A Multi-Source Entity-Level Sentiment Corpus for the Financial Domain: The FinLin Corpus

FinLin is multi-source multi-entity corpus for fine-grained financial sentiment analysis. It covers a period of 3 months and a set of automobile companies as sentiment targets. In addition to fine-grained sentiment, it comes with a relevance measure aimed at supporting the capture of a market sentiment by giving each item an individual weight. FinLin covers microblogs from StockTwit, news articles, company reports, and analyst reports. If you want, then you can add figure 4 from the paper here but I dont think its needed.

A full description of the corpus can be found in the original paper.

The data distribution is as follows:

FinLin Structure

The FinLin files come in the following structure:

• "type" : The data type [SW,NA,AR,CR].
• "text" : The body of text.
• "title" : The title/headline.
• "created_at" : The timestamp.
• "entity" : The sentiment target (abbreviations are defined in the original paper).
• "id" : The unique identifier.
• "sentiment" : The fine-grained sentiment score in the range [-1,1].
• "relevance" : The fine-grained relevance score in the range [0,1].
• "annotations" : The list of each individual annotation per annotator.
• "spans" : The list of "text_spans" and "title_spans".
• "text_spans" : The substring of "text" as highlighted by the corresponding annotator.
• "title_spans" : The substring of "title" as highlighted by the corresponding annotator.

Reference

If you use FinLin in your research, please cite this paper:

@article{daudert2021multi,
  title={A multi-source entity-level sentiment corpus for the financial domain: the FinLin corpus},
  author={Daudert, Tobias},
  journal={Language Resources and Evaluation},
  pages={1--24},
  year={2021},
  publisher={Springer},
  doi = {10.1007/s10579-021-09555-3},
  url = {https://doi.org/10.1007/s10579-021-09555-3}
}

License

This corpus is openly available for non-commercial use under the Attribution-NonCommercial-ShareAlike 4.0 International.

Contact & Obtaining this Corpus

If you have any questions, suggestions, or want to obtain a copy of this corpus, you can contact Tobias Daudert ([email protected]).

Disclaimer

FinLin is created for research purposes only. The author is not responsible for any damages related to or inflicted by its use.