Datasets:
The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: ValueError
Message: Bad split: train. Available splits: ['test']
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 118, in get_rows
ds = safe_load_dataset(
dataset,
...<4 lines>...
download_config=download_config,
)
File "/src/services/worker/src/worker/utils.py", line 531, in safe_load_dataset
return load_dataset(
path,
...<5 lines>...
token=token,
)
File "/usr/local/lib/python3.14/site-packages/datasets/load.py", line 1684, in load_dataset
return builder_instance.as_streaming_dataset(split=split)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1165, in as_streaming_dataset
raise ValueError(f"Bad split: {split}. Available splits: {list(splits_generators)}")
ValueError: Bad split: train. Available splits: ['test']Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
wikipunk/fibo2023Q3(複製)
データの内容
金融業界ビジネスオントロジー(FIBO)を三つ組(主語・述語・目的語)に直したデータ。1 行が 1 つの三つ組で、subject、predicate、object はいずれも N3 表記の文字列であり、プレフィックスの対応付けは含まれない。オントロジーは Protege で読み込み、ELK で推論して推論された公理を書き出し、N-Triples に変換して gzip で圧縮された。業種分類の代わりにはならない。
データの概要
| 項目 | 値 |
|---|---|
| 行数 | 236,579(1 構成・1 分割の合計) |
| 列数 | 3 |
| 容量 | 15.03 MB(ファイル 379 本) |
| ファイルの形式 | rdf 187、md 2、nt.gz 1、py 1、txt 1、(拡張子なし) 1 |
| 言語 | en |
| 公開 | public |
構成と分割
| 構成(config) | 分割(split)ごとの行数 | 列数 |
|---|---|---|
default |
train 236,579 |
3 |
列(構成 default)
| 列 | 型 |
|---|---|
subject |
string |
predicate |
string |
object |
string |
行数と列は元のカードの YAML(dataset_info)の記載。データセットビューアでは読めなかった。全データセットの一覧は HYUKI5/dataset-catalog にある。
複製について
wikipunk/fibo2023Q3 をリビジョン 250a2e6ab51421a2bcf1f47c4957d76a55ad0c8f の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(convaiinnovations/laya)の微調整に使う候補として、調査メモ(Laya 学習用データセット調査)に載せたデータを集めている。
出典
- 元のデータセット: https://huggingface.co/datasets/wikipunk/fibo2023Q3
- 複製したリビジョン:
250a2e6ab51421a2bcf1f47c4957d76a55ad0c8f(2023-10-04 20:03:28+00:00) - 元のライセンス表記:
mit
論文
- 元のカードに論文へのリンクは見つからなかった
アノテーションの方法
構築と検証:
- 構築: オントロジーは AboutFIBOProd-IncludingReferenceData から Protege バージョン 5.6.1 に読み込まれた。
- 推論: オントロジーが大きいため、ELK リーザープラグインを使って、オントロジー内の推論を具体化(明示化)した。
- 整合性の検査: Protege の Debug Ontology プラグインを使って、オントロジーの整合性と一貫性を確かめた。
- 書き出し: 検証後、推論された公理は、宣言された公理やアノテーションとともに、Protege を使って書き出された。
- 符号化と圧縮: Apache Jena の riot を使って結果を ntriples に変換し、その後 gzip で圧縮した。この圧縮された成果物は、Hugging Face の datasets ライブラリによってダウンロードされ展開され、データセットの例が得られる。
微調整に使うときの注意
- 調査メモの台帳での区分: B(継承条件・gated・モデル由来の条件・上流の規約などが付く)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
- Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
- 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。
調査メモの記述
| メモ | 節 | ライセンス(記載) | 確認 | 注記 |
|---|---|---|---|---|
ontology_kg_entity.md |
Q3. 業種分類(企業説明 → SIC / NAICS / GICS)にどんなデータがあり、GICS 相当をどう用意するか (L99) | fibo2023Q3: MIT(HF タグ)。FinSim の配布先・ライセンスは未確認 | 確認済 | 上流/条件注記 / 確認印なし(各ファイルの凡例により確認済扱い) / 原文注記: 業種分類の代替にならない。FinSim の実データは取得していない |
元のカード(日本語訳)
以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。
FIBO: 金融業界ビジネスオントロジー
概要
金融テクノロジーの世界では、データの膨大さと金融商品の複雑さが、課題と機会の両方をもたらしている。金融業界ビジネスオントロジー(FIBO)は、理論的な金融の概念と実世界のデータとの間の隔たりを埋める、構造化された枠組みを提供する。金融分野に関心を持つ機械学習の研究者は、FIBO の関係を金融の特徴量エンジニアリングに活かし、既存のモデルを微調整したり新しいモデルを構築したりできると考える。
オープンソース
FIBO のオントロジーは GitHub で開発されている: https://github.com/edmcouncil/fibo/
ユースケース
- 包括的なデータ構造: FIBO はデリバティブから証券まで、幅広い金融の概念を提供する。この設計は知識表現と金融の両分野の専門知識に基づき、金融商品の深い理解を保証する。
- 複雑な関係の解読: 金融の領域は、入り組んだ相互依存を特徴とする。FIBO の構造化されたアプローチはこれらの関係を明確にし、機械学習のアルゴリズムが大規模なデータセット内のパターンや相関を見つけられるようにする。
- 実世界のデータとの連携: FIBO の際立った特徴は、金融の概念を実世界の金融データや統制語彙(controlled vocabularies)と結び付けられることである。この結び付けは、既存のデータを持つ金融企業の実務の文脈で理論的な知見を応用しようとする研究者にとって重要である。
- Retrieval Augmented Generation: 大規模言語モデル、特に Retrieval Augmented Generation(RAG)と組み合わせた場合、金融データの処理と解釈の仕方を変革する見込みがある。
- 文書分類: 金融文書が急増するなか、RAG を使って FIBO の概念で分類された金融データセットを分類すれば、巧みなプロンプトによって、金融アナリストがデータ解釈の精度と深さを高められる。
構築と検証:
- 構築: オントロジーは AboutFIBOProd-IncludingReferenceData から Protege バージョン 5.6.1 に読み込まれた。
- 推論: オントロジーが大きいため、ELK リーザープラグインを使って、オントロジー内の推論を具体化(明示化)した。
- 整合性の検査: Protege の Debug Ontology プラグインを使って、オントロジーの整合性と一貫性を確かめた。
- 書き出し: 検証後、推論された公理は、宣言された公理やアノテーションとともに、Protege を使って書き出された。
- 符号化と圧縮: Apache Jena の riot を使って結果を ntriples に変換し、その後 gzip で圧縮した。この圧縮された成果物は、Hugging Face の datasets ライブラリによってダウンロードされ展開され、データセットの例が得られる。
使い方
まず必要なパッケージがインストールされていることを確認する:
pip install datasets
pip install rdflib
次の Python コードで、Hugging Face Datasets ライブラリを使ってデータセットを読み込める:
from datasets import load_dataset
dataset = load_dataset('wikipunk/fibo2023Q3', split='train')
特徴
FIBO データセットは、金融の異なる概念と、市場参加者・企業・契約上の主体などの名前付きの個体(named individuals)の間の関係を表す三つ組(triples)からなる。
形式についての注記:
subject、predicate、object の各フィーチャーは、プレフィックスの対応付けなしの N3 表記で格納されている。これにより、RDFLib Python ライブラリの rdflib.util.from_n3 を使って各要素を解析できる。
1. Subject(string)
三つ組の subject は、その文の主たる実体または焦点である。このデータセットでは、subject は特定の金融商品や実体を表すことが多い。例えば:
<https://spec.edmcouncil.org/fibo/ontology/SEC/Equities/EquitiesExampleIndividuals/XNYSListedTheCoca-ColaCompanyCommonStock>
は、NYSE に上場している The Coca-Cola Company の普通株を指す。
2. Predicate(string)
三つ組の predicate は、subject と object の間の関係の性質を示す。subject の特定の性質、特徴、または結び付きを記述する。今回の例では:
<https://spec.edmcouncil.org/fibo/ontology/SEC/Securities/SecuritiesListings/isTradedOn>
は、金融商品(subject)が特定の取引所(object)で取引されていることを示す。
3. Object(string)
三つ組の object は、predicate を介して subject と関連付けられた実体または値である。別の金融の概念、取引プラットフォーム、またはその他の関連する実体でありうる。今回の例の文脈では:
<https://spec.edmcouncil.org/fibo/ontology/FBC/FunctionalEntities/NorthAmericanEntities/USMarketsAndExchangesIndividuals/NewYorkStockExchange>
は、前述の Coca-Cola の普通株が取引されているニューヨーク証券取引所を表す。
続き
データセット内の三つ組の別の例を示す:
- Subject:
"<https://spec.edmcouncil.org/fibo/ontology/FBC/FunctionalEntities/MarketsIndividuals/ServiceProvider-L-JEUVK5RWVJEN8W0C9M24>" - Predicate:
"<http://www.w3.org/1999/02/22-rdf-syntax-ns#type> - Object:
"<https://spec.edmcouncil.org/fibo/ontology/BE/FunctionalEntities/FunctionalEntities/FunctionalEntity>"
この三つ組は、市場の個体 ServiceProvider-L-JEUVK5RWVJEN8W0C9M24 の型が FunctionalEntity であるという記述を表す。
注記:
データセットには、オントロジーから取った例示的な個体が参照点として含まれる。これらの例は、金融領域内の関係と実体を理解するための構造化された枠組みを与える。ただし、含まれる個体は網羅的ではない。特に RAG を含む大規模言語モデルの進歩により、これらの例を生成して拡張し、より構造化されたデータと知見でデータセットを豊かにできる可能性がある。
FIBO Viewer
FIBO Viewer を使えば、ウェブ上でオントロジーを探索できる。FIBO の最も優れた特徴の一つは、https://spec.edmcouncil.org/fibo/ontology/ という接頭辞を持つ実体が、その URL をブラウザや任意の HTTP クライアントで開くだけで、ウェブ上で参照できることである。
FIBO から Graph Neural Network の特徴量を導く案:
Graph Neural Networks(GNN)は、構造化されたデータに対する機械学習の強力な道具として登場した。構造化されたオントロジーを持つ FIBO は、GNN の特徴量を導くために活用できる。
ノードの特徴量:
rdf:type: FIBO の各実体には、そのクラスやカテゴリを示す
rdf:type(<http://www.w3.org/1999/02/22-rdf-syntax-ns#type>)が 1 つ以上関連付けられている。これは、符号化すべき主要なノードの特徴量として使える。実体の属性: 名前や説明など、各実体の属性を追加のノードの特徴量として使える。説明文は、意味的なテキストの埋め込みモデルを使って埋め込むことを検討すること。
エッジの特徴量:
- RDF Predicates: FIBO の実体間の関係は RDF の predicate で表される。これらの predicate は GNN のエッジの特徴量として使え、ノード間の関係の性質を捉える。
想定される応用:
- 実体の分類: 導出したノードとエッジの特徴量を使うと、GNN は実体をさまざまな金融のカテゴリに分類でき、金融データ分析の細かさを高める。
- 関係の予測: GNN は実体間で起こりうる関係を予測でき、金融データに潜む隠れたパターンや相関の発見を助ける。
- 異常検知: FIBO の構造化データと相互にリンクされた金融データセットで GNN を学習させることで、それらに含まれる異常や不規則性を検出でき、データの完全性と正確性を保証できる。
謝辞
FIBO の貢献者による知識表現における丹念な取り組みに、心からの感謝を述べる。彼らの専門性と献身は、金融業界のイノベーションの礎となる包括的で示唆に富む枠組みの形成に不可欠であった。
金融業界のモデル化に関心がある場合は、FIBO への貢献 を検討すること。
引用
@misc{fibo2023Q3,
title={Financial Industry Business Ontology (FIBO)},
author={Object Management Group, Inc. and EDM Council, Inc. and Various Contributors},
year={2023},
note={Available as OWL 2 ontologies and UML models compliant with the Semantics for Information Modeling and Federation (SMIF) draft specification. Contributions are open on GitHub, consult the repository for a list of contributors.},
howpublished={\url{https://spec.edmcouncil.org/fibo/}},
abstract={The Financial Industry Business Ontology (FIBO) is a collaborative effort to standardize the language used to define the terms, conditions, and characteristics of financial instruments; the legal and relationship structure of business entities; the content and time dimensions of market data; and the legal obligations and process aspects of corporate actions.},
license={MIT License, \url{https://opensource.org/licenses/MIT}}
}
原文(英語)
FIBO: The Financial Industry Business Ontology
Overview
In the world of financial technology, the vastness of data and the complexity of financial instruments present both challenges and opportunities. The Financial Industry Business Ontology (FIBO) offers a structured framework that bridges the gap between theoretical financial concepts and real-world data. I believe machine learning researchers interested in the financial sector could use the relationships in FIBO to innovate in financial feature engineering to fine-tune existing models or build new ones.
Open Source
The FIBO ontology is developed on GitHub at https://github.com/edmcouncil/fibo/.
Use-cases
- Comprehensive Data Structure: FIBO offers a broad spectrum of financial concepts, ranging from derivatives to securities. This design, rooted in expert knowledge from both the knowledge representation and financial sectors, ensures a profound understanding of financial instruments.
- Decoding Complex Relationships: The financial domain is characterized by its intricate interdependencies. FIBO's structured approach provides clarity on these relationships, enabling machine learning algorithms to identify patterns and correlations within large datasets.
- Linkage with Real-world Data: A distinguishing feature of FIBO is its capability to associate financial concepts with real-world financial data and controlled vocabularies. This connection is crucial for researchers aiming to apply theoretical insights in practical contexts in financial enterprises with their existing data.
- Retrieval Augmented Generation: The advent of Large Language Models, particularly in conjunction with Retrieval Augmented Generation (RAG), holds promise for revolutionizing the way financial data is processed and interpreted.
- Document Classification: With the surge in financial documents, utilizing RAG to categorize financial datasets classifed by FIBO concepts can assist financial analysts in achieving enhanced accuracy and depth in data interpretation, facilitated by intelligent prompting.
Building and Verification:
- Construction: The ontology was imported from AboutFIBOProd-IncludingReferenceData into Protege version 5.6.1.
- Reasoning: Due to the large size of the ontology I used the ELK reasoner plugin to materialize (make explicit) inferences in the ontology.
- Coherence Check: The Debug Ontology plugin in Protege was used to ensure the ontology's coherence and consistency.
- Export: After verification, inferred axioms, along with asserted axioms and annotations, were exported using Protege.
- Encoding and Compression: Apache Jena's riot was used to convert the result to ntriples, which was then compressed with gzip. This compressed artifact is downloaded and extracted by the Hugging Face datasets library to yield the examples in the dataset.
Usage
First make sure you have the requirements installed:
pip install datasets
pip install rdflib
You can load the dataset using the Hugging Face Datasets library with the following Python code:
from datasets import load_dataset
dataset = load_dataset('wikipunk/fibo2023Q3', split='train')
Features
The FIBO dataset is composed of triples representing the relationships between different financial concepts and named individuals such as market participants, corporations, and contractual agents.
Note on Format:
The subject, predicate, and object features are stored in N3 notation
with no prefix mappings. This allows users to parse each component
using rdflib.util.from_n3 from the RDFLib Python library.
1. Subject (string)
The subject of a triple is the primary entity or focus of the statement. In this dataset, the subject often represents a specific financial instrument or entity. For instance:
<https://spec.edmcouncil.org/fibo/ontology/SEC/Equities/EquitiesExampleIndividuals/XNYSListedTheCoca-ColaCompanyCommonStock>
refers to the common stock of The Coca-Cola Company that is listed on
the NYSE.
2. Predicate (string)
The predicate of a triple indicates the nature of the relationship between the subject and the object. It describes a specific property, characteristic, or connection of the subject. In our example:
<https://spec.edmcouncil.org/fibo/ontology/SEC/Securities/SecuritiesListings/isTradedOn>
signifies that the financial instrument (subject) is traded on a
particular exchange (object).
3. Object (string)
The object of a triple is the entity or value that is associated with the subject via the predicate. It can be another financial concept, a trading platform, or any other related entity. In the context of our example:
<https://spec.edmcouncil.org/fibo/ontology/FBC/FunctionalEntities/NorthAmericanEntities/USMarketsAndExchangesIndividuals/NewYorkStockExchange>
represents the New York Stock Exchange where the aforementioned
Coca-Cola common stock is traded.
Continued
Here is an another example of a triple in the dataset:
- Subject:
"<https://spec.edmcouncil.org/fibo/ontology/FBC/FunctionalEntities/MarketsIndividuals/ServiceProvider-L-JEUVK5RWVJEN8W0C9M24>" - Predicate:
"<http://www.w3.org/1999/02/22-rdf-syntax-ns#type> - Object:
"<https://spec.edmcouncil.org/fibo/ontology/BE/FunctionalEntities/FunctionalEntities/FunctionalEntity>"
This triple represents the statement that the market individual ServiceProvider-L-JEUVK5RWVJEN8W0C9M24 has a type of FunctionalEntity.
Note:
The dataset contains example individuals from the ontology as reference points. These examples provide a structured framework for understanding the relationships and entities within the financial domain. However, the individuals included are not exhaustive. With advancements in Large Language Models, especially Retrieval Augmented Generation (RAG), there's potential to generate and expand upon these examples, enriching the dataset with more structured data and insights.
FIBO Viewer
Use the FIBO Viewer to explore the ontology on the web. One of the coolest features about FIBO is that entities with a prefix of https://spec.edmcouncil.org/fibo/ontology/ can be looked up in the web just by opening its URL in a browser or in any HTTP client.
Ideas for Deriving Graph Neural Network Features from FIBO:
Graph Neural Networks (GNNs) have emerged as a powerful tool for machine learning on structured data. FIBO, with its structured ontology, can be leveraged to derive features for GNNs.
Node Features:
rdf:type: Each entity in FIBO has one or more associated
rdf:type,<http://www.w3.org/1999/02/22-rdf-syntax-ns#type>, that indicates its class or category. This can serve as a primary node feature to encode.Entity Attributes: Attributes of each entity, such as names or descriptions, can be used as additional node features. Consider embedding descriptions using a semantic text embedding model.
Edge Features:
- RDF Predicates: The relationships between entities in FIBO are represented using RDF predicates. These predicates can serve as edge features in a GNN, capturing the nature of the relationship between nodes.
Potential Applications:
Entity Classification: Using the derived node and edge features, GNNs can classify entities into various financial categories, enhancing the granularity of financial data analysis.
Relationship Prediction: GNNs can predict potential relationships between entities, aiding in the discovery of hidden patterns or correlations within the financial data.
Anomaly Detection: By training GNNs on the structured data from FIBO and interlinked financial datasets, anomalies or irregularities in them may be detected, ensuring data integrity and accuracy.
Acknowledgements
We extend our sincere gratitude to the FIBO contributors for their meticulous efforts in knowledge representation. Their expertise and dedication have been instrumental in shaping a comprehensive and insightful framework that serves as a cornerstone for innovation in the financial industry.
If you are interested in modeling the financial industry you should consider contributing to FIBO.
Citation
@misc{fibo2023Q3,
title={Financial Industry Business Ontology (FIBO)},
author={Object Management Group, Inc. and EDM Council, Inc. and Various Contributors},
year={2023},
note={Available as OWL 2 ontologies and UML models compliant with the Semantics for Information Modeling and Federation (SMIF) draft specification. Contributions are open on GitHub, consult the repository for a list of contributors.},
howpublished={\url{https://spec.edmcouncil.org/fibo/}},
abstract={The Financial Industry Business Ontology (FIBO) is a collaborative effort to standardize the language used to define the terms, conditions, and characteristics of financial instruments; the legal and relationship structure of business entities; the content and time dimensions of market data; and the legal obligations and process aspects of corporate actions.},
license={MIT License, \url{https://opensource.org/licenses/MIT}}
}
- Downloads last month
- 41