Enhancing the Accuracy of Recommender System Using Graph Databases

Abstract

In recent years size of the data available to user over online media has increased exponentially. Due to this large amount of data, people face problem in viewing all available data due to lack of time. To overcome such type of problem, the recommender system plays an important role. In recent years, the recommender systems are most popular tool and have been used in a many types of applications such as facebook, twitter, youtube.com, Amazon.com etc. Recommender systems broadly fall into two groups: Collaborative and content based recommender systems. The collaborative-filtering methods generate the recommendation to user on the basis of the nearest user which is most similar to them. User-based Collaborative-filtering (CF) which uses the matrix to store the ratings of the user is the most frequently used recommender technique, widely used because of its simplicity and efficient performance. Although it is extensively used, one of its major problems is that its performance decreases when the user-item matrix becomes sparse. One of the proposed solutions is to the usage of combination of graph data base and Locality Sensitive Hashing (LSH). Graph database provide the flexibility to developer to design database without performing any normalization and LSH provides a faster method to find the nearest neighbor for the recommendation to user. The proposed system concludes with comparison of traditional approach with the proposed approach.