SAABM: A framework for combining sentiment analysis and agent-based modelling for dynamic marketplace analysis

Abstract

The evolution of social media as a digital platform to share consumers experiences of purchasing products has gained a considerable amount of attention from researchers and businesses in recent years. However, the accumulation of large volumes of textual data has presented some challenges in analyzing such unstructured data and capturing complex social phenomena. In light of companies utilizing online platforms to sell their products, the concept of dynamic pricing has been prevalent as a pricing strategy in which the price of a product is continuously adjusted in response to changing market conditions and demand patterns. This element of complexity at which the pace of price fluctuations has evolved over time has resulted in creating advanced models for dynamic marketplace analysis and can be used to understand consumer behaviour and enhance competitiveness for companies. The intersection of dynamic pricing and sentiment analysis presents a unique opportunity to investigate the influence of consumer sentiment on pricing strategies and decision-making processes. This thesis proposes the Sentiment Analysis and Agent-Based Modelling (SAABM) framework which implements a combination approach of sentiment analysis and Agent-Based Modelling (ABM) to model behavioural complexity. Sentiment analysis and topic modelling is applied in this study based on a case study approach of 100 trainers products including Nike, Puma and Timberland, sourced from Amazon UK consumer reviews. Key insights which were extracted from the data include developing wordclouds for positive, negative and neutral sentiment and applying topic modelling to list the top 10 common topics that were being discussed among consumers about a particular trainers product. Correlation analysis was performed to determine whether there is a correlation between sentiment and price which resulted in a positive correlation for 6 Nike products, 11 Puma products and 11 Timberland products. This exploratory data analysis was used to create an agent-based model to observe interactions between consumers and an Amazon UK seller in a simulation environment. By incorporating the Bass model (Bass, 1969), coefficient of innovation and social influence were included to investigate what-if scenarios. Visualizations were created to examine how consumers react in the following what-if scenarios: high sentiment and high price, low sentiment and low price, neutral sentiment and neutral price. In addition, to test the robustness of the model, parameter sweeping was implemented which indicated a faster rate of adoption in a smaller market size, increasing the number of innovators increases the social influence diffusion rate and the higher the coefficient of innovation and social influence values result in a higher rate of adoption. The key contribution of this study is the SAABM framework which is evaluated through the case study approach. One of the findings of the SAABM framework is the integration of sentiment and price variables which influences the adoption threshold of adopting trainers products in a simulated dynamic marketplace environment. Moreover, it provides data-driven insights to be extracted which can aid in data-driven decision making (DDD) to better understand consumer behaviour and to model complex, heterogeneous systems to observe emergent social interactions.