Annotation Process for the Dialog Act Classification of a Taglish E-commerce Q&A Corpus

Proceedings of the Second Workshop on Economics and Natural Language Processing, pages 61–68 Hong Kong, November 4. c⃝2019 Association for Computational Linguistics 61 Annotation Process for the Dialog Act Classification of a Taglish E-commerce Q&A Corpus Jared Rivera, Jan Caleb Oliver Pensica, Jolene Valenzuela, Alfonso Secuya, Charibeth Cheng De La Salle University, Manila, Philippines {jared rivera, jan pensica, jolene valenzuela, alfonso secuya, charibeth.cheng}@dlsu.edu.ph Abstract With conversational agents or chatbots mak- ing up in quantity of replies rather than qual- ity, the need to identify user intent has become a main concern to improve these agents. Di- alog act (DA) classification tackles this con- cern, and while existing studies have already addressed DA classification in general con- texts, no training corpora in the context of e-commerce is available to the public. This research addressed the said insufficiency by building a text-based corpus of 7,265 posts from the question and answer section of prod- ucts on Lazada Philippines. The SWBD- DAMSL tagset for DA classification was mod- ified to 28 tags fitting the categories applicable to e-commerce conversations. The posts were annotated manually by three (3) human anno- tators and preprocessing techniques decreased the vocabulary size from 6,340 to 1,134. Af- ter analysis, the corpus was composed domi- nantly of single-label posts, with 34% of the corpus having multiple intent tags. The anno- tated corpus allowed insights toward the struc- ture of posts created with single to multiple in- tents. 1 Introduction An essential part of social media is the messag- ing feature which is easily adopted due to its con- venience and speed in comparison to other com- munication methods (Alison Bryant et al., 2006). In the Philippines, most online sellers prefer us- ing social media as an e-commerce platform for their businesses for exactly this reason (Marcelo, 2018). However, for social media to be effective as an e-commerce platform, active participation of the seller and the customer in the conversation is required. A general drawback in e-commerce is the lack or unavailability of sales clerks (i.e. on- line shop moderators) to interact with customers online. This problem is commonly evident among solo retailers, which provides an opportunity for the use of conversational agents to act as sales clerks on behalf of these sellers (Bogdanovych et al., 2005). Taglish (Tagalog-English) is often comfortably used on Philippine social media, and since natural language is noted as the most nat- ural means of communication between humans (Weischedel et al., 1989), interaction using Taglish as natural language, is seen as a feasible option to connect these conversational agents with Filipino customers online (Hill et al., 2015). However, misunderstandings are common in conversational interactions, more so when an on- line platform is used and transactions are con- ducted online than in-person, and the problem may be more complex once a machine is on one end of the conversation. Despite its capability to partic- ipate in a conversation, conversational agents still fail to simulate and capture the essence of the full range of an intelligent human …

Content truncated to 500 words