Modeling user behavior using a search-engine

Reviewer: Alvin Chin

Ever since Google came onto the search engine scene with PageRank, and changed the search landscape, researchers have been looking for better ways to improve search. Many times, search results are not the ones that the user is looking for. This paper describes a model of how users behave when interacting with a search engine such as Google, and then implements that model to improve search results. Experiments are conducted between the model and the basic search engine results, to demonstrate the viability of the model. Most search research deals with link analysis, based on how many links a particular Web site has from others. The results are then aggregated. However, there is little or no feedback when the search results are returned that can help to refine the search. The problem is that user feedback is not included as part of the search results analysis. This paper correctly acknowledges the cognitive notion of search, how we as humans perform searches, and how search is personalized on an individual level, which is missing in search research. There could be more background literature cited here. This would strengthen the merits of this paper, and the authors' approach. A more extensive literature search would help the reader to understand the proposed model. More literature needs to be added regarding cognition and its relation to search. The authors talk about eye-tracking evidence, but there should be other references to early research on cognitive search and how people navigate Web pages. The authors note that they use the "ACT-R" cognitive architecture for their model. Even though they do mention the reference where they talk about the ACT-R architecture, it would be beneficial to briefly explain (perhaps in one or two sentences) what exactly it is, to provide context to the model section. In addition, the authors should briefly explain the data that they used from Huberman et al., because, when reading the rest of the results, the reader cannot understand what the data is and what the results refer to. There is some confusion in the "Relevance Topology" subsection of the "Confounding Factors" section. For example, the authors mention that users visit lower-ranked links more frequently. Are these links ordered near the top of the list in the search results page__?__ At the end of this subsection, the last paragraph seems to come out of the blue, and it is difficult to relate it to the results explained immediately above. Finally, in the last section on implications for search interfaces, the authors mention that the development of explicit cognitive models could lead to a number of improvements in user-Web interaction. It would have been helpful to list some of those improvements. Online Computing Reviews Service