A CEP-driven framework for real-time news impact prediction on financial markets

Real-time news impact prediction on financial markets is a challenging task for finance experts with limited IT expertise. Many practitioners build machine learning models trained with many low-level features extracted from multiple event-based streams (news and financial market data), which often l...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Service oriented computing and applications Jg. 17; H. 2; S. 129 - 144
Hauptverfasser: Chen, Weisi, El Majzoub, Ahmad, Al-Qudah, Islam, Rabhi, Fethi A.
Format: Journal Article
Sprache:Englisch
Veröffentlicht: London Springer London 01.06.2023
Springer Nature B.V
Schlagworte:
Algorithms
Classification
Computer Appl. in Administrative Data Processing
Computer Science
Computer Systems Organization and Communication Networks
Data collection
Data transmission
Datasets
Digital currencies
e-Commerce/e-business
Finance
Impact prediction
IT in Business
Machine learning
Management of Computing and Information Systems
Manufacturing
News
News media
Pharmaceutical industry
Real time
Securities markets
Sentiment analysis
Social networks
Software
Software Engineering/Programming and Operating Systems
Special Issue Paper
Subject specialists
Windows (intervals)
Data quality
Real-time analytics
Service-oriented architecture
Finance
Complex event processing
Machine learning
ISSN:1863-2386, 1863-2394
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Real-time news impact prediction on financial markets is a challenging task for finance experts with limited IT expertise. Many practitioners build machine learning models trained with many low-level features extracted from multiple event-based streams (news and financial market data), which often leads to poor outcomes. State-of-the-art solutions either ignore domain-specific contexts or are customised to merely one type of static datasets rather than real-time data streams. In most cases, the domain expert would have to manually conduct data collection, data cleaning and aggregation, and machine learning step by step with the assistance of IT experts, which is time-consuming and complicated. To address these limitations, we propose a technique that uses real-time data pre-processing in accordance with domain-specific event patterns to generate better-quality datasets. This technique is supported by a systematic framework featuring a data model capturing domain-specific event patterns, an SOA-based architecture and processes that integrate the capabilities of sentiment analysis, complex event processing with automated machine learning (AutoML), facilitating event pattern detection and continual learning with sliding time windows. The benefit of adopting an SOA architecture is to ensure the flexibility of the selection and seamless integration of components. This solution allows domain experts to define domain-specific event patterns via a user-friendly interface and prepare better-quality datasets by pre-processing real-time data streams accordingly via the complex event processing component, aiming to generate meaningful prediction results by the downstream AutoML component. The AutoML component allows for minimal machine learning skills by the domain expert to conduct the prediction tasks. A prototype was implemented to evaluate its feasibility and functionality on a real-life price movement prediction scenario involving 3 years of news and financial market data. The results demonstrate that finance experts are able to complete news impact prediction tasks in real time without the intervention of IT experts, which saves a large amount of time compared with traditional machine learning processes.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1863-2386
1863-2394
DOI:10.1007/s11761-023-00358-8