In wireless caching networks, users' content request behavior is a compelling aspect for maximizing the achievable revenue. Multinomial logit (MNL) choice model is commonly used to characterize the relationship between users' request behavior and the assortment decision. However, conventional MNL model assumes that the systems show an assortment of content items to users, and a user can purchase the items among the assortment or leave without consuming anything. Yet in most cases, users tend to observe the assorted items, and select within the list in accordance with their personal preference or just search for the items they are interested in by closing the assortment set directly. To address this issue, a revised MNL (RevMNL) choice model is proposed in this paper, wherein we presume that all the remaining items will be shown to the user if the pre-determined assortment set is unsatisfactory. Under which, we mathematically derive the content demanding probability distribution per user. Thereafter, the assortment decision-making problem is studied to maximize system's revenue, which is a non-convex integer programming problem. By using structure-oriented geometric properties, we design an iterative algorithm with quadratic time complexity to obtain the globally optimal solution to the formulated optimization problem. Extensive simulation results validate the superiority of our devised scheme in terms of system revenue and cache hit ratio when compared against various baselines under the conventional MNL model.