Personalized Search Based on Context-Centric Model Mingyang Liu, Shufen Liu, Changhong Hu Dept. College of Computer Science and Technology, Jilin University, Changchun, Jilin, China Email: mingyangliu2012@yeah.net Ramana Reddy, and Sumitra Reddy Department of Computer Science & Electrical Engineering, West Virginia University, Morgantown USA Abstract—With the rapid development of the World Wide Web, huge amount of data has been growing exponentially in our daily life. Users will spend much more time on searching the information they really need than before. Even when they make the exactly same searching input, different users would have various goals. Otherwise, users commonly annotate the information resources or make search query according to their own behaviors. As a matter of fact, this process will bring fuzzy results and be time- consuming. Based on the above problems, we propose our methodology that to combine user’s context, users’ profile with users’ Folksonomies together to optimize personal search. At the end of this paper, we make an experiment to evaluate our methodology and from which we can conclude that our work performs better than other samples. Index Terms—Search, Context, Query, Folksonomy I. INTRODUCTION When it comes to knowledge workers searching for research papers or technique reports, they normally adopt search engines such as Google Scholar [1], Citeulike [2], ProQuest [3] or some other corpus databases. Those technologies are mainly based on keywords or key phrases’ searching strategy, which can’t fully provide accurate answers to users’ query most of the time. Especially for those who have similar interests in the same field and intend to get the similar targeted results, but instead, because of their various behaviors when doing queries’ input. As a matter of fact, they would not get the contented aimed searching results. We name the queries that are origin from different behaviors as Folksonomies. To solve the stated problem in last paragraph, we propose our strategy to construct user’s profiles model, user’s context model and the resource model collaboratively for better constructing personalized search [4] based on Computer Science (Shortly named as CS) domain. The reason that we choose CS is that just for exemplifying our methodology in the chosen domain, which would be suitable for other domains afterwards. The user’s profile model which is more comparatively static and needs to be updated after searching process, and it will work mainly for enriching the referring information for the researching step. And the resources here we mainly referred as knowledge units called JANs [7], that should be not only semantically annotated by ontologies, but also include the elements such as gender, major, preference of papers, etc exist in user ’s profile. Besides, the context within user’s searching session should be modeled, which is comparatively dynamic by following each searching operation. In the traditional Information Retrieval (IR) approaches, which more rely on the rate of match between the input terms and the resources (normally we use the document collections as resources). But the limitations are that the diversity exists in the user’s context, such as users more apt to take the input terms (e.g., acronyms, homonyms, synonyms, folksonomies, etc.) they prefer according to their individual behaviors. Since the Semantic Web emerged [5], these problems have been addressed by taking into account the semantic relation between the terms in user ’s context. Semantic IR approaches are an attempt to go beyond simple term matching by relaxing the strong assumption of term independence and also to cope with term variation in documents/queries [6]. The reason that we adopt context model is aiming at identifying terms within which how to describe the domain concepts in resources or queries. In our work, we present a novel context model combined with user profile model to prepare for better searching results in our CS domain knowledge base. After searching, we will recycle the existing models for building the resource model, which will be three models for smoothly cooperating. That is, we combine the resource (documents, e-mails, etc.), the context (query input) and user profile (expansion based on user’s original profile) to reduce the gap between the user ’s query and the results in the documents base by query context analysis model (shortly named as qCAM), user profile associated model (shortly named as uPAM) and knowledge resource expansion model (shortly named as kREM). About qCAM, the core within which is to get the most relevant information where the current user ’s context is. It mainly depends on certain number of returned results from the knowledge base, then according to the statistical mechanism to extract the keywords as concepts to learn. After this when users do searching process, which will automatically enlarge the user’s input JOURNAL OF NETWORKS, VOL. 8, NO. 7, JULY 2013 1551 © 2013 ACADEMY PUBLISHER doi:10.4304/jnw.8.7.1551-1557