Browsing by Author "Mendoza Rocha, Marcelo Gabriel"

Now showing 1 - 10 of 10
  • Thumbnail Image
    Item
    A New Content-Based Image Retrieval System for SARS-CoV-2 Computer-Aided Diagnosis
    (2022) Molina, Gabriel; Mendoza Rocha, Marcelo Gabriel; Loayza, Ignacio; Núñez, Camilo; Araya, Mauricio; Castañeda, Víctor; Solar, Mauricio
  • No Thumbnail Available
    Item
    Augmenting BERT-style Models with Predictive Coding to Improve Discourse-level Representations
    (ASSOC COMPUTATIONAL LINGUISTICS-ACL, 2021) Araujo Vasquez, Vladimir Giovanny; Villa, Andres; Mendoza Rocha, Marcelo Gabriel; Moens, Marie-Francine; Soto, Alvaro
    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.
  • Thumbnail Image
    Item
    Bimodal Style Transference from Musical Composition to Image Using Deep Generative Models
    (2023) Apolo, María José; Mendoza Rocha, Marcelo Gabriel
  • Thumbnail Image
    Item
    CLNews: The First Dataset of the Chilean Social Outbreak for Disinformation Analysis
    (Association for Computing Machinery, 2022) Providel, Eliana; Toro, Daniel; Riquelme, Fabián; Mendoza Rocha, Marcelo Gabriel; Puraivan, E.
    Disinformation is one of the main threats that loom on social networks. Detecting disinformation is not trivial and requires training and maintaining fact-checking teams, which is labor-intensive. Recent studies show that the propagation structure of claims and user messages allows a better understanding of rumor dynamics. Despite these findings, the availability of verified claims and structural propagation data is low. This paper presents a new dataset with Twitter claims verified by fact-checkers along with the propagation structure of retweets and replies. The dataset contains verified claims checked during the Chilean social outbreak, which allows for studying the phenomenon of disinformation during this crisis. We study propagation patterns of verified content in CLNews, showing differences between false rumors and other types of content. Our results show that false rumors are more persistent than the rest of verified contents, reaching more people than truthful news and presenting low barriers of readability to users. The dataset is fully available and helps understand the phenomenon of disinformation during social crises being one of the first of its kind to be released.
  • Thumbnail Image
    Item
    Exploring the Impact of Generative AI for StandUp Report Recommendations in Software Capstone Project Development
    (2024) Neyem, Hugo Andrés; Sandoval Alcocer, Juan Pablo; Mendoza Rocha, Marcelo Gabriel; Centellas-Claro, Leonardo; González, Luis A.; Paredes Robles, Carlos Daniel
    StandUp Reports play an important role in capstone software engineering courses, facilitating progress tracking, obstacle identification, and team collaboration. However, despite their significance, students often grapple with the challenge of creating StandUp Reports that are clear, concise, and actionable. This paper investigates the impact of the use of generative AI in producing StandUp report recommendations, aiming to assist students in enhancing the quality and effectiveness of their reports. In a semester-long capstone course, 179 students participated in 16 real-world software development projects. They submitted weekly StandUp Reports with the assistance of an AI-powered Slack, which analyzed their initial reports and provided suggestions for enhancing them using both GPT-3.5 and the early access GPT-4 API. After each submitted report, students voluntarily answered a survey about usability and suggestion preference. Furthermore, we conducted a linguistic analysis of the recommendations made by the algorithms to gauge reading ease and comprehension complexity. Our findings indicate that the AI-based recommendation system helped students improve the overall quality of their StandUp Reports throughout the semester. Students expressed a high level of satisfaction with the tool and exhibited a strong willingness to continue using it in the future. The survey reveals that students perceived a slight improvement when using GPT-4 compared to GPT-3.5. Finally, a computational linguistic analysis performed on the recommendations demonstrates that both algorithms significantly improve the alignment between the generated texts and the students' educational level, thereby improving the quality of the original texts.
  • Thumbnail Image
    Item
    Inspecting the concept knowledge graph encoded by modern language models
    (Association for Computational Linguistics (ACL), 2021) Aspillaga, Carlos; Soto, Alvaro; Mendoza Rocha, Marcelo Gabriel
    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.
  • No Thumbnail Available
    Item
    Inspecting the concept knowledge graph encoded by modern language models
    (Association for Computational Linguistics (ACL), 2021) Aspillaga C.; Soto A.; Mendoza Rocha, Marcelo Gabriel
    © 2021 Association for Computational LinguisticsThe 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.
  • Thumbnail Image
    Item
    SpatialCluster: A Python library for urban clustering
    (Elsevier B.V., 2024) Reyes A.; Mendoza Rocha, Marcelo Gabriel; Vera, Camila; Lucchini Wortzman, Francesca; Dimter J.; Gutierrez F.; Bro N.; Lobel Díaz, Hans Albert; Reyes A.
    This paper introduces SpatialCluster, a Python library developed for clustering urban areas using geolocated data. The library integrates a range of methods for urban clustering, including Deep Modularity Networks, Gaussian Mixtures, K-Nearest Neighbours, Self Organized Maps, and Information-Theoretic Clustering, providing a comprehensive framework. These methods are evaluated using indices such as the Adjusted Rand Index and Adjusted Mutual Information, and the library includes features for detailed map visualization. SpatialCluster's online documentation offers examples, making the library accessible to researchers and urban planners. The library aims to facilitate urban data analysis and contribute to the field of urban studies.
  • Thumbnail Image
    Item
    Supporting Users in Refining and Comparing Topic Models: An Experimental Study
    (2023) González-Pizarro, F.; Moncada, C.L.; Milios, E.V.; Paulovich, F.; Mendoza Rocha, Marcelo Gabriel
  • Thumbnail Image
    Item
    Towards an AI Knowledge Assistant for Context-aware Learning Experiences in Software Capstone Project Development
    (2024) Neyem, Hugo Andres; González, Luis A.; Mendoza Rocha, Marcelo Gabriel; Sandoval Alcocer, Juan Pablo; Centellas, Leonardo; Paredes, Carlos
    Software assistants have significantly impacted software development for both practitioners and students, particularly in capstone projects. The effectiveness of these tools varies based on their knowledge sources; assistants with localized, domain-specific knowledge may have limitations, while tools like ChatGPT, using broad datasets, might offer recommendations that do not always match the specific objectives of a capstone course. Addressing a gap in current educational technology, this paper introduces an AI Knowledge Assistant specifically designed to overcome the limitations of existing tools by enhancing the quality and relevance of Large Language Models (LLMs). It achieves this through the innovative integration of contextual knowledge from a local “lessons learned” database tailored to the capstone course. We conducted a study with 150 students using the assistant during their capstone course. Integrated into the Kanban project tracking system, the assistant offered recommendations using different strategies: direct searches in the lessons learned database, direct queries to a GPT model, query enrichment with lessons learned before submission to GPT and LLaMa models, and query enhancement with Stack Overflow data before GPT processing. Survey results underscored a strong preference among students for direct LLM queries and those enriched with local repository insights, highlighting the assistant's practical value. Further, our linguistic analysis conclusively demonstrated that texts generated by the LLM closely mirrored the linguistic standards and topical relevance of university course requirements. This alignment not only fosters a deeper understanding of course content but also significantly enhances the material's applicability to real-world scenarios.