Iberian Health and Food Language Technologies

Held in conjunction with the XXXVII International Conference of the Spanish Society for Natural Language Processing (SEPLN2021)

News:

[Sep 13]: schedule updated
[Sep 9]: the proceedings are now available
[Aug 31]: preliminary version of the schedule and talks was published
[Mai 7]: website of the IberHealth2021 workshop launched

Important Dates

Workshop will be held on Tuesday, September 21, 10.00-14:30 (CEST timezone)

IBERIAN HEALTH AND FOOD LANGUAGE TECHNOLOGIES - Schedule.pdf

IBERIAN HEALTH AND FOOD LANGUAGE TECHNOLOGIES - Proceedings.pdf

Overview

One of the most active application domains of language technologies and NLP is health and biomedicine. This scenario has been accentuated even more due to current COVID-19 pandemic, where more efficient access, processing and exploitation of multiple types of unstructured data sources, including the biomedical literature, electronic healthcare records, clinical trials, patents, news articles or even social media is of critical importance. Not to mention other data sources like clinical practice guidelines, clinical trials documentation, medical questionnaires, medical informed consent documents, etc.). Nevertheless, despite the considerable progress made in the last two years, most of the current resources and research in clinical and biomedical NLP and text mining is still done mostly on data in English. Adapting and applying such tools to healthcare data in Spanish, as well as other languages like Portuguese or Catalan is not straightforward even under the increasing promise of multilingual NLP applications. Among the main driving forces of clinical NLP research and development one can highlight scientific venues in the form of workshops devoted specifically to this topic, such as BioCreative, BioNLP, AMIA NLP workshops, i2b2, (and its successor n2c2). Other scientific venues like LOUHI, HealTAC or eHealth CLEF, although theoretically being multi-lingual, did only marginally cover content in Spanish. Considering the number of native Spanish speakers and the large community of healthcare-related professionals it is key to provide a more stable and specific scientific environment to present and exchange NLP research results, resources and tools not only to NLP professionals but also to the industrial sector, pharma industry, healthcare professionals including medical informatics and healthcare technology experts. The objective of this workshop is to foster research both at the theoretical as well as at the level of practical real-world applications of NLP technologies applied to a diversity of textual data types. This workshop will also specially cover contributions from industry and companies working on healthcare textual data and results obtained by pharmaceutical industry.Finally, in addition to the healthcare NLP session we will also include a second session covering language technologies applied to the domain of food, nutrition and agriculture. Despite the economic importance of this sector for Spain and Latin-American countries, the potential of NLP applications was only marginally explored.

Schedule (CEST timezone)

9/21/2021 10:00
Welcoming to the Iberian Health and Food Language Technologies and BSC/Plan TL resources
Martin Krallinger - Barcelona Supercomputing Center (BSC)

9/21/2021 10:30
Mining the Sociome for Insights on Chronic Conditions and Healthy Lifestyles
Anália Lourenço - Universidade de Vigo, Spain

9/21/2021 11:15
Food Nutrition and Security Cloud: A case usage of NLP technologies to extract food – drug interactions from scientific and clinical texts.
Enrique Carrillo de Santa Pau - IMDEA Food Institute, Madrid. Spain

9/21/2021 11:45
Use of NLP and Text Mining for health, nutrition, and food: Plan TL/BSC resources, components, corpora and use cases
Antonio Miranda/Eulalia Farre - Barcelona Supercomputing Center (BSC)

9/21/2021 12:15 coffee break

9/21/2021 12:30
Food Information Extraction and Normalization: the Past, the Present, and the Future
Tome Eftimov, Gjorgjina Cenikj - Jožef Stefan Institute, Ljubljana, Slovenia

9/21/2021 13:30
Text mining as a way to select microbial strains to ferment new food products
Claire Nédellec - University Paris-Saclay, INRAE

9/21/2021 14:00
EFSA's initiative for exchanging content and news on Data Science for Food Safety Risk Assessment
Carsten Behring - European Food Safety Authority

9/21/2021 14:15
Wrap-up final session
Martin Krallinger - Barcelona Supercomputing Center (BSC)

Invited Talks (Preliminary)

Welcoming to the Iberian Health and Food Language Technologies and BSC/Plan TL resourcesAbstract: During the welcome session, the importance, current research and existing resources for NLP and text mining applications applied to health, nutrition, and food sciences will be summarized together with a short outline of the workshop purpose, content and talks. Some of the ongoing and past efforts in promoting the development of health NLP tools, in particular for content in Spanish by the Plan TL will be presented.
Title: Food Information Extraction and Normalization: the Past, the Present, and the FutureName: Tome Eftimov, Gjorgjina CenikjAffiliation: Computer Systems Department, Jožef Stefan Institute, Ljubljana, SloveniaAbstract: In the last decades, a great amount of work has been done in predictive modelling of issues related to human and environment health. This is made possible by the existence of several available biomedical vocabularies and standards, which play a crucial role for understanding health information, together with a large amount of health data. In 2019, Lancet Planetary Health noted that the focus of future improvements in our wellbeing and societies will depend on investigating the links between food systems, human health, and the environment. However, despite the large number of available resources and work done in the health and environmental domains, there is a lack of resources that can be utilized in the food and nutrition domain, as well as their interconnections. This talk will provide a summary of the past, the present and the future related to food information extraction and normalization. It will cover different NLP approaches already developed for tracing the food information in textual data, resources that allow to make food and nutrition data interoperable, and NLP/ML pipelines for exploring relations between food and biomedical entities. In particular, this is important during the current pandemics of COVID-19, when food provision and security, as well as healthy nutrition and environment, are tremendously needed for quick recovery and long-term sustainable development of our societies.
Title: Food Nutrition and Security Cloud: A case usage of NLP technologies to extract food – drug interactions from scientific and clinical texts.Authors: Marco Garranzo Asensio, Teresa Laguna Lobo, Enrique Carrillo de Santa PauAffiliation: Computational Biology Group. Precision nutrition and cancer program. IMDEA Food Institute, Madrid. SpainPresenter: Enrique Carrillo de Santa PauAbstract: Food Nutrition and Security cloud (FNS-Cloud) is a European project bringing together 34 organizations from 14 European countries, trying to solve fragmentation, lack critical mass and unequal access to food and nutrition data. Thus, the objective of FNS-Cloud is to develop an infrastructure of services in the cloud, to manage data on food and nutrition, and safety in a more efficient way, which allows health professionals or researchers to consult them quickly.FNS-Cloud will use and develop text mining applications to organize and distribute food knowledge from structured and unstructured data, curate and annotate with descriptors, matched with other data, classified, and described in different ways. To annotate FNS data with descriptors, pre-processing will be based on state-of-the-art Natural Language Processing and Machine Learning methods. Data structuring will also be applied to extract semantics used for curation and annotation in a form of descriptors, enabling data linkage and harmonisation. Herein, we will present our work in the FNS-Cloud project developing an NLP pipeline to extract food -drug interactions (FDIs) mentioned in scientific article abstracts, with the intent of creating resources for clinicians and nutritionists to use when searching for FDIs. Food- drug interaction extraction is formulated as a relationship extraction problem. First entities related to food, drugs and food chemicals are recognized in a sentence and for every pair of food-drug entities it is determined if a relationship of “interaction” exists between them. The pipelines main hurdle is the poor amount of labelled text data for the food domain, which it overcomes with the combination of labelled data from different sources, ontologies, and food composition databases. The pipeline under development uses a mixture of deep learning, dictionary-based and ontology-based approaches for the recognition and linking of food, chemical and drug entities in the text, and in linking chemicals to foods. Recognition of abbreviations and co-reference resolution approaches are also used to link all different expressions of the same entity together. A transfer learning approach is used to recognize FDI, by training a deep learning relationship extractor to recognize anonymized drug-drug interactions on a corpus of drug-drug interactions and using it on anonymized food - drug interactions.Health and nutrition are intimately related. The consumption of specialized food with the goal of improving health, preventing disease or as alternative complementary treatment has risen in the last years increasing the risk of potential FDIs. Therefore, it becomes essential the development of resources with high standards and quality data in a centralized way to provide health care workers with the best information and tools for giving advice about FDIs.Food Nutrition Security Cloud (FNS-Cloud) has received funding from the European Union’s Horizon 2020 Research and Innovation programme (H2020-EU.3.2.2.3. – A sustainable and competitive agri-food industry) under Grant Agreement No. 863059
Title: Text mining as a way to select microbial strains to ferment new food productsPresenter: Claire NédellecAffiliation: University Paris-Saclay, INRAEAbstract: The on-line Omnicrobe kowledge base gathers information on food microbiology from multiple sources expressed in natural language, e.g. papers, genetic databases, biological resource center catalogues. The textual data is processed by information extraction methods that automatically identify entities, relationships and categories from relevant taxonomies and ontologies. Food innovation is an example of application. An example on plant juice fermentation illustrates how data linking of world wide scattered information by NLP and semantic web methods is a powerful solution to speed-up food and beverage innovation. Omnicrobe is used as a tool to select a limited set of microbe strains that present all expected properties and availability in Biological Resource Center catalogues.These strains are then tested for their functionalities in different media.
Title: Use of NLP and Text Mining for health, nutrition, and food: Plan TL/BSC resources, components, corpora and use cases.Names: Eulalia Farre, Anotnio MirandaAffiliation: Barcelona Supercomputing Center (BSC)Abstract: There is a pressing need to generate more efficient access to food and nutrition-related information applied to health-related content through text mining and natural language processing technologies, not only for data in English but also for other languages like Spanish. For instance, the recent COVID-19 pandemic has also caused noticeable changes in food consumption patterns with potential effects on population health and wellbeing. Most of the previous food-related NLP applications were applied to gastronomy, processing menus, ingredients, and recipes, with far less research on actual clinical and medical application scenarios and content types. The BSC Text Mining Unit, in the context of the Spanish National Plan for Advancement of Language Technologies (Plan TL), has characterized a set of key concepts and entity types of relevance for health and clinical food language technology applications such as food safety (food poisoning, contamination, allergies, food-intolerance, food-drug interactions, food-borne diseases, and patient dietary records). This talk will summarize resources, annotated data types, corpora, and components generated for medical data in Spanish, with potential adaptation to other languages and other application fields (veterinary medicine, agriculture, and environmental health). Particular emphasis will be placed on a novel annotated dataset for the extraction, recognition, and normalization of species mentions, one of the critical concept types for clinical food sciences, and its use for a community evaluation shared task. We will also present the Spanish Food and Health corpus, a dataset annotated with species, diseases, procedures, among other entity types to foster the development of novel language technology applications.
Title: Mining the Sociome for Insights on Chronic Conditions and Healthy LifestylesName: Anália LourençoAffiliation: Universidade de Vigo, SpainAbstract: The environmental and lifestyle impacts of diet-related and chronic diseases is a main, timely health topic. Notably, research on diabetes, gluten intolerance, Crohn's syndrome or colorectal cancer attracts considerable attention. At the same time, social media have become an important vehicle to disseminate health information as well as to gather unbiased reports on personal and professional health experiences. Understanding how patients and relatives, health workers and the general public perceive the problematic of these conditions and interact through social media can be key to raise awareness and promote healthy lifestyles.Our work is focused on the application of machine learning and graph analysis methodologies to the mining of the sociome. Apart from ontological resources, we are exploring the use of deep learning techniques for entity recognition, namely the extraction of disease symptoms, foods and diets. In addition, we are interested in conversational contexts rather than post-level analysis. Understanding social interaction through conversations or as a post-comment relationship enables a more comprehensive depiction of user behaviour, namely in terms of user influence and support.This talk summarises the lessons we have learned in different works and introduces our new research project on the application of machine learning-based models to detect and curb Health misinformation on social media.Following recent efforts of the Plan TL on processing medical and social media texts in Spanish, we will also take this opportunity to prompt discussion on current limitations and challenges on mining the sociome for Spanish and Galician Health-related contents.
Title: EFSA's initiative for exchanging content and news on Data Science for Food Safety Risk AssessmentCarsten BehringAffiliation: European Food Safety AuthorityAbstract: Combined food safety and data science knowledge is rare across industry , academics and governmental organizations. We need to work closely together to be successful in our aim of making food safety risk assessment more efficient.
Final Wrap-up sessionThe workshop organizers, speakers, and attendants will have the opportunity to discuss needs, resources, and efforts to promote the development of health and food language technology tools for a variety of languages, content types, and application scenarios.

Topics

Natural language processing of healthcare text including EHRs, literature, patents, clinical trials, clinical practice guidelines, social media, health forums, or informed consents
Information extraction of clinical variables, named entities, values and clinical modifiers like negations
Speech analytics and voice processing for healthcare applications
Medical controlled vocabularies and ontologies
Machine-learning and transfer learning for healthcare text analytics
Information fusion, integration and semantic interoperability of NLP results with structured data or image data
Explainable models and interpretability of clinical NLP results
Real-world application and use cases of healthcare text analytics
Evaluation and benchmarking of clinical NLP methods
Knowledge discovery and hypothesis generation for health, agriculture, food or nutrition applications
Information fusion, i.e. integrating data from various sources, e.g. structured and narrative documentation
Anonymization of sensitive health data
NLP applied to agriculture, food (food poisoning, allergies, food-medication interactions food contamination, etc.…), nutrition
Multilingual health, agriculture or food NLP approaches (methods, applications, annotated data and terminologies).
Shared tasks and evaluation campaigns
Machine translation of specialized content and document types applied to health, agriculture and food/nutrition domains.

Organizers

Martin Krallinger
Text Mining unit at the Barcelona Supercomputing Center (BSC), Spain
Jocelyn Dunstan
Center for Mathematical Modeling, Faculty of Physical and Mathematical Sciences, University of Chile
Francisco M. Couto
LASIGE, Departamento de Informática, Faculdade de Ciências, Universidade de Lisboa, Portugal

Journal Special Issue

The planned workshop functions as a venue for the different types of contributors, mainly task providers and solution providers, to meet together and exchange their experiences.

We expect that investigation on the topics of the task will continue after the workshop, based on new insights obtained through discussions during the workshop.