由於近年來越來越多的檢索形式不再侷限於關鍵字搜索，而是以完整的多個句子為基礎進行檢索，比如在實證醫學中的相關研究檢索。為了以句子進行檢索，多個研究模型擷取上下文資訊，意圖在建構每個字的詞向量時，考慮到所有會影響到該字意義的詞彙，進而學習到該字的語意概念。但由於這些上下文資訊大多是自動學習而來的，很可能會考慮到在意義上無關聯或忽略在文法結構上相互關聯的詞彙。由於文法結構能夠指明在意義上相關聯的詞彙，因此，在本論文中，為了取得文中的語意概念，我們以文法結構為本，建造在文法上相關聯的詞彙形態去取得。使用這些語意概念進行檢索和文件之間的匹配，並建構一個匹配信號，和現存模型結合，進而強調檢索與文件內意義上相關聯詞彙的匹配程度。實驗結果顯示，藉由結合論文中的匹配信號，以彌補現存模型的缺陷，我們可以在模型精確度及相關文件檢索率上贏過現存的檢索模型。

In recent years, fewer and fewer queries utilize keywords for information retrieval; instead, full sentences are used for searching, for instance, with the relevant document retrieval task in Evidence-Based Medicine (EBM). For the purpose of searching with sentences, many of the previous models capture the contextualized information, intending to consider all of the terms which have a connection with the meaning of each term in the same sentence while constructing the word vector of each term. However, for most of the works contextualized information is learnt automatically. As there is possibility that the contextualized information considers the terms that are not semantically related or ignores the terms that are grammatically related. Inherent within the reason that grammar relation can help indicate which terms are semantically related in medical data, we extracted the semantic concepts in text based on grammar relation. In this thesis, to extract the semantic concepts in the text, we constructed contextualized patterns which record the grammar relation between terms. Contextualized patterns are then utilized to measure the relevant degree between a document and a query using a matching signal. This matching signal is thereafter combined with existing IR models to emphasize the matching degree between the semantically related terms of a query and those of a document. The experimental results show that after combining the existing model with our matching signal, we are able to outperform the existing IR models on model accuracy and relevant document retrieval evaluations.

中文摘要
Abstract
目錄
Introduction..............1
Related Work..............5
Term-based model..............5
Context-based models..............7
Data Collection..............10
Methodology..............12
Overview..............12
Contextualized Patterns Construction..............14
Grammatically Dependent Terms Extraction..............15
Pattern Candidates..............17
Matching Scores Construction..............24
Matching Signal..............30
Model and Matching Signal Combination..............32
Experiments..............33
Experimental Setup..............33
Data Preprocessing..............33
Training Data..............34
Baseline Methods..............34
Evaluation Methodology..............36
Experimental results and Discussion..............39
Conclusion and Future Work..............42
Reference..............44

[1] Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473, 2014.
[2] Haolan Chen, Fred X Han, Di Niu, Dong Liu, Kunfeng Lai, Chenglin Wu, and Yu Xu. Mix: Multi-channel information crossing for text matching. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Min-ing, pages 110–119, 2018.
[3] Zhuyun Dai, Chenyan Xiong, Jamie Callan, and Zhiyuan Liu. Convolutional neural networks for soft-matching n-grams in ad-hoc search. In Proceedings of the eleventh ACM international conference on web search and data mining, pages 126–134, 2018.
[4] Paul N Gorman, Joan Ash, and Leslie Wykoff. Can primary care physicians’ questions be answered using the medical journal literature? Bulletin of the Medical Library Association, 82(2):140, 1994.
[5] Jiafeng Guo, Yixing Fan, Qingyao Ai, and W Bruce Croft. A deep relevance match-ing model for ad-hoc retrieval. In Proceedings of the 25th ACM International on Conference on Information and Knowledge Management, pages 55–64, 2016.
[6] Baotian Hu, Zhengdong Lu, Hang Li, and Qingcai Chen. Convolutional neural net-work architectures for matching natural language sentences. In Advances in neural information processing systems, pages 2042–2050, 2014.
[7] Paul Neculoiu, Maarten Versteegh, and Mihai Rotaru. Learning text similarity with siamese recurrent networks. In Proceedings of the 1st Workshop on Representation Learning for NLP, pages 148–157, 2016.
[8] Liang Pang, Yanyan Lan, Jiafeng Guo, Jun Xu, Shengxian Wan, and Xueqi Cheng. Text matching as image recognition. In Thirtieth AAAI Conference on Artificial Intel-ligence, 2016.
[9] Rudolf Schneider, Sebastian Arnold, Tom Oberhauser, Tobias Klatt, Thomas Steffek, and Alexander L¨oser. Smart-md: Neural paragraph retrieval of medical topics. In Companion Proceedings of the The Web Conference 2018, pages 203–206, 2018.
[10] Shengxian Wan, Yanyan Lan, Jiafeng Guo, Jun Xu, Liang Pang, and Xueqi Cheng. A deep architecture for semantic matching with multiple positional sentence represen-tations. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 30, 2016.
[11] Chenyan Xiong, Zhuyun Dai, Jamie Callan, Zhiyuan Liu, and Russell Power. End-to-end neural ad-hoc ranking with kernel pooling. In Proceedings of the 40th Interna-tional ACM SIGIR conference on research and development in information retrieval, pages 55–64, 2017.
[12] Liu Yang, Qingyao Ai, Jiafeng Guo, and W Bruce Croft. anmm: Ranking short answer texts with attention-based neural matching model. In Proceedings of the 25th ACM international on conference on information and knowledge management, pages 287–296, 2016.
[13] Wenpeng Yin and Hinrich Sch¨utze. Multigrancnn: An architecture for general matching of text chunks on multiple levels of granularity. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 63–73, 2015.