以深度學習方法探索人物互動關係之研究

A Study of Deep Neural Network for Person Interaction Discovery

林庭宇(Ting-Yu Lin)

指導教授 ： 許聞廉

共同指導教授 ： 張智星(Jyh-Shing Jang)

國立臺灣大學/電機資訊學院/資料科學學位學程/碩士(2019年)

https://doi.org/10.6342/NTU201901569

摘要

本文的研究主題為人物互動關係之探索，我們試圖識別社交媒體中提到的不同人物之間的互動關係，藉此幫助讀者建構出在某個主題下，不同人物之間的關係背景，加快讀者理解不同主題的文本內容。此研究基於 Chang et al.提出的傳統內核方法，我們以深度學習方法做改良，並將傳統的自然語言特徵與樹結構融合進神經網路模型中，其中利用了實體嵌入、豐富互動樹嵌入、詞性嵌入、句子類別和依賴特徵，藉此完成人物互動關係探索中的兩個任務－關係偵測任務與關係擷取任務，另外我們還對多任務模型進行探討，希望透過兩任務模型之間的互相輔助來提升彼此的效能，我們的方法在關係偵測任務中，最終在F1分數上超越了原作者論文約7%，達到了中文人物互動關係偵測到目前為止最好的效能表現，同時我們實作了原作者論文中所沒有實作的關係擷取任務，並且在效能方面有不錯的表現，這對於建構人物互動網絡的知識庫很有用。

關鍵字

人物互動關係探索 ； 關係擷取 ； Open IE ； 深度學習 ； 豐富互動樹

並列摘要

The research topic of this paper is person interaction discovery. We are trying to identify interactions between different people mentioned in social media. To help readers construct a relationship between people under a certain topic, so that readers can quickly understand the text content of different topics. This study is based on the traditional kernel method proposed by Chang et al. We use the deep learning method to improve and integrate the traditional natural language features and tree structure into the neural network model. It utilizes entity embedding, rich interactive tree embedding, part of speech embedding, sentence categories, and dependency features. In this way, two tasks in the person interaction discovery - relation detection task and relation extraction task are completed. In addition, we also explore the multitasking model and hope to improve each other's effectiveness through mutual assistance between task models. Our method in the relation detection task, eventually surpassed the original author's paper by about 7% on the F1 score. At the same time, we have implemented a relation extraction model which the original author didn't implement. It demonstrates superior performances on the person interaction extraction task. This is useful for building a knowledge base for people's interactive networks.

並列關鍵字

Person Interaction Discovery ； Relation Extraction ； Open IE ； Deep Learning ； Rich Interactive Tree

參考文獻

[1] Yung-Chun Chang, C. C. Chen, and W. Hsu. 2016. SPIRIT: A tree kernel-based method for topic person interaction detection. IEEE Transactions on Knowledge and Data Engineering, 28(9):2494–2507.

[2] Yung-Chun Chang, Chien Chin Chen, and Wen-Lian Hsu, "A Composite Kernel Approach for Detecting Interactive Segments in Chinese Topic Documents," the 9th Asia Information Retrieval Societies Conference (AIRS 2013), Lecture Notes in Computer Science, pages 215-226, December 2013.

[3] Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matthew Gardner, Christopher T Clark, Kenton Lee, and Luke S. Zettlemoyer. 2018. Deep contextualized word representations. CoRR, abs/1802.05365.

[4] Cai, R., Zhang, X., Wang, H.. Bidirectional Recurrent Convolutional Neural Network for Relation Classification. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (ACL 2016) 2016;:756–765.

[5] Chia-Wei Wu, Shyh-Yi Jan, Tzong-Han Tsai, Wen-Lian Hsu, “On Using Ensemble Methods for Chinese Named Entity Recognition”, Proceedings of the Fifth SIGHAN Workshop on Chinese Language Processing, Sydney, July 2006, pp. 142–145.