近幾年來，越來越多具備自然語言處理能力的人工智慧代理人問世，比起其他用人來提供服務的方式，人工智慧代理人可以服務大量的請求，除此之外，在資料的更新以及維護上也更容易。在傳統的代理人服務上，大多都是用按鈕或是制式的回答問題，假如代理人沒有支援自然語言的能力，是無法提供使用者最好的使用體驗。此外，在商業用途中服務顧客的機器人，對於顧客所說的話需要具有準確的回應。所謂的準確的回應，指的是當機器人接收到顧客對話可以馬上知道使用者想表達的意圖以及希望得到的答案類型以及顧客對於句子中的語意實體的喜好。
本研究主要有兩點貢獻，第一點提出了一個意圖框架概念，針對了不同的意圖包含WH問句,YesNO問句以及一般的直述句與直述名詞進行分析並歸納出了各個意圖的意圖框架; 第二點，為了要找出語句中提出的實體意見目標與數值範圍，我們提出一個方法，藉由歸納語句中的意見影響因子，並根據語法規則建立出意見影響路徑，並針對意見影響路徑上的意見影響因子去推算語句中實體意見目標。

In recent years, more and more AI agents, based on natural language processing techniques, jump out to the market. Compared to human involved service, AI agents can respond huge requests at the same time and it can extend the knowledge more easily by adding knowledge into knowledge base. In conversational agents, they rely on the button and exactly slot filling. Without supporting natural language processing, the bot can’t provide the highest satisfaction to user, because button and exactly slot filling are not the approach which users get used to. Moreover, the AI agents for commercial purpose should be sensitive to the meaning of sentence. When an AI agent receives a sentence from customer, it must know the intent of customer, the response type user want to know and the interest to the entity in the sentence.
Hence, in this research we provide two main contributions. Firstly, we propose an intent frame concept which focuses on various intents such as 5W1H, Yes-No Question and statement to design their intent frame. Secondly, in order to find out the positive/native score of opinion target in the sentence, we propose an approach that can infer the positive/native score of opinion target by tracing the Sentiment path to record the opinion influenced factor.

[1] A. Sun and M. Hu., Query-guided event detection from news and blog streams. IEEE Transactions on Systems, Man, and Cybernetics, 2011.

[2] Jonathan Berant and Percy Liang, Imitation Learning of Agenda-based Semantic Parsers. Transactions of the Association for Computational Linguistics, 2015.

[3] Dipanjan Das, Desai Chen, André F. T. Martins, Nathan Schneider, and Noah A. Smith Frame-Semantic Parsing Computational Linguistics 40(1), March 2014.

[4] Yushi Wang, Jonathan Berant and Percy Liang. Building a Semantic Parser Overnight. Association for Computational Linguistics, 2012.

[5] Meghana Kshirsagar, Sam Thomson, and Nathan Schneider. Frame-Semantic Role Labeling with Heterogeneous Annotations. Association for Computational Linguistics, 2015.

[6] Convolutional Neural Networks for Sentence Classification. Yoon Kim The 2014 Conference on Empirical Methods on Natural Language Processing, 2014

[7] The application of semantic classification tree to Natural Language. R. Kuhn and R. De Mori. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 17, NO. 5, MAY 1995.

[8] Adding redundant features for CRFs-based sentence sentiment classification Honolulu, Hawaii. The 2008 Conference on Empirical Methods on Natural Language Processing, 2008

[9] Emotion Classification Using Web Blog Corpora. Changhua Yang, Kevin Hsin-Yih Lin and Hsin-Hsi Chen. IEEE/WIC/ACM International Conference on Web Intelligence 2007

[10] Baselines and Bigrams: Simple, Good Sentiment and Topic Classification. Sida Wang and Christopher D. Manning. Annual Meeting of the Association for Computational Linguistics 2012.

[11] Recurrent Neural Network for Text Classification with Multi-Task Learning. Pengfei Liu, Xipeng Qiu and Xuanjing Huang International Joint Conference on Artificial Intelligence, 2016

[12] Lei Zhang , Bing Liu. Aspect and Entity Extraction for Opinion Mining. Springer Berlin: Heidelberg 2014

[13] Min Song, Won Chul Kim, Dahee Lee, Go Eun Heo, Keun Young Kang. PKDE4J: Entity and relation extraction for public knowledge discovery. Journal of Biomedical Informatics, Volume 57, October 2015

[14] Jason P.C. Chiu, Eric Nichols. Sequential Labeling with Bidirectional LSTM-CNNs.
Transactions of the Association for Computational Linguistics, 2016

[15] Marie-Catherine de Marneffe and Christopher D. Manning. Stanford typed dependencies manual. California: Stanford, 2008

[16] Richard Johansson, Pierre Nugues. Dependency-based Semantic Role Labeling of PropBank. Association for Computing Machinery, 2008

[17] Andrea Esuli, Fabrizio Sebastiani. SENTIWORDNET: A Publicly Available Lexical Resource for Opinion Mining. The International Conference on Language Resources and Evaluation 2006.

[18] Ugan Yasavur, Jorge Travieso, Christine Lisetti, Naphtali Rishe. Opinion analysis Using Dependency Trees and Named-Entities. Proceedings of the Twenty-Seventh International Florida Artificial Intelligence Research Society Conference 2014.

[19] Poria S., Ofek N., Gelbukh A., Hussain A., Rokach L. (2014) Dependency Tree-Based Rules for Concept-Level Aspect-Based Opinion analysis. In: Presutti V. et al. Semantic Web Evaluation Challenge. SemWebEval 2014. Communications in Computer and Information Science, vol 475. Springer, Cham

[20] Minlie Huang, Yujie Cao, Chao Dong. Modeling Rich Contexts for Sentiment Classification with LSTM. arXiv:1605.01478, 2016.

[21] Sabine Buchholz, Erwin Marsi. CoNLL-X shared task on Multilingual Dependency Parsing. Annual Meeting of the Association for Computational Linguistics 2006.

[22] Felix A. Gers, Jurgen Schmidhuber, Fred Cummins. Learning to Forget: Continual Prediction with LSTM. Switzerland: Dalle Molle Institute for Artificial Intelligence Research. 1999

[23] Yoshua Bengil, Patrice Simard, and Paolo Frasconi. Learning long-term dependencies with gradient descent is difficult. IEEE Transaction on Neural Networks Journal Vol. 5, No 2. March 1994.

[24] Sepp Hochreiter and . Jurgen Schmidhuber. LONG SHORT-TERM MEMORY. Switzerland: Dalle Molle Institute for Artificial Intelligence Research. 1997

[25] Anthony Khoo, Yuval Marom and David Albrecht. Experiments with Sentence Classification. Proceedings of the 2006 Australasian Language Technology Workshop, 2006