Much of the human activity defines information context and it can be effectively used for a matching process of context-aware mobile advertising systems. Although prior context-aware mobile advertising systems employed personalized and context sensitive approaches, they have mostly used simply aggregated user-side context information from sensors and have relied on previously prepared inference rules when performing match-making. There exist limitations in identifying more diverse variations that can happen in the real world due to the lack of considerations on daily human lives. As a novel solution, in this paper, we address a natural language processing (NLP) combined Web mining approach. Especially, we propose an object-activity-location-feedback (OALF) model which describes what objects are used for a specific activity in a specific location. Most of all, time variant feedback valences were employed to estimate users' responses to the advertisement triggering entities (objects, activities, locations, and their combinations) in terms of long-term and short-term basis. The model can be realized with a set of web mining procedures including web crawling, data refinement, and sentiment analysis. In addition, we describe how the OALF model can be applied into context-aware mobile advertising and discuss its business model design issues.