webleads-tracker

Winning a financial data numeral understanding competition - Fortia
Data Science
fortia_admin fortia_admin
31 May 2019

Winning a financial data numeral understanding competition

NTCIR-14 FinNum Task

Last week, our team of Data Scientists successfully won the Fine-grained Numeral Understanding of financial tweets’ competition. We are so proud of them!

 

Introduction

In order to understand the details of a financial document, investors do not only need to analyze the text, but also the fundamental and technical numeric information. The numerical data plays a crucial part in the financial domain, for instance in the evaluation of security value or in the prediction of a stock market. Thus, the identification, understanding, and analysis of this financial data form the backbone of the industry, with data being the most important resource for investors.

Today, social media is becoming an important hub for investors as it contains a large volume of financial data impacting the way their opinions are shaped. Twitter is one of these platforms gathering massive numerical-based tweets creating new challenges to process the data manually.

National Taiwan University has organized FinNum, a task for fine-grained numeral understanding in financial social media data, to identify the category of a numeral.

Details on the data provided and tasks

Fortia data science team had to work mainly with unstructured data. The taxonomy of numerals was provided. It was classified into 7 categories and some subcategories. The FinNum shared task was to classify these numerals into the predefined categories and sub-categories.

Subtask 1: Classify a numeral into 7 categories, i.e., Monetary, Percentage, Option, Indicator, Temporal, Quantity and Product/Version Number.

Subtask 2: Extend the classification task to the subcategory level, and classify numerals into 17 classes, including Indicator, Quantity, Product/Version Number, and all subcategories shown in Table 1.

Using the provided dataset, Fortia’s Data Scientists created a new deep learning architecture to tackle this challenge of multi-class classification. Stay tuned! Our team will present their model at the 14th NTCIR Conference in Japan next June!

Link to the FinNum participants result

By continuing your navigation, you accept the use of cookies for statistical purposes and personalization. Learn more