Natural Language Processing
Research Areas
Intro
Today, we live in a time in which talking to machines is normal. In the morning, many of us are
woken up by a virtual personal assistant. In the afternoon, a chatbot processes the civil complaint
we filed. There may even be times we are fooled into thinking that the machine actually
understands our emotions.
How then does a machine understand human language?
Natural Language Processing(NLP) refers to the realm of AI technology in which a computer
understands, processes, and “speaks” a human language. Research on NLP began in the 1950s,
Warren Weaver first suggested the idea of translation between English and French based on a
dictionary, which is known as “machine translation.” In the 1990s, this segued into studies on
statistical methods and machine learning-based methods that utilize massive amounts of textual
data. Recently, the advancement of deep learning technology has also been leading to international
interest in the extraction of meaningful information from an immense quantity of text corpora and
achieving state-of-the-art advancements in many NLP tasks.
What is NLP?
Since 2018, HIT has worked very hard to study diverse NLP technologies specializing in the financial
domain to deliver the latest developments to Hana Financial Group(such as deep learning).
The first key technology group in NLP is core foundational technologies, which carry out basic
processing. The foundation technology in NLP is analyzing syntax of the text, which includes
morpheme analysis, part-of-speech tagging, and entity recognition, from diverse kinds of irregular
domain text (counseling logs, regulations, etc.) news and community data. The second area is
Natural Language Understanding, which is made up of: meaning analysis/classification (analyzes
what user wants or intends), topic/event extraction technology (analyzes periodic events or a
certain incident or situation), machine learning, and sentiment analysis. The third area, text mining,
includes keyword extraction, inter-textual relationship extraction, text similarity analysis, and text
clustering technologies. Finally, information retrieval involves research of optimized index/search
algorithms. This area also studies and develops technologies that: 1) define and learn ranking
models of search outcomes, 2) search for information based on a Q&A system and 3) extract
textual data from diverse types of internal or external data.
-
NLP core
foundational
technologies - · Morpheme analysis
- · Semantic analysis
- · Syntax analysis
-
· Pragmatic semantics
analysis
-
Natural Language
-
· Irregular financial
text data - · News data
- · Social media text data
- Natural Language Processing
- · Intent analysis/classification
-
· Machine Reading
Comprehension (MRC) - · Topic/event extraction
- · Sentiment analysis
- Text Mining
- · Keyword extraction
- · Text similarity analysis
-
· Inter-textual relationship
extraction - · Text clustering
- Information Retrieval
- · Search optimization algorithm
- · IR-based Q&A
- · Ranking model/learning
- · Textual information extraction
Future Research
HIT is searching for ways to analyze and improve NLP technologies currently being used for
financial chatbots and counseling assistants. Ultimately, the goal is to internalize NLP technology
in a more sophisticated format and to create a text-based data providing environment that
satisfies both customers and employees.
To achieve these goals, HIT will continue to develop new technologies that apply deep learning
and reinforcement learning while also creating and packaging solutions for the establishment
and operation of systematic services. Cooperation among the HIT investment, data science, and
computer vision divisions is expected to create synergy and result in the development of better
financial services.