Target-based Sentiment Annotation in Chinese Financial News

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5040–5045 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5040 Target-based Sentiment Annotation in Chinese Financial News Chaofa Yuan1,†, Yuhan Liu1,†, Rongdi Yin1,†, Jun Zhang2 Qinling Zhu1, Ruibin Mao2, Ruifeng Xu1,3,‡ 1Harbin Institute of Technology (Shenzhen), China, 2Shenzhen Securities Information Co.,Ltd, China 3Joint Lab of Harbin Institute of Technology and RICOH {bruceyuan123, rongdiyin}@gmail.com liuyuhan hitsz@163.com zhuqinglin@stu.hit.edu.cn {zhangjun, maoruibin}@cninfo.com.cn, xuruifeng@hit.edu.cn Abstract This paper presents the design and construction of a large-scale target-based sentiment annotation corpus on Chinese financial news text. Different from the most existing paragraph/document-based annotation corpus, in this study, target-based fine-grained sentiment anno- tation is performed. The companies, brands and other financial entities are regarded as the targets. The clause reflecting the profitability, loss or other business status of financial entities is regarded as the sentiment expression for determining the polarity. Based on high quality annotation guideline and effective quality control strategy, a corpus with 8,314 target-level sentiment annotation is constructed on 6,336 paragraphs from Chinese financial news text. Based on this corpus, several state-of-the-art sentiment analysis models are evaluated. Keywords: Target-based Sentiment, Annotation, Chinese Financial Text 1. Introduction With the rapid development of the Internet, it becomes eas- ier to collect a large scale of public textual data. Identifying the sentiment corresponding to the financial entities such as companies from the financial text is crucial to financial in- vestment and decision making. Thus, sentiment analysis techniques on financial text attracted much research inter- est in the recent years. Sentiment analysis (SA) is a hot research topic in natural language processing. It has gained increas- ing popularity due to the its widely applications (Pang et al., 2008). (Kim, 2014; Li et al., 2015; Zhou et al., 2016; Yang et al., 2016; Chen et al., 2017; Li et al., 2015) focused on classifying the given text as positive, negative, neutral at document level and sentence level. However, the accuracy of sentiment classification on multiple opinion aspects/targets level is unsatisfactory. Especially, such kind of fine-grained sentiment analysis is essential to many practical applications. Hence aspect-based sentiment analy- sis (ABSA) (Pontiki et al., 2014; Pontiki et al., 2016; Poria et al., 2016), target-based sentiment analysis (TBSA) (Tang et al., 2016a; Tang et al., 2016b; Dong et al., 2014; Wang et al., 2017) and targeted aspect-based sentiment analysis (TABSA) (Liang et al., 2019; Ma et al., 2018) had been proposed to tackle the subtler sentiment in term of different aspects/targets in a sentence. For TBSA, it aims to identify the sentiment expression and determine the polarities of different target entities. Take the follow- ing Ex.1, as an example, there are three entities in this paragraph, 乐 融 致 新(LeMall), 乐 视 网(LETV) and 乐 视控股(LeEco). 乐融致新(LeMall) and 乐视网(LETV) appear more than once and their corresponding sentiment polarities are negative, while for 乐视控股(LeEco), the polarity is neutral. †Authors equally contributed to this work ‡Corresponding Author: xuruifeng@hit.edu.cn Ex.1 虽然乐 乐 乐融 融 融致 致 致新 新 新增资获得通过， …

Content truncated to 500 words