利用深度學習之笑話辨識與生成

Humor Recognition and Generation Using Deep Learning

陳鵬宇(Peng-Yu Chen)

指導教授 ： 蘇豐文

國立清華大學/電機資訊學院/資訊工程學系所/碩士(2018年)

摘要

幽默作為一個特殊的語意表達方式，是生活中活躍氣氛、化解尷尬的重要元 素。近年來隨著人工智慧的快速發展，深度學習在自然語言處理的許多任務 中，取得了不錯的成果，如何利用電腦技術識別和生成幽默，也成為自然語 言處理領域熱門的研究內容之一。在本論文中，我們構建並收集了四個具 有不同笑話類型的中英文語料庫，並進行了幽默識別及幽默生成的研究。 我們在本論文中提出了一個基於深度學習裡的卷積神經網路 (Convolutional Neural Network)的模型，並結合Highway Networks的技術訓練深層的網路來進 行幽默辨識的研究，實驗結果顯示，我們的深度學習模型在識別不同類型及 語言的幽默方面，表現皆優於以前的基準，並達到了大約九成的準確率。除 了幽默辨識之外，我們也進行了幽默生成相關研究，利用生成式對抗網路 (Generative Adversarial Networks)與強化學習相結合 (Reinforcement Learning)的模 型，來產生蘊含幽默語意的文章，我們改進並提出了一個包含了兩個鑑別器 (Discriminator)的生成式對抗網路架構，來使模型更好地生成幽默文章，除此之 外，並進行了嚴謹的比較評估，來探討如何使用深度學習進行幽默生成。

關鍵字

幽默辨識 ； 幽默生成 ； 深度學習 ； 自然語言處理

並列摘要

Computational humor has been a fascinating topic that poses great challenge to artificial intelligence. For computers to understand and tell jokes does not seem to be an trivial task that remains to be a mystery. There have been very few attempts in literature that discuss how to build computational models in either discovering the structures of hu- mor, recognizing humor or even generating humor. In this thesis, I construct and collect four datasets with distinct joke types in both English and Chinese and conduct learn- ing experiments on humor recognition. I implement a Convolutional Neural Network (CNN) with extensive filter size, number and Highway Networks to increase the depth of networks. Results show that our model outperforms in recognition of different types of humor with benchmarks collected in both English and Chinese languages on accu- racy, precision, and recall in comparison to previous works. In addition to recognition, we also conduct research on humor generation that utilize adversarial networks combine with reinforcement learning (policy gradient) to generate humorous text. We purpose a two discriminators architecture that indicate more precisely rewards for generator to improve learning and produce quality jokes.

並列關鍵字

Humor Recongnition ； Humor Generation ； Deep Learning ； Natural Language Processing

參考文獻

Salvatore Attardo. Linguistic Theories of Humor. Walter de Gruyter, 1996.

Dario Bertero and Pascale Fung. A long short-term memory framework for predicting humor in dialogues. In NAACL HLT 2016, The 2016 Conference of the North Amer- ican Chapter of the Association for Computational Linguistics: Human Language Technologies, San Diego California, USA, June 12-17, 2016, pages 130–135, 2016. URL http://aclweb.org/anthology/N/N16/N16-1016.pdf.

Kim Binsted, Benjamin Bergen, and Justin McKay. Pun and non-pun humour in second- language learning. In Workshop Proceedings, CHI, 2003.

Lei Chen and Chong MIn Lee. Predicting Audience’s Laughter Using Convolutional Neural Network. ArXiv e-prints:1702.02584, February 2017.

Peng-Yu Chen and Von-Wun Soo. Humor recognition using deep learning. In Pro- ceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Pa- pers), volume 2, pages 113–117, 2018.