Effects of different field weights on search performance in digital libraries: A preliminary study

The objective of this research was to examine the effect of major metadata fields on fielded searching in digital libraries. The effectiveness of title, abstract plus full-text was investigated in an IEEE Xplore test bed. Weights on the three components were varied in five schemes (i.e. five runs) to retrieve relevant documents for twenty different search questions in five subject areas. The five weighting schemes were combinations of low, middle (default), and high weight on title, abstract and full-text. The top 50 search results of each search question, retrieved by each run, were evaluated for relevance by five human assessors, each responsible for one subject area. Average Precision (AP) scores as the performance measure were calculated and then compared among these runs. The direct comparisons of APs did not find significant differences among the five runs. However, the results demonstrate a better performance for the run with low weight on title but high weight on both abstract and full-text. This scheme also retrieved the least portion of ephemeral documents. A significant difference was found between the general search question and specific question from one of the runs. This research was a preliminary work towards optimizing weighting assignments among metadata fields and ultimately improving information retrieval performance in digital libraries.