arXiv subjects

Marcelo Mendoza

Publications and source records attributed to Marcelo Mendoza.

17 recordsLinked to original sources

arXiv2026

Measuring the Predictability of Recommender Systems using Structural Complexity Metrics

Recommender Systems (RS) shape the filtering and curation of online content, yet we have limited understanding of how predictable their recommendation outputs are. We propose data-driven metrics that quantify the predictability of recommendation datasets by measuring the structural complexity of the user-item interaction matrix. High complexity indicates intricate interaction patterns that are harder to predict; low complexity indicates simpler, more predictable structures. We operationalize structural complexity via data perturbations, using singular value decomposition (SVD) to assess how stable the latent structure remains under perturbations. Our hypothesis is that random perturbations minimally affect highly organized data, but cause substantial structural disruption in intrinsically complex data. By analyzing prediction errors on perturbed interactions, we derive metrics that quantify this sensitivity at both the dataset and the interaction levels, yielding a principled measure of inherent predictability. Experiments on real-world datasets show that our structural complexity metrics correlate with the performance of state-of-the-art recommendation algorithms. We also demonstrate structure-aware data selection: in low-data settings, models trained on a carefully chosen subset of interactions with low structural perturbation error consistently outperform models trained on the full dataset. Thus, structural complexity serves both as a precise diagnostic of dataset complexity and as a principled foundation for efficient, data-centric training of RS.

cs.IR↗

arXiv2024

Unveiling Social Media Comments with a Novel Named Entity Recognition System for Identity Groups

While civilized users employ social media to stay informed and discuss daily occurrences, haters perceive these platforms as fertile ground for attacking groups and individuals. The prevailing approach to counter this phenomenon involves detecting such attacks by identifying toxic language. Effective platform measures aim to report haters and block their network access. In this context, employing hate speech detection methods aids in identifying these attacks amidst vast volumes of text, which are impossible for humans to analyze manually. In our study, we expand upon the usual hate speech detection methods, typically based on text classifiers, to develop a Named Entity Recognition (NER) System for Identity Groups. To achieve this, we created a dataset that allows extending a conventional NER to recognize identity groups. Consequently, our tool not only detects whether a sentence contains an attack but also tags the sentence tokens corresponding to the mentioned group. Results indicate that the model performs competitively in identifying groups with an average f1-score of 0.75, outperforming in identifying ethnicity attack spans with an f1-score of 0.80 compared to other identity groups. Moreover, the tool shows an outstanding generalization capability to minority classes concerning sexual orientation and gender, achieving an f1-score of 0.77 and 0.72, respectively. We tested the utility of our tool in a case study on social media, annotating and comparing comments from Facebook related to news mentioning identity groups. The case study reveals differences in the types of attacks recorded, effectively detecting named entities related to the categories of the analyzed news articles. Entities are accurately tagged within their categories, with a negligible error rate for inter-category tagging.

cs.CL↗

arXiv2023

A study on information disorders on social networks during the Chilean social outbreak and COVID-19 pandemic

Information disorders on social media can have a significant impact on citizens' participation in democratic processes. To better understand the spread of false and inaccurate information online, this research analyzed data from Twitter, Facebook, and Instagram. The data was collected and verified by professional fact-checkers in Chile between October 2019 and October 2021, a period marked by political and health crises. The study found that false information spreads faster and reaches more users than true information on Twitter and Facebook. Instagram, on the other hand, seemed to be less affected by this phenomenon. False information was also more likely to be shared by users with lower reading comprehension skills. True information, on the other hand, tended to be less verbose and generate less interest among audiences. This research provides valuable insights into the characteristics of misinformation and how it spreads online. By recognizing the patterns of how false information diffuses and how users interact with it, we can identify the circumstances in which false and inaccurate messages are prone to become widespread. This knowledge can help us develop strategies to counter the spread of misinformation and protect the integrity of democratic processes.

cs.SI↗

arXiv2023

Predicting affinity ties in a surname network

From administrative registers of last names in Santiago, Chile, we create a surname affinity network that encodes socioeconomic data. This network is a multi-relational graph with nodes representing surnames and edges representing the prevalence of interactions between surnames by socioeconomic decile. We model the prediction of links as a knowledge base completion problem, and find that sharing neighbors is highly predictive of the formation of new links. Importantly, We distinguish between grounded neighbors and neighbors in the embedding space, and find that the latter is more predictive of tie formation. The paper discusses the implications of this finding in explaining the high levels of elite endogamy in Santiago.

cs.SI↗

arXiv2023

Surname affinity in Santiago, Chile: A network-based approach that uncovers urban segregation

Based on a geocoded registry of more than four million residents of Santiago, Chile, we build two surname-based networks that reveal the city's population structure. The first network is formed from paternal and maternal surname pairs. The second network is formed from the isonymic distances between the city's neighborhoods. These networks uncover the city's main ethnic groups and their spatial distribution. We match the networks to a socioeconomic index, and find that surnames of high socioeconomic status tend to cluster, be more diverse, and occupy a well-defined quarter of the city. The results are suggestive of a high degree of urban segregation in Santiago.

cs.SI↗

arXiv2022

Evaluation Benchmarks for Spanish Sentence Representations

Due to the success of pre-trained language models, versions of languages other than English have been released in recent years. This fact implies the need for resources to evaluate these models. In the case of Spanish, there are few ways to systematically assess the models' quality. In this paper, we narrow the gap by building two evaluation benchmarks. Inspired by previous work (Conneau and Kiela, 2018; Chen et al., 2019), we introduce Spanish SentEval and Spanish DiscoEval, aiming to assess the capabilities of stand-alone and discourse-aware sentence representations, respectively. Our benchmarks include considerable pre-existing and newly constructed datasets that address different tasks from various domains. In addition, we evaluate and analyze the most recent pre-trained Spanish language models to exhibit their capabilities and limitations. As an example, we discover that for the case of discourse evaluation tasks, mBERT, a language model trained on multiple languages, usually provides a richer latent representation than models trained only with documents in Spanish. We hope our contribution will motivate a fairer, more comparable, and less cumbersome way to evaluate future Spanish language models.

cs.CL↗

arXiv2021

Augmenting BERT-style Models with Predictive Coding to Improve Discourse-level Representations

Current language models are usually trained using a self-supervised scheme, where the main focus is learning representations at the word or sentence level. However, there has been limited progress in generating useful discourse-level representations. In this work, we propose to use ideas from predictive coding theory to augment BERT-style language models with a mechanism that allows them to learn suitable discourse-level representations. As a result, our proposed approach is able to predict future sentences using explicit top-down connections that operate at the intermediate layers of the network. By experimenting with benchmarks designed to evaluate discourse-related knowledge using pre-trained sentence representations, we demonstrate that our approach improves performance in 6 out of 11 tasks by excelling in discourse relationship detection.

cs.CL↗

arXiv2021

Neural Abstractive Unsupervised Summarization of Online News Discussions

Summarization has usually relied on gold standard summaries to train extractive or abstractive models. Social media brings a hurdle to summarization techniques since it requires addressing a multi-document multi-author approach. We address this challenging task by introducing a novel method that generates abstractive summaries of online news discussions. Our method extends a BERT-based architecture, including an attention encoding that fed comments' likes during the training stage. To train our model, we define a task which consists of reconstructing high impact comments based on popularity (likes). Accordingly, our model learns to summarize online discussions based on their most relevant comments. Our novel approach provides a summary that represents the most relevant aspects of a news item that users comment on, incorporating the social context as a source of information to summarize texts in online social networks. Our model is evaluated using ROUGE scores between the generated summary and each comment on the thread. Our model, including the social attention encoding, significantly outperforms both extractive and abstractive summarization methods based on such evaluation.

cs.CL↗

arXiv2021

Inspecting the concept knowledge graph encoded by modern language models

The field of natural language understanding has experienced exponential progress in the last few years, with impressive results in several tasks. This success has motivated researchers to study the underlying knowledge encoded by these models. Despite this, attempts to understand their semantic capabilities have not been successful, often leading to non-conclusive, or contradictory conclusions among different works. Via a probing classifier, we extract the underlying knowledge graph of nine of the most influential language models of the last years, including word embeddings, text generators, and context encoders. This probe is based on concept relatedness, grounded on WordNet. Our results reveal that all the models encode this knowledge, but suffer from several inaccuracies. Furthermore, we show that the different architectures and training strategies lead to different model biases. We conduct a systematic evaluation to discover specific factors that explain why some concepts are challenging. We hope our insights will motivate the development of models that capture concepts more precisely.

cs.AI↗

arXiv2021

A data-driven strategy to combine word embeddings in information retrieval

Word embeddings are vital descriptors of words in unigram representations of documents for many tasks in natural language processing and information retrieval. The representation of queries has been one of the most critical challenges in this area because it consists of a few terms and has little descriptive capacity. Strategies such as average word embeddings can enrich the queries' descriptive capacity since they favor the identification of related terms from the continuous vector representations that characterize these approaches. We propose a data-driven strategy to combine word embeddings. We use Idf combinations of embeddings to represent queries, showing that these representations outperform the average word embeddings recently proposed in the literature. Experimental results on benchmark data show that our proposal performs well, suggesting that data-driven combinations of word embeddings are a promising line of research in ad-hoc information retrieval.

cs.IR↗

arXiv2018

Unsupervised Learning of Structure in Spectroscopic Cubes

We consider the problem of analyzing the structure of spectroscopic cubes using unsupervised machine learning techniques. We propose representing the target's signal as a homogeneous set of volumes through an iterative algorithm that separates the structured emission from the background while not overestimating the flux. Besides verifying some basic theoretical properties, the algorithm is designed to be tuned by domain experts, because its parameters have meaningful values in the astronomical context. Nevertheless, we propose a heuristic to automatically estimate the signal-to-noise ratio parameter of the algorithm directly from data. The resulting light-weighted set of samples ($\leq 1\%$ compared to the original data) offer several advantages. For instance, it is statistically correct and computationally inexpensive to apply well-established techniques of the pattern recognition and machine learning domains; such as clustering and dimensionality reduction algorithms. We use ALMA science verification data to validate our method, and present examples of the operations that can be performed by using the proposed representation. Even though this approach is focused on providing faster and better analysis tools for the end-user astronomer, it also opens the possibility of content-aware data discovery by applying our algorithm to big data.

astro-ph.IM↗

arXiv2018

Viscovery: Trend Tracking in Opinion Forums based on Dynamic Topic Models

Opinions in forums and social networks are released by millions of people due to the increasing number of users that use Web 2.0 platforms to opine about brands and organizations. For enterprises or government agencies it is almost impossible to track what people say producing a gap between user needs/expectations and organizations actions. To bridge this gap we create Viscovery, a platform for opinion summarization and trend tracking that is able to analyze a stream of opinions recovered from forums. To do this we use dynamic topic models, allowing to uncover the hidden structure of topics behind opinions, characterizing vocabulary dynamics. We extend dynamic topic models for incremental learning, a key aspect needed in Viscovery for model updating in near-real time. In addition, we include in Viscovery sentiment analysis, allowing to separate positive/negative words for a specific topic at different levels of granularity. Viscovery allows to visualize representative opinions and terms in each topic. At a coarse level of granularity, the dynamic of the topics can be analyzed using a 2D topic embedding, suggesting longitudinal topic merging or segmentation. In this paper we report our experience developing this platform, sharing lessons learned and opportunities that arise from the use of sentiment analysis and topic modeling in real world applications.

cs.IR↗

arXiv2016

Revisiting Link Prediction: Evolving Models and Real Data Findings

The explosive growth of Web 2.0, which was characterized by the creation of online social networks, has reignited the study of factors that could help us understand the growth and dynamism of these networks. Various generative network models have been proposed, including the Barabasi-Albert and Watts-Strogatz models. In this study, we revisit the problem from a perspective that seeks to compare results obtained from these generative models with those from real networks. To this end, we consider the dating network Skout Inc. An analysis is performed on the topological characteristics of the network that could explain the creation of new network links. Afterwards, the results are contrasted with those obtained from the Barabasi-Albert and Watts-Strogatz generative models. We conclude that a key factor that could explain the creation of links originates in its cluster structure, where link recommendations are more precise in Watts-Strogatz segmented networks than in Barabasi-Albert hierarchical networks. This result reinforces the need to establish more and better network segmentation algorithms that are capable of clustering large networks precisely and efficiently.

cs.SI↗

arXiv2016

The Research Space: using the career paths of scholars to predict the evolution of the research output of individuals, institutions, and nations

In recent years scholars have built maps of science by connecting the academic fields that cite each other, are cited together, or that cite a similar literature. But since scholars cannot always publish in the fields they cite, or that cite them, these science maps are only rough proxies for the potential of a scholar, organization, or country, to enter a new academic field. Here we use a large dataset of scholarly publications disambiguated at the individual level to create a map of science-or research space-where links connect pairs of fields based on the probability that an individual has published in both of them. We find that the research space is a significantly more accurate predictor of the fields that individuals and organizations will enter in the future than citation based science maps. At the country level, however, the research space and citations based science maps are equally accurate. These findings show that data on career trajectories-the set of fields that individuals have previously published in-provide more accurate predictors of future research output for more focalized units-such as individuals or organizations-than citation based science maps.

cs.DL↗

arXiv2014

Revealing Comparative Advantages in the Backbone of Science

Mapping Science across countries is a challenging task in the field of Scientometrics. A number of efforts trying to cope with this task has been discussed in the state of the art, addressing this challenge by processing collections of scientific digital libraries and visualizing author-based measures (for instance, the h-index) or document-based measures (for instance, the averaged number of citations per document). A major drawback of these approaches is related to the presence of bias. The bigger the country, the higher the measure value. We explore the use of an econometric index to tackle this limitation, known as the Revealed Comparative Advantage measure (RCA). Using RCA, the diversity and ubiquity of each field of knowledge is mapped across countries. Then, a RCA-based proximity function is explored to visualize citation and h-index ubiquity. Science maps relating 27 knowledge areas and 237 countries are introduced using data crawled from Scimago that ranges from 1996 to 2011. Our results shows that the proposal is feasible and can be extended to ellaborate a global scientific production characterization.

cs.DL↗

arXiv2014

Affinity Prediction in Online Social Networks

Link prediction is the problem of inferring whether potential edges between pairs of vertices in a graph will be present or absent in the near future. To perform this task it is usual to use information provided by a number of available and observed vertices/edges. Then, a number of edge scoring methods based on this information can be created. Usually, these methods assess local structures of the observed graph, assuming that closer vertices in the original period of observation will be more likely to form a link in the future. In this paper we explore the combination of local and global features to conduct link prediction in online social networks. The contributions of the paper are twofold: a) We evaluate a number of strategies that combines global and local features tackling the locality assumption of link prediction scoring methods, and b) We only use network topology-based features, avoiding the inclusion of informational or transactional based features that involve heavy computational costs in the methods. We evaluate our proposal using real-world data provided by Skout Inc., an affinity online social network with millions of users around the world. Our results show that our proposal is feasible.

cs.SI↗

arXiv2013

Says who? Automatic Text-Based Content Analysis of Television News

We perform an automatic analysis of television news programs, based on the closed captions that accompany them. Specifically, we collect all the news broadcasted in over 140 television channels in the US during a period of six months. We start by segmenting, processing, and annotating the closed captions automatically. Next, we focus on the analysis of their linguistic style and on mentions of people using NLP methods. We present a series of key insights about news providers, people in the news, and we discuss the biases that can be uncovered by automatic means. These insights are contrasted by looking at the data from multiple points of view, including qualitative assessment.

cs.CL↗