Scalable news recommendation using multi-dimensional similarity and Jaccard–Kmeans clustering

• We propose a Jaccard–Kmeans fast clustering method.
• We propose a multi-dimensional similarity calculation method.
• The methods enhance the scalability of news recommendation system.
• The methods improve the accuracy in condition of data sparsity.
• The methods improve the timeliness of news recommendation.

In order to solve the scalability problem in news recommendation, a scalable news recommendation method is proposed. The method includes the multi-dimensional similarity calculation, the Jaccard–Kmeans fast clustering and the Top-N recommendation. The multi-dimensional similarity calculation method is used to compute the integrated similarity between users, which considers abundant content feature of news, behaviors of users, and the time of these behaviors occurring. Based on traditional K-means algorithm, the Jaccard–Kmeans fast clustering method is proposed. This clustering method first computes the above multi-dimensional similarity, then generates multiple cluster centers with user behavior feature and news content feature, and evaluates the clustering results according to cohesiveness. The Top-N recommendation method integrates a time factor into the final recommendation. Experiment results prove that the proposed method can enhance the scalability of news recommendation, significantly improve the recommendation accuracy in condition of data sparsity, and improve the timeliness of news recommendation.