臺灣博碩士論文加值系統

面試在升學中是一種常見的管道，所以對學生而言面試是很重要的，然而真正尋求面試專家進行面試的人少之又少。練習面試最簡單的方式就是請專業人士擔任面試官，然而這需要花費大量的金錢和時間的配合，所以學生難以反覆訓練。本論文希望開發一個可重複訓練面試的面試對話系統，讓學生可以更容易地練習面試和技巧。
本論文之研究主題是追問句生成，在對話系統中，追問句生成可以將對話決策的決策動作(action)轉換成文字，藉此代表系統的回應。傳統的追問句生成使用模板或規則來生成語句，但必須事先定義模板或規則，耗時又耗力，而本論文可以透過詞分類的方式將語句自動轉換成模板，並可同時降低語句的複雜度，但當對話回合由多個模板組成時，我們必須找到可被追問的模板，本論文使用卷積張量神經網路(CNTN)，找出適當的可被追問的模板，再透過帶注意力的序列到序列模型(Seq2seq with attention)，學習模板與模板之間的關聯，藉此生成追問語句的模板，接著將追問語句的模板透過詞分類的詞彙填回模板中，產生各種候選追問語句，最後透過語言模型排序候選追問語句，把最適當的追問語句回應給使用者。
本論文共收集3390個可追問與追問組合，並採用五次交叉驗證進行實驗評估。從實驗結果顯示，本論文提出的方法與傳統方法，在客觀評估中，有較好的表現，在主觀評估方面也有明顯的差異。

Participation in an admission interview is a common, critical procedure before entering a school. However, students rarely seek professional help even if the best approach is to ask questions of an expert interviewer. To some degree, students lack interview experience and are unable to perform well in such important interviews. The reason for this may include both time and financial considerations.
In light of this, in this thesis, an interview system is developed that repeatedly trains a student with questions that frequently appear in interview situations. In the dialogue system, follow-up questions are able to transform policy into words as a representative system response. Traditional follow-up question generation focuses on generating sentences, which takes a significant amount of time in regard to defining the template and patterns as well as setting the rules.
Compared to the traditional method, word clustering is adopted here to automatically transform the sentences into patterns, which simultaneously decreases the complexity of the sentences. When answers are composed of numerous patterns, CNTN (convolution neural tensor network) is adopted to select the appropriate answer pattern for follow-up pattern generation.
As in the thesis, we apply the CNTN for tracing the proper pattern. In order to generate the pattern for a follow-up question, we also apply Seq2seq model to learn the relations between the patterns. To acquire the candidate follow-up questions, we then make the follow-up question patterns fill up with words using the Word Class Table. For the last steps, we use a Language Model to rank the candidate follow-up questions and choose the most suitable follow-up question as the response for an interviewee.
In this study, 3,390 answer and follow-up question pairs were collected. Five-fold cross validation was employed for evaluation. The results show that the proposed method performed better than the traditional method. Also, the results had positive value in terms of both objective and subjective assessment.

摘要 I
Abstract II
誌謝 IV
Contents V
List of Tables VII
List of Figures VIII
Chapter 1 Introduction 1
1.1 Background 1
1.2 Motivation 2
1.3 Goal 3
1.4 Literature Review 3
1.4.1 Natural Language Generation 3
1.4.2 Follow-up Question Generation 5
1.4.3 Word Clustering 6
1.4.4 Sequence-to-Sequence Model 8
1.5 Problems and Proposed Methods 9
1.6 Research Framework 11
Chapter 2 MHMC Interview Database 12
2.1 Data Collection 12
2.2 Corpus Introduction 12
Chapter 3 Proposed Methods 16
3.1 Word Clustering 17
3.2 Word Class Embedding 22
3.3 Sentence Selection 23
3.3.1 Convolutional Neural Tensor Network (CNTN) 24
3.3.2 Sentence Selection – CNTN 27
3.4 Follow-up Question Pattern Generation 28
3.4.1 Long Short-Term Memory 28
3.4.2 LSTM-based Sequence-to-Sequence with Attention 30
3.5 Word Filling 34
3.5.1 Template Matching 35
3.5.2 Word Filling 35
3.6 Language Model 37
Chapter 4 Experimental Results and Discussion 38
4.1 Effect of Hidden Node and Tensor Dimension for Sentence selection 38
4.2 Objective measure of follow-up question Generation 39
4.2.1 Pattern-based vs. Word-based in response generation 40
4.2.2 Effect of Number of Training Data on Pattern-based Response Generation 40
4.2.3 Template Matching Accuracy 41
4.2.4 Effect of Template Matching in Response Generation 42
4.3 Subjective measure of Response Generation 43
4.4 Discussion 45
Chapter 5 Conclusion and Future Work 47
5.1.1 Conclusion 47
5.1.2 Future Work 48
References 49

[1]J. Williams, A. Raux, and M. Henderson, The dialog state tracking challenge series: A review, Dialogue ＆ Discourse, vol. 7, no. 3, pp. 4-33, 2016.
[2](2016). 今年起大學個人申請最高占7成. Available: http://www.chinatimes.com/realtimenews/20160222005034-260405
[3]Palladian. Available: http://www.palladiancr.com/
[4]M. Agarwal, R. Shah, and P. Mannem, Automatic question generation using discourse cues, in Proceedings of the 6th Workshop on Innovative Use of NLP for Building Educational Applications, 2011, pp. 1-9.
[5]W. Chen, Aist, G., Mostow, J. Generating questions automatically from informational text, in Proceedings of the 2nd Workshop on Question Generation, held at the Conference on AI in, Education, 2009, pp. 17-24.
[6]Y. C. S. A. Hasan, Towards Automatic Topical Question Generation.
[7]S. S. Pradhan, W. Ward, K. Hacioglu, J. H. Martin, and D. Jurafsky, Shallow Semantic Parsing using Support Vector Machines, in HLT-NAACL, 2004, pp. 233-240.
[8]L. Ratinov and D. Roth, Design challenges and misconceptions in named entity recognition, in Proceedings of the Thirteenth Conference on Computational Natural Language Learning, 2009, pp. 147-155.
[9]A. H. Oh and A. I. Rudnicky, Stochastic language generation for spoken dialogue systems, in Proceedings of the 2000 ANLP/NAACL Workshop on Conversational systems-Volume 3, 2000, pp. 27-32.
[10]F. Mairesse and S. Young, Stochastic language generation in dialogue using factored language models, Computational Linguistics, 2014.
[11]T.-H. Wen, M. Gasic, D. Kim, N. Mrksic, P.-H. Su, D. Vandyke, et al., Stochastic language generation in dialogue using recurrent neural networks with convolutional sentence reranking, arXiv preprint arXiv:1508.01755, 2015.
[12]T.-H. Wen, M. Gasic, N. Mrksic, P.-H. Su, D. Vandyke, and S. Young, Semantically conditioned lstm-based natural language generation for spoken dialogue systems, arXiv preprint arXiv:1508.01745, 2015.
[13]Vibhu O. Mittal Johanna D. Moore , Dynamic Generation of Follow up Question Menus: Facilitating Interactive Natural Language Dialogues, CHI '95 Proceedings of the SIGCHI Conference on Human Factors in Computing Systems Pages 90-97.
[14]Xiaohua Guo, Xinhua Zhu, Fei Li and Qi Li, A New Semantic Similarity Measurement Based on HowNet Concept, International Forum on Management, Education and Information Technology Application (IFMEITA 2016).
[15]Rui Zheng, Huan Zhao, Xixiang Zhang, A Word Similarity Algorithm with Sememe Probability Density Ratio Based on HowNet, International Journal of Hybrid .Information Technology Vol.8, No.10 (2015), pp.417-426.
[16]Hang Li and Naoki Abe, Word Clustering and Disambiguation Based on Co-occurrence Data, Proceeding COLING '98 Proceedings of the 17th internation conference on Computational linguistics-vol 2 Pages 749-755.
[17]Tokunaga Takenobu, Iwayama Makoto, Tanaka Hozumi, Automatic Thesaurus Construction based on Grammatical Relations, Proceeding IJCAI '95 Proceedings of the 14th international joint conference on Artifical intellignece -vol 2 Pages 1308-1313.
[18]LUO Jie, WANG Qinglin, LI Yuan, Word Clustering based on Word2vec and Semantic Similarity, Proceeding of the 33rd Chinese Control Conference 2014.
[19]Shan Feng, Ruifang Liu, Qinlong Wang, Ruisheng Shi, Word Distribution Representation Based Text Clustering, Proceeding of CCIS2014.
[20]Qiu Lin, Xu Jungang, A Chinese Word Clustering Method Using Latent Dirichlet Allocation and K-means, 2nd International Conference on Advances in Computer Science and Engineering (CSE 2013).
[21]Nikoletta Bassiou and Constantine Kotropoulos, Word Clustering using PLSA enhanced with Long Distance Bigrams, 2010 International Conference on Pattern Recognition.
[22]Sabir Ismail, M. Shahidur Rahman, Bangla Word Clustering Based on N-gram Language Model, International Conference on Electrical Engineering and Information ＆ Communication Technology (ICEEICT) 2014 .
[23]Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Fethi Bougares, Dzmitry Bahdanau, Holger Schwenk, Yoshua Bengio, Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation, arXiv:1406.1078v3 [cs.CL] 3 Sep 2014.
[24]Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Fethi Bougares, Dzmitry Bahdanau, Holger Schwenk, Yoshua Bengio, Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation, arXiv:1406.1078v3 [cs.CL] 3 Sep 2014.
[25]Dzmitry Bahdanau, KyungHyun Cho, Yoshua Bengio, NEURAL MACHINE TRANSLATION BY JOINTLY LEARNING TO ALIGN AND TRANSLATE, arXiv:1409.0473v7 [cs.CL] 19 May 2016.
[26]X. Qiu and X. Huang, Convolutional Neural Tensor Network Architecture for Community-Based Question Answering, in IJCAI, 2015, pp. 1305-1311.
[27]R. Socher, D. Chen, C. D. Manning, and A. Ng, Reasoning with neural tensor networks for knowledge base completion, in Advances in neural information processing systems, 2013, pp. 926-934.
[28]S. Hochreiter and J. Schmidhuber, Long short-term memory, Neural computation, vol. 9, pp. 1735-1780, 1997.
[29]C. Xing, W. Wu, Y. Wu, J. Liu, Y. Huang, M. Zhou, W. Ma,Topic Augmented
Neural Response Generation with a Joint Attention Mechanism,
arXiv:1606.08340v1 [cs.CL] 21 Jun 2016.