PingAnLifeInsurance at SemEval-2023 Task 12: Sentiment Analysis for Low-resource African Languages with Multi-Model Fusion

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 679–685 July 13-14, 2023 ©2023 Association for Computational Linguistics PingAnLifeInsurance at SemEval-2023 Task 12: Sentiment Analysis for Low-resource African Languages with Multi-Model Fusion MeiZhi Jin and Cheng Chen and MengYuan Zhou and MengFei Yuan and XiaoLong Hou and XiYang Du and LianXin Jiang and JianYu Li Ping An Life Insurance Company of China, Ltd. {JINMEIZHI005, CHENCHENG498, ZHOUMENGYUAN425, YUANMENGFEI854, HOUXIAOLONG430, DUXIYANG037, JIANGLIANXIN769, LIJIANYU002} @pingan.com.cn Abstract This paper describes our system used in the SemEval-2023 Task12: Sentiment Analysis for Low-resource African Languages using Twitter Dataset. The AfriSenti-SemEval Shared Task 12 is based on a collection of Twitter datasets in 14 African languages for sentiment classifi- cation. It consists of three sub-tasks. Task A is a monolingual sentiment classification which covered 12 African languages. Task B is a mul- tilingual sentiment classification which com- bined training data from Task A (12 African languages). Task C is a zero-shot sentiment classification. We utilized various strategies, including monolingual training, multilingual mixed training, and translation technology, and proposed a weighted voting method that com- bined the results of different strategies. Sub- stantially, in the monolingual subtask, our sys- tem achieved Top-1 in two languages (Yoruba and Twi) and Top-2 in four languages (Nigerian Pidgin, Algerian Arabic, and Swahili, Multilin- gual). In the multilingual subtask, Our system achived Top-2 in publish leaderBoard. 1 Introduction Sentiment Analysis (SA) refers to the identification of sentiment and opinion contained in the inpfut texts that are often user-generated comments. The problem of automatic sentiment analysis is grow- ing research topic. Although sentiment analysis is an important area and already has a wide range of applications, it clearly is not a straightforward task and has many challenges related to Natural Language Processing (NLP). In recent years, sentiment analysis has gained significant attention due to its numerous vital appli- cations. The growth of social networks has resulted in the generation of more complex and interrelated information, enabling various artificial intelligence (AI) applications such as sentiment analysis, ma- chine translation, and detection of harmful con- tent. Social media platforms have become impor- tant sources of data for sentiment analysis. How- ever, most sentiment analysis research focuses on high-resource languages such as English, while low-resource languages like African languages re- main underrepresented. According to UNESCO (2003), around 2,058, or 30%, of all living lan- guages are African languages. However, most of these languages lack curated datasets required for the development of AI applications, and this issue is not unique to sentiment analysis, but affects NLP research as a whole. Sentiment analysis has garnered increasing inter- est due to its applicability across many domains, in- cluding public health, commerce/business, art and literature, social sciences, neuroscience, and psy- chology. Previous shared sentiment analysis tasks have been conducted, such as those by (Rosen- thal et al., 2019), (Nakov et al., 2019), (Pontiki et al., 2016), and (Ghosh et al., 2015), among oth- ers. However, none of these tasks have included African languages, highlighting the need for …

Content truncated to 500 words