An Annotated Dataset of Discourse Modes in Hindi Stories

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 112–119 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 112 Affection Driven Neural Networks for Sentiment Analysis Rong Xiang1, Yunfei Long34, Mingyu Wan25, Jinghang Gu2, Qin Lu1, Chu-Ren Huang2 The Hong Kong Polytechnic University12, University of Nottingham3, Peking University4, University of Essex5 Department of Computing, 11 Yuk Choi Road, Hung Hom, Hong Kong (China)1 Chinese and Bilingual Studies, 11 Yuk Choi Road, Hung Hom, Hong Kong (China)2 NIHR Nottingham Biomedical Research Centre, Nottingham (UK) 3 Engineering, University of Essex, Colchester (UK)4 School of Foreign Languages, 5 Yiheyuan Road, Haidian, Beijing (China)5 School of Computer Science and Electronic xiangrong0302@gmail.com1, qin.lu@comp.polyu.edu.hk1, {mingyu.wan,jinghgu,churen.huang}@polyu.edu.hk2, yunfei.long@nottingham.ac.uk3 Abstract Deep neural network models have played a critical role in sentiment analysis with promising results in the recent decade. One of the essential challenges, however, is how external sentiment knowledge can be effectively utilized. In this work, we propose a novel affection-driven approach to incorporating affective knowledge into neural network models. The affective knowledge is obtained in the form of a lexicon under the Affect Control Theory (ACT), which is represented by vectors of three-dimensional attributes in Evaluation, Potency, and Activity (EPA). The EPA vectors are mapped to an affective influence value and then integrated into Long Short-term Memory (LSTM) models to highlight affective terms. Experimental results show a consistent improvement of our approach over conventional LSTM models by 1.0% to 1.5% in accuracy on three large benchmark datasets. Evaluations across a variety of algorithms have also proven the effectiveness of leveraging affective terms for deep model enhancement. Keywords: Sentiment analysis, Affective knowledge, Attention mechanism 1. Introduction Texts are collections of information that may encode emo- tions and deliver impacts to information receivers. Recog- nizing the underlying emotions encoded in a text is essen- tial to understanding the key information it conveys. On the other hand, emotion-embedded text can also provide rich sentiment resources for relevant natural language pro- cessing tasks. As such, sentiment analysis (SA) has gained increasing interest among researchers who are keen on the investigation of natural language processing techniques as well as emotion theories to identify sentiment expressions in a natural language context. Typical SA studies analyze subjective documents from the author’s perspective using high-frequency word representations and mapping the text (e.g., sentence or document) to categorical labels, e.g., sen- timent polarity, with either a discrete label or a real number in a continuum. Recently, the rising use of neural network models has further elevated the performance of SA without involv- ing laborious feature engineering. Typical neural network models such as Convolutional Neural Network (CNN) (Kim, 2014), Recursive auto-encoders (Socher et al., 2013), Long-Short Term Memory (LSTM) (Tang et al., 2015a) have shown promising results in a variety of sentiment anal- ysis tasks. In spite of this, neural network models still face two main problems. First, neural network approaches lack direct mechanisms to highlight important components in a text. Second, external resources such as linguistic knowledge, …

Content truncated to 500 words