Use this URL to cite or link to this record in EThOS:
Title: Relevance Feedback Using Weight Propagation
Author: Yamout, Fadi
ISNI:       0000 0001 3574 2184
Awarding Body: University of Sunderland
Current Institution: University of Sunderland
Date of Award: 2008
Availability of Full Text:
Access from EThOS:
A new relevance feedback technique called Weight Propagation has been developed which provides greater retrieval effectiveness than previously described techniques. Documents judged relevant by the user propagate positive Weights to documents close by in vector similarity space, while documents judged not relevant propagate Negative Weights to such neighbouring documents. Variants of Weight Propagation are also described, namely WPI and WPR (inspired by Ide and Rocchio respectively), and WPY which is the main focus of this thesis, where only the maximum weight propagated to each document is counted. Weight Propagation was further enhanced by introducing a second-order propagation (documents that receive weights themselve!propagate weights to related documents) which increased the precision of the results. WPY is compared against the Rocchio and Ide techniques in the vector model based on the tf.idf weighting scheme, and against the Information-theoretic query expansion technique based on the Kullback-Leibler divergence measure using the DB2 weighting model of the Divergence From Randomness framework. Different RF models were employed such as pseudo relevance feedback in addition to both simulated positive and negative feedback using residual collection technique. The experiments are performed on different test collections such as MED, CISI, Cranfield, LISA, NPL, WTIOG and GOV. Small collections such as MED, CISI, and Cranfield were also tested in the semantic space using Latent Semantic Indexing and the optimal number of dimensions that captures the underlying semantics that exists between the documents is determined for these collections. Retrieval effectiveness is improved since the documents are treated as independent vectors rather than being merged into a single vector as is the case with traditional vector model relevance feedback techniques, or by determining the documents' relevancy based on the lengths of all the documents as with the Kullback-Leibler divergence measure used in traditional probabilistic relevance feedback techniques. In addition, the Weight Propagation technique does not expand terms as in the case with traditional approaches.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available