Page 63 - Plastics News - April 2026
P. 63
BUSINESS NEWS
With the foundation of the knowledge base, Learning from the LCEI Database
researchers created an automated workflow The resultant plastics LCEI database provides in-
combining prompt engineering and LLM-driven sights into research priorities. By plotting gaps
information extraction. This process resulted in in the data, researchers can visualize where re-
a structured database for LCEI in the plastics in- search is lacking. For example, emerging tech-
dustry. nologies such as chemical recycling have scarce
data. For policymakers, this lack of evidence
Overcoming Bias in Data may be a barrier to incentivizing the diversifica-
tion of plastics end-of -life strategies.
A database is only as useful as its source data.
Visualizations resultant from this LLM-based
The researchers conducted multi-dimensional
statistical analyses to visualize the data. Figure framework can give a top-down view of re-
courtesy of Artificial intelligence-driven frame- search gaps. Figure courtesy of Artificial in-
work for science-policy interface on global telligence-driven framework for science-policy
plastic life cycle environmental impacts. interface on global plastic life cycle environ-
mental impacts.
Research literature may introduce bias into the
workflow, underscoring the importance of data Another large gap is evident for engineering
quality control. Uncertainty analysis is a path- plastics, such as polyphenylene oxide, and spe-
way to improving LLM results. Out of three test- cialty engineering plastics (liquid crystal polymer,
ed machine learning (ML) models, researchers polyacrylate). Research around additives in the
chose eXtreme Gradient Boosting (XGBoost) for context of LCEI also remains limited. Data de-
further training and analysis. Afterwards, they ficiencies may affect policy outcomes due to a
screened the data from this model, which re- lack of scientific evidence. Reliable LCEI data can
sulted in the retention of higher-quality records. help policymakers identify intervention points,
These records were only 12.76% of the original appropriate substitutions, and design targeted
data volume, highlighting the need for data qual- incentives for industry stakeholders.
ity control. Source – Plastics Engineering
April 2026 PLASTICS NEWS 65
