Carmen Vaca

A time-based collective factorization for topic discovery and monitoring in news

Discovering and tracking topic shifts in news constitutes a new challenge for applications nowadays. Topics evolve,emerge and fade, making it more difficult for the journalist -or the press consumer- to decrypt the news. For instance, the current Syrian chemical crisis has been the starting point of the UN Russian initiative and also the revival of the US France alliance. A topical mapping representing how the topics evolve in time would be helpful to contextualize information. As far as we know, few topic tracking systems can provide such temporal topic connections. In this paper, we introduce a novel framework inspired from Collective Factorization for online topic discovery able to connect topics between different time-slots. The framework learns jointly the topics evolution and their time dependencies. It offers the user the ability to control, through one unique hyper-parameter, the tradeoff between the past accumulated knowledge and the current observed data. We show, on semi-synthetic datasets and on Yahoo News articles, that our method is competitive with state-of-the-art techniques while providing a simple way to monitor topics evolution (including emerging and disappearing topics).

Geo-localized social media data to improve characterization of international travelers

The motivation for traveling across political borders might be influenced by the socio-economic condition and the educational level of the traveler. Ecuador is both a highly popular touristic destination and the home country of at least 1 million migrants. The relatively high availability and low cost of online social network data may be a complement to migration data in the study of travelers economical means and motivations. Such travel variables are relevant for countries as Ecuador in which the economy partly relies on the tourism industry and remittances. In this empirical study, we collected 63 millions of tweets, extracted the modal country for each user and studied the tweets generated. We first demonstrated, that the volume of local travelers leaving a country grouped by destination, as well as the volume of foreign travelers visiting the same country and grouped by provenance origin location, can be accurately estimated (r = 0.85) from geo-located records extracted from Twitter. Next, we characterized each group of travelers and computed spatial clusters, revealing popular locations of national and foreign travelers while in Ecuador. Finally, we showed that Ecuadorians with higher educational level have more chances to travel abroad, validating such findings with official data.

Who You Should Not Follow: Extracting Word Embeddings from Tweets to Identify Groups of Interest and Hijackers in Demonstrations

In the latest years, a number of citizen movements and protests have spread across the world. One of the characteristics of such events is that demonstrations have been aroused by the use of social networking channels such as Twitter, Facebook, and Whatsapp, among others. Different scholars are currently analyzing this phenomenon to better understand its impact on societies. Furthermore, the use of the Internet as a driver or tool for organizing different groups and demonstrations leaves traces of social changes that have been addressed by technology. Nevertheless, it is important to define ways of identifying different movements, as well as possible misuse by so-called Internet trolls or hijackers, whose objective is to start arguments and confuse or upset other users. In this work, the authors present the case of demonstrations in Ecuador from March 2015 to April 2016 and use data from Twitter users who engaged in those demonstrations. Ecuador has a long history of demonstrations against different governments, which makes this scenario very attractive for more in depth study. Moreover, the authors present a framework for identifying political interest groups as well as possible hashtag hijackers. Specifically, this work focuses on the problem of giving recommendations to groups in which a group of users with the same political view receives suggestions of users they should not follow because they have opposing political views but use hijacked hashtags. Experiments on real-world data collected from the previously mentioned demonstrations show the effectiveness of this approach in automatically identifying hijackers so that they can be effectively recommended to a group as people they should not follow.

Crisis management on Twitter: Detecting emerging leaders

Twitter has become an effective communication platform during crisis management. Volunteers coordinate humanitarian relief initiatives that spread effectively through this platform, producing real-time data useful for the civil society. In some cases, this works faster than regular government protocols. If over time volunteers gain popularity and credibility, they become emerging leaders. This study proposes a method to detect emerging leaders on Twitter after a catastrophe. It consisted of collecting 4M of tweets using keywords related to Ecuadors earthquake on April 2016. Secondly, we quantifying relevant interactions in order to analyze the centrality of each user belonging to the active post-disaster twitter community. Consequently, a total of 63 emerging leaders were found, where 49% correspond to media accounts, 12% to citizen accounts and the remain to other entities. Finally, five information topics were identified in the tweets published by the emerging leaders, and their individual contribution was quantified. This result shows that it is possible to identify the specific aid in which each emerging leader focused.

Back to #6D: Predicting Venezuelan states political election results through Twitter

The large adoption of Twitter during electioneering has created an unprecedented opportunity to capture the citizens behaviour nationwide. The real-time access to information published by citizens has motivated researchers to design methods in order to enrich traditional political polling with insights from this rich source of data. However, less work has been done to capture the political scenario in Latin American countries, given that some methods rely on the use of English words, the reproducibility of such studies in Spanish speaking countries is a challenging task. Therefore, we propose a framework in which we apply social network analysis techniques and unsupervised machine learning to infer the political alignment at state level during Venezuelan Parliamentary election, which were performed on December 6, 2015. This electoral process took place in the middle of an acute political polarization in the country, the masses were organized around two political coalitions with opposite ideology: Government and opposition. In order to discover automatically the corresponding state political preferences, we analyze 60K tweets posted within the Venezuelan geographic boundaries during one week before the election day. Applying our framework, we are able to infer a given state political alignment starting from the quantified differences in communication patterns and linguistic profiles of the state aggregated tweets. We demonstrate that the online political atmosphere reflects the offline tendency at state scale given that we are able to predict the election tendency in Venezuela states with an accuracy of 87.5% with respect to official election results publicly available.

Requiem for online harassers: Identifying racism from political tweets

During the last five years, the amount of users of online social networks has increased exponentially. With the growing of users, social problems also arise. Due to the nature of these platforms, specifically Twitter, users can express their ideas in the way they prefer no matter if it is racist or not. As the Twitter CEO says, one of the most difficult things for them is to detect and ban people who harass others. Researches have addressed this issue in recent years. However, it is needed a wider range of strategies to detect racist users and content. In this work, we collect tweets produced by the ego networks of the two former 2016 US Presidential Candidates: Hillary Clinton and Donald Trump, grouped in four datasets. After deleting spammers, we get 84,371 unique users labeled by using two different metrics: Sentiment Word Count and Racist Score. Both of them let us not only to identify users as racists, but also to detect the level of negativism by analyzing their most recent 200 tweets, increasing the effectiveness of the method. Using it, we find the most negative and racist user and the most positive and non-racist user from all datasets. Taking advantage of the topological properties of the ego networks we analyzed, we also verify that our results satisfy the sociologist theory of homophily; where the followers of each candidate represent their homophilous. For a nation as the United States of America, detecting online harassers might help to decrease racism and cyberbullying, social problems that affect their society. A world without online harassers is an utopia, but this is one step to achieve it.

Where to go in Brooklyn: NYC Mobility Patterns from Taxi Rides.

Urban centers attractive for local citizens commonly house local cuisine restaurants or commercial areas. Local authorities are interested in discovering pattern to explain why city residents go to different areas of the city at a given time of the day. We explore a massive dataset of taxi rides, 69 million records in New York city, to uncover attractive places for local residents when going to Brooklyn. First, we obtain the origin destination matrix for New York boroughs. Second, we apply a density based clustering algorithm to detect popular drop-off locations. Next, we automatically find the closest venue, using the Foursquare API, to the most popular destination in each cluster. Our methodology let us to uncover popular destinations in urban areas in any city for which taxi rides information is available.

Uncovering Aspects of Places for Fitness Activities Through Social Media.

Nowadays, a growing number of people publicly share information about their fitness activities on social media platforms like Twitter or Facebook. These social networks can furnish people with useful information to get an overview of different geographic areas where people can practice different sport-related activities. In this study, we analyze 14 million tweets to identify places to perform fitness activities and uncovering their aspects from twitterers’ opinions. To this end, we apply clustering analysis to uncover places where twitterers perform fitness activities, and then train a text classifier that achieves a score F1 of \(76\%\) to discriminate the aspects of fitness places. Using this information, recommender systems can provide useful information to local people or tourists that look for places to do exercise.

Affinity Groups: A Linguistic Analysis for Social Network Groups Identification

Socially cohesive groups tend to share similar ideas and express themselves in similar ways when posting their thoughts in online social networks. Therefore, some researchers have conducted studies to uncover the issues discussed by groups who are structurally connected in a network. In this study, we take advantage of the language usage patterns present in online communication to unveil affinity groups, i.e. like-minded people, who are not necessarily interacting in the network currently. We analyze 735K tweets written by 620 unique users and compute scores for 14 grammatical categories using the linguistic inquiry word count software (LIWC). With the LIWC scores, we build a vector for each user, apply a similarity measure and feed an affinity propagation clustering algorithm to find the affinity groups. Following the proposed method, clusters of religious activists, journalists, entrepreneurs, among others emerge. We automatically characterize each cluster using a topic modeling algorithm and validate the generated topics with a user study conducted with 200 people. As a result, more than 70% of the participants agreed on their selection. These results confirm that communities share certain similarities in the use of language, traits that characterize their behavior and grouping.

Secrets of Quito: Discovering a city through TripAdvisor

People generate online content everyday at every hour at social networks. Social networks are a medium in which people can give their opinion on different topics and obtain new information. The content people create can be useful for researchers to understand human behavior in cities such as Quito. In this work, we are going to describe how Quito city is described on the travel network TripAdvisor by users based on geo referenced and lexical information. To describe this, we are going to combine a clustering algorithm(k-means) and a numerical statistic for information retrieval (TF-IDF). The results are later compared to information the municipality obtained from field studies to determine if they have similar behaviors.