only the stock preference due to personal risk tolerance, but also the temporal order of stock investment to maximize the profit because of the temporal effects of supply chain positions. Based on the above ideas, this study defines a novel model and proposes its associated algorithm for discovering consensus patterns combining preference ranking and temporal sequence. A two-phase experiment was designed to collect authentic datasets, execute the algorithm for its effectiveness via user rating, and demonstrate its managerial meaning.

並列關鍵字

Data Mining ； Heterogeneous Database ； Consense Patterns ； Temporal Sequences

參考文獻

Hong, Tzung-Pei, Huang, Cheng-Ming, & Horng, Shi-Jinn. (2008). Linguistic object-oriented web-usage mining. International Journal of Approximate Reasoning, 48(1), 47-61.

Alonso, Sergio, Herrera-Viedma, Enrique, Chiclana, Francisco, & Herrera, Francisco. (2010). A web based consensus support system for group decision making problems and incomplete preferences. Information Sciences, 180(23), 4477-4495.

Chen, Shyi-Ming, Cheng, Shou-Hsiung, & Lin, Tsung-En. (2015). Group decision making systems using group recommendations based on interval fuzzy preference relations and consistency matrices. Information Sciences, 298, 555-567.

Chen, Yen-Liang, Chen, Shih-Sheng, & Hsu, Ping-Yu. (2002). Mining hybrid sequential patterns and sequential rules. Information Systems, 27(5), 345-362.

Chen, Yen-Liang, & Cheng, Li-Chen. (2009). Mining maximum consensus sequences from group ranking data. European Journal of Operational Research, 198(1), 241-251.