A Mobile Context-Aware Recommendation System Based on DBpedia

Abstract

Context-aware recommendation systems generate more relevant recommendations by adapting them to the specific contextual situation of the user. They have been widely used for tourism to recommend locations relevant to the user's needs. A plenty of efforts have presented different approaches to capture the user's needs and contextual information, and then suggest tourist locations that fulfill the user's requirements. These approaches, however, often rely on conventional mappings services, like Google maps, to identify locations and their corresponding categories (e.g. restaurants, libraries, shopping, etc). For example, if the user is looking for museums, the recommendation system will refer to the locations' database and categorization offered by the mapping service. Unless the mapping service provides sufficient information about locations on the map, recommendation approaches will not be able to identify target locations. In this work, we proposed a context-aware recommendation approach that leverages Linked Open Data (LOD), and DBpedia in particular, to provide recommendations that are semantically related to the user's needs. The proposed approach prompts the user to input a search query (i.e. keywords to search for locations of interest). Then, it will use the user's GPS location and keywords to query DBpedia for locations that best match the user's interests. The proposed approach has three contributions: First, it can offer recommendations of geographical locations not covered by traditional mapping services. Our approach does not use the location's database of the mapping service. Instead, it seeks to extract location's details by directly querying DBpedia. Second, it uses the search keywords submitted by the user to search DBpedia for locations that best match the user's needs. Third, it presents an algorithm that ranks the recommendations in a way that balances between proximity (i.e. distance to the user) and relevance to the user's interests. A prototype mobile application was also developed to demonstrate the use of the proposed recommendation approach. The performance and efficiency of the proposed approach was assessed by using a dataset of 41 queries covering three cities. We used two evaluation metrics to assess the performance and the ranking of results. The system achieved 77.77% using the MAP metric with SD of 0.029, and 91.377% Normalized Discount Cumulative Gain with SD 0.337, and average of 6.27 second in query execution with 2.5 SD. The archived results as shown in the previous values indicates that nDCG value was greater than the MAP value. This result proves that the system achieved better results in ranking than in generating accurate results. Keywords: Recommendation System, semantic search, user context, LOD, DBpedia.