Show simple item record

dc.contributor.authorChen, Sitong
dc.date.accessioned2017-08-15T13:10:07Z
dc.date.available2017-08-15T13:10:07Z
dc.identifier.urihttp://hdl.handle.net/10222/73104
dc.descriptionChoosing an appropriate conference or symposium is the key in publishing a scientific work. Nowadays, the large amount of the call for papers poses a big challenge to researchers who intend to find a suitable venue for their soon-to-be-published papers. Although existing search engines provide convenience for researchers to search relevant call for papers based on the input keyword, those engines retrieving potential documents by keywords matching are kind of incomprehensive and lacking of useful infomation. Besides, retrieval algorithms based on the standard Bag of Word representation is the most direct way, however, they are unable to uncover the semantic meaning of words which is a barrier to enhance retrieval quality. In this thesis, we propose a Call-For-Papers Retrieval System (CRS) which can provide a list of relevant calls for papers given the textual content of a paper as the input query. The core of the system is a binary classification model which is trained by the user feedback collected in a continuous active learning strategy. It allows the user to leverage a term weighting strategy to emphasize key terms of the query. Retrievals are initially made based on a modified Bag of Word (BoW) representation which introduces the \textit{WordNet} Synset to enrich its semantic background. Then a semantic ranking module prioritizes the retrievals using semantic similarity algorithm. The final ranked list is generated by merging the intermediate lists and then displayed to the user for labeling. The system is interactive where the user can submit the feedback to get updated result iteratively or terminate the iteration directly. Our experimental results show that active learning significantly improves the performance of retrieval compared to the search-based retrieval system. The term weighting strategy and semantic similarity further enhance the performance of the system.en_US
dc.description.abstractChoosing an appropriate conference or symposium is the key in publishing a scientific work. We propose a Call-For-Papers Retrieval System (CRS) which provides a list of relevant calls for papers given the textual content of a paper as the input query. The core of the system is a binary classification model trained by the user feedback collected in a continuous active learning strategy. It allows the user to leverage term weighting strategy to emphasize keyterms of the query. Retrievals are initially made based on a modified Bag of Word (BoW) representation which introduces the WordNet Synset to enrich its semantic background. Then a semantic ranking module prioritizes the retrievals using semantic similarity algorithm. The final ranked list is generated by merging the intermediate lists and then displayed to the user for labeling. The user submits the feedback to update result iteratively or terminate the iteration directly.en_US
dc.language.isoen_USen_US
dc.subjectretrieval systemen_US
dc.subjectactive learningen_US
dc.subjectSemantic Similarityen_US
dc.titleCall-For-Papers Retrieval System based on Active Learning and Semantic Similarityen_US
dc.date.defence2017-07-14
dc.contributor.departmentFaculty of Computer Scienceen_US
dc.contributor.degreeMaster of Computer Scienceen_US
dc.contributor.external-examinern/aen_US
dc.contributor.graduate-coordinatorNorbert Zehen_US
dc.contributor.thesis-readerVlado Keseljen_US
dc.contributor.thesis-readerSeyednaser Nourashrafeddinen_US
dc.contributor.thesis-supervisorEvangelos E.Miliosen_US
dc.contributor.manuscriptsNot Applicableen_US
dc.contributor.copyright-releaseNot Applicableen_US
 Find Full text

Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record