Lite-BERT-SL is a
BERT-based sequence labeling model fine-tuned on the
HiFi-KPI Lite dataset. This model was introduced in the paper
HiFi-KPI: A Dataset for Hierarchical KPI Extraction from Earnings Filings.
The model is designed for the hierarchical extraction of Key Performance Indicators (KPIs) from financial earnings filings (SEC 10-K and 10-Q reports). While the full HiFi-KPI dataset contains a massive taxonomy of iXBRL tags, Lite-BERT-SL is fine-tuned on a manually curated subset focusing on four expert-mapped KPI clusters:
-
Revenues
-
Earnings
-
EPS (Earnings Per Share)
-
EBIT (Earnings Before Interest and Taxes)
-
Developed by: Rasmus Aavang, Giovanni Rizzi, Rasmus Bøggild, Alexandre Iolov, Mike Zhang, Johannes Bjerva
-
Model type: Token Classification (Sequence Labeling)
-
Base Model: bert-base-uncased
-
Language: English
According to the paper, encoder-based models achieve over 0.906 macro-F1 on the HiFi-KPI Lite classification task. For detailed performance metrics, please refer to the
paper and the
HiFi-KPI Lite dataset page.
1@inproceedings{aavang-etal-2026-hifi,
2 title = "{H}i{F}i-{KPI}: A Dataset for Hierarchical {KPI} Extraction from Earnings Filings",
3 author = "Aavang, Rasmus T. and
4 Rizzi, Giovanni and
5 Tjalk-B{\o}ggild, Rasmus and
6 Iolov, Alexandre and
7 Zhang, Mike and
8 Bjerva, Johannes",
9 editor = "Piperidis, Stelios and
10 Bel, N{\'u}ria and
11 van den Heuvel, Henk and
12 Ide, Nancy and
13 Krek, Simon and
14 Toral, Antonio",
15 booktitle = "Proceedings of the Fifteenth Language Resources and Evaluation Conference",
16 month = may,
17 year = "2026",
18 address = "Palma de Mallorca, Spain",
19 publisher = "ELRA Language Resource Association",
20 url = "https://aclanthology.org/2026.lrec-1.30/",
21 doi = "10.63317/2nbsp7zzfb3g",
22 pages = "441--455",
23 abstract = "Accurate tagging of earnings reports can yield significant short-term returns for stakeholders. The machine-readable inline eXtensible Business Reporting Language (iXBRL) is mandated for public financial filings. Yet, its complex, fine-grained taxonomy limits the cross-company transferability of tagged Key Performance Indicators (KPIs). To address this, we introduce the Hierarchical Financial Key Performance Indicator (HiFi-KPI) dataset, a large-scale corpus of 1.65M paragraphs and 198k unique, hierarchically organized labels linked to iXBRL taxonomies. HiFi-KPI supports multiple tasks and we evaluate three: KPI classification, KPI extraction, and structured KPI extraction. For rapid evaluation, we also release HiFi-KPI-Lite, a manually curated 2.5K-instance subset. Baselines on HiFi-KPI-Lite show that encoder-based models achieve over 0.906 macro-F1 on classification, while Large Language Models (LLMs) reach 0.440 F1 on structured extraction. Finally, a qualitative analysis reveals that extraction errors primarily relate to dates. We open-source all code and data at Anonymous."
24}
