臺灣博碩士論文加值系統

隨著人口結構的改變與高齡社會的來臨，為年長者提供的服務或裝置也越來越受到重視，根據研究顯示孤獨感和寂寞感會加速老年人的失能情況。為了增加年長者的情感交流，本論文打造了一個情境感知對話機器人陪年長者對話，並藉此降低他們的寂寞感。
過去有不少學者提出對話系統，但是多是利用社群媒體或網路語料，用詞與居家對話有相當大差異，也缺乏情境的考量。因此本論文收集了口語對話語料，希望系統更貼近日常對話，因為口語句變化性很高，在前處理部份我們先將所有的語句轉換為模板句，以涵蓋更多語句變化，接著利用LSTM-based多層特徵轉換模型和情境追蹤模型來擷取字對字、句對句的單回合語意向量以及跨回合語句之間的語意濃縮作為情境感知向量。最後利用神經張量網路(Neural Tensor Network, NTN)做回應句挑選，並將模板填充為最後的回應句。
本論文所使用的MHMC聊天對話語料庫一共由40人收集而成，其中包含65場主題對話，以及2239組信息-回應組合，並採用5次交叉驗證做實驗評估。實驗結果顯示本論文提出之系統可達到69.9%的辨識率，相較於Okapi有顯著的提升。

According to demographic changes, the services designed for elderly are more needed and important. Previous studies show that there is a high relationship between loneliness, gloom, and disability for elderly. In order to increase the affective interaction and reduce the loneliness and isolation of elderly, this thesis develops a context-aware chatbot system for elderly care.
In previous work, social media or community-based question-answer data were generally used to build the chatbots. This kind of data is different from our daily conversation and lack of context information. For elderly care, the chatbot system should respond to the users with the sentences considering the causal context. Therefore, we collected the MHMC chatting dataset from daily communication. Since people are free to say anything to the system, the collected sentences are converted into patterns in the pre-processing part to cover the variability of spoken language. Then, an LSTM-based multi-layer embedding model is used to embed the semantic information between words and sentences in a single turn for context tracking. Finally, the Neural Tensor Network (NTN) is employed to select a proper response pattern, which will be further filled with suitable words based on a rule-based method as the response to the elderly.
For performance evaluation, this study collected an MHMC chatting dataset, consisting of 65 topic-based dialogues and 2239 message-response pairs, from 40 subjects. Five-fold cross-validation scheme was employed for training and evaluation. Experimental results show that the proposed method achieved an accuracy of 69.8%, which outperformed the traditional Okapi model.

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 and Goal.......................................................................................... 2
1.3 Literature Review.............................................................................................. 3
1.3.1 Question Answering and Dialogue System........................................... 3
1.3.2 Chatbot System ..................................................................................... 5
1.3.3 Neural Sentence Modeling.................................................................... 6
1.3.4 Response Selection................................................................................ 8
1.4 Problems and Proposed Ideas.......................................................................... 10
1.5 Research Framework....................................................................................... 12
Chapter 2 Database Design and Collection ............................................................. 14
2.1 MHMC Chatting Database.............................................................................. 14
2.1.1 Single-turn Dialogue Database............................................................ 14
2.1.2 Topic-based Dialogue Database .......................................................... 14
2.2 Database .......................................................................................................... 16
2.2.1 Patternation.......................................................................................... 16
2.2.2 Categorization ..................................................................................... 17
Chapter 3 Proposed Method..................................................................................... 18
3.1 Pre-process ...................................................................................................... 19
3.1.1 Chinese Sentence/Word Segmentation................................................ 19
3.1.2 Sentence to Pattern .............................................................................. 20
3.1.3 Word Embedding................................................................................. 21
3.2 Turn Embedding.............................................................................................. 23
VI
3.2.1 TF-IDF ................................................................................................ 23
3.2.2 Long Short-Term Memory .................................................................. 24
3.3 Context Tracking............................................................................................. 28
3.3.1 Data Process ........................................................................................ 28
3.3.2 Model Training.................................................................................... 28
3.4 Response Selection.......................................................................................... 29
3.4.1 Neural Tensor Network ....................................................................... 30
3.4.2 Message-Message Matching (MM) .................................................... 31
3.4.3 Message-Response Matching (MR) .................................................... 32
3.4.4 Pattern Filling...................................................................................... 33
Chapter 4 Experimental Results and Discussion ..................................................... 34
4.1 Experimental Setup ......................................................................................... 34
4.1.1 Epoch Setting ...................................................................................... 34
4.1.2 Training Data for NTN Model ............................................................ 35
4.1.3 Baseline ............................................................................................... 36
4.1.4 Effect of LSTM Hidden Size............................................................... 37
4.2 Evaluation of Turn Embedding ....................................................................... 38
4.3 Message-Message Matching (MM) ................................................................ 40
4.3.1 Representative Sentence Selection...................................................... 41
4.3.2 Experimental results............................................................................ 42
4.4 Message-Response Matching (MR) ................................................................ 47
4.4.1 MR....................................................................................................... 47
4.4.2 MM+MR............................................................................................. 48
4.4.3 Evaluation of Context Tracking .......................................................... 49
4.5 Discussion ....................................................................................................... 49
Chapter 5 Conclusion and Future Work................................................................... 51
5.1 Conclusion....................................................................................................... 51
5.2 Future Work..................................................................................................... 51
Reference..................................................................................................................... 53

[1](106年). 內政統計通報 106年第三週.
[2](105). 中華民國人口推估 (105 至 150 年).
[3](2017). 長照十年計畫2.0. Available: http://www.digi.ey.gov.tw/hot_topic.aspx?n=A1C2B2C174E64DE7＆sms=AB6812391DC74DB8
[4]A. Karki, Loneliness among elderly women: A literature review, 2009.
[5]E. M. Voorhees, The TREC-8 Question Answering Track Report, in Trec, 1999, vol. 99, pp. 77-82.
[6]Y. Sasaki, Overview of the NTCIR-5 Cross-Lingual Question Answering Task (CLQA1), December 6-9 2005.
[7]M. Vargas-Vera and M. D. Lytras, AQUA: A Closed-Domain Question Answering System, Information Systems Management, vol. 27, no. 3, pp. 217-225, 2010/07/16 2010.
[8]E. Cabrio, J. Cojan, A. P. Aprosio, B. Magnini, A. Lavelli, and F. Gandon, QAKiS: an open domain QA system based on relational patterns, in Proceedings of the 2012th International Conference on Posters ＆ Demonstrations Track-Volume 914, 2012, pp. 9-12: CEUR-WS. org.
[9]S. Quarteroni and S. Manandhar, Designing an interactive open-domain question answering system, Natural Language Engineering, vol. 15, no. 01, pp. 73-95, 2009.
[10]S. Auer, C. Bizer, G. Kobilarov, J. Lehmann, R. Cyganiak, and Z. Ives, Dbpedia: A nucleus for a web of open data, The semantic web, pp. 722-735, 2007.
[11]K. Bollacker, C. Evans, P. Paritosh, T. Sturge, and J. Taylor, Freebase: a collaboratively created graph database for structuring human knowledge, in Proceedings of the 2008 ACM SIGMOD international conference on Management of data, 2008, pp. 1247-1250: AcM.
[12]S. He, K. Liu, Y. Zhang, L. Xu, and J. Zhao, Question Answering over Linked Data Using First-order Logic, in EMNLP, 2014, pp. 1092-1103.
[13]J. Berant, A. Chou, R. Frostig, and P. Liang, Semantic Parsing on Freebase from Question-Answer Pairs.
[14]R. Higashinaka et al., Towards an open-domain conversational system fully based on natural language processing, in COLING, 2014, pp. 928-939.
[15]H. Sugiyama, T. Meguro, R. Higashinaka, and Y. Minami, Open-domain utterance generation for conversational dialogue systems using web-scale dependency structures, in Proc. SIGDIAL, 2013, pp. 334-338.
[16]O. Vinyals and Q. Le, A neural conversational model, arXiv preprint arXiv:1506.05869, 2015.
[17]B. Hu, Z. Lu, H. Li, and Q. Chen, Convolutional neural network architectures for matching natural language sentences, in Advances in neural information processing systems, 2014, pp. 2042-2050.
[18]Y. Wu, W. Wu, Z. Li, and M. Zhou, Response Selection with Topic Clues for Retrieval-based Chatbots, arXiv preprint arXiv:1605.00090, 2016.
[19]S. Hochreiter and J. Schmidhuber, Long short-term memory, Neural computation, vol. 9, no. 8, pp. 1735-1780, 1997.
[20]B. Liu, M. Huang, S. Liu, X. Zhu, and X. Zhu, A Sentence Interaction Network for Modeling Dependence between Sentences.
[21]R. S. Wallace, The anatomy of ALICE, Parsing the Turing Test, pp. 181-210, 2009.
[22]R. E. Banchs and H. Li, IRIS: a chat-oriented dialogue system based on the vector space model, in Proceedings of the ACL 2012 System Demonstrations, 2012, pp. 37-42: Association for Computational Linguistics.
[23]S. Jafarpour, C. Burges, and A. Ritter, Filter, rank, and transfer the knowledge: Learning to chat, Advances in Ranking, vol. 10, 2010.
[24]G. Salton, A. Wong, and C.-S. Yang, A vector space model for automatic indexing, Communications of the ACM, vol. 18, no. 11, pp. 613-620, 1975.
[25]W.-Y. Ma and K.-J. Chen, Introduction to CKIP Chinese word segmentation system for the first international Chinese Word Segmentation Bakeoff, in Proceedings of the second SIGHAN workshop on Chinese language processing-Volume 17, 2003, pp. 168-171: Association for Computational Linguistics.
[26]J. Sun, ‘Jieba’Chinese word segmentation tool, 2012.
[27]J. Pennington, R. Socher, and C. D. Manning, Glove: Global vectors for word representation, in EMNLP, 2014, vol. 14, pp. 1532-1543.