Diversified recommendation on graphs

摘要: Result diversification has gained a lot of attention as way to answer ambiguous queries and tackle the redundancy problem in results. In last decade, been applied on or integrated into process PageRank- eigenvector-based methods that run various graphs, including social networks, collaboration networks academia, web product co-purchasing graphs. For these applications, is usually addressed bicriteria objective optimization relevance diversity. However, such an approach questionable since query-oblivious algorithm recommends most its results without even considering query may perform best commonly used measures. this paper, we show deficiencies popular evaluation techniques methods, investigate multiple diversity measures understand whether they have any correlations. Next, propose novel measure called expanded which combines both single function order coverage relevant part graph. We also present new greedy BestCoverage, optimizes result set with (1-1/e)-approximation. With rigorous experimentation graphs from proposed method efficient effective for many use cases.