A Model for processing public participation feedback using topic Modeling

Abstract

Governments acknowledge the need to involve citizens, through different platforms like public task forces, when public policies are drafted. Public task forces are expected to consider views from different stakeholders in the process of drafting their recommendations and policy proposals. They tend to receive large amounts of submissions and in most cases, the submissions are contained in massive documentations. The amount of material received is beyond the task forces’ processing ability causing input from critical stakeholders being ignored completely leading to biased output. This provides a situation where topic modeling can be applied. Topic modeling is an unsupervised machine learning algorithm that can process large corpora of data by classifying them by identifying themes in those corpora. In our study, we set out to develop a model that task forces can use to process the received feedback. The model was validated by comparing the topics identified by the model against those identified by a human expert. An experiment was conducted where we built and trained an LDA topic model with 15 submissions. We then presented 7 submissions both to the trained model for processing and also to a human expert to manually identify the topics contained in those submissions. The topics generated by the model were compared to the topics identified by the human expert. The model generated topics that are similar to the topics identified by a human expert. Distinctive topics are contained in submissions. These topics are few and trying to extract a higher number of topics, will lead to more overlapping and nonsensical topics being generated. Extraction of topics contained in feedback from the public can be automated. This study contributes to practice by enabling task forces to objectively identify topics and themes covered by submissions which results into more acceptance of outputs and recommendations of task forces by the citizens.