Presentation is loading. Please wait.

Presentation is loading. Please wait.

Re-ranking Documents Segments To Improve Access To Relevant Content in Information Retrieval Gary Madden Applied Computational Linguistics Dublin City.

Similar presentations


Presentation on theme: "Re-ranking Documents Segments To Improve Access To Relevant Content in Information Retrieval Gary Madden Applied Computational Linguistics Dublin City."— Presentation transcript:

1 Re-ranking Documents Segments To Improve Access To Relevant Content in Information Retrieval Gary Madden Applied Computational Linguistics Dublin City University Gary Madden Applied Computational Linguistics Dublin City University

2 Problems Facing IR Users need to sift through documents to find relevant information Low ranked documents may contain small relevant sections Users need to sift through documents to find relevant information Low ranked documents may contain small relevant sections

3 Project Aim Extract relevant sections of documents Re-rank these segments so the most relevant segments appear at the top of ranked list Extract relevant sections of documents Re-rank these segments so the most relevant segments appear at the top of ranked list

4 Implementation Input is a ranked list from an Information Retrieval system TextTiling segments the documents Segments are re-ranked according to ‘centrality’ Input is a ranked list from an Information Retrieval system TextTiling segments the documents Segments are re-ranked according to ‘centrality’

5 What is Centrality? Similar to PageRank Documents are ranked highly if they render other highly ranked documents well Maximum-Likelihood Estimation used to calculate how well a document renders another Similar to PageRank Documents are ranked highly if they render other highly ranked documents well Maximum-Likelihood Estimation used to calculate how well a document renders another

6 Evaluation TREC-8 data set used Topics 401- 425 were processed Manual relevance assessment TREC-8 data set used Topics 401- 425 were processed Manual relevance assessment

7 Results Output depends on the quality of the input ranked lists The search query is not always the most central theme Output depends on the quality of the input ranked lists The search query is not always the most central theme

8 Further Work Treat the search query as the most highly ranked document segment Many possible applications Treat the search query as the most highly ranked document segment Many possible applications


Download ppt "Re-ranking Documents Segments To Improve Access To Relevant Content in Information Retrieval Gary Madden Applied Computational Linguistics Dublin City."

Similar presentations


Ads by Google