| Summary: | 碩士 === 輔仁大學 === 資訊管理學系碩士班 === 104 === The study aims to analyze users’ search performance and associated behaviors in a refined interface with the aid of the Terrier information retrieval (IR) platform. We used TREC-6 documents and topics provided by Text Retrieval Conference (TREC) as our evaluation dataset. We adopted the language model (Hiemstra_LM) provided by the Terrier IR platform and tested different parameters of the model for retrieving, and then investigating, the effectiveness of the language model (Hiemstra_LM). Furthermore, we applied the LDA (Latent Dirichlet Allocation) clustering method to assist users to conduct search tasks. This research aims to know the effectiveness of the language model by evaluating the precision of the search results. In addition, it also aims to know if grouping documents by the clustering method can help users search efficiently. We recorded users’ search processes to analyze their behaviors.
The evaluation results reveal that when the parameter λ (lambda) of the language model is set to 0.5, it can achieve the best retrieval results. In addition, the interface with the clustering function and term suggestions can improve users’ search performance in terms of precision metric. Moreover, although the difficulty of the tasks seems to influence users’ performance during the search process, those that used the interface with the clustering assistance achieved better performance compared to the users without the clustering function. Our preliminary evaluation results show that the clustering and term suggestion functions can improve the users’ search performance. We will conduct a large-scale experiment in the future to validate the results.
|
|---|
