| Abstract |
Managingwaste plastic is a serious global challenge since mostof this waste is either landfilled, incinerated, burned in the open,or littered. Each of these approaches has a large environmental impact.Establishing a circular economy of plastics requires its recoveryand recycling, and much effort is now focused in this direction. Thebody of literature on approaches for managing the end of life of plasticsis growing exponentially, making it increasingly difficult to segregatethe most relevant information across multiple articles. Such workis extremely time- and effort-consuming, particularly when performedmanually. To address this issue, in this study, we propose a methodologybased on natural language processing (NLP) for automatically extractingand compiling information thatis most relevant to a selected category of plastics. In the developedmethodology, the research articles are first extracted with the helpof a science-direct Elsevier Application Programming Interface keyby utilizing a set of keywords such as "polyethylene recyclemethods", "polyethylene terephthalate recycle methods","polypropylene recycle methods", and "polystyrenerecycle methods" for relevant articles. Extracted articlesare processed to address two fundamental problems; (i) classificationand (ii) question and answer (Q & A) related to literature pertainingto plastic waste recycling. To this extent, we developed a bundleof NLP tools called Recycle-Bidirectional Encoder Representationsfrom Transformers (BERT). Under the hood, Recycle-BERT comprised fivelanguage models, (1) Class-BERT, for classifying the literature asrelevant or nonrelevant; (2) Catalyst-BERT, for extracting catalystdetails for recycling; (3) Method-BERT, for finding the methods enlistedin the literature for recycling; (4) Reactant-BERT to identify thereactants used for waste recycling; and (5) Product-BERT to pinpointproducts obtained from recycling. We have evaluated the performanceof the developed models based on the metrics such as accuracy andF1-score. For the classification task, an accuracy metric value of0.974 is obtained for the test data set. Similarly, the metric F1-scorevalues for the Q & A task are 0.7646, 0.8014, 0.8221, and 0.8512for the test data set for Catalyst-BERT, Method-BERT, Reactant-BERT,and Product-BERT, respectively. The results indicate the proposedNLP-based model's ability to extract essential informationfrom the literature related to plastic waste processing, aiding suitablerecommendations to assist transformation to a sustainable circulareconomy. Reliable literature on plastic wasterecycling was usedto build a natural language processing-based Q & A framework, Recycle-BERT,as a knowledge extractor. |