Semantic Search Guidance: Learn From History

User queries submitted to web search engines are not always informative enough for retrieving the related pages to the user intention. The main problem is that users may not know the best query items they should enter to get the most related web pages to their intentions. They may not be familiar with the specific keywords in that domain knowledge. A user may remember only a part of the phrase that he/she wants to use in the query string. Sometimes the user does not know how to order the keywords (most web search engines are sensitive to the order of the keywords) or even does not know the correct spelling of a specific keyword in the query string. A novice user sometimes sends an imperfect query and scans the returned web pages (even reads a number of the returned documents) to prepare a more precise query by finding new related keywords in the documents. To assist the users in formulating their search queries, we propose two query recommendation algorithms which are based on frequent query phrase similarity, and latent semantic indexing, respectively. The proposed models have been introduced in this paper. They have been evaluated in order to analyze their performance based on the contribution and point-wise mutual information metrics. The algorithms show a promising performance and can serve as a nice starting point for further investigation.