Dataset Card for FoodClimNER

Expert-annotated NER corpus containing food practices entities and climate change entities.

Dataset Details

Dataset Description

We present a new corpus for zero-shot NER, containing food practices related to climate change. The goal is to be able to detect and categorize food practices that could pose a risk in the future due to climate change (example: oyster can develop bacteries after floods, so climate change => more floods => oyster consumption becomes more dangerous).

Uses

This corpus has been produced for research purposes mainly.

Direct Use

Evaluation of models in zero-shot setting to test their ability to find food practices entities linked to climate change.

Dataset Structure

Contains brat annotations (at the document level) in the brat folder (.ann and .txt files, as well as config files). Contains the preprocessed data (sentence splitted) for running experiments with our code in the test.jsonl file, the entities.json and rules.json are also needed to run our prompting method. As the annotation contains discontinuous entities, the format of test.jsonl objects is as follows :

{
  "id":"<id_of_article_and_sentence_nb>",
  "text":"<annotated_text>",
  "entities": [
    {
      "type":"<entity_type>",
      "text":"<entity_text>",
      "offsets":[
        [<first_span_start_int>,<first_span_end_int>],
        [<second_span_start>,<second_span_end>],
        ...
      ]
    }
  ]
}

Entity types

• climate_greenhouse_gases : Gases that cause heating of the atmosphere. Examples : carbon dioxide, CH4, GHG emission
• climate_hazards : Hazards with potential negative impact on climate. Where a hazard is named in more detail in a text, the entire term is annotated. Examples : floods, wildfires, droughts, heatwaves, surface water flood, soil liquefaction
• climate_impacts : Effects of hazards, primarily negative effects on humans. Climate impacts on livestock are also considered impacts, as they indirectly affect humans. COVID-19 is annotated as impact only when the text explicitly establishes a link between climate change and the pandemic. Examples : Vector-borne diseases, diseases explicitly described as climate-sensitive
• climate_mitigations : Activities to reduce climate change or to better deal with the consequences. If an increase or decrease in activity reduces climate change, it should be annotated. Transitions, strategies, policies, or transformations explicitly aimed at the sustainability of a system should be annotated even when climate change is not explicitly mentioned. Examples : food waste reduction, transitions aimed at sustainability, strategies aimed at sustainability, policies aimed at sustainability
• climate_problem_origins : Problems that describe why the climate is changing. Sectors cited as causes of energy use are also annotated as problem origins. Examples : fossil fuel, deforestation, transport sector
• environmental_concern : An intermediate/residual category within the environmental sphere for which the link to climate is not established or direct. It does not specifically refer to climate change. Includes generic or non-actionable references to sustainability, environmental awareness, or trends in consumption that do not describe explicit consumer actions. Examples : biodiversity, ecological footprint, pollution, water
• food_consumption_uses : Covers consumption practices related to eating and meals. Examples : portion sizes, choice of utensils
• food_culinary_preparation : All attitudes and practices related to food preparation and cooking. It also includes skills and actions that we perform when cooking. Examples : raw food diets, water conservation during cooking, reducing energy consumption in cooking, minimizing waste during preparation
• food_repertoire : Covers practices and factors related to food choices (in terms of quality or quantity) that may or may not be part of an individual's diet. Mentions of behaviour related to food or diet should be annotated only if the text links these behaviours explicitly to climate change. Examples : meat consumption, food choices
• food_logistics_travel : Includes pre-purchase practices and travel to the place of purchase. Examples : checking existing resources before shopping, meal planning, using low-carbon transportation to go grocery shopping
• food_storage : Covers food storage practices after purchase and before preparation. Examples : choice of packaging, adherence to expiration dates, disregard of expiration dates, preservation techniques
• food_supply_purchases : Covers purchasing practices (product selection, choice of retail locations, interactions with marketing) as well as non-market procurement strategies. Examples : product selection, choice of retail locations, interactions with marketing, self-production
• food_waste_management : Covers practices that take place after a meal, with consideration for the end of a food item's life cycle. Examples : storing leftovers, composting, sorting food waste, expiration date reasoning

Dataset Creation

Curation Rationale

We created this dataset to benchmark zero-shot NER methods for the extraction of food practices entities in the context of climate change, in scientific articles.

Source Data

The documents are open access PubMed articles on food science and climate change.

Data Collection and Processing

The base documents were retrieved using a PubMed query designed by two experts in food science / microbiology & sociology. Some parts which were judged as noise by the experts were manually removed to have the final documents on which the annotations are made.

Annotation process

• Annotation guideline creation : two experts produced the annotation schema
• Selection of articles: they designed a query to find relevant articles and then manually selected 10 of them
• Annotation : a single expert annotator annotated the documents with the defined schema using the brat tool

Who are the annotators?

We have a single expert annotator which holds a PhD in microbiology and have previous experience in text data annotation on scientific articles.