Matches in SemOpenAlex for { <https://semopenalex.org/work/W3083077690> ?p ?o ?g. }
Showing items 1 to 67 of
67
with 100 items per page.
- W3083077690 abstract "The production of high quality ranking results is the main goal of web search engines. An important aspect of modern search engines is the use of a large number of distinct sources of relevance evidence to build the learning to rank (L2R) model. Collectively, they determine whether the document is relevant to a query or not. The ranking of query results is computed by fusing all sources of evidence into a single document score, for each document in the final ranking. In the past few decades, most of the works on evidence fusion has been done with the implementation of L2R methods.L2R methods use examples of queries and their respective results to train supervised learning models that determine the relative position of the documents in the result list. Once trained, the model can be used during query processing to determine the final ranking. This approach, however, inadvertently adds computational costs to query processing, which may lead to a drop in time performance. To mitigate this problem, an alternative approach was proposed in literature — Learn to Precompute Evidence Fusion (LePrEF), based on supervised learning techniques with GP (Genetic Programming). LePrEF proposes to implement the bulk of the evidence fusion during indexing time, generating a single inverted index containing unified entries representing all sources of evidence. These unified entries are called Unified Term Impacts (UTIs). Each unified term impact replaces several features with a single value in the document index, thereby reducing the effort to compute the document scores at query processing time because the system fetches and processes fewer values. The adoption of UTI values produces competitive ranking results. However, the lack of features available only at query time might lead to accuracy loss.In this dissertation we study and propose a modified LambdaMART, named UTI-LambdaMART, a gradient boosting algorithm to generate unified term impacts (UTI) values at indexing time. We also propose and evaluate a hybrid model that uses UTI values with query-dependent features. We demonstrate that our hybrid methods can deliver high-quality results on par with those of the existing state-of-the-art neural ranking models. The experimental results show that our best hybrid model, HLambdaMART, achieves an NDCG@10 value of 0.495 using only 36 features at query processing time when applied to the MQ2007 collection, while the best baseline achieves 0.490 using a larger set of features at query processing time. The use of our hybrid framework reduces the time to run LambdaMART to about 35% of the time to run it without using our proposals. In addition, we study and propose a simple method to obtain significant gains in UTI-index compression with virtually no loss in the quality of search results. Our approach was able to achieve 79% compression rate of the index, while keeping the quality of results on par with methods that do not use compression. We also conduct experiments that demonstrate the use of the UTI-LambdaMART as a base ranker." @default.
- W3083077690 created "2020-09-11" @default.
- W3083077690 creator A5051740073 @default.
- W3083077690 date "2020-07-30" @default.
- W3083077690 modified "2023-09-26" @default.
- W3083077690 title "Applying machine learning to relevance evidence fusion at indexing time" @default.
- W3083077690 hasPublicationYear "2020" @default.
- W3083077690 type Work @default.
- W3083077690 sameAs 3083077690 @default.
- W3083077690 citedByCount "0" @default.
- W3083077690 crossrefType "journal-article" @default.
- W3083077690 hasAuthorship W3083077690A5051740073 @default.
- W3083077690 hasConcept C119857082 @default.
- W3083077690 hasConcept C124101348 @default.
- W3083077690 hasConcept C130590232 @default.
- W3083077690 hasConcept C154945302 @default.
- W3083077690 hasConcept C158154518 @default.
- W3083077690 hasConcept C164120249 @default.
- W3083077690 hasConcept C17744445 @default.
- W3083077690 hasConcept C189430467 @default.
- W3083077690 hasConcept C199539241 @default.
- W3083077690 hasConcept C23123220 @default.
- W3083077690 hasConcept C41008148 @default.
- W3083077690 hasConcept C75165309 @default.
- W3083077690 hasConcept C86037889 @default.
- W3083077690 hasConcept C97854310 @default.
- W3083077690 hasConceptScore W3083077690C119857082 @default.
- W3083077690 hasConceptScore W3083077690C124101348 @default.
- W3083077690 hasConceptScore W3083077690C130590232 @default.
- W3083077690 hasConceptScore W3083077690C154945302 @default.
- W3083077690 hasConceptScore W3083077690C158154518 @default.
- W3083077690 hasConceptScore W3083077690C164120249 @default.
- W3083077690 hasConceptScore W3083077690C17744445 @default.
- W3083077690 hasConceptScore W3083077690C189430467 @default.
- W3083077690 hasConceptScore W3083077690C199539241 @default.
- W3083077690 hasConceptScore W3083077690C23123220 @default.
- W3083077690 hasConceptScore W3083077690C41008148 @default.
- W3083077690 hasConceptScore W3083077690C75165309 @default.
- W3083077690 hasConceptScore W3083077690C86037889 @default.
- W3083077690 hasConceptScore W3083077690C97854310 @default.
- W3083077690 hasLocation W30830776901 @default.
- W3083077690 hasOpenAccess W3083077690 @default.
- W3083077690 hasPrimaryLocation W30830776901 @default.
- W3083077690 hasRelatedWork W118772747 @default.
- W3083077690 hasRelatedWork W1492942844 @default.
- W3083077690 hasRelatedWork W1551074560 @default.
- W3083077690 hasRelatedWork W1551258999 @default.
- W3083077690 hasRelatedWork W155594457 @default.
- W3083077690 hasRelatedWork W2066273153 @default.
- W3083077690 hasRelatedWork W2119163536 @default.
- W3083077690 hasRelatedWork W2121243408 @default.
- W3083077690 hasRelatedWork W2131840460 @default.
- W3083077690 hasRelatedWork W2513946593 @default.
- W3083077690 hasRelatedWork W2589809277 @default.
- W3083077690 hasRelatedWork W2755770860 @default.
- W3083077690 hasRelatedWork W2798611185 @default.
- W3083077690 hasRelatedWork W2807487138 @default.
- W3083077690 hasRelatedWork W2898640137 @default.
- W3083077690 hasRelatedWork W2955375559 @default.
- W3083077690 hasRelatedWork W3152647757 @default.
- W3083077690 hasRelatedWork W3164387052 @default.
- W3083077690 hasRelatedWork W600555174 @default.
- W3083077690 hasRelatedWork W72279380 @default.
- W3083077690 isParatext "false" @default.
- W3083077690 isRetracted "false" @default.
- W3083077690 magId "3083077690" @default.
- W3083077690 workType "article" @default.