Refine viewpoints into opportunities through NLP

In everyday life, we often rely on numerical scores to impart quick judgement on things. However, it is often difficult for human judgement to tie down complex subjects, analysis and experiences into a single number. This is where Natural Language Processing (NLP) can help.

While we humans often don’t have the time to read every single written analysis to get the complete picture, NLP can do it with ease. NLP can intelligently scan the text of every single analysis and distil all qualitative text into a single quantitative score that predicts a much more accurate and reliable reflection of the collective viewpoint.

We at AllianzGI envisage NLP playing a key role in the active asset management of tomorrow. NLP can help to provide insights that cuts through the common biases and distortions that arise from the standard investment metrics used in the industry.

What is NLP?
Natural language processing (NLP) made possible by advances in artificial intelligence and computational linguistics utilises computers to analyse and understand language, based on keywords, which can be trained to come up with a highly predictive result.
For instance, the process can distill written text into a quantifiable score which is useful in the measurement and independent comparison of language.
Why NLP?

Numbers, regardless of their relevancy or accuracy, can be easily compared to one another.
In contrast, a descriptive text or opinion often captures emotions and “gut feel”, which is much more subjective and open to interpretation.

As a result, qualitative language is much more cumbersome to compare.

How NLP works?
Let’s take restaurant rating as an example and see how NLP works:
AllianzGI
AllianzGI
AllianzGI
AllianzGI
Turn sentences into words to process the text
  • Adjust the language into a format the computer algorithm can analyse
  • Clean the sentences to remove words with no productive power such as “I”, “to”, “we”, etc.
AllianzGI
Analyse the text by turning words into vectors
  • Necessary step to run regression models required to analyse the words and come up with a solution
AllianzGI
Translate vectors into a regression model to calculate a final star score
At times, comments provided in reviews may differ from their actual star score, which is where things get more interesting.
AllianzGI

User review #1:

“My Favorite place to eat in Las Vegas! The BBQ Ribs are phenomenal, along with the $1 beers. Such a great Brewery!”

User review #1:

“My Favorite place to eat in Las Vegas! The BBQ Ribs are phenomenal, along with the $1 beers. Such a great Brewery!”
AllianzGI
AllianzGI
User's stars: AllianzGI
AllianzGI Predicted stars: AllianzGI
AllianzGI

User review #2:

“I have great expectations, but they let me down, i order the carne asada salad, w beans, rice, tomatoes letuce and steak, but the dressing is to watery, too runny, the steak was blend and the rice mushy, also get the guac and chips, the chips way to salty, and the guac was bland, need a little lime , cilantro onions or seasoning, just taste like a mash avocados, the staff was very friendly, thats why i give the 5 stars, i hope you fix it all and for my next visit u will get the corn, (they didnt have it this time), good luck ...”

User review #2:

“I have great expectations, but they let me down, i order the carne asada salad, w beans, rice, tomatoes letuce and steak, but the dressing is to watery, too runny, the steak was blend and the rice mushy, also get the guac and chips, the chips way to salty, and the guac was bland, need a little lime , cilantro onions or seasoning, just taste like a mash avocados, the staff was very friendly, thats why i give the 5 stars, i hope you fix it all and for my next visit u will get the corn, (they didnt have it this time), good luck ...”
AllianzGI
AllianzGI User's stars: AllianzGI
AllianzGI Predicted stars: AllianzGI
The difference in predicted score from actual score can be from simple user error... or because of behavioral biases which cannot be as easily spotted.
How is it applied into active asset management?

The investment world carries similar distortion of commonly used investment metrics from intended or unintended bias.

AllianzGI
Natural language processing for the first time should allow our investment team to every day read through information, comprehend and capture the sentiment of
over 10,000 stocks
in the investment universe.

We have devised models that strip away biases in an analyst’s ratings, estimates and target prices using NLP.

Instances where actual score differs from predicted score offer differentiated viewpoint and opportunities for active management.

AllianzGI
Our vision

Do you know…
AI is a theme you can invest in? Find out more about our thematic investing offering in different locations:

Hong Kong (Chinese only)

Singapore