Cobertura de diarios en Twitter de las elecciones autonómicas de Madrid 2021. Procesamiento de lenguaje natural y algoritmos de aprendizaje automático

Abstract

The aim of this work is to study the news coverage by the main Spanish generalist newspapers on Twitter during the campaign for the Community of Madrid elections in 2021 (n = 2,709). Natural language processing techniques and machine learning algorithms are applied to identify the predominant topic related to the elections and the mentions of candidates and political parties by each media, and to calibrate the emotional value of the messages published by El país, El mundo, Eldiario.es, and El confidencial. Among other findings, the results reveal how the media coverage focused mainly on campaign events and electoral debates. Despite the detection of minor differences between the newspapers, a general pattern emerges through this content, with a notable dominance of Isabel Díaz Ayuso but little influence of Rocío Monasterio as the lead candidates for their party. The sentiment analysis reveals the political alignment of each newspaper, using mainly negative messages with the aim of reducing the importance of a candidate or political party. While El país and Eldiario.es focused their criticism on Vox and the Partido Popular, El mundo and El confidencial criticized the actions of the national government, the PSOE’s proposal to join forces with Unidas Podemos, and Vox’s position, as well as emphasizing the disaster faced by Ciudadanos. It can be deduced that the media contributed to Ayuso’s success and to the consolidation of her image as an individual distinct from her own party. Resumen Se examina la cobertura informativa en las principales cabeceras generalistas españolas en Twitter durante la campaña de las elecciones a la Comunidad de Madrid en 2021 (n=2.709). Mediante técnicas de procesamiento de lenguaje natural y algoritmos de aprendizaje automático se identifica el tema preponderante vinculado a los comicios, se señala la incidencia de los candidatos y los partidos en cada medio y se calibra el valor emocional de los mensajes publicados por El país, El mundo, Eldiario.es y El confidencial. Entre otras pesquisas, los resultados evidencian cómo la cobertura mediática se centra, principalmente, en los actos de campaña y en los debates electorales. Pese a detectar pequeñas disimilitudes entre los diarios, se intuye un patrón generalizado: la notable incidencia de Isabel Díaz Ayuso y la escasa influencia de Rocío Monasterio como cabeza de lista en el contenido. A partir del análisis de sentimientos se constata la alineación partidista de las cabeceras, exponiendo, sobre todo, mensajes en tono negativo con la intención de mermar la relevancia de algún candidato o formación política. Mientras El país y Eldiario.es situaron en el centro de sus críticas a Vox y al Partido Popular, El mundo y El confidencial recriminaron la actuación del gobierno central, menospreciaron la proposición del PSOE a Unidas Podemos en aras de aunar votos, censuraron la postura de Vox e, incluso, insistieron en la debacle de Ciudadanos. Se deduce que los propios medios han contribuido al modelo de éxito de Ayuso y a la consolidación de su figura, propiciando ese individualismo y el distanciamiento de las siglas de su propio partido.

Clustering Algorithms in Sentiment Analysis Techniques in Social Media – A Rapid Literature Review

“… Lexicon-based methods [19]. This approach uses a lexicon (a collection of words and their associated sentiments) to classify the sentiment of text.…”

“… ML-based methods [30], [31], [19]. In this case, it trains a model using labeled data, where the labeled data contains text and the corresponding sentiment (positive, negative, neutral) and then the model is used to predict the sentiment of un-seen text.…”

“… Emotion detection [43], [19]: SA algorithms identify emotions in text, such as happiness, sadness, anger, and surprise. Similarly with Opinion mining this endeavor is pursued by NLP techniques as well.…”

“… Opinion Spam detection [19,38]: SA algorithms may be employed to detect fake or biased reviews or opinion from text. This can be achieved by comparing the sentiment of different text and detect any suspicious patterns.…”

“…While labeled data is often used in SA, there are several techniques that can be used to estimate the sentiment of text without prior labeled data, depending on the specific use case and available resources. It can be mentioned here Unsupervised SA [2], [19], Lexicon-based SA [9], [19], Transfer learning [20] and Active learning [21]. The goals of SA are accomplished through a number of phases.…”