Maria Mora-Cross
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Maria Mora-Cross has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- William Ulate
Missouri Botanical Garden St. Louis MO United States of America
3 shared publications Manuel Vargas
2 shared publicationsAdriana Morales-Carmiol
1 shared publication- Adriana Morales-Carmiol1 shared publication
Brandon Retana Chacón
1 shared publication- Camila Plata1 shared publication
Latest Publications
- Open Access
- 13 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 213-216
Multiobjective Evolutionary Calibration of Uncertainty Quantification for Large Language Models
- 13 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 213-216
Open AccessMultiobjective Evolutionary Calibration of Uncertainty Quantification for Large Language Models
- 4 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Relation Extraction from Unstructured Species Descriptions Using TaxonNERD and Llama 7B
Ontologies have become essential tools in knowledge representation, particularly in organizing information on taxonomy, ecology, and inter-species relationships. Their integration with Natural Language Processing (NLP) methods, such as Named Entity Recognition (NER) and Relation Extraction (RE), has significantly advanced the extraction of relevant ecological data from scientific texts. This research explores the fusion of ontological frameworks and NLP methodologies, focusing on animal feeding relationships. Using a dataset from the National Biodiversity Institute (INBio), we applied TaxoNERD for entity extraction, followed by fine-tuning the LLaMA2-7b model to analyze complex feeding behaviors and predator-prey interactions. Our results show a recall of 0.73 and a precision of 0.68, highlighting the model's effectiveness in capturing feeding relationships. These structured insights into species' diets offer valuable contributions to ecological research and conservation efforts.
- 4 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Relation Extraction from Unstructured Species Descriptions Using TaxonNERD and Llama 7B
Ontologies have become essential tools in knowledge representation, particularly in organizing information on taxonomy, ecology, and inter-species relationships. Their integration with Natural Language Processing (NLP) methods, such as Named Entity Recognition (NER) and Relation Extraction (RE), has significantly advanced the extraction of relevant ecological data from scientific texts. This research explores the fusion of ontological frameworks and NLP methodologies, focusing on animal feeding relationships. Using a dataset from the National Biodiversity Institute (INBio), we applied TaxoNERD for entity extraction, followed by fine-tuning the LLaMA2-7b model to analyze complex feeding behaviors and predator-prey interactions. Our results show a recall of 0.73 and a precision of 0.68, highlighting the model's effectiveness in capturing feeding relationships. These structured insights into species' diets offer valuable contributions to ecological research and conservation efforts.
- Open Access
- 22 November 2024
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 8, e142382
Relation Extraction From Unstructured Species Descriptions Using TaxonNERD and LLaMA 2 7B
Ontologies are essential tools for organizing information on taxonomy, ecology, and inter-species relationships, helping to standardize ecological data and facilitate integration of large datasets. Combining ontologies with advanced Natural Language Processing (NLP) techniques, such as Named Entity Recognition (NER) and Relation Extraction (RE), has greatly improved the discovery of insights from unstructured scientific texts, particularly in biodiversity (Gabud et al. 2023, Abdelmageed et al. 2022, Hearst 1992). This study combines ontologies and NLP to analyze complex trophic interactions among animal species (Gabud et al. 2023), using a dataset (National Biodiversity Institute of Costa Rica (INBio) 2015) containing species descriptions in English and Spanish. We applied TaxoNERD to identify taxonomic entities (Le Guillarme and Thuiller 2021) and we fine-tuned the Large Language Model Meta AI (LLaMA 2 7B) to extract feeding interactions and predator-prey relationships (CheeKean 2023), due to its effectiveness in handling complex language patterns and its adaptability to diverse scientific domains. Our results (Fig. 1) showed a recall of 0.73 and a precision of 0.68, indicating that the model effectively identifies feeding relationships in most cases. However, the lower precision suggests that the model may still capture some unrelated interactions, highlighting an area for improvement to reduce false positives and increase accuracy (Touvron et al. 2023). Previous studies also emphasize the need for further refinement of relation extraction models to enhance accuracy (Mora-Cross et al. 2023). The structured dataset offers valuable insights into species’ diets and roles, contributing to biodiversity research and conservation efforts (Mora-Cross et al. 2023, Touvron et al. 2023). Moreover, this research highlights the potential of integrating AI-driven tools with ontological frameworks to manage and analyze biodiversity data at scale (Abdelmageed et al. 2022). By transforming unstructured text into structured data, we make ecological information more accessible, supporting better decision-making in conservation strategies (Abdelmageed et al. 2022, Hearst 1992). This approach scales well with the growing volume of biodiversity data, offering a more efficient and accurate method for analyzing species interactions, which are crucial for ecosystem management and endangered species protection (Gabud et al. 2023).
- 22 November 2024
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 8, e142382
Open AccessRelation Extraction From Unstructured Species Descriptions Using TaxonNERD and LLaMA 2 7B
Ontologies are essential tools for organizing information on taxonomy, ecology, and inter-species relationships, helping to standardize ecological data and facilitate integration of large datasets. Combining ontologies with advanced Natural Language Processing (NLP) techniques, such as Named Entity Recognition (NER) and Relation Extraction (RE), has greatly improved the discovery of insights from unstructured scientific texts, particularly in biodiversity (Gabud et al. 2023, Abdelmageed et al. 2022, Hearst 1992). This study combines ontologies and NLP to analyze complex trophic interactions among animal species (Gabud et al. 2023), using a dataset (National Biodiversity Institute of Costa Rica (INBio) 2015) containing species descriptions in English and Spanish. We applied TaxoNERD to identify taxonomic entities (Le Guillarme and Thuiller 2021) and we fine-tuned the Large Language Model Meta AI (LLaMA 2 7B) to extract feeding interactions and predator-prey relationships (CheeKean 2023), due to its effectiveness in handling complex language patterns and its adaptability to diverse scientific domains. Our results (Fig. 1) showed a recall of 0.73 and a precision of 0.68, indicating that the model effectively identifies feeding relationships in most cases. However, the lower precision suggests that the model may still capture some unrelated interactions, highlighting an area for improvement to reduce false positives and increase accuracy (Touvron et al. 2023). Previous studies also emphasize the need for further refinement of relation extraction models to enhance accuracy (Mora-Cross et al. 2023). The structured dataset offers valuable insights into species’ diets and roles, contributing to biodiversity research and conservation efforts (Mora-Cross et al. 2023, Touvron et al. 2023). Moreover, this research highlights the potential of integrating AI-driven tools with ontological frameworks to manage and analyze biodiversity data at scale (Abdelmageed et al. 2022). By transforming unstructured text into structured data, we make ecological information more accessible, supporting better decision-making in conservation strategies (Abdelmageed et al. 2022, Hearst 1992). This approach scales well with the growing volume of biodiversity data, offering a more efficient and accurate method for analyzing species interactions, which are crucial for ecosystem management and endangered species protection (Gabud et al. 2023).
- Open Access
- 20 August 2024
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 8, e135064
Plinian Core: A Data Specification for Species Pages in the Real World
Plinian Core (PliC), a standard in development by a Biodiversity Information Standards (TDWG) task group, is a set of vocabulary terms that can be used to describe different aspects of biological species information, including all kinds of properties or traits related to taxa—biological and non-biological. PliC incorporates terms pertaining to descriptions, legal aspects, conservation, management, uses, demographics, nomenclature, or related resources (Plinian Core Task Group 2021). Having a data specification is just a small part of what it takes to prepare and publish species information. The specification provides a structured way to describe species-related data, ensuring consistency and interoperability across different platforms and databases. Different aspects of Plinian Core and its development have been presented in the past (Pando 2018; Pando et al. 2022). In this contribution, we shift the focus out of the specification itself, to other key aspects of species information dissemination. Plinian Core has been already put to use in various real-world scenarios. Here we select three examples to present the most significant lessons learned: Cross-Nature Project This project developed a Resource Description Framework for PliC to create an ontology-based endpoint for the Species Information System of the Spanish Ministry for the Environment (EIDOS). It provides a structured framework for describing and integrating species data following the Linked Open Data principles. The system provides information for over 85,000 species. Biodiversity Catalogue of SiB Colombia This national catalogue provides information on the natural history, conservation, threats, and biology of Colombian species. Aimed at a broad audience, it maintains scientific rigor while being accessible to anyone interested in Colombia's biodiversity. PliC has been essential in this endeavor, enabling the inclusion of both highly structured technical information and rich, engaging textual descriptions for over 6,000 species. EncicloVida Developed by CONABIO, EncicloVida aims to publicize the Mexican biodiversity. The platform integrates biological species information on more than 114,000 species of plants, animals, fungi, bacteria, and protozoa that CONABIO has gathered through the National Information System on Biodiversity (SNIB) in collaboration with Mexican researchers. These implementations are based on the PliC Core abstract Model that is expressed as an XML Schema definitiion (XSD), and the supporting documentation is provided in the PliC Wiki. Automatic aggregation of species pages is challenging and often ineffective as differences in taxonomic treatments, or in local populations (either biological or otherwise: threat status, laws and regulations, etc.) frequently create inconsistencies or even render false information. Conversely, sharing a standard across projects offers immediate benefits, such as consistency of navigation and information, or the facility for sharing software solutions. Plinian Core terms, vocabularies and structure have been developed in collaboration with implementers and in response to real-world demands. That has resulted in a set of fairly large data specifications, but one flexible and easy to understand by domain experts who may not work comfortably with more technical approaches. Each implementation puts emphasis on different aspects, applying different levels of granularity depending on the requirements. This is done by choosing between "unstructured" terms (for low granularity) or the "atomized classes'' (for highly structured information). For instance, dispersal information can be recorded as a text block ("DispersalUnstructured") or using a set of specific terms: Purpose, DispersalMeans, StructureDisperse ("DispersalAtomized"). PliC reuses a number of terms from other standards (see Borrowed terms section) following a well-established best practice. Using XSD files as a basis, we link borrowed terms from external sources to their original specifications, which were also expressed as XSD files. This should provide consistency and reliability. However, this has proven to be problematic to maintain and validate, as links are broken and files changed. To keep the PliC XSD file valid, we opted for maintaining local copies of the XSD of the respective reused standards. We also aim to present some current and future developments of PliC, especially in the context of the Living Atlases community. Thus, we will explore other elements involved in sharing species information such as the Global Biodiversity Information Facility's Integrated Publishing Toolkit extensions, SPARQL Endpoints, Frictionless Data Packages, Research Object-Crates, and integration with Atlas of Living Australia modules.
- 20 August 2024
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 8, e135064
Open AccessPlinian Core: A Data Specification for Species Pages in the Real World
Plinian Core (PliC), a standard in development by a Biodiversity Information Standards (TDWG) task group, is a set of vocabulary terms that can be used to describe different aspects of biological species information, including all kinds of properties or traits related to taxa—biological and non-biological. PliC incorporates terms pertaining to descriptions, legal aspects, conservation, management, uses, demographics, nomenclature, or related resources (Plinian Core Task Group 2021). Having a data specification is just a small part of what it takes to prepare and publish species information. The specification provides a structured way to describe species-related data, ensuring consistency and interoperability across different platforms and databases. Different aspects of Plinian Core and its development have been presented in the past (Pando 2018; Pando et al. 2022). In this contribution, we shift the focus out of the specification itself, to other key aspects of species information dissemination. Plinian Core has been already put to use in various real-world scenarios. Here we select three examples to present the most significant lessons learned: Cross-Nature Project This project developed a Resource Description Framework for PliC to create an ontology-based endpoint for the Species Information System of the Spanish Ministry for the Environment (EIDOS). It provides a structured framework for describing and integrating species data following the Linked Open Data principles. The system provides information for over 85,000 species. Biodiversity Catalogue of SiB Colombia This national catalogue provides information on the natural history, conservation, threats, and biology of Colombian species. Aimed at a broad audience, it maintains scientific rigor while being accessible to anyone interested in Colombia's biodiversity. PliC has been essential in this endeavor, enabling the inclusion of both highly structured technical information and rich, engaging textual descriptions for over 6,000 species. EncicloVida Developed by CONABIO, EncicloVida aims to publicize the Mexican biodiversity. The platform integrates biological species information on more than 114,000 species of plants, animals, fungi, bacteria, and protozoa that CONABIO has gathered through the National Information System on Biodiversity (SNIB) in collaboration with Mexican researchers. These implementations are based on the PliC Core abstract Model that is expressed as an XML Schema definitiion (XSD), and the supporting documentation is provided in the PliC Wiki. Automatic aggregation of species pages is challenging and often ineffective as differences in taxonomic treatments, or in local populations (either biological or otherwise: threat status, laws and regulations, etc.) frequently create inconsistencies or even render false information. Conversely, sharing a standard across projects offers immediate benefits, such as consistency of navigation and information, or the facility for sharing software solutions. Plinian Core terms, vocabularies and structure have been developed in collaboration with implementers and in response to real-world demands. That has resulted in a set of fairly large data specifications, but one flexible and easy to understand by domain experts who may not work comfortably with more technical approaches. Each implementation puts emphasis on different aspects, applying different levels of granularity depending on the requirements. This is done by choosing between "unstructured" terms (for low granularity) or the "atomized classes'' (for highly structured information). For instance, dispersal information can be recorded as a text block ("DispersalUnstructured") or using a set of specific terms: Purpose, DispersalMeans, StructureDisperse ("DispersalAtomized"). PliC reuses a number of terms from other standards (see Borrowed terms section) following a well-established best practice. Using XSD files as a basis, we link borrowed terms from external sources to their original specifications, which were also expressed as XSD files. This should provide consistency and reliability. However, this has proven to be problematic to maintain and validate, as links are broken and files changed. To keep the PliC XSD file valid, we opted for maintaining local copies of the XSD of the respective reused standards. We also aim to present some current and future developments of PliC, especially in the context of the Living Atlases community. Thus, we will explore other elements involved in sharing species information such as the Global Biodiversity Information Facility's Integrated Publishing Toolkit extensions, SPARQL Endpoints, Frictionless Data Packages, Research Object-Crates, and integration with Atlas of Living Australia modules.
- Open Access
- 21 September 2023
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 7, e113055
Structuring Information from Plant Morphological Descriptions using Open Information Extraction
Taxonomic literature keeps records of the planet's biodiversity and gives access to the knowledge needed for research and sustainable management. The number of publications generated is quite large: the corpus of biodiversity literature includes tens of millions of figures and taxonomic treatments. Unfortunately, most of the taxonomic descriptions are from scientific publications in text format. With more than 61 million digitized pages in the Biodiversity Heritage Library (BHL), only 467,265 taxonomic treatments are available in the Biodiversity Literature Repository. To obtain highly structured texts from digitized text has been shown to be complex and very expensive (Cui et al. 2021). The scientific community has described over 1.2 million species, but studies suggest that 86% of existing species on Earth and 91% of species in the ocean still await description (Mora et al. 2011). The published descriptions synthesize observations made by taxonomists over centuries of research and include detailed morphological aspects (i.e., shape and structure) of species useful to identify specimens, to improve information search mechanisms, to perform data analysis of species having particular characteristics, and to compare species descriptions. To take full advantage of this information and to work towards integrating it with repositories of biodiversity knowledge, the biodiversity informatics community first needs to convert plain text into a machine-processable format. More precisely, there is a need to identify structures and substructure names and the characters that describe them (Fig. 1). Open information extraction (OIE) is a research area of Natural Language Processing (NLP), which aims to automatically extract structured, machine-readable representations of data available in unstructured text; usually the result is handled as n-ary propositions, for instance, triples of the form (Shen et al. 2022). OIE is continuously evolving with advancements in NLP and machine learning techniques. The state of the art in OIE involves the use of neural approaches, pre-trained language models, and integration of dependency parsing and semantic role labeling. Neural solutions mainly formulate OIE as a sequence tagging problem or a sequence generation problem. Ongoing research focuses on improving extraction accuracy; handling complex linguistic phenomena, for instance, addressing challenges like coreference resolution; and more open information extraction, because most existing neural solutions work in English texts (Zhou et al. 2022). The main objective of this project is to evaluate and compare the results of automatic data extraction from plant morphological descriptions using pre-trained language models (PLM) and a language model trained on data from plant morphological descriptions written in Spanish. The research data for this study were sourced from the species records database of the National Biodiversity Institute of Costa Rica (INBio). Specifically, the project focused on selecting records of morphological descriptions of plant species written in Spanish. The system processes the morphological descriptions using a workflow that includes phases like data selection and pre-processing, feature extraction, test PLM, local language model training, and test and evaluate results. Fig. 2 shows the general workflow used in this research. Pre-processing and Annotation: Descriptions were standardized by removing special characters like double and single quotes, replacing abbreviations, tokenizing text, and other transformations. Some records of the dataset were annotated with the ground-truth structured information in the form of triples that were extracted from each paragraph. Additionally, structured data from the project carried out by Mora and Araya (Mora and Araya 2018) were included in the dataset. Feature extraction: The token vectorization was done using word embedding directly by the language models. Test PLM: The evaluation process of PLM models used the zero-shot approach and involved applying the models to the test dataset, extracting information, and comparing it to annotated ground truth. Local Language Model Training: The annotated data was split into 80% training data and 20% test data. Using the training data, a language model based on the Transformers architecture was trained. Evaluate results: Evaluation metrics such as precision, recall, and F1 (a meaure of the model's accuracy) were calculated comparing the extracted information and the ground truth. The results were analyzed to understand the models' performance, identify strengths and weaknesses, and gain insights into their ability to extract accurate and relevant information. Based on the analysis, the evaluation process iteratively improved models results. The main contributions of this project are: A Transformers-based language model to extract information from morphological descriptions of plants written in Spanish available on the project website.*1 A corpus of morphological descriptions of plants, written in Spanish, labeled for information extraction, and made available on the project website. The results of the project, the first of its kind applied to morphological descriptions of plants written in Spanish, published on the project website. A Transformers-based language model to extract information from morphological descriptions of plants written in Spanish available on the project website.*1 A corpus of morphological descriptions of plants, written in Spanish, labeled for information extraction, and made available on the project website. The results of the project, the first of its kind applied to morphological descriptions of plants written in Spanish, published on the project website.
- 21 September 2023
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 7, e113055
Open AccessStructuring Information from Plant Morphological Descriptions using Open Information Extraction
Taxonomic literature keeps records of the planet's biodiversity and gives access to the knowledge needed for research and sustainable management. The number of publications generated is quite large: the corpus of biodiversity literature includes tens of millions of figures and taxonomic treatments. Unfortunately, most of the taxonomic descriptions are from scientific publications in text format. With more than 61 million digitized pages in the Biodiversity Heritage Library (BHL), only 467,265 taxonomic treatments are available in the Biodiversity Literature Repository. To obtain highly structured texts from digitized text has been shown to be complex and very expensive (Cui et al. 2021). The scientific community has described over 1.2 million species, but studies suggest that 86% of existing species on Earth and 91% of species in the ocean still await description (Mora et al. 2011). The published descriptions synthesize observations made by taxonomists over centuries of research and include detailed morphological aspects (i.e., shape and structure) of species useful to identify specimens, to improve information search mechanisms, to perform data analysis of species having particular characteristics, and to compare species descriptions. To take full advantage of this information and to work towards integrating it with repositories of biodiversity knowledge, the biodiversity informatics community first needs to convert plain text into a machine-processable format. More precisely, there is a need to identify structures and substructure names and the characters that describe them (Fig. 1). Open information extraction (OIE) is a research area of Natural Language Processing (NLP), which aims to automatically extract structured, machine-readable representations of data available in unstructured text; usually the result is handled as n-ary propositions, for instance, triples of the form (Shen et al. 2022). OIE is continuously evolving with advancements in NLP and machine learning techniques. The state of the art in OIE involves the use of neural approaches, pre-trained language models, and integration of dependency parsing and semantic role labeling. Neural solutions mainly formulate OIE as a sequence tagging problem or a sequence generation problem. Ongoing research focuses on improving extraction accuracy; handling complex linguistic phenomena, for instance, addressing challenges like coreference resolution; and more open information extraction, because most existing neural solutions work in English texts (Zhou et al. 2022). The main objective of this project is to evaluate and compare the results of automatic data extraction from plant morphological descriptions using pre-trained language models (PLM) and a language model trained on data from plant morphological descriptions written in Spanish. The research data for this study were sourced from the species records database of the National Biodiversity Institute of Costa Rica (INBio). Specifically, the project focused on selecting records of morphological descriptions of plant species written in Spanish. The system processes the morphological descriptions using a workflow that includes phases like data selection and pre-processing, feature extraction, test PLM, local language model training, and test and evaluate results. Fig. 2 shows the general workflow used in this research. Pre-processing and Annotation: Descriptions were standardized by removing special characters like double and single quotes, replacing abbreviations, tokenizing text, and other transformations. Some records of the dataset were annotated with the ground-truth structured information in the form of triples that were extracted from each paragraph. Additionally, structured data from the project carried out by Mora and Araya (Mora and Araya 2018) were included in the dataset. Feature extraction: The token vectorization was done using word embedding directly by the language models. Test PLM: The evaluation process of PLM models used the zero-shot approach and involved applying the models to the test dataset, extracting information, and comparing it to annotated ground truth. Local Language Model Training: The annotated data was split into 80% training data and 20% test data. Using the training data, a language model based on the Transformers architecture was trained. Evaluate results: Evaluation metrics such as precision, recall, and F1 (a meaure of the model's accuracy) were calculated comparing the extracted information and the ground truth. The results were analyzed to understand the models' performance, identify strengths and weaknesses, and gain insights into their ability to extract accurate and relevant information. Based on the analysis, the evaluation process iteratively improved models results. The main contributions of this project are: A Transformers-based language model to extract information from morphological descriptions of plants written in Spanish available on the project website.*1 A corpus of morphological descriptions of plants, written in Spanish, labeled for information extraction, and made available on the project website. The results of the project, the first of its kind applied to morphological descriptions of plants written in Spanish, published on the project website. A Transformers-based language model to extract information from morphological descriptions of plants written in Spanish available on the project website.*1 A corpus of morphological descriptions of plants, written in Spanish, labeled for information extraction, and made available on the project website. The results of the project, the first of its kind applied to morphological descriptions of plants written in Spanish, published on the project website.
- Open Access
- 21 September 2023
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 7, e113057
Comparative Study: Evaluating the effects of class balancing on transformer performance in the PlantNet-300k image dataset
Image-based identification of plant specimens plays a crucial role in various fields such as agriculture, ecology, and biodiversity conservation. The growing interest in deep learning has led to remarkable advancements in image classification techniques, particularly with the utilization of convolutional neural networks (CNNs). Since 2015, in the context of the PlantCLEF (Conference and Labs of the Evaluation Forum) challenge (Joly et al. 2015), deep learning models, specifically CNNs, have consistently achieved the most impressive results in this field (Carranza-Rojas 2018). However, recent developments have introduced transformer-based models, such as ViT (Vision Transformer) (Dosovitskiy et al. 2020) and CvT (Convolutional vision Transformer) (Wu et al. 2021), as a promising alternative for image classification tasks. Transformers offer unique advantages such as capturing global context and handling long-range dependencies (Vaswani et al. 2017), which make them suitable for complex recognition tasks like plant identification. In this study, we focus on the image classification task using the PlantNet-300k dataset (Garcin et al. 2021a). The dataset consists of a large collection of 306,146 plant images representing 1,081 distinct species. These images were selected from the Pl@ntNet citizen observatory database. The dataset has two prominent characteristics that pose challenges for classification. First, there is a significant class imbalance, meaning that a small subset of species dominates the majority of the images. This imbalance creates bias and affects the accuracy of classification models. Second, many species exhibit visual similarities, making it tough, even for experts, to accurately identify them. These characteristics are referred to by the dataset authors as long-tailed distribution and high intrinsic ambiguity, respectively (Garcin et al. 2021b). In order to address the inherent challenges of the PlantNet-300k dataset, we employed a two-fold approach. Firstly, we leveraged transformer-based models to tackle the dataset's intrinsic ambiguity and effectively capture the complex visual patterns present in plant images. Secondly, we focused on mitigating the class imbalance issue through various data preprocessing techniques, specifically class balancing methods. By implementing these techniques, we aimed to ensure fair representation of all plant species in order to improve the overall performance of image classification models. Our objective is to assess the effects of data preprocessing techniques, specifically class balancing, on the classification performance of the PlantNet-300k dataset. By exploring different preprocessing methods, we addressed the class imbalance issue and through precise evaluation, conducted a comparison of the performance of transformer-based models with and without class balancing techniques. Through these efforts, our ultimate goal is to assert if these techniques allow us to achieve more accurate and reliable classification results, particularly for underrepresented species in the dataset. In our experiment, we compared the performance of two transformer-based models, ViT and CvT, using two versions of the PlantNet-300k dataset: one with class balancing and the other without class balancing. This setup results in a total of four sets of metrics for evaluation. To assess the classification performance, we utilized a wide range of commonly used metrics including recall, precision, accuracy, AUC (Area Under the Curve), ROC (Receiver Operating Characteristic), and others. These metrics provide insights into each models' ability to correctly classify plant species, identify false positives and negatives, measure overall accuracy, and assess the models' discriminatory power. By conducting this comparative study, we seek to contribute to the advancement of plant identification research by providing empirical evidence of the benefits and effectiveness of class balancing techniques in improving the performance of transformer-based models on the PlantNet-300k dataset and any other similar ones.
- 21 September 2023
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 7, e113057
Open AccessComparative Study: Evaluating the effects of class balancing on transformer performance in the PlantNet-300k image dataset
Image-based identification of plant specimens plays a crucial role in various fields such as agriculture, ecology, and biodiversity conservation. The growing interest in deep learning has led to remarkable advancements in image classification techniques, particularly with the utilization of convolutional neural networks (CNNs). Since 2015, in the context of the PlantCLEF (Conference and Labs of the Evaluation Forum) challenge (Joly et al. 2015), deep learning models, specifically CNNs, have consistently achieved the most impressive results in this field (Carranza-Rojas 2018). However, recent developments have introduced transformer-based models, such as ViT (Vision Transformer) (Dosovitskiy et al. 2020) and CvT (Convolutional vision Transformer) (Wu et al. 2021), as a promising alternative for image classification tasks. Transformers offer unique advantages such as capturing global context and handling long-range dependencies (Vaswani et al. 2017), which make them suitable for complex recognition tasks like plant identification. In this study, we focus on the image classification task using the PlantNet-300k dataset (Garcin et al. 2021a). The dataset consists of a large collection of 306,146 plant images representing 1,081 distinct species. These images were selected from the Pl@ntNet citizen observatory database. The dataset has two prominent characteristics that pose challenges for classification. First, there is a significant class imbalance, meaning that a small subset of species dominates the majority of the images. This imbalance creates bias and affects the accuracy of classification models. Second, many species exhibit visual similarities, making it tough, even for experts, to accurately identify them. These characteristics are referred to by the dataset authors as long-tailed distribution and high intrinsic ambiguity, respectively (Garcin et al. 2021b). In order to address the inherent challenges of the PlantNet-300k dataset, we employed a two-fold approach. Firstly, we leveraged transformer-based models to tackle the dataset's intrinsic ambiguity and effectively capture the complex visual patterns present in plant images. Secondly, we focused on mitigating the class imbalance issue through various data preprocessing techniques, specifically class balancing methods. By implementing these techniques, we aimed to ensure fair representation of all plant species in order to improve the overall performance of image classification models. Our objective is to assess the effects of data preprocessing techniques, specifically class balancing, on the classification performance of the PlantNet-300k dataset. By exploring different preprocessing methods, we addressed the class imbalance issue and through precise evaluation, conducted a comparison of the performance of transformer-based models with and without class balancing techniques. Through these efforts, our ultimate goal is to assert if these techniques allow us to achieve more accurate and reliable classification results, particularly for underrepresented species in the dataset. In our experiment, we compared the performance of two transformer-based models, ViT and CvT, using two versions of the PlantNet-300k dataset: one with class balancing and the other without class balancing. This setup results in a total of four sets of metrics for evaluation. To assess the classification performance, we utilized a wide range of commonly used metrics including recall, precision, accuracy, AUC (Area Under the Curve), ROC (Receiver Operating Characteristic), and others. These metrics provide insights into each models' ability to correctly classify plant species, identify false positives and negatives, measure overall accuracy, and assess the models' discriminatory power. By conducting this comparative study, we seek to contribute to the advancement of plant identification research by providing empirical evidence of the benefits and effectiveness of class balancing techniques in improving the performance of transformer-based models on the PlantNet-300k dataset and any other similar ones.
- Open Access
- 12 September 2022
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 6, e94718
Sharing Species Pages through Living Atlases using Plinian Core: Current state and new developments
Living Atlases (living-atlases.gbif.org) is a community of developers and organizations built around the software platform developed by the Atlas of Living Australia (ALA, ala.org.au). ALA includes a number of open source reusable modules designed to assemble biodiversity data portals. Nowadays, more than 25 institutional and national biodiversity portals around the world have been implemented with the ALA tools (living-atlases.gbif.org/participants). The community organizes workshops and prepares training materials to present ALA modules to other institutions that want to implement it, to improve already existing portals and to learn from each other’s achievements (living-atlases.gbif.org/about). ALA based portals integrate data from multiple sources using biodiversity data standards ruled by TDWG, such as Darwin Core for species occurrences. The Living Atlases community is working in the integration of other data types, like species records and sampling events. Plinian Core (PliC) is a generalistic specification, oriented to share descriptions and nomenclature, as well as many other aspects (e.g. legal, conservation, and management details) of species level information from local and regional sources (Pando 2018). Following several years of development and earlier versions, PliC version 3.1 was made publicly available in 2012. The following year, after the approval of the “Species Information Interest Group'' by the TDWG Executive, a TDWG Task Group was created to start aligning Plinian Core with the TDWG Standards Documentation Standard (SDS). Several full-fledged implementations of Plinian Core were produced, between 2014 and 2019, including the Atlas of Living Costa Rica (CRBio.cr), the Cross-Nature Project (datos.iepnb.es), the Vasque Country (Spain) Official Nature Information System (ivap.euskadi.eus), the Colombian Biodiversity Catalog (catalogo.biodiversidad.co) and Enciclovida (Mexican Species Catalog, enciclovida.mx). These last three years, the Plinian Core Task Group has reviewed definitions of terms and formal declarations, assisted by the TDWG Technical Architecture Group (TAG), with the objective of ratifying PliC as a TDWG Standard. One of the most used tools to share standardized biodiversity data is the GBIF Integrated Publishing Toolkit (IPT, gbif.org/en/ipt). The IPT is an open source web application that allows institutions to standardize, to publish and share biodiversity data sets through Living Atlases, GBIF, and other biodiversity networks. The toolkit simplifies the process of publishing core data types (e.g. taxon occurrence, taxon checklists, sampling event) and, by means of extensions, it is possible to associate additional data with records of the core type. As part of our project, the extensions of PliC, accessible in the IPT Sandbox, were validated to show how they can be used by institutions to publish species pages through the IPT. In order to fully integrate Plinian Core based species pages into the ALA architecture, the Biodiversity Informatics Research Center (CRBio) has been working on the adoption of the Biodiversity Information Explorer (BIE), an ALA module which manages taxonomic and species contents by integrating global resources like EOL or Wikipedia. This adoption includes the required modifications of the Portal codebase to use the data model of the Plinian Core (github.com/tdwg/PlinianCore) to display a much richer content. This year, with assistance from the Capacity Enhancement Support Programme (CESP) from GBIF, we have revisited the integration of species pages information and expanded the content and type of information provided by the Atlas of Living Costa Rica. We present here the updated ALA Species Pages (ALA-BIE) and Index (BIE-Index) Modules and general documentation that could be applied to any ALA installation in order to prepare species records using the IPT and PliC standard, index and display them through the ALA-BIE Module.
- 12 September 2022
- Published by Pensoft Publishers in Biodiversity Information Science and Standards
- Vol. 6, e94718
Open AccessSharing Species Pages through Living Atlases using Plinian Core: Current state and new developments
Living Atlases (living-atlases.gbif.org) is a community of developers and organizations built around the software platform developed by the Atlas of Living Australia (ALA, ala.org.au). ALA includes a number of open source reusable modules designed to assemble biodiversity data portals. Nowadays, more than 25 institutional and national biodiversity portals around the world have been implemented with the ALA tools (living-atlases.gbif.org/participants). The community organizes workshops and prepares training materials to present ALA modules to other institutions that want to implement it, to improve already existing portals and to learn from each other’s achievements (living-atlases.gbif.org/about). ALA based portals integrate data from multiple sources using biodiversity data standards ruled by TDWG, such as Darwin Core for species occurrences. The Living Atlases community is working in the integration of other data types, like species records and sampling events. Plinian Core (PliC) is a generalistic specification, oriented to share descriptions and nomenclature, as well as many other aspects (e.g. legal, conservation, and management details) of species level information from local and regional sources (Pando 2018). Following several years of development and earlier versions, PliC version 3.1 was made publicly available in 2012. The following year, after the approval of the “Species Information Interest Group'' by the TDWG Executive, a TDWG Task Group was created to start aligning Plinian Core with the TDWG Standards Documentation Standard (SDS). Several full-fledged implementations of Plinian Core were produced, between 2014 and 2019, including the Atlas of Living Costa Rica (CRBio.cr), the Cross-Nature Project (datos.iepnb.es), the Vasque Country (Spain) Official Nature Information System (ivap.euskadi.eus), the Colombian Biodiversity Catalog (catalogo.biodiversidad.co) and Enciclovida (Mexican Species Catalog, enciclovida.mx). These last three years, the Plinian Core Task Group has reviewed definitions of terms and formal declarations, assisted by the TDWG Technical Architecture Group (TAG), with the objective of ratifying PliC as a TDWG Standard. One of the most used tools to share standardized biodiversity data is the GBIF Integrated Publishing Toolkit (IPT, gbif.org/en/ipt). The IPT is an open source web application that allows institutions to standardize, to publish and share biodiversity data sets through Living Atlases, GBIF, and other biodiversity networks. The toolkit simplifies the process of publishing core data types (e.g. taxon occurrence, taxon checklists, sampling event) and, by means of extensions, it is possible to associate additional data with records of the core type. As part of our project, the extensions of PliC, accessible in the IPT Sandbox, were validated to show how they can be used by institutions to publish species pages through the IPT. In order to fully integrate Plinian Core based species pages into the ALA architecture, the Biodiversity Informatics Research Center (CRBio) has been working on the adoption of the Biodiversity Information Explorer (BIE), an ALA module which manages taxonomic and species contents by integrating global resources like EOL or Wikipedia. This adoption includes the required modifications of the Portal codebase to use the data model of the Plinian Core (github.com/tdwg/PlinianCore) to display a much richer content. This year, with assistance from the Capacity Enhancement Support Programme (CESP) from GBIF, we have revisited the integration of species pages information and expanded the content and type of information provided by the Atlas of Living Costa Rica. We present here the updated ALA Species Pages (ALA-BIE) and Index (BIE-Index) Modules and general documentation that could be applied to any ALA installation in order to prepare species records using the IPT and PliC standard, index and display them through the ALA-BIE Module.
- Open Access
- 23 August 2022
- Published by Pensoft Publishers in Research Ideas and Outcomes
- Vol. 8, e86012
Essential Biodiversity Variables: extracting plant phenological data from specimen labels using machine learning
Essential Biodiversity Variables (EBVs) make it possible to evaluate and monitor the state of biodiversity over time at different spatial scales. Its development is led by the Group on Earth Observations Biodiversity Observation Network (GEO BON) to harmonize, consolidate and standardize biodiversity data from varied biodiversity sources. This document presents a mechanism to obtain baseline data to feed the Species Traits Variable Phenology or other biodiversity indicators by extracting species characters and structure names from morphological descriptions of specimens and classifying such descriptions using machine learning (ML). A workflow that performs Named Entity Recognition (NER) and Classification of morphological descriptions using ML algorithms was evaluated with excellent results. It was implemented using Python, Pytorch, Scikit-Learn, Pomegranate, Python-crfsuite, and other libraries applied to 106,804 herbarium records from the National Biodiversity Institute of Costa Rica (INBio). The text classification results were almost excellent (F1 score between 96% and 99%) using three traditional ML methods: Multinomial Naive Bayes (NB), Linear Support Vector Classification (SVC), and Logistic Regression (LR). Furthermore, results extracting names of species morphological structures (e.g., leaves, trichomes, flowers, petals, sepals) and character names (e.g., length, width, pigmentation patterns, and smell) using NER algorithms were competitive (F1 score between 95% and 98%) using Hidden Markov Models (HMM), Conditional Random Fields (CRFs), and Bidirectional Long Short Term Memory Networks with CRF (BI-LSTM-CRF).
- 23 August 2022
- Published by Pensoft Publishers in Research Ideas and Outcomes
- Vol. 8, e86012
Open AccessEssential Biodiversity Variables: extracting plant phenological data from specimen labels using machine learning
Essential Biodiversity Variables (EBVs) make it possible to evaluate and monitor the state of biodiversity over time at different spatial scales. Its development is led by the Group on Earth Observations Biodiversity Observation Network (GEO BON) to harmonize, consolidate and standardize biodiversity data from varied biodiversity sources. This document presents a mechanism to obtain baseline data to feed the Species Traits Variable Phenology or other biodiversity indicators by extracting species characters and structure names from morphological descriptions of specimens and classifying such descriptions using machine learning (ML). A workflow that performs Named Entity Recognition (NER) and Classification of morphological descriptions using ML algorithms was evaluated with excellent results. It was implemented using Python, Pytorch, Scikit-Learn, Pomegranate, Python-crfsuite, and other libraries applied to 106,804 herbarium records from the National Biodiversity Institute of Costa Rica (INBio). The text classification results were almost excellent (F1 score between 96% and 99%) using three traditional ML methods: Multinomial Naive Bayes (NB), Linear Support Vector Classification (SVC), and Logistic Regression (LR). Furthermore, results extracting names of species morphological structures (e.g., leaves, trichomes, flowers, petals, sepals) and character names (e.g., length, width, pigmentation patterns, and smell) using NER algorithms were competitive (F1 score between 95% and 98%) using Hidden Markov Models (HMM), Conditional Random Fields (CRFs), and Bidirectional Long Short Term Memory Networks with CRF (BI-LSTM-CRF).
- Open Access
- 9 May 2022
- Published by Pensoft Publishers in ARPHA Preprints
- Vol. 3, e86014
Essential Biodiversity Variables: Extracting Plant Phenological Data from Specimen Labels Using Machine Learning
Essential Biodiversity Variables (EBVs) make it possible to evaluate and monitor the state of biodiversity over time at different spatial scales. Its development is led by the Group on Earth Observations Biodiversity Observation Network (GEO BON) to harmonize, consolidate, and standardize biodiversity data from varied biodiversity sources. This document presents a mechanism to get baseline data to feed the Species Traits Variable Phenology or other biodiversity indicators by extracting species characters and structure names from morphological descriptions of specimens and classifying such descriptions using machine learning (ML). A workflow that performs Named Entity Recognition (NER) and Classification of morphological descriptions using ML algorithms was evaluated with excellent results. It was implemented using Python, Pytorch, Scikit-Learn, Pomegranate, Python-crfsuite, and other libraries applied to 106,804 herbarium records from the National Biodiversity Institute of Costa Rica (INBio). The text classification results were almost excellent (F1 score between 96% and 99%) using three traditional ML methods: Multinomial Naive Bayes (NB), Linear Support Vector Classification (SVC), and Logistic Regression (LR). Furthermore, results extracting names of species morphological structures (e.g., leaves, trichomes, flowers, petals, sepals) and character names (e.g., length, width, pigmentation patterns, and smell) using NER algorithms were competitive (F1 score between 95% and 98%) using Hidden Markov Models (HMM), Conditional Random Fields (CRFs), and Bidirectional Long Short Term Memory Networks with CRF (BI-LSTM-CRF).
- 9 May 2022
- Published by Pensoft Publishers in ARPHA Preprints
- Vol. 3, e86014
Open AccessEssential Biodiversity Variables: Extracting Plant Phenological Data from Specimen Labels Using Machine Learning
Essential Biodiversity Variables (EBVs) make it possible to evaluate and monitor the state of biodiversity over time at different spatial scales. Its development is led by the Group on Earth Observations Biodiversity Observation Network (GEO BON) to harmonize, consolidate, and standardize biodiversity data from varied biodiversity sources. This document presents a mechanism to get baseline data to feed the Species Traits Variable Phenology or other biodiversity indicators by extracting species characters and structure names from morphological descriptions of specimens and classifying such descriptions using machine learning (ML). A workflow that performs Named Entity Recognition (NER) and Classification of morphological descriptions using ML algorithms was evaluated with excellent results. It was implemented using Python, Pytorch, Scikit-Learn, Pomegranate, Python-crfsuite, and other libraries applied to 106,804 herbarium records from the National Biodiversity Institute of Costa Rica (INBio). The text classification results were almost excellent (F1 score between 96% and 99%) using three traditional ML methods: Multinomial Naive Bayes (NB), Linear Support Vector Classification (SVC), and Logistic Regression (LR). Furthermore, results extracting names of species morphological structures (e.g., leaves, trichomes, flowers, petals, sepals) and character names (e.g., length, width, pigmentation patterns, and smell) using NER algorithms were competitive (F1 score between 95% and 98%) using Hidden Markov Models (HMM), Conditional Random Fields (CRFs), and Bidirectional Long Short Term Memory Networks with CRF (BI-LSTM-CRF).
- Open Access
- 26 June 2018
- Published by Pensoft Publishers in Biodiversity Data Journal
- Vol. 6 (6) , e21282
Semi-automatic Extraction of Plants Morphological Characters from Taxonomic Descriptions Written in Spanish
Taxonomic literature keeps records of the planet's biodiversity and gives access to the knowledge needed for its sustainable management. Unfortunately, most of the taxonomic information is available in scientific publications in text format. The amount of publications generated is very large; therefore, to process it in order to obtain high structured texts would be complex and very expensive. Approaches like citizen science may help the process by selecting whole fragments of texts dealing with morphological descriptions; but a deeper analysis, compatible with accepted ontologies, will require specialised tools. The Biodiversity Heritage Library (BHL) estimates that there are more than 120 million pages published in over 5.4 million books since 1469, plus about 800,000 monographs and 40,000 journal titles (12,500 of these are current titles). It is necessary to develop standards and software tools to extract, integrate and publish this information into existing free and open access repositories of biodiversity knowledge to support science, education and biodiversity conservation. This document presents an algorithm based on computational linguistics techniques to extract structured information from morphological descriptions of plants written in Spanish. The developed algorithm is based on the work of Dr. Hong Cui from the University of Arizona; it uses semantic analysis, ontologies and a repository of knowledge acquired from the same descriptions. The algorithm was applied to the books Trees of Costa Rica Volume III (TCRv3), Trees of Costa Rica Volume IV (TCRv4) and to a subset of descriptions of the Manual of Plants of Costa Rica (MPCR) with very competitive results (more than 92.5% of average performance). The system receives the morphological descriptions in tabular format and generates XML documents. The XML schema allows documenting structures, characters and relations between characters and structures. Each extracted object is associated with attributes like name, value, modifiers, restrictions, ontology term id, amongst other attributes. The implemented tool is free software. It was developed using Java and integrates existing technology as FreeLing, the Plant Ontology (PO), the Plant Glossary, the Ontology Term Organizer (OTO) and the Flora Mesoamericana English-Spanish Glossary.
- 26 June 2018
- Published by Pensoft Publishers in Biodiversity Data Journal
- Vol. 6 (6) , e21282
Open AccessSemi-automatic Extraction of Plants Morphological Characters from Taxonomic Descriptions Written in Spanish
Taxonomic literature keeps records of the planet's biodiversity and gives access to the knowledge needed for its sustainable management. Unfortunately, most of the taxonomic information is available in scientific publications in text format. The amount of publications generated is very large; therefore, to process it in order to obtain high structured texts would be complex and very expensive. Approaches like citizen science may help the process by selecting whole fragments of texts dealing with morphological descriptions; but a deeper analysis, compatible with accepted ontologies, will require specialised tools. The Biodiversity Heritage Library (BHL) estimates that there are more than 120 million pages published in over 5.4 million books since 1469, plus about 800,000 monographs and 40,000 journal titles (12,500 of these are current titles). It is necessary to develop standards and software tools to extract, integrate and publish this information into existing free and open access repositories of biodiversity knowledge to support science, education and biodiversity conservation. This document presents an algorithm based on computational linguistics techniques to extract structured information from morphological descriptions of plants written in Spanish. The developed algorithm is based on the work of Dr. Hong Cui from the University of Arizona; it uses semantic analysis, ontologies and a repository of knowledge acquired from the same descriptions. The algorithm was applied to the books Trees of Costa Rica Volume III (TCRv3), Trees of Costa Rica Volume IV (TCRv4) and to a subset of descriptions of the Manual of Plants of Costa Rica (MPCR) with very competitive results (more than 92.5% of average performance). The system receives the morphological descriptions in tabular format and generates XML documents. The XML schema allows documenting structures, characters and relations between characters and structures. Each extracted object is associated with attributes like name, value, modifiers, restrictions, ontology term id, amongst other attributes. The implemented tool is free software. It was developed using Java and integrates existing technology as FreeLing, the Plant Ontology (PO), the Plant Glossary, the Ontology Term Organizer (OTO) and the Flora Mesoamericana English-Spanish Glossary.