Overview of FinancES 2023: Financial Targeted Sentiment Analysis in Spanish

Abstract

This paper presents the FinancES 2023 shared task, organized in the IberLEF 2023 workshop, within the framework of the 39th International Conference of the Spanish Society for Natural Language Processing (SEPLN 2023). The aim of this task is to extend the challenge of sentiment analysis in Spanish to the financial domain, in order to extract the sentiment that a piece of financial information can have for several actors, including the main economic target (i.e., the specific company or asset where the economic fact applies), other companies (i.e., the entities producing the goods and services that others consume) and consumers (i.e., households/individuals). Specifically, two tasks are proposed and evaluated separately. One to identify the main target and to determine the sentiment polarity towards such target, and a second task to assess the sentiment towards both other companies and consumers. The ranking includes results for 10 different teams proposing novel approaches, mostly based on Transformers and generative language models.

Este artículo resume la tarea FinancES 2023, organizada en el taller IberLEF 2023, dentro del marco de la 39ª Conferencia Internacional de la Sociedad Española de Procesamiento del Lenguaje Natural (SEPLN 2023). El objetivo de esta tarea es mejorar la materia de la minería de opiniones en español dentro del ámbito financiero realizando el análisis de sentimientos desde distintos puntos de vista. En concreto, se proponen y estudian dos tareas que son evaluadas de forma independiente. La primera tarea consiste en (i) identificar el actor principal asociado a una noticia financiera, y (ii) el sentimiento expresado hacia dicho actor. La segunda tarea consiste en determinar el sentimiento de la noticia (i) hacia otras empresas (i.e., otros agentes económicos), y (ii) hacia los consumidores (i.e., la sociedad). El ranking incluye los resultados de 10 equipos diferentes que proponen enfoques novedosos, en su mayoría basados en Transformers y modelos generativos del lenguaje.