Item representation is a significant building block of recommender systems (RSs). Most existing methods are based on the inherent features of the item, such as its type, shape, and color. In this paper, we propose a new perspective to characterize items by building ideal user groups (IUGs). We define the IUG as the most suitable user group for an item, such as athletes are the IUG of fitness equipment. The construction of an IUG begins with separating user groups according to every single demographic attribute (gender, occupation, etc.), such as male or female groups by gender attribute. Subsequently, we calculate the average rating of each group based on historical records. The next step is to compare the average ratings of different attribute values for the same attribute (e.g., male and female in gender) of the demographics. Meanwhile , we employ Bayesian averaging to address the issue of certain user groups having fewer members but stronger preferences to achieve a fair comparison. Finally, we combine the attribute values corresponding to the group with the maximum average rating of each attribute to obtain the demographics of an IUG, such as female students in their 20s. To validate the effectiveness of the IUG, we propose an IUG-based neural Collaborative Filtering model. Experiments are taken on two real-world datasets in comparison with nine state-of-the-art methods. Results show that the application of the IUG is effective in terms of two metrics.