README.md

---
license: apache-2.0
task_categories:
- question-answering
- text-classification
language:
- en
- fr
- es
size_categories:
- 10K<n<100K
tags:
- climate
- policy
---

This dataset is curated by [GIZ Data Service Center](https://www.giz.de/expertise/html/63018.html). The source dataset for this
comes from Internal GIZ team (IKI_Tracs) and [Climatewatchdata](https://www.climatewatchdata.org/data-explorer/historical-emissions?historical-emissions-data-sources=climate-watch&historical-emissions-gases=all-ghg&historical-emissions-regions=All%20Selected&historical-emissions-sectors=total-including-lucf%2Ctotal-including-lucf&page=1),
where Climatewatch has analysed Intended nationally determined contribution (INDC), NDC and Revised/Updated NDC of the countries to answer some important questions related to Climate change.

Specifications
- Dataset size: ~85k
- Language: English, French, Spanish

# Columns 
- **index (type:int)**: Unique Response ID
- **ResponseText (type:str)**: Annotated answer/response to query
- **Alpha3 (type:str)**:country alpha-3 code (ISO 3166)
- **Country (type:str)**: country name
- **Document (type:str)**:Name of type of Policy document from which response is provided
- **IkiInfo (type: list[dict])**: Responsetext can appear/occur as answer/response for different kind of query, therefore in that case we preserve all raw information for each occurences.
  Each dictionary object represents one such occurrence for response and provides all raw metadata for an occurrence.In case of None, it means
  the entry belongs to Climate data and not IKI Tracs data)
- **CWInfo (type: list[dict])**:Responsetext can appear/occur as answer/response for different kind of query, therefore in that case we preserve all raw information for each occurences.
  Each dictionary object represents one such occurrence for response and provides all raw metadata for an occurrence. In case of None, it means
  the entry belongs to Iki tracs data and not CW)
- **Source (type:list[str])**: Contains the name of source
- **Target (type:list)**: Value at index 0, represents number of times ResponseText appears as 'Target', and not-Target (value at index 1 )
- **Action (type:list)**: Value at index 0, represents number of times ResponseText appears as 'Action', and not-Action (value at index 1 )
- **Policies_Plans (type:list)**: Value at index 0, represents number of times ResponseText appears as 'Policy/Plan', and not-Policy/Plan (value at index 1 )
- **Mitigation (type:list)**: Value at index 0, represents number of times ResponseText appears in reference to Mititgation and not-Mitigation (value at index 1 )
- **Adaptation (type:list)**: Value at index 0, represents number of times ResponseText appears in reference to Adaptation and not-Adaptation (value at index 1 )
- **language (type:str)**: ISO code of language of ResponseText.
- **context (type:list[str])**: List of paragraphs/textchunk from the document of country which contains the ResponseText. These results are based on Okapi bm25 retriever,
 and hence dont represent ground truth. 
- **context_lang (type:str)**:  ISO code of language of ResponseText. In some cases context and ResponseText are different as annotator have provided the translated response, rather than original text from document.
- **matching_words(type:list[list[[words]])**:For each context, finds the matching words from ResponseText (stopwords not considered).
- **response_words(type:list[words])**:Tokens/Words from ResponseText (stopwords not considered)
- **context_wordcount (type:list[int])**: Number of tokens/words in each context (remember context itself is list of multiple strings, and stopwords not considered)
- **strategy (type:str)**: Can take either of *small,medium,large* value. Represents the length of paragraphs/textchunk considered for finding the right context for ResponseText
- **match_onresponse (type:list[float])**: Percentage of overlapping words between Response and context with respect to the length of ResponseText.
- **candidate (type:list[list[int]])**: Candidate within context which corresponds (fuzzy matching/similarity) to ResponseText. Value at index(0,1) represents (start,end) of string within context
- **fetched_text (type:list[str])**: Candidate within context which corresponds (fuzzy matching/similarity) to ResponseText.
- **response_translated(type:str)**:Translated ResponseText
- **context_translated(type:str)**: Translated Context 
- **candidate_translated(type:str)**: Translated Candidate index values (check column 'candidate')
- **fetched_text_translated(type:str)**: Translated Candidates (check column 'candidate')
- **QA_data(type:dict)**: Metadata about ResponseText, highlighting nature of query to which ResponseText corresponds as 'answer/response'
- **match_onanswer (type:list[float])**: Represents percentage match between Response and candidate text ( from statistics it is recommended to keep only values above 0.3% as
answer and consider the context for 'No answer' for SQUAD2 data format)