Abstract
Nowadays, smart phones and tablets have been getting more and more popular. Many applications in various mobile platforms can provide location based service to users but seldom take personalized information into consideration which is usually important for some kinds of applications. In order to solve the popular problem of searching a available classroom on campus to do individual study and provide students with personalized services, in this paper, we design and implementation of a context-aware based available classroom searching system-CACS, which can not only use the user's location information but also takes personal preferences into consideration to make this type of applications more convenient and personalized. In this paper, the system model is constructed and some key issues and methods such as weigh based classroom evaluation algorithm are proposed. The prototype system shows that this system can benefit users with good experiences.
Introduction
Finding an available classroom for individual study is a perennial challenge for university students, particularly during peak periods like final exams. While many mobile applications offer Location-Based Services (LBS) to find the "nearest" destination, they often ignore the broader context.
A classroom might be physically close but only available for a short duration before a lecture starts, or it might be located on a high floor when a student prefers the ground level. Traditional systems fail to account for these nuances—available time, room capacity, and personal habits—leading to a frustrating "trial and error" search process. The Context-Aware Classroom searching System (CACS) addresses this by integrating personal preferences and real-time environmental data.
System Architecture and Methodology
The authors propose a system architecture that supports both Client/Server (mobile app) and Browser/Server (web) structures. The core of CACS is a Weight-Based Available Classroom Searching Algorithm.
The system relies on real-time classroom state data retrieved from university databases regarding course schedules and room availability. User history records are also maintained to build and update "Personal Profiles." To handle privacy concerns while respecting the limited processing power of mobile devices, a hybrid storage approach is used: sensitive data is encrypted on the server, while key identifying tokens remain on the client.
The Weight-Based Recommendation Process
The recommendation engine evaluates classrooms based on "Common Factors" (e.g., available time duration, current student density) and "Personal Factors" (e.g., GPS location, floor preference, and building preference).
Each candidate classroom is assigned a score using a linear combination: $$Classroom Marks = \sum_{i=1}^{n} w_i * m_i$$ Where $w_i$ represents the weight of a factor and $m_i$ is the normalized mark for that factor.
A critical feature of CACS is the Feedback Loop. When a user selects a classroom from the recommended top 10 list, the system treats this as a label. It updates the weights ($w_i$) in the user's personal profile, effectively learning that user's preferences over time. This allows the system to distinguish between implicit and explicit preferences, providing a more seamless, automated user experience.
Key Insights and Results
A key insight of this research is that "context" is multi-dimensional, encompassing physical situation (availability), social situation (crowdedness), and historical situation (past preferences). For instance, a classroom's utility is heavily tied to its remaining availability; a room free for only 20 minutes is a poor recommendation for deep study.
Initial prototype testing indicated that CACS successfully generates personalized "Top 10" lists tailored to individual user profiles. Once a room is selected, the system integrates navigation to provide the shortest path to the destination. Qualitative feedback from students suggests that the system is an efficient alternative to manual searching.
Conclusion and Future Work
While the current weight-based model is effective, the authors acknowledge the potential for more robust machine learning algorithms to improve weight update accuracy. Future iterations could also expand the set of contextual factors to include environmental data like ambient noise levels, power outlet availability, or indoor temperature. Additionally, further research is needed to address scalability for large-scale campus deployments.
Summary updated