Improving Relevance of Information Retrieval Systems and User’s Preferred Search Language
Ghada Refaat El Said · Information Technology Journal · 2017
Background and Objective: Enhancing relevance of search results is becoming a crucial challenge for search engines.Collection of implicit and explicit feedback as indicators of search result relevancy is currently a growing interest in information systems research.While there is some evidence that individual differences affect the effectiveness of implicit indicators, little work investigated the potential effect of userʼs preferred search language in such context.The current study pioneered investigation of this effect on the relationship between a number of implicit indicators (dwell time, number of clicks and amount of scrolling) and user explicit rating. Materials and Methods:The experiment included 48 Arabic users divided in two groups, where only one group was given the option to select a preferred search language.The relationship between implicit indicators and explicit relevance ratings was examined using Pearson correlation.Significance testing was employed to ensure results from Pearson correlation are not random, where a confidence interval of 95% and a statistical significant coefficient, p<0.05, is accepted.To provide more confidence to the obtained results, t-value and standard deviation were calculated in the two groups.Significant differences between (number of clicks, amount of scrolling, dwell time) and relevance were (t = 4.32, t = 5.11, t = 2.62; p<0.0005) respectively.Results: Results suggested that the prediction level of implicit feedback for result relevance is enhanced when users are given the option to select a preferred language.The results also show that both the amount of scrolling and number of mouse clicks have higher precision with post-retrieval document relevancy compared to dwell time.Finally, the present study suggests that the prediction performance of dwell time varies from factual to intellectual task type. Conclusion:The current study provides a cost-effective method for understanding user behavior in the context of different languages through the use of implicit feedback.Findings of this study can be used to enhance the degree of result relevance for search-based recommender systems.