本研究提出一個基於詞嵌入實現之槽填充之對話系統，利用詞嵌入模型將輸入語句之詞作詞向量化後，將槽填充所缺乏之資訊用餘弦相似度做匹配，再進行病理之推估，最後將最適合之商品回傳給使用者。將輸入語音透過ASR轉換成文字後，透過Jieba斷詞系統進行斷詞，接著利用已經預處理之詞嵌入模型進行向量化。由於使用者意圖和意圖所需槽填充之資訊皆以建在系統內部，包含了診斷與購買兩個意圖，利用餘弦相似度將各個詞做比對，就能得到所需之資訊，本篇提出有別於使用遞歸神經網路實現槽填充，而是使用詞嵌入模型和餘弦相似度去實現，解決了遞歸神經網路使用上需要大量資料庫作以訓練之文本，且在擴充與泛化性，更勝遞歸神經網路，在不同之領域使用不用像遞歸神經網路重新用新的資料庫去訓練，只需建造新的意圖文本即可。槽填充取出所需之資訊之後，借鏡了詞頻與反詞頻演算法之概念，訓練了資料庫中的症狀及疾病之權重，越容易在生活中出現之疾病權重越高，越多疾病有此種症狀，則該症狀之權重越低，在擁有了症狀及疾病之權重，即可得之目前最有可能之疾病為何，最後測試槽填充與診斷模擬之實驗結果準確率分別為88%與86%，成功得到正確疾病後去藥品與疾病關聯資料庫將符合此項疾病之藥品取出，填入模板句子，合成成回應句，藉由語音合成並且輸出。

We proposed a medical dialogue system based on word embedding and slot filling. We vectorized the words of the input sentence with the word embedding model, extract information with slot filling based on the cosine similarity, and then the diagnostic reasoning simulation is performed. Finally, the most suitable product is returned to the user. After the input speech is converted into text through ASR, the sentences are cut into words through the Jieba word segmentation system, and then vectorize the word based on word embedding. Since the format of user’s intent and relative slots of intent were constructed in the slot and intent corpus, including the two intents: “diagnosis” and “product selection”. We apply the cosine similarity to compare the words between input sentence and corpus, and the required information of slot can be obtained. Different from using recursive neural network to implement slot filling, we use word embedding model and cosine similarity which solves the problem that recursive neural network requires a huge number of databases for training corpus, and our approach is flexible and scalable. It is better than recursive neural networks. In different domain of dialogue system, we don't need to collect another database to train model like a recurrent neural network, just build new intent and slot corpus. After filling all the required information, we adopt the concept of TF-IDF algorithm to train the weight of the symptoms and diseases in the database. The more common the disease, the higher the weight. As for symptom, the more the disease has this symptom, the lower the weight of the symptom. After knowing the weight of disease and symptom, we can start to calculate the score of disease and get the most likely disease. In the experimental result, the accuracy of slot filling and diagnostic reasoning simulation was 88% and 86% respectively. After successfully obtaining the correct disease, system will search medical product and related disease database to find out the medical product that meets the user need. Then system will fill all the blank in the template sentence, synthesize sentence into a speech sentence, return to user.

[1] J. Liu, J. Wang, and C. Wang, "Spoken Language Understanding in Dialog Systems for Olympic Game Information," in 2006 4th IEEE International Conference on Industrial Informatics, 2006, pp. 1042-1045.
[2] Z. Yan, N. Duan, P. Chen, M. Zhou, J. Zhou, and Z. Li, "Building Task-Oriented Dialogue Systems for Online Shopping," in AAAI, 2017, pp. 4618-4626.
[3] P.-H. Su et al., "On-line active reward learning for policy optimisation in spoken dialogue systems," arXiv preprint arXiv:1605.07669, 2016.
[4] 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.
[5] M. Henderson, B. Thomson, and J. D. Williams, "The second dialog state tracking challenge," in Proceedings of the 15th Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGDIAL), 2014, pp. 263-272.
[6] T. Mikolov, K. Chen, G. Corrado, and J. Dean, "Efficient estimation of word representations in vector space," arXiv preprint arXiv:1301.3781, 2013.
[7] X. Guan, Q. Peng, J. Zhang, and X. Zhang, "Renovating word vectors to build Chinese sentiment lexicon," in 2015 IEEE International Conference on Information and Automation, 2015, pp. 2977-2982.
[8] J. Allen, Natural language understanding. Pearson, 1995.
[9] D. Goddeau, H. Meng, J. Polifroni, S. Seneff, and S. Busayapongchai, "A form-based dialogue manager for spoken language applications," in Spoken Language, 1996. ICSLP 96. Proceedings., Fourth International Conference on, 1996, vol. 2, pp. 701-704: IEEE.
[10] E. Reiter and R. Dale, Building natural language generation systems. Cambridge university press, 2000.
[11] J. Sun, "‘Jieba’Chinese word segmentation tool," ed, 2012.
[12] C. Silva and B. Ribeiro, "The importance of stop word removal on recall values in text categorization," in Neural Networks, 2003. Proceedings of the International Joint Conference on, 2003, vol. 3, pp. 1661-1666: IEEE.
[13] I. Abdelaziz, A. Fokoue, O. Hassanzadeh, P. Zhang, and M. Sadoghi, "Large-scale structural and textual similarity-based mining of knowledge graph to predict drug–drug interactions," Web Semantics: Science, Services and Agents on the World Wide Web, vol. 44, pp. 104-117, 2017/05/01/ 2017.
[14] T. Mikolov, I. Sutskever, K. Chen, G. S. Corrado, and J. Dean, "Distributed representations of words and phrases and their compositionality," in Advances in neural information processing systems, 2013, pp. 3111-3119.
[15] H. Huang, Y. Wang, C. Feng, Z. Liu, and Q. Zhou, "Leveraging Conceptualization for Short-Text Embedding," IEEE Transactions on Knowledge and Data Engineering, vol. 30, no. 7, pp. 1282-1295, 2018.
[16] D. J. Brenes, D. Gayo-Avello, and K. Pérez-González, "Survey and evaluation of query intent detection methods," in Proceedings of the 2009 workshop on Web Search Click Data, 2009, pp. 1-7: ACM.
[17] J. K. Kim, G. Tur, A. Celikyilmaz, B. Cao, and Y. Y. Wang, "Intent detection using semantically enriched word embeddings," in 2016 IEEE Spoken Language Technology Workshop (SLT), 2016, pp. 414-419.
[18] M. Steinbach, G. Karypis, and V. Kumar, "A comparison of document clustering techniques," in KDD workshop on text mining, 2000, vol. 400, no. 1, pp. 525-526: Boston.
[19] G. Mesnil et al., "Using recurrent neural networks for slot filling in spoken language understanding," IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 23, no. 3, pp. 530-539, 2015.
[20] J. Ramos, "Using tf-idf to determine word relevance in document queries," in Proceedings of the first instructional conference on machine learning, 2003, vol. 242, pp. 133-142.
[21] R. Davis, "Diagnostic reasoning based on structure and behavior," Artificial Intelligence, vol. 24, no. 1, pp. 347-410, 1984/12/01/ 1984.
[22] 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.
[23] W. B. Frakes and R. Baeza-Yates, Information retrieval: Data structures & algorithms. prentice Hall Englewood Cliffs, NJ, 1992.
[24] G. Salton and M. J. McGill, "Introduction to modern information retrieval," 1986.
[25] A. Raux and M. Eskenazi, "A finite-state turn-taking model for spoken dialog systems," in Proceedings of Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the Association for Computational Linguistics, 2009, pp. 629-637: Association for Computational Linguistics.
[26] E. D. Brill and M. R. Richardson, "Table querying," ed: Google Patents, 2010.
[27] T. Dutoit, An introduction to text-to-speech synthesis. Springer Science & Business Media, 1997.