Quote Recommendation for Dialogs and Writings


Title Quote Recommendation for Dialogs and Writings
Yeonchan Ahn, Hanbit Lee, Heesik Jeon, Seungdo Ha, Sang-goo Lee
Year 2016 / 09
Keywords Quote recommendation, Context matching, Random forest, Convolutional Neural Network, Recurrent Neural Network, Rank aggregation
Acknowledgement Samsung Electronics
Publication Type International Conference
Publication 3rd Workshop on New Trends in Content-Based Recommender Systems (CBRecsys2016), Volume 1673, pp. 39-42
Link url


Citing proverbs and (famous) statements of other people can provide support, shed new perspective, and/or add humor to one’s arguments in writings or dialogs. Recommending quote for dialog or writing can be done by considering the various features of the current text called context. We present five new approaches to quote recommendation: 1) methods to adjust the matching granularity for better context matching, 2) random forest based approach that utilizes word discrimination, 3) convolutional neural network based approach that captures important local semantic features, 4) recurrent neural network based approach that reflects the ordering of sentences and words in the context, and 5) rank aggregation of these algorithms for maximum performance. We adopt as baseline state-of-the-arts in citation recommendation and quote recommendation. Experiments show that our rank aggregation method outperforms the best baseline by up to 46.7%. As candidate quotes, we use famous proverbs and famous statement of other person in dialogs and writings. The quotes and their contexts were extracted from Twitter, Project Gutenberg, and Web blog corpus.