ECNU_ICA at SemEval-2022 Task 10: A Simple and Unified Model for Monolingual and Crosslingual Structured Sentiment Analysis

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 1336 - 1342 July 14-15, 2022 ©2022 Association for Computational Linguistics ECNU ICA at SemEval-2022 Task 10: A Simple and Unified Model for Monolingual and Crosslingual Structured Sentiment Analysis Qi Zhang1, Jie Zhou2∗, Qin Chen1, Qingchun Bai3, Jun Xiao3, and Liang He1 1School of Computer Science and Technology East China Normal University, China 2School of Computer Science, Fudan Univerisity, China 3Shanghai Open University, China qzhang@stu.ecnu.edu.cn, jzhou@ica.stc.sh.cn {qchen,lhe}@cs.ecnu.edu.cn, qc_bai@foxmail.com, xiaoj@sou.edu.cn Abstract In this paper, we focus on the structured senti- ment analysis task that is released on SemEval- 2022 Task 10. The task aims to extract the structured sentiment information (e.g., holder, target, expression and sentiment polarity) in a text. We propose a simple and unified model for both the monolingual and crosslingual struc- tured sentiment analysis tasks. We translate this task into an event extraction task by regarding the expression as the trigger word and the other elements as the arguments of the event. Particu- larly, we first extract the expression by judging its start and end indices. Then, to consider the expression, we design a conditional layer normalization algorithm to extract the holder and target based on the extracted expression. Finally, we infer the sentiment polarity based on the extracted structured information. We conduct the experiments on seven datasets in five languages. It attracted 233 submissions in monolingual subtask and crosslingual subtask from 32 teams. Finally, we obtain the top 5 place on crosslingual tasks. 1 Introduction The identification of sentiment in the text is an im- portant field of study. Users’ opinions on products, events, topics, and so on are valuable for both the company and government to improve products or policies. Recently, more and more researchers fo- cus on fine-grained sentiment analysis tasks, such as structured sentiment analysis, which can be formulated into tuple extraction from the context (Wiebe et al., 2005). We focus on structured sentiment analysis, which is released by SemEval 2022 task 10. For- mally, the task aims to extract all of the opinion tuples O = {Oi, ..., On} in a text. A tuple (h, t, e, p) can represent the structure of sentiment in context. As shown in Figure 1, “Some other " (h) is a holder who expresses a polarity “positive" (p) ∗ Jie Zhou is the corresponding author of this paper. Figure 1: A structured sentiment graph is composed of a holder, target, sentiment expression, their relationships and a polarity attribute. Holders and targets can be null. towards a target “the new UMUC" (t) through a sentiment expression “5 stars" (e), implicitly defin- ing the relationships between the elements of a sentiment graph. Moreover, the SemEval task 10 is divided into two settings of increasing complexity: Setting A trains and predicts the sentiment polarity on monolingual. Setting B trains on the source lan- guage and tests on the target language. Particularly, we can train on any of the other datasets, as well as any other resource that does not …

Content truncated to 500 words