Project

SENTiVENTL event extraction and sentiment analysis for financial applications.

Acronym
SentEvent
Duration
01 January 2017 → Ongoing
Funding
Research Foundation - Flanders (FWO)
Research disciplines
  • Humanities
    • Computational linguistics
    • Corpus linguistics
  • Natural sciences
    • Data mining
    • Machine learning and decision making
    • Natural language processing
    • Information retrieval and web search
    • Ontologies, data curation and text mining
  • Social sciences
    • Knowledge representation and machine learning
    • Business economics
    • Applied economics not elsewhere classified
Keywords
sentiment analysis event extraction financial analysis applications information retrieval text mining machine learning natural language processing
 
Project description

In economic news, journalists and analysts give objective information on recent events while also
discussing the implications of events in an implicitly subjective manner. We investigate text mining
approaches for extracting structured factual data alongside subjective information from Dutch and
English economic news reporting. Event extraction obtains detailed information about economic
events such as acquisitions, CEO changes, or product launches: it summarizes an event and tells us
who is involved in what event with which event properties. Aspect-based sentiment analysis gives
us an overview about what negative or positive opinion is expressed about what part of an event
or entity. In standard sentiment analysis, only explicitly expressed sentiment is detected ("This
movie is fantastic.") and current systems do not not handle common-sense implicit sentiments
which is connotationally attached to certain events or situations. For instance, "Motorola sees an
increase in revenue" implies a positive sentiment towards the company. This implicit sentiment
makes up half of the opinion expressions in economic news, so processing it is important for
making financial technology applications. As validation, the extracted factual and opinion data is
compared to judgments of financial analysts and is used in stock price prediction experiments
where we automatically predict the price movement of stocks.