---
base_model: bert-base-uncased
datasets:
- AAU-NLP/HiFi-KPI
language:
- en
library_name: transformers
license: apache-2.0
model_name: BERT-SL1000
pipeline_tag: text-classification
tags:
- financial NLP
- named entity recognition
- sequence labeling
- structured extraction
- hierarchical taxonomy
- XBRL
- iXBRL
- SEC filings
- financial-information-extraction
task_categories:
- text-classification
- token-classification
task_ids:
- named-entity-recognition
- financial-information-extraction
pretty_name: 'BERT-SL1000: Sequence Labeling for Financial KPI Extraction'
size_categories: 1M<n<10M
languages:
- en
dataset_name: HiFi-KPI
model_description: "BERT-SL1000 is a **BERT-based sequence labeling model** fine-tuned\
  \ on the **HiFi-KPI dataset** for extracting \n**financial key performance indicators\
  \ (KPIs)** from **SEC earnings filings (10-K & 10-Q)**. It specializes in identifying\
  \ \nentities, such as revenue, earnings, and financial ratios, using **token classification**.\n\
  \nThis model is part of the **HiFi-KPI benchmark** and is optimized for **hierarchical\
  \ label consistency**.\n"
dataset_link: https://huggingface.co/datasets/AAU-NLP/HiFi-KPI
repo_link: https://github.com/aaunlp/HiFi-KPI
---

## **BERT-SL1000**

### **Model Description**  
BERT-SL1000 is a **BERT-based sequence labeling model** fine-tuned on the **[HiFi-KPI dataset](https://huggingface.co/datasets/AAU-NLP/HiFi-KPI)** for extracting **financial key performance indicators (KPIs)** from **SEC earnings filings (10-K & 10-Q)**. It specializes in identifying entities, such as revenue, earnings etc. 

This model was introduced in the paper [HiFi-KPI: A Dataset for Hierarchical KPI Extraction from Earnings Filings](https://huggingface.co/papers/2502.15411) by Rasmus Aavang, Giovanni Rizzi, Rasmus Bøggild, Alexandre Iolov, Mike Zhang, and Johannes Bjerva.

### **Use Cases**
- Extracting **financial KPIs** from SEC **10-K and 10-Q** reports  
- **Financial document parsing** with iXBRL-based entity recognition   

### **Performance**
- Trained on **1,000 most frequent labels** from the **[HiFi-KPI dataset](https://huggingface.co/datasets/AAU-NLP/HiFi-KPI)**  

### **Dataset & Code**
- **Dataset**: [HiFi-KPI on Hugging Face](https://huggingface.co/datasets/AAU-NLP/HiFi-KPI)  
- **Code**: [HiFi-KPI GitHub Repository](https://github.com/aaunlp/HiFi-KPI)  

### **Citation**

```bibtex
@inproceedings{aavang-etal-2026-hifi,
    title = "{H}i{F}i-{KPI}: A Dataset for Hierarchical {KPI} Extraction from Earnings Filings",
    author = "Aavang, Rasmus T.  and
      Rizzi, Giovanni  and
      Tjalk-B{\o}ggild, Rasmus  and
      Iolov, Alexandre  and
      Zhang, Mike  and
      Bjerva, Johannes",
    editor = "Piperidis, Stelios  and
      Bel, N{\'u}ria  and
      van den Heuvel, Henk  and
      Ide, Nancy  and
      Krek, Simon  and
      Toral, Antonio",
    booktitle = "Proceedings of the Fifteenth Language Resources and Evaluation Conference",
    month = may,
    year = "2026",
    address = "Palma de Mallorca, Spain",
    publisher = "ELRA Language Resource Association",
    url = "https://aclanthology.org/2026.lrec-1.30/",
    doi = "10.63317/2nbsp7zzfb3g",
    pages = "441--455",
    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."
}
```