判斷、實體名詞覆蓋率、數字覆蓋率和詞彙相依相似度並藉由 Linearly Weighted Functions 建構出推論的模型，我們將各項特徵的參數根據準確率的變化調整其 參數值、最後將最佳準確率的組合視為最後的參數組合，在訓練語料中我們獲得 約 60%左右的準確率。
而透過上述的建構我們參加 NTCIR-RITE VAL 子任務，並透過訓練語料中 其中三組較佳的參數組合，作為我們參賽的選擇，在測試語料中我們獲得 51%
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
61
在分類的部分，除了採用原先的 Linearly Weighted Functions 方法外，未來也可以 嘗試使用其他機器學習的分類方法，例如使用支持向量機器(Support Vector Machine)[22]、決策樹(Decision Tree)[17]等等的方法，交互比對其優劣，藉以選 擇出最適合此推論系統的分類器。
我們也希望透過此推論系統的技術，未來可以延伸到其他相關的應用，例如 自動問答系統，可實際應用在客服系統的自動化，或者自動答題系統，可幫助在 教學上驗證題目的難易，藉以提升教學的品質，也可在自動摘要上使用推論的技 術，以及某些關於自然語言的相關應用，都可能有直接或間接的幫助。
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
62
參考文獻
[1] R. Adams, “Textual Entailment Through Extended Lexical Overlap,”
Proceedings of the Second PASCAL Challenges Workshop on Recognising Textual Entailment, pp. 128-133, 2006.
[2] BLEU, http://en.wikipedia.org/wiki/BLEU
[3] A. Budanitsky and G. Hirst, Semantic distance in WordNet: An experimental, application-oriented evaluation of five measures, Workshop on WordNet and Other Lexical Resources, Second Meeting of the North American Chapter of the Association for Computational Linguistics, Pittsburgh, Pennsylvania, USA, 2001.
[4] S. Cohen and N. Or, "A general algorithm for subtree similarity-search," Data Engineering (ICDE), IEEE 30th International Conference. pp. 928-939, 2014.
[5] Grid search, http://scikit-learn.org/stable/modules/grid_search.html
[6] S. Hattori and S. Sato, “Team SKL’s Strategy and Experience in RITE2,”
Proceedings of the 10th NTCIR Conference, pp. 435-442, 2013.
[7] A. Hickl, J. Bensley, J. Williams, K. Roberts, B. Rink, and Y. Shi, “Recognizing Textual Entailment with LCC’s GROUNDHOG System,” Proceedings of the Second PASCAL Challenges Workshop on Recognising Textual Entailment, pp.
80-85, 2006.
[8] Heuristic function, http://en.wikipedia.org/wiki/Heuristic_function
‧
Syntactic, and Semantic Features for the RITE Tasks,” Proceedings of the 10th NTCIR Conference, pp. 430-434, 2013.[10] G. Li, X. Liu, J. Feng, and L. Zhou, “Efficient Similarity Search for
Tree-Structured Data, Author Affiliations: Department of Computer Science and Technology,” Proceedings of the 20th Scientific and Statistical Database
Management Conference, pp. 131-149, 2008.
[11] Linearly Weighted Functions, http://en.wikipedia.org/wiki/Weight_function [12] Longest Common Strings,
http://en.wikipedia.org/wiki/Longest_common_substring_problem
[13] Lucene, http://lucene.apache.org/core/
[14] Named Entity Recognition,
http://alias-i.com/lingpipe/demos/tutorial/ne/read-me.html
[15] NTCIR RITE-VAL, http://research.nii.ac.jp/ntcir/index-en.html [16] RTE, http://research.microsoft.com/en-us/groups/nlp/rte.aspx
[17] S. Rasoul and D. Landgrebe, “A Survey of Decision Tree Classifier Methodology,” IEEE Transactions on Systems, Man, and Cybernetics, Vol. 21, No. 3, pp 660-674, May 1991.
[18] Stanford Corenlp , http://nlp.stanford.edu/software/corenlp.shtml [19] Stanford Named Entity Recognizer,
http://www-nlp.stanford.edu/software/CRF-NER.shtml
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
64
[20] Stanford Parser, http://nlp.stanford.edu/software/lex-parser.shtml [21] Stanford Typed Dependencies,
http://nlp.stanford.edu/software/stanford-dependencies.shtml
[22] SVM, http://en.wikipedia.org/wiki/Support_vector_machine
[23] Textual Entailment , http://en.wikipedia.org/wiki/Textual_entailment [24] Total commander, http://www.ghisler.com/
[25] Wikipedia, http://en.wikipedia.org/wiki/Main_Page [26] WordNet, http://wordnet.princeton.edu/
[27] S.-H. Wu, S.-S. Yang, L.-P. Chen, H.-S. Chiu, and R.-D. Yang, “CYUT Chinese Textual Entailment Recognition System for NTCIR-10 RITE-2.” Proceedings of the 10th NTCIR Conference, pp. 443-448, 2013.
[28] S.-H. Wu, W.-C. Huang, L.-P. Chen, and T. Ku, “Binary-class and Multi-class Chinese Textural Entailment System Description in NTCIR-9 RITE,”
Proceedings of the 9th NTCIR Conference, pp. 422-426, 2011.
[29] Y. Y. Zhang, J. Xu, C.-L. Liu, X.-L. Wang, R.-F. Xu, Q.-C. Chen, X. Wang, Y.-S.
Hou, and B. Tang, “ICRC_HITSZ at RITE: Leveraging Multiple Classifiers Voting for Textual Entailment Recognition,” Proceedings of the 9th NTCIR Conference, pp. 325-329, 2011.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
65
附錄 相關文章與相關句範例
在附錄 I 中，我們以第六章實驗為範本，以第 19 道題目 Cola is popular in the early 1970s.為範例，展示出依據該論述句的語文資訊，所擷取出的相關文章如下所 示，礙於版面有限，僅列出標題及維基百科網址。
維基百科文章: Coca-Cola
網址：http://en.wikipedia.org/wiki/Coca-Cola 維基百科文章: Inca Kola
網址：http://en.wikipedia.org/wiki/Inca_Kola 維基百科文章: Caffeine
網址：http://en.wikipedia.org/wiki/Caffeine 維基百科文章: Cola
網址：http://en.wikipedia.org/wiki/Cola 維基百科文章: Cuba Libre
網址：http://en.wikipedia.org/wiki/Cuba_Libre
在相關句中，我們以第六章實驗為範本，以第 19 道題目 Cola is popular in the early 1970s.為範例，展示出依據該論述句所擷取的相關文章中，透過詞彙覆蓋率的排 名所擷取出的 15 個相關句如下所示：
‧
1. Clear cola is a colorless variety of cola popular in the early 1990s
2. In the early 20th century a fatwa was created in Egypt to discuss the question of whether Muslims were permitted to drink Coca-Cola and Pepsi cola
3. Trace flavorings may include nutmeg and a wide variety of ingredients but the base flavorings that most people identify with a cola taste remain vanilla and cinnamon
4. In the Netherlands the drink is usually served without lime and commonly referred to as Baco from the two ingredients of Bacardi rum and cola
5. In Mexico it is one of the most popular alcoholic drinks and it is usually referred to simply as a Cuba
6. Campa Cola was India s most popular brand prior to the introduction of Coca-cola and Pepsi to the Indian market in 1991
7. A variety of different sweeteners may be added to cola often partly dependent on local agricultural policy
8. In the Dominican Republic it is a popular drink poured with a generous amount of locally produced Dominican Rum and cola topped off with a slice of lime
9. Jarritos Cola is a brand of cola from Mexico while popular and native to Mexico it is widely distributed mainly to Latino citizens of the United States
10. Many of these early television commercials for Coca-Cola featured movie stars sports heroes and popular singers
11. Zam Zam Cola popular in Iran and parts of the Arab world
12. In Greece Thessaloniki there is another variant that consists of retsina and cola
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
67
named tumba libre
13. Virgin Cola was popular in South Africa and Western Europe in the 1990s but has waned in availability
14. Coca Cola is also one of the associate sponsor of Delhi Daredevils in Indian Premier League receding Wikipedia
15. It has tried to maintain the exclusive right to sell products using the Coca-Cola name and its diminutive form Coke by suggesting the alternative of cola drink as a generic name for similar types of carbonated soft drinks