Skip to main content

Rodrigo Alfaro

Pontificia Universidad Catolica de Valparaiso, Informatics engineering, Faculty Member

Followers

47

Following

6

Public Views

Sebastián Ventura

Universidad de Córdoba

Alfonso E. Romero

Royal Holloway, University of London

Dejan Gjorgjevikj

Ss. Cyril & Methodius University in Skopje

Aswani Kumar Cherukuri

VIT University

University of Southampton

Vasant G Honavar

Pennsylvania State University

Oliver Ferschke

Carnegie Mellon University

Kati (Katalin) Prajda

University of Vienna

Viacheslav Kuleshov

Stockholm University

University of Technology Sydney

InterestsView All (6)

Uploads

Papers by Rodrigo Alfaro

Sourcing Pandemic News: A Cross-National Computational Analysis of Mainstream Media Coverage of COVID-19 on Facebook, Twitter, and Instagram

Digital Journalism, 2021

This article explores the uses of sources in coverage of the COVID-19 pandemic in social media po... more This article explores the uses of sources in coverage of the COVID-19 pandemic in social media posts of mainstream news organizations in Brazil, Chile, Germany, Mexico, Spain, the U.K., and the U.S. Based on computational content analysis, our study analyzes the sources and actors present in more than 940,000 posts on COVID-19 published in the 227 Facebook, Instagram, and Twitter accounts of 78 sampled news outlets between January 1 and December 31 of 2020, comparing their relative importance across countries, across media platforms, and across time as the pandemic evolved in each country. The analysis shows the dominance of political sources across countries and platforms, particularly in Latin America, demonstrating a strong role of the state in constructing pandemic news and suggesting that mainstream news organizations' social media posts maintain a strong elite orientation. Health sources were also prominentconsistent with the defining role of biomedical authority in health coverage-, while significant diversity of sources, including citizen sources, emerged as the pandemic went on. Our results also revealed that the use of specific sources significantly varied over time. These variations tend to go hand in hand with specific global milestones of the pandemic.

Análisis y resumen automático de políticas de privacidad

Actas III Congreso Internacional de Lingüística Computacional y de Corpus - CILCC 2020 y V Workshop en Procesamiento Automatizado de Textos y Corpus - WoPATeC 2020. Universidad de Antioquia, Medellín, 21-23 octubre de 2020.: Libro de resumenes, 2020, pág. 286, 2020

Information Gathering and Classification for Collaborative Logistics Decision Making

Supply Chain Management - New Perspectives, 2011

Multilabel Text Classification with Label-Dependent Representation

Applied Sciences

Assigning predefined classes to natural language texts, based on their content, is a necessary co... more Assigning predefined classes to natural language texts, based on their content, is a necessary component in many tasks in organizations. This task is carried out by classifying documents within a set of predefined categories using models and computational methods. Text representation for classification purposes has traditionally been performed using a vector space model due to its good performance and simplicity. Moreover, the classification of texts via multilabeling has typically been approached by using simple label classification methods, which require the transformation of the problem studied to apply binary techniques, or by adapting binary algorithms. Over the previous decade, text classification has been extended using deep learning models. Compared to traditional machine learning methods, deep learning avoids rule design and feature selection by humans, and automatically provides semantically meaningful representations for text analysis. However, deep learning-based text cl...

Analyzing #LasTesis Feminist Movement in Twitter Using Topic Models

Social Computing and Social Media. Design, Ethics, User Behavior, and Social Network Analysis, 2020

Nowadays, social networks have created a massive mean of communication, that was unthinkable many... more Nowadays, social networks have created a massive mean of communication, that was unthinkable many years ago. Informal communication, blogging, and online discussions have transformed the Web into a huge repository of remarks on numerous themes, producing a potential wellspring of data for various areas. In this paper we analyze, using Topic Models, a recent widespread feminist movement. Las Tesis is a feminist collective that initiated a protest against sexual abuse, and that was replicated in more than dozen different countries in matter of days. We use LDA and BTM to detect automatically the topics in over 627643 tweets that were gathered from the 25th November until the 5th January. The resulting topics obtained, from tweets in Spanish and English, show that these algorithms are able to capture the real-world events that occurred in Chile and Turkey.

Intent Classification of Social Media Texts with Machine Learning for Customer Service Improvement

Social Computing and Social Media. Applications and Analytics, 2017

Social media platforms in the last few years have facilitated the development of communities that... more Social media platforms in the last few years have facilitated the development of communities that discuss real-world events, and have shaped the way users interact. The content generated in these platforms reflect a variety of intentions, ranging from social interaction to commercial interest, among many others. The present study aims at the implementation of an automatic intent classification system for a Chilean electricity company social media account. The dataset was created from 5000 tweets that were manually classified by 5 people. If discrepancies were detected, a majority voting scheme was used in order to tag the tweets' intentions. In order to perform the experimental validation of the automatic classification with the machine learning algorithms, several text representations were used (tf-idf, tf-rfl and bin-rfl). The results obtained from the various tests that were conducted yielded satisfactory results. We also analyzed how to assign automatic responses to frequently asked questions, and obtained promising results.

ARAPP: Análisis y Resumen Automático de Políticas de Privacidad

Linguamática

Un derecho fundamental de los usuarios de aplicaciones informáticas es que puedan conocer las pol... more Un derecho fundamental de los usuarios de aplicaciones informáticas es que puedan conocer las políticas de privacidad (PP) que tales aplicaciones establecen, en particular es relevante que conozcan acerca del tratamiento que aceptan sobre el uso de sus datos. No obstante, estas PP son muy extensas y escritas en un lenguaje administrativo-jurídico y comercial, lo que dificulta su lectura y comprensión. El objetivo de este artículo es resumir automatizadamente las PP de cinco aplicaciones de redes sociales (Facebook, Twitter, TikTok, Snapchat e Instagram) en español, a través de técnicas extractivas y abstractivas. Para ello se utilizan tres aproximaciones de representación desde el Procesamiento de Lenguaje Natural, estas son: Teoría de Grafos, TF-IDF y Gensim. A partir de ellas, se generan automáticamente 15 resúmenes, los que son evaluados por un experto en derecho, para medir la legibilidad y relevancia en base a 20 preguntas confeccionadas por un estudio de la Universidad de Aust...

Automatic Tweets Classification Under an Intelligent Agents Framework

Social Computing and Social Media. Applications and Analytics, 2017

Twitter is a microblogging platform that allows users to share opinions with a restricted amount ... more Twitter is a microblogging platform that allows users to share opinions with a restricted amount of characters. Given the social characteristic of Twitter, it is a potential source for sentiment analysis. For this reason, opinions of certain subjects such as people or brands can change in short periods of time. A traditional approach of a sentiment classifier implementation performs poorly since it depends on how it is trained. We propose a novel method for tackling this problem, with the implementation of a multi-agent system for classifying and corpus analysis mechanism for retraining the classifier. This mechanism consists of a critic agent which compares the trained corpus of the classifier agent with new collections of documents from future time steps, using primarily two methods: hypothesis test analysis and histogram differences. A Naïve-Bayes based classifier was implemented with this mechanism with multiple configurations. The results of experimental data show that the mechanism boosts its performance, when compared to a pure Naïve Bayes classifier.

Text Representation in Multi-label Classification: Two New Input Representations

Lecture Notes in Computer Science, 2011

Automatic text classification is the task of assigning unseen documents to a predefined set of cl... more Automatic text classification is the task of assigning unseen documents to a predefined set of classes. Text representation for classification purposes has been traditionally approached using a vector space model due to its simplicity and good performance. On the other hand, multi-label automatic text classification has been typically addressed either by transforming the problem under study to apply binary techniques or by adapting binary algorithms to work with multiple labels. In this paper we present two new representations for text documents based on label-dependent term-weighting for multi-label classification. We focus on modifying the input. Performance was tested with a wellknown dataset and compared to alternative techniques. Experimental results based on Hamming loss analysis show an improvement against alternative approaches.

Content and Style Features for Automatic Detection of Users’ Intentions in Tweets

Lecture Notes in Computer Science, 2014

The aim of this paper is to evaluate the use of content and style features in automatic classific... more The aim of this paper is to evaluate the use of content and style features in automatic classification of intentions of Tweets. For this we propose different style features and evaluate them using a machine learning approach. We found that although the style features by themselves are useful for the identification of the intentions of tweets, it is better to combine such features with the content ones. We present a set of experiments, where we achieved a 9.46 % of improvement on the overall performance of the classification with the combination of content and style features as compared with the content features.

Clasificación automática de textos usando redes de palabras

Revista signos, 2014

El objetivo de este trabajo es proponer un algoritmo para la clasificación automática de textos, ... more El objetivo de este trabajo es proponer un algoritmo para la clasificación automática de textos, como una alternativa a los tradicionalmente utilizados en esta tarea. El clasificador propuesto considera la dependencia entre las variables predictoras (palabras o términos), algo que los clasificadores de texto comúnmente utilizados no hacen. La dependencia entre estas variables queda plasmada en forma de enlaces en grafos de palabras co-ocurrentes, objetos utilizados para entrenar el clasificador y además estimar la categoría de un texto desconocido. Los resultados obtenidos al clasificar automáticamente el sentido positivo, negativo o neutral de más de 1.000 mensajes de Twitter escritos en español, en distintos contextos (temas), muestran que el algoritmo, además de ser una propuesta novedosa para la clasificación automática de textos, tiene un desempeño, al menos, similar al de otros tradicionalmente utilizados en este tipo de problemas, como las Máquinas de Soporte Vectorial o algoritmos de estadística Bayesiana. Palabras Clave: Clasificación automática de textos, redes de palabras, algoritmo, inteligencia artificial, inteligencia computacional. Clasificación automática de textos usando redes de palabras* Automatic text classification using words networks

Clasificación de Textos Multi-etiquetados con Modelo Bernoulli Multi-variado y Representación Dependiente de la Etiqueta

Revista signos, 2020

The allocation of natural language texts to one or more predefined categories or classes based on... more The allocation of natural language texts to one or more predefined categories or classes based on their content is an important component and a recent need in many information organization and management tasks. Automatic text classification is the task of categorizing documents to a predefined set of classes by a computational method or model. Text representation for classification purposes has been traditionally approached using a vector space model due to its simplicity and good performance. On the other hand, multi-label automatic text classification has been typically addressed either by transforming the problem under study to apply binary techniques or by adapting binary algorithms to work with multiple labels. In this article, the objective is to evaluate a term-weighting factor in the Boolean model for text representation in multi-label classification, using a mix of two approaches: problem transformation and model adaptation. This term-weighting factor and the combination of approaches in the automatic text classification was tested with four different sets of textual data used in the specialized literature and compared with alternative techniques by means of three measures of evaluation. The results present improvements of more than 10% in the performance of the classifiers, attributed to our proposal, in all the cases analyzed.

La Dualidad de Automatizar/Aumentar con Tecnologías de la Información

eii.ucv.cl

Resumen Este artículo presenta la visión de una dualidad entre automatizar y aumentar cuando se i... more

Automated Text Binary Classification Using Machine Learning Approach

2010 XXIX International Conference of the Chilean Computer Science Society, 2010

Automated Text Binary Classification Using Machine Learning Approach

2010 XXIX International Conference of the Chilean Computer Science Society, 2010

Sourcing Pandemic News: A Cross-National Computational Analysis of Mainstream Media Coverage of COVID-19 on Facebook, Twitter, and Instagram

Digital Journalism, 2021

This article explores the uses of sources in coverage of the COVID-19 pandemic in social media po... more This article explores the uses of sources in coverage of the COVID-19 pandemic in social media posts of mainstream news organizations in Brazil, Chile, Germany, Mexico, Spain, the U.K., and the U.S. Based on computational content analysis, our study analyzes the sources and actors present in more than 940,000 posts on COVID-19 published in the 227 Facebook, Instagram, and Twitter accounts of 78 sampled news outlets between January 1 and December 31 of 2020, comparing their relative importance across countries, across media platforms, and across time as the pandemic evolved in each country. The analysis shows the dominance of political sources across countries and platforms, particularly in Latin America, demonstrating a strong role of the state in constructing pandemic news and suggesting that mainstream news organizations' social media posts maintain a strong elite orientation. Health sources were also prominentconsistent with the defining role of biomedical authority in health coverage-, while significant diversity of sources, including citizen sources, emerged as the pandemic went on. Our results also revealed that the use of specific sources significantly varied over time. These variations tend to go hand in hand with specific global milestones of the pandemic.

Análisis y resumen automático de políticas de privacidad

Actas III Congreso Internacional de Lingüística Computacional y de Corpus - CILCC 2020 y V Workshop en Procesamiento Automatizado de Textos y Corpus - WoPATeC 2020. Universidad de Antioquia, Medellín, 21-23 octubre de 2020.: Libro de resumenes, 2020, pág. 286, 2020

Information Gathering and Classification for Collaborative Logistics Decision Making

Supply Chain Management - New Perspectives, 2011

Multilabel Text Classification with Label-Dependent Representation

Applied Sciences

Assigning predefined classes to natural language texts, based on their content, is a necessary co... more Assigning predefined classes to natural language texts, based on their content, is a necessary component in many tasks in organizations. This task is carried out by classifying documents within a set of predefined categories using models and computational methods. Text representation for classification purposes has traditionally been performed using a vector space model due to its good performance and simplicity. Moreover, the classification of texts via multilabeling has typically been approached by using simple label classification methods, which require the transformation of the problem studied to apply binary techniques, or by adapting binary algorithms. Over the previous decade, text classification has been extended using deep learning models. Compared to traditional machine learning methods, deep learning avoids rule design and feature selection by humans, and automatically provides semantically meaningful representations for text analysis. However, deep learning-based text cl...

Analyzing #LasTesis Feminist Movement in Twitter Using Topic Models

Social Computing and Social Media. Design, Ethics, User Behavior, and Social Network Analysis, 2020

Nowadays, social networks have created a massive mean of communication, that was unthinkable many... more Nowadays, social networks have created a massive mean of communication, that was unthinkable many years ago. Informal communication, blogging, and online discussions have transformed the Web into a huge repository of remarks on numerous themes, producing a potential wellspring of data for various areas. In this paper we analyze, using Topic Models, a recent widespread feminist movement. Las Tesis is a feminist collective that initiated a protest against sexual abuse, and that was replicated in more than dozen different countries in matter of days. We use LDA and BTM to detect automatically the topics in over 627643 tweets that were gathered from the 25th November until the 5th January. The resulting topics obtained, from tweets in Spanish and English, show that these algorithms are able to capture the real-world events that occurred in Chile and Turkey.

Intent Classification of Social Media Texts with Machine Learning for Customer Service Improvement

Social Computing and Social Media. Applications and Analytics, 2017

Social media platforms in the last few years have facilitated the development of communities that... more Social media platforms in the last few years have facilitated the development of communities that discuss real-world events, and have shaped the way users interact. The content generated in these platforms reflect a variety of intentions, ranging from social interaction to commercial interest, among many others. The present study aims at the implementation of an automatic intent classification system for a Chilean electricity company social media account. The dataset was created from 5000 tweets that were manually classified by 5 people. If discrepancies were detected, a majority voting scheme was used in order to tag the tweets' intentions. In order to perform the experimental validation of the automatic classification with the machine learning algorithms, several text representations were used (tf-idf, tf-rfl and bin-rfl). The results obtained from the various tests that were conducted yielded satisfactory results. We also analyzed how to assign automatic responses to frequently asked questions, and obtained promising results.

ARAPP: Análisis y Resumen Automático de Políticas de Privacidad

Linguamática

Un derecho fundamental de los usuarios de aplicaciones informáticas es que puedan conocer las pol... more Un derecho fundamental de los usuarios de aplicaciones informáticas es que puedan conocer las políticas de privacidad (PP) que tales aplicaciones establecen, en particular es relevante que conozcan acerca del tratamiento que aceptan sobre el uso de sus datos. No obstante, estas PP son muy extensas y escritas en un lenguaje administrativo-jurídico y comercial, lo que dificulta su lectura y comprensión. El objetivo de este artículo es resumir automatizadamente las PP de cinco aplicaciones de redes sociales (Facebook, Twitter, TikTok, Snapchat e Instagram) en español, a través de técnicas extractivas y abstractivas. Para ello se utilizan tres aproximaciones de representación desde el Procesamiento de Lenguaje Natural, estas son: Teoría de Grafos, TF-IDF y Gensim. A partir de ellas, se generan automáticamente 15 resúmenes, los que son evaluados por un experto en derecho, para medir la legibilidad y relevancia en base a 20 preguntas confeccionadas por un estudio de la Universidad de Aust...

Automatic Tweets Classification Under an Intelligent Agents Framework

Social Computing and Social Media. Applications and Analytics, 2017

Twitter is a microblogging platform that allows users to share opinions with a restricted amount ... more Twitter is a microblogging platform that allows users to share opinions with a restricted amount of characters. Given the social characteristic of Twitter, it is a potential source for sentiment analysis. For this reason, opinions of certain subjects such as people or brands can change in short periods of time. A traditional approach of a sentiment classifier implementation performs poorly since it depends on how it is trained. We propose a novel method for tackling this problem, with the implementation of a multi-agent system for classifying and corpus analysis mechanism for retraining the classifier. This mechanism consists of a critic agent which compares the trained corpus of the classifier agent with new collections of documents from future time steps, using primarily two methods: hypothesis test analysis and histogram differences. A Naïve-Bayes based classifier was implemented with this mechanism with multiple configurations. The results of experimental data show that the mechanism boosts its performance, when compared to a pure Naïve Bayes classifier.

Text Representation in Multi-label Classification: Two New Input Representations

Lecture Notes in Computer Science, 2011

Automatic text classification is the task of assigning unseen documents to a predefined set of cl... more Automatic text classification is the task of assigning unseen documents to a predefined set of classes. Text representation for classification purposes has been traditionally approached using a vector space model due to its simplicity and good performance. On the other hand, multi-label automatic text classification has been typically addressed either by transforming the problem under study to apply binary techniques or by adapting binary algorithms to work with multiple labels. In this paper we present two new representations for text documents based on label-dependent term-weighting for multi-label classification. We focus on modifying the input. Performance was tested with a wellknown dataset and compared to alternative techniques. Experimental results based on Hamming loss analysis show an improvement against alternative approaches.

Content and Style Features for Automatic Detection of Users’ Intentions in Tweets

Lecture Notes in Computer Science, 2014

The aim of this paper is to evaluate the use of content and style features in automatic classific... more The aim of this paper is to evaluate the use of content and style features in automatic classification of intentions of Tweets. For this we propose different style features and evaluate them using a machine learning approach. We found that although the style features by themselves are useful for the identification of the intentions of tweets, it is better to combine such features with the content ones. We present a set of experiments, where we achieved a 9.46 % of improvement on the overall performance of the classification with the combination of content and style features as compared with the content features.

Clasificación automática de textos usando redes de palabras

Revista signos, 2014

El objetivo de este trabajo es proponer un algoritmo para la clasificación automática de textos, ... more El objetivo de este trabajo es proponer un algoritmo para la clasificación automática de textos, como una alternativa a los tradicionalmente utilizados en esta tarea. El clasificador propuesto considera la dependencia entre las variables predictoras (palabras o términos), algo que los clasificadores de texto comúnmente utilizados no hacen. La dependencia entre estas variables queda plasmada en forma de enlaces en grafos de palabras co-ocurrentes, objetos utilizados para entrenar el clasificador y además estimar la categoría de un texto desconocido. Los resultados obtenidos al clasificar automáticamente el sentido positivo, negativo o neutral de más de 1.000 mensajes de Twitter escritos en español, en distintos contextos (temas), muestran que el algoritmo, además de ser una propuesta novedosa para la clasificación automática de textos, tiene un desempeño, al menos, similar al de otros tradicionalmente utilizados en este tipo de problemas, como las Máquinas de Soporte Vectorial o algoritmos de estadística Bayesiana. Palabras Clave: Clasificación automática de textos, redes de palabras, algoritmo, inteligencia artificial, inteligencia computacional. Clasificación automática de textos usando redes de palabras* Automatic text classification using words networks

Clasificación de Textos Multi-etiquetados con Modelo Bernoulli Multi-variado y Representación Dependiente de la Etiqueta

Revista signos, 2020

The allocation of natural language texts to one or more predefined categories or classes based on... more The allocation of natural language texts to one or more predefined categories or classes based on their content is an important component and a recent need in many information organization and management tasks. Automatic text classification is the task of categorizing documents to a predefined set of classes by a computational method or model. Text representation for classification purposes has been traditionally approached using a vector space model due to its simplicity and good performance. On the other hand, multi-label automatic text classification has been typically addressed either by transforming the problem under study to apply binary techniques or by adapting binary algorithms to work with multiple labels. In this article, the objective is to evaluate a term-weighting factor in the Boolean model for text representation in multi-label classification, using a mix of two approaches: problem transformation and model adaptation. This term-weighting factor and the combination of approaches in the automatic text classification was tested with four different sets of textual data used in the specialized literature and compared with alternative techniques by means of three measures of evaluation. The results present improvements of more than 10% in the performance of the classifiers, attributed to our proposal, in all the cases analyzed.

La Dualidad de Automatizar/Aumentar con Tecnologías de la Información

eii.ucv.cl

Resumen Este artículo presenta la visión de una dualidad entre automatizar y aumentar cuando se i... more

Automated Text Binary Classification Using Machine Learning Approach

2010 XXIX International Conference of the Chilean Computer Science Society, 2010

Automated Text Binary Classification Using Machine Learning Approach

2010 XXIX International Conference of the Chilean Computer Science Society, 2010