jechen2140/cos-iw-financial-narratives

SENTiVENT × X Financial Narratives

A dataset linking financial event annotations from the SENTiVENT benchmark corpus with X (Twitter) posts scraped around the same events, used to study how financial narratives propagate from institutional news to social media.

Dataset Contents

File	Description	Rows
data/dataset_event_type.tsv	SENTiVENT sentence-level annotations with 18 event type multi-labels	6,883
data/dataset_event_subtype.tsv	SENTiVENT sentence-level annotations with 60 event subtype multi-labels	6,883
data/x_sentivent_master_lowcredit_50each.csv	X posts scraped for each SENTiVENT event (up to 50 posts/event, filtered for low-credit accounts)	20,968
data/x_sentivent_cleaned.csv	Cleaned subset of X posts used in downstream analysis	1,917
data/x_sentivent_event_counts.csv	Post counts per event after filtering	93
data/event_registry_sentivent.csv	Registry mapping each SENTiVENT event to its X scraping query, WSJ/official doc references, and metadata	93
data/type_classes_multilabelbinarizer.json	Ordered label list for the 18 event type classes	—
data/subtype_classes_multilabelbinarizer.json	Ordered label list for the 60 event subtype classes	—

SENTiVENT Annotations (dataset_event_type.tsv / dataset_event_subtype.tsv)

Column	Description
document_id	Article identifier (e.g. aal00)
document_title	Article filename
sentence_idx	Sentence position within the article
text	Sentence text
types_event	List of event types present in the sentence
subtypes_event	List of event subtypes (may be multi-label)
types_event_unq	Unique event types
subtypes_event_unq	Unique event subtypes
token_cnt	Token count
event_cnt	Number of event mentions
dataset	Split tag (silver / gold)
labels	Binary multi-label vector (matches *_classes_multilabelbinarizer.json)

Event type classes (18): CSR/Brand, Deal, Dividend, Employment, Expense, Facility, FinancialReport, Financing, Investment, Legal, Macroeconomics, Merger/Acquisition, Product/Service, Profit/Loss, Rating, Revenue, SalesVolume, SecurityValue

X Posts (x_sentivent_master_lowcredit_50each.csv)

X (Twitter) posts scraped using queries derived from each SENTiVENT event. Columns:

Column	Description
event	SENTiVENT event identifier
event_label	Human-readable event label
event_date	Date of the underlying financial event
thread_source	Source platform (x)
thread_url	Direct URL to the post
id	Post ID
parent_id	Parent post ID (for replies)
author	X username
body	Post text
score	Relevance score assigned during scraping
date	Post timestamp (ISO 8601)
query_used	X API search query that retrieved this post
lang	Language code
record_type	post or reply

Event Registry (event_registry_sentivent.csv)

Maps each event to its scraping metadata:

Column	Description
event	Event identifier
event_date	Date of event
event_label	Human-readable label
event_type	SENTiVENT event type category
x_query	X API query used
x_language	Language filter
wsj_file	Linked WSJ article (if any)
official_file	Linked official document (if any)
analysis_note	Provenance note

Usage

from datasets import load_dataset

# SENTiVENT event type annotations
sentivent = load_dataset("jechen2140/cos-iw-financial-narratives", data_files="data/dataset_event_type.tsv", delimiter="\t")

# X posts linked to SENTiVENT events
x_posts = load_dataset("jechen2140/cos-iw-financial-narratives", data_files="data/x_sentivent_master_lowcredit_50each.csv")

Citation

If you use this dataset, please cite the original SENTiVENT corpus:

@inproceedings{jacobs2018economeria,
  title={Economic Event Detection in Company-Specific News Text},
  author={Jacobs, Gilles and Lefever, Els and Hoste, Véronique},
  booktitle={Proceedings of the First Workshop on Economics and Natural Language Processing},
  year={2018}
}

License

Dataset card and X post collection: CC BY 4.0.
Original SENTiVENT annotations are subject to their own license; see the SENTiVENT project.