Complaint Analysis and Classification for Economic and Food Safety

Proceedings of the Second Workshop on Economics and Natural Language Processing, pages 51–60 Hong Kong, November 4. c⃝2019 Association for Computational Linguistics 51 Complaint Analysis and Classification for Economic and Food Safety Jo˜ao Filgueiras*, Lu´ıs Barbosa*, Gil Rocha*, Henrique Lopes Cardoso*, Lu´ıs Paulo Reis*, Jo˜ao Pedro Machado+, Ana Maria Oliveira+ *Laborat´orio de Inteligˆencia Artificial e Ciˆencia de Computadores (LIACC) Faculdade de Engenharia da Universidade do Porto Rua Dr. Roberto Frias, s/n, 4200-465 Porto, Portugal +Autoridade de Seguranc¸a Alimentar e Econ´omica (ASAE), Rua Rodrigo da Fonseca, 73, 1269-274 Lisboa, Portugal {filgueiras, up201405729, gil.rocha, hlc, lpreis}@fe.up.pt {jpmachado, amoliveira}@asae.pt Abstract Governmental institutions are employing ar- tificial intelligence techniques to deal with their specific problems and exploit their huge amounts of both structured and unstructured information. In particular, natural language processing and machine learning techniques are being used to process citizen feedback. In this paper, we report on the use of such techniques for analyzing and classifying com- plaints, in the context of the Portuguese Eco- nomic and Food Safety Authority. Grounded in its operational process, we address three different classification problems: target eco- nomic activity, implied infraction severity level, and institutional competence. We show promising results obtained using feature-based approaches and traditional classifiers, with ac- curacy scores above 70%, and analyze the shortcomings of our current results and av- enues for further improvement, taking into ac- count the intended use of our classifiers in helping human officers to cope with thousands of yearly complaints. 1 Introduction Artificial intelligence (AI) techniques are nowa- days widespread in virtually every sector of human activity. Not only the private sector but also pub- lic administration institutions and governments are looking into ways of taking advantage of AI to deal with their specific problems and exploit their substantial amounts of both structured and un- structured information. Natural language process- ing (NLP) techniques are being employed in this regard to handle text available in the web (such as in social networks or newswires) and, most im- portantly, written forms of direct interaction be- tween citizens and governmental institutions (Eg- gers, 2019). Several governmental institutions provide pub- lic services electronically. Moreover, such in- stitutions are responsible for processing citizen feedback (such as requests or complaints), often materialized through email or contact forms in so-called virtual counters. The amount of such contacts can become intractable in a short pe- riod of time, depending on the size of the coun- try/administrative region. Based on such informa- tion, NLP techniques can be used to improve pub- lic services (Kowalski et al., 2019). This paper focuses on the needs of the Por- tuguese Economic and Food Safety Authority (ASAE)1, a national administrative authority spe- cialized in the context of food safety and economic surveillance, responsible for monitoring and en- forcing regulatory legislation. One of the main in- puts of this institution is comprised of citizen com- plaints on the activity of economic agents, with more than twenty thousand complaints being re- ceived annually. Usually, more than 30% of these are found not to be …

Content truncated to 500 words