💬 Natural Language Processing (NLP)

Development of multi-dimensional hierarchical classification (MHC) methods for NLP tasks. In MHC, each instance is assigned to one category from each of multiple hierarchical classification trees simultaneously. This is a more general setting than standard multi-label or hierarchical classification, and arises naturally in spoken dialogue systems, product categorization, and medical coding.

The research addresses the characterization of MHC problems, the development of dedicated solving strategies (local vs. global approaches, label correlation modeling), and the design of appropriate performance measures for evaluating MHC systems.

Research on data generation approaches for topic classification in multilingual spoken dialogue systems. Development of methods to augment training data for NLP classifiers in low-resource languages, including back-translation, paraphrasing, and generation of synthetic dialogue turns.

Exploration of large language models (LLMs) as tools for feature selection and anomaly classification in industrial settings. Research on stacking predictions from multiple LLM models (ensemble of LLMs) to improve accuracy and explainability in anomaly detection tasks, particularly for building management systems.

Use of large language models to generate causal explanations for anomaly classification decisions. Research on how LLMs can provide human-readable explanations of why a particular observation is classified as anomalous, improving the transparency and trustworthiness of anomaly detection systems.

Development and annotation of a dialogue-act taxonomy for a virtual coach system designed to improve the quality of life of elderly people. Research on how dialogue acts can be used to structure the conversational behavior of virtual assistants that support elderly users in their daily activities.

Empirical comparison of distance and similarity measures for natural language processing tasks. Evaluation of different measures (including edit distance, cosine similarity, and neural embedding distances) for tasks such as sentence similarity, question answering, and dialogue act classification.

Research comparing word embedding models and computer vision models in their ability to predict fMRI brain activity during visual word recognition. The study examines how well different computational representations of word meaning correspond to the neural representations observed in human brains during language processing.

Participation in the ENIA (Estrategia Nacional de Inteligencia Artificial) Chair at UPV/EHU, hosted within the HiTZ Center for Language Technology. The chair focuses on artificial intelligence and language technology, with activities covering research, teaching, and knowledge transfer in AI for language processing tasks.

• Santana R (2017). Reproducing and learning new algebraic operations on word embeddings using genetic programming. GECCO 2017.
• Santana R (2021). Semantic Composition of Word-Embeddings with Genetic Programming. GECCO 2021.
• Magalhães RPL, Santana R and Pozo A (2019). An empirical comparison of distance/similarity measures for Natural Language Processing. BRACIS 2019.
• Montenegro M, Marafioti A, Santana R, Alcaide JB, Bolaños M and Cuayahuitl H (2019). Data generation approaches for topic classification in multilingual spoken dialog system. INTERSPEECH 2019.
• Montenegro M, Santana R, Bolaños M and Cuayahuitl H (2020). Transfer learning in hierarchical dialogue topic classification with neural networks. IberSPEECH 2020.
• Montenegro M, Santana R, Bolaños M and Cuayahuitl H (2021). Analysis of the sensitivity of the End-Of-Turn Detection task to errors generated by the Automatic Speech Recognition module. INTERSPEECH 2021.
• Roman I, Santana R, Mendiburu A and Lozano JA (2019). Sentiment analysis with genetically evolved Gaussian kernels. GECCO 2019.
• Roman I, Santana R, Mendiburu A and Lozano JA (2021). Evolution of Gaussian Process kernels for machine translation post-editing effort estimation. Applied Soft Computing.
• Mei N, Carreiras M, Santana R and Pylkkänen L (2019). How the brain encodes meaning: Comparing word embedding and computer vision models to predict fMRI data. NeurIPS Workshop.
• Santana R (2022). An embedding space for SARS-CoV-2 epitope-based vaccines. European Journal of Clinical Investigation.
• Romero M and Santana R (2022). Creating wordless meaning in word-embedding. In preparation.