Desktop Data Driven Approach to Personalize Query Recommendation

摘要

Query recommendation is an effective method to help users describe their search intentions.In a personalized system,cold-start and the data sparsity were unavoidable,which directly lead to deficient performance of personalizing.As a significant part of a user’s personal information space,a personal computer owns lots of documents relevant to his or her interest.Therefore,desktop data was introduced to construct a user’s preference model.Furthermore,considering the variety of desktop data,relationship between search task and work task was simultaneously exploited to predict a user’s specific information need.Ten volunteers joined experiments to evaluate the potential of desktop data.A series of experiments were conducted and the results proved that desktop data greatly contributed to providing effective personalized reference words.Besides,the results demonstrated that a user’s long-term interest model performed steadier than work task context,but the most valuable words were the top-3 words extracted from the work context.