A Recommendation Framework Based on Three Stages for Social Network
Annotatsiya
During the last few years, the issue of creating efficient recommendation systems has posed a big challenge especially to the upcoming or smaller social networks. The main issues in this regard are lack of information and the sparsity of data, which are the major obstacles that delays the effectiveness of the traditional recommendation models. Smaller networks have low user interaction and thus cannot get enough data to be able to train recommendation algorithms. Also, users usually access a small part of the content, which contributes to increasing the data sparsity issue. In a bid to solve these concerns, a new three-staged recommendation framework has been proposed in this paper to apply on smaller and new social networks. The framework is a combination of anchor links prediction, sentiment analysis, and neural collaborative filtering (NCF) in improving the accuracy of the recommendations. Particularly, the framework takes advantage of sentiment analysis to reduce data scarcity through the use of unigram and bigram sentiment lexicons and anchor link prediction to bridge users in explicit networks such as Twitter, which allows a richer recommendation model. The effectiveness of the presented framework is shown based on the extensive experiments conducted on the real-life datasets, including Twitter and Foursquare, to address the issue of data scarcity and enhance the effectiveness of the recommendation. The results of the experiment indicate that the proposed system is better than the state-of-the-art methods, especially when it comes to such measures as accuracy, nDCG, and hit ratio. The work also presents an encouraging solution to recommendation systems of small and emerging social networks as it makes a considerable contribution to the topic.