XML Retrieval More Efficient Using Double Scoring Scheme

摘要

This is the first year of Kasetsart University's participation in INEX. We participated in three tracks: the Ad Hoc, the Data Centric, and the Web Service Discovery tracks. In the Ad hoc and Data Centric tracks, the BM25F function has performed well in the past evaluations. However, there is an issue that needs more attention; what is the weight for each field? Previously, the weights are given manually to the fields. In general, many experts feel that uncontrolled selected fields. We proposed an unsupervised implementation of the BM25F scoring function, which we call the Double Scoring function. This scoring function assigns the weight for each field by an extended indexing scheme that handles the tuned weight parameter for each selected field. In the Web Service Discovery track, the standard tokens cannot be utilised directly by IR models, and need to be converted to natural language before indexing. We applied the Capitalization function to solve the tokenisation issue.