KNU CI System at SemEval-2018 Task4: Character Identification by Solving Sequence-Labeling Problem

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 655–659 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics KNU CI System at SemEval-2018 Task4: Character Identification by Solving Sequence-Labeling Problem Cheoneum Park*, Heejun Song**, Changki Lee* *Department of Computer Science, Kangwon National University, South Korea **Samsung Research, Samsung Electronics Co., Ltd., South Korea *{parkce, leeck}@kangwon.ac.kr **heejun7.song@samsung.com Abstract Character identification is an entity-linking task that finds words referring to the same person among the nouns mentioned in a conversation and turns them into one entity. In this paper, we define a sequence-labeling problem to solve character identification, and propose an attention-based recurrent neural network (RNN) encoder–decoder model. The input document for character identification on multiparty dialogues con- sists of several conversations, which in- crease the length of the input sequence. The RNN encoder–decoder model suffers from poor performance when the length of the in- put sequence is long. To solve this problem, we propose applying position encoding and the self-matching network to the RNN en- coder–decoder model. Our experimental results demonstrate that of the four models proposed, Model 2 showed an F1 score of 86.00% and a label accuracy of 85.10% at the scene-level. 1 Introduction In this paper, we define character identification (CI) (Chen et al., 2017) as a sequence-labeling problem and use a recurrent neural network (RNN) en- coder–decoder (Enc–Dec) model (Cho et al., 2014) based on the attention mechanism (Bahdanau et al., 2015) to solve it. An Enc–Dec is an extension of the RNN model; it generates an encoding vector using an RNN in the encoder when an input se- quence is given and performs decoding using the encoding vector. The attention mechanism calcu- lates the alignment score for the two sequences and performs the input sequence and weighted sum so that they can focus more on the position that affects the output result. The self-matching network (Wang et al., 2017) is used to calculate an attention weight for itself and a context vector by using a weighted sum, after which the weights of similar words in the RNN sequence can be applied to aid in coreference resolution. Position encoding (PE) (Sukhbaatar et al., 2015, Park and Lee, 2017, Vas- wani et al., 2017) is a method of applying weights differently, according to the positions of words ap- pearing in a sequence. Training and prediction are performed by multiplying a weight vector by a vec- tor of positions to be identified in a given input se- quence. In an Enc–Dec model, a long input sequence re- sults in performance degradation due to loss of in- formation in the front portion of the input sequence when encoding is performed. In this paper, we pro- pose four models that apply PE, attention mecha- nism, and self-matching network to Enc–Dec mod- els to solve the problem of performance degrada- tion due to long input sequences. To summarize, the main contributions of this pa- per are as follows: 1. In this paper, we define CI task as …

Content truncated to 500 words