danaxlinux.blogg.se

However, the use of randomness provides advantages as well. 1ĭimitrios Serpanos, Tilman Wolf, in Architecture of Network Systems, 2011 Use of randomizationĭeterministic scheduling algorithms are widespread in all problems where conflict resolution is required because they achieve predictable behavior and lead to low-cost implementations. In the editorial annotation dataset, we treated the grade “Good” and above as relevant for precision-based metrics, and we used discounted cumulative gain (DCG) as an evaluation metric. Definitions of these metrics can be found in standard texts (e.g., ). In click data, we treat all the clicked URLs as relevant and calculate the corresponding Precision at 1 ( ), Precision at 2 ( ), Mean Average Precision at 3 ( ), Mean Average Precision at 4 ( ), and Mean Reciprocal Rank (MRR). To quantitatively compare different ranking algorithms’ retrieval performance, we employed a set of standard evaluation metrics in information retrieval. In addition to compare the models trained on clicks with those trained on editorial judgments, we also used Dong et al.’s freshness-demotion-trained GBRank model as our baseline and denoted it “FreshDem.” Because these two algorithms do not explicitly model relevance and freshness aspects for ranking, we fed them with the concatenation of all our URL relevance/freshness and query features. Since the proposed JRFL model works in a pairwise learning-to-rank manner, we employed two classic pairwise learning-to-rank algorithms, RankSVM and GBRank, as our baseline methods. Bo Long, Yi Chang, in Relevance Ranking for Vertical Search Engines, 2014 2.2.3.5 Baselines and Evaluation Metrics