SENTIMENT ANALYSIS MODEL FOR ONLINE PUBLIC PARTICIPATION FORUMS

Abstract

Public participation (PP) is a key constitutional principle outlined in the Constitution of Kenya. It promotes democratic and accountable exercise of power. It gives the citizens opportunity to enhance self-development and service delivery while accounting for their leaders‘ actions. However, lack of/insufficient public participation in Kenyan county governments is impeding effective devolution process. Among the reasons advanced for this development are inadequate communications. Still even in cases where PP has been successfully carried out, capturing and analysing the sentiments of the participants still remain a serious challenge. Therefore, an online PP tool with embedded sentiment analysis algorithms specifically designed for the counties can be quite resourceful under the circumstances. The main objective of the study was to develop a sentiment analysis model for use in public participation forums in County Governments in Kenya. The specific objectives are to; evaluate the difficulty in obtaining sentiments; determine the challenges faced in the design of an effective sentiment analysis model for public participation forums; design a sentiment model for public participation forums in county governments, and; evaluate the performance of sentiment analysis model for public participation forums in county governments. The study was conducted through the design thinking process. The population of interest in this study comprised of county management and staff also area residents in Nakuru, Busia and Baringo counties who have participated in public participation forums before. A sample size of 106 respondents comprising 23 county administrators and 83 residents were purposively sampled for the project. The sentiment analysis model was developed by implementing cloud NLP package and Bidirectional Encoder Representations from Transformers (BERT) algorithm to get magnitudes of user sentiments. Cross validation was then used to evaluate the performance metrics at the design stage and users participated in the evaluation of the model. The overall conclusion of validation is that the model performed as expected and recorded instrumental results in increasing effective public participation in county governments in Kenya and strengthen the devolution process. This study recommends that the model can be cascaded to all the counties in Kenya to improve the efficiency of public participation.