A Multimodal Fusion Framework for Low Resource Game Analytics

Ludo king (LK) is one of the most popularly downloaded games from Appstore and is regarded as low resource game. In this paper, we propose a novel framework to rate LK using multimodal features and ratings derived from user, expert and their fusion. The User feature grid comprises of 18 UX and 4 UI features and serves as input to collect ratings from 40 gamers whose response lies in {-1, 0, +1}. We scrape 9 expert reviews from reliably curated online sources and use these to formulate expert feature grid and ratings. We process user and expert feature grids for distinctive and common features to formulate fusion rating. The proposed Fusion approach is expected to overcome typical limitations such as class imbalance, limited sample size and poor quality samples in user reviews. Simulations on the proposed framework yield following performance related insights for LK: Average UI ratings: 0.14 for User features, 1 for Expert features and 0.80 for Overlap features; Average UX ratings: 0.29 for User features, 0.63 for Expert features and 0.63 for Overlap features; Effective ratings combining UI and UX: 0.26 for User features, 0.70 for Expert features and 0.68 for Overlap features; and finally, the rating is 0.44 for Fusion features. The proposed framework though demonstrated for LK can be made applicable for performance analysis of other e-commerce products.