TyCo: A Novel Approach to Collaborative Filtering Recommendation Based on User Typicality
Ajmeera Kiran, Sudesh Rao, Pravin B. Waghmare, Jalari Somasekar, Dharmesh Dhabliya, Ankur Gupta · 2023
Collaborative filtering (CF), which uses user-based and item-based approaches, is a popular methodology for recommender systems. Item-based CF focuses on making suggestions based on strong correlations, whereas user-based CF focuses on discovering comparable user preferences and recommending things to consumers. Currently, the majority of CF approaches employ corated items to compare users. This study suggests a typicality-based technique called TyCo that locates users' “neighbors” by using user typicality in user groups. A user is represented by TyCo by a user typicality vector, which reflects their preference for various goods. It chooses “neighbors” by comparing individuals based on how typical they are, which can get around the drawbacks of conventional collaborative filtering techniques. This is the first piece of work to use typicality to collaborative filtering, and it runs experiments to assess and show off its benefits. The cost of preprocessing steps, which include creating user prototypes and assessing user typicality in user groups, varies based on the clustering approach selected. Simulation result concludes that in case of Movie lens, MAE is 0.732, Precision is 0.820, and recall value is 0.738 when method is TyCo. On other hand MAE is 0.854, Precision is 0.702, recall value is 0.623 when method is UPCC and MAE is 0.863, Precision is 0.716, recall value is 0.612 when method is case of IPCC. Simulation result concludes that in case of Book-Crossing, MAE is 0.828, Precision is 0.782, and recall value is 0.712 when method is TyCo. On other hand MAE is 0.943, Precision is 0.671, recall value is 0.589 when method is UPCC and MAE is 0.958, Precision is 0.683, recall value is 0.586 when method is case of IPCC.