A Publicly Available Indonesian Corpora for Automatic Abstractive and Extractive Chat Summarization

A Publicly Available Indonesian Corpora for Automatic Abstractive and Extractive Chat Summarization Fajri Koto Advanced Research Lab Samsung R&D Institute Indonesia, INDONESIA fajri.fajri@samsung.com Abstract In this paper we report our effort to construct the first ever Indonesian corpora for chat summarization. documents of multi-participant chat from a well known online instant messaging application, WhatsApp.Specifically, we utilized We construct the gold standard by asking three native speakers to manually summarize 300 chat sections (152 of them contain images). As result, three reference summaries in extractive and either abstractive form are produced for each chat sections. The corpus is still in its early stage of investigation, yielding exciting possibilities of future works. Keywords: chat, summarization, corpora, indonesian 1. Introduction The growth of Internet around the world engenders new trends for people to actively communicate online by using various application and services involving text, audio, pic- ture, and video. According to Statista1 the active user of WhatsApp2 - the most well-known mobile chat application in the world, has reached 600 million by 2014. Similar data is also shown by other applications, such as Viber3, WeChat4, LINE5, and KakaoTalk6. Consequently, summa- rization system becomes important to ease people in ex- tracting the information automatically and quickly from chat conversation. Even though this situation will trigger many researchers to investigate methods and technique related to chat summa- rization, literatures show that only few studies specifically discuss about it. Whereas, chat summarization has many potential uses. It is applicable to various domains includ- ing multinational companies (Handel and Herbsleb, 2002), open source meetings (Shihab et al., 2009) (Zhou and Hovy, 2005), distance learning (Osman and Herring, 2007), and even military (Uthus and Aha, 2011). Challenge that is mostly discussed is finding suitable chat corpus that can be used for testing and evaluating summa- rization applications (Uthus and Aha, 2011). Most chat corpora do not have any summaries associated with them to use for a gold standard, making evaluations difficult. To ad- dress this issue, we start our investigation by first construct- ing a corpus containing chat conversation and its associ- ated summary in two forms: abstractive and extractive sum- mary. Extractive summary is obtained by selecting impor- tant sentences from the given chat conversation. Whereas, abstractive summarization is a more advanced summariza- tion which involves natural language technique in building 1http://www.statista.com/ 2https://web.whatsapp.com/ 3http://www.viber.com/ 4http://www.wechat.com/ 5http://line.me/en/ 6http://www.kakao.com/talk new sentences by considering pattern and grammar of lan- guage. In order to achieve research development on chat summa- rization, we initialize this work by constructing a suitable corpora that can be used for automatic qualitative evalua- tion. Here we use Indonesian language (Bahasa Indone- sia), by considering two factors: 1) Indonesia is the 5th biggest population and the 13th biggest Internet user in the world. According to the International Telecommunication Union (Geneva)7, Indonesia has more than 35 million Inter- net users by 2014. Consequently, understanding and pro- viding summary for chat in Bahasa Indonesia is not only useful for user but also for business need like market analy- sis, …

Content truncated to 500 words