Desirability/Rarity-based Ranks in the Prediction of Acceptability as a Presurfacing Step for Products

Artificially intelligent systems simulate humanoid qualities in terms of problem-solving by learning, precluding the need for explicit programming. This mimicry of human behavior can be greatly useful in Business Intelligence owing to its potential of predicting how a user group may react to a particular product. This constitutes a Machine Learning task (T), which improves its predictive performance (P) with the increase in experience (E). This paper attempts to build a User Experience-centric supply rating system to estimate Customer Satisfaction on a scale of one to five by taking into account three factors: (a) the opinions of the users from (b) variegated perspectives of the listings and (c) tangible features thereof. The research starts out by hard, statistical computing in that it initially constructs a ranking system that adjusts weights of five different criteria based on rarity, only after assuring their distinctiveness by a statistical ANOVA. Inferential statistics is further applied to sort attributes affecting the ranks after which a labeled dataset is prepared and divided by clubbing certain sequential ranks. Finally a learning-based soft computing algorithm is applied to recognize the patterns and assign the ratings, with zero rankspecific priors. A Neural Network was employed which could approximate with above 97% accuracy on each of its k-fold cross-validations. Thus the efficacy of the qualitative desirability (also, scarcity)-based ranker was established by a case study on travelogue data, showing great potential to be extended over other businesses as well.