A High Precision Pipeline for Financial Knowledge Graph Construction

Abstract

Motivated by applications such as question answering, fact checking, and data integration, there is significant interest in constructing knowledge graphs by extracting information from unstructured information sources, particularly text documents. Knowledge graphs have emerged as a standard for structured knowledge representation, whereby entities and their inter-relations are represented and conveniently stored as (subject, predicate, object) triples in a graph that can be used to power various downstream applications. The proliferation of financial news sources reporting on companies, markets, currencies, and stocks presents an opportunity for extracting valuable knowledge about this crucial domain. In this paper, we focus on constructing a knowledge graph automatically by information extraction from a large corpus of financial news articles. For that purpose, we develop a high precision knowledge extraction pipeline tailored for the financial domain. This pipeline combines multiple information extraction techniques with a financial dictionary that we built, all working together to produce over 342,000 compact extractions from over 288,000 financial news articles, with a precision of 78% at the top-100 extractions. The extracted triples are stored in a knowledge graph making them readily available for use in downstream applications.

Knowledge Graphs Representation for Event-Related E-News Articles

“…Fifteen years ago, researchers at the University of Twente and the University of Groningen, Netherlands, initiated the researches on KG theory and continued its focus on KG applications to analyze a text [5]. Over the years, several open KBs or ontologies have been created, including WordNet, DBpedia, YAGO, Freebase, BabelNet, and NELL, covering millions of real-world entities and relations in different domains [3,10,11,13]. Among these, traditional KGs such as ConceptNet and Freebase usually contain known facts and assertions about entities [12].…”

“…Usually, KGs are constructed using predefined databases or ontologies built by domain experts, human-annotated training data or semi-structured textual sources, or unstructured text on the web using manual curation approaches and automated techniques. Manual curation requires domain experts to study text sources and annotate sentences that include relationships, which is a tiresome and time-consuming process [3,13]. Manual approaches limit the scalability and applicability of KGs across various domains, while automated approaches which depend on machine learning or NLP techniques rapidly discover sentences of interest.…”

“…Manual approaches limit the scalability and applicability of KGs across various domains, while automated approaches which depend on machine learning or NLP techniques rapidly discover sentences of interest. The Open Information Extraction (OpenIE) technique, which extracts all semantic relational tuples in source text without human supervision, has been introduced to overcome the limitations in manual curation [3]. Later, researchers attempted to address the limitations in manually curated KGs by developing a procedure to populate a KB semi-automatically or automatically with values extracted from a corpus.…”

“…To address the limitations in manually curated domain-specific knowledge representations, automatic knowledge extractions and knowledge-representation methods have been introduced. Recent research by Elhammadi et al [3] designed a high-precision automatic knowledge extraction pipeline to build a KB semi-automatically or automatically with quintuples of the form subject-predicate-object-monetary value-date extracted from a financial news domain using Semantic Role Labeling (SRL) and pattern-based information extraction to extract noun-and verb-mediated relations. This pipeline contains over 380,000 extracted triples, which are readily queried.…”

A General Method for Transferring Explicit Knowledge into Language Model Pretraining

“…For example, "Xiaomi is a science and technology company" is kind of typical explicit knowledge. It could be well stored in knowledge bases, represented in the form of SPO (subject, predicate, and object) triplet [17], where subject and object are entities and predicate is a relation between those entities. is paper seeks to find the impact of explicit knowledge on transformer pretraining.…”