Inferring User Search Goals with Weakly Supervised Methodology

Pratima Kadam, Prof.Sandeep B.Vanjale

Abstract Major Challenge Search Engine Face is ambiguity of word sense which gives rise to large subjectâ€™s .search goals differ with user and Examining this goals to retrieve information with relevance is problem scenario in Information retrieval Systems. Ranking algorithms present relevant information orderly and present user search subjects. Relevance can be optimized by inferring search goals. Feedback System is proposed which incorporates user one click to select relevant category of information. Proposed system unsupervisely retrieves information from GSON API and is clustered with enhanced k-means algorithm in supervised fashion building a weakly or partial supervised system mapping user search goals. Research work has been presented from supervised system development to unsupervised system development and further developing best weakly supervised system. Research work has been presented on data extracted log files of commercial search engine available freely. System is been tested completely on web data. System is been evaluated on parameters of precision recall along with VAP(Voted Average precision),MAP(Mean Average Precision) and CAP(cumulative Average precision),a set of feedback tags are been used for manual evaluation of system which is different point of evaluation used in research. Research work has been Presented in Three tasks initially supervised System, and then search log based system and finally softly Supervised System which found to be best. Researches demonstrate better outcomes Keywords: Clustering, Supervised, Weakly Supervised, Search Engine, Information retrieval.

Title:	Inferring User Search Goals with Weakly Supervised Methodology
Author Name:	Pratima Kadam, Prof.Sandeep B.Vanjale
Abstract:	Abstract Major Challenge Search Engine Face is ambiguity of word sense which gives rise to large subjectâ€™s .search goals differ with user and Examining this goals to retrieve information with relevance is problem scenario in Information retrieval Systems. Ranking algorithms present relevant information orderly and present user search subjects. Relevance can be optimized by inferring search goals. Feedback System is proposed which incorporates user one click to select relevant category of information. Proposed system unsupervisely retrieves information from GSON API and is clustered with enhanced k-means algorithm in supervised fashion building a weakly or partial supervised system mapping user search goals. Research work has been presented from supervised system development to unsupervised system development and further developing best weakly supervised system. Research work has been presented on data extracted log files of commercial search engine available freely. System is been tested completely on web data. System is been evaluated on parameters of precision recall along with VAP(Voted Average precision),MAP(Mean Average Precision) and CAP(cumulative Average precision),a set of feedback tags are been used for manual evaluation of system which is different point of evaluation used in research. Research work has been Presented in Three tasks initially supervised System, and then search log based system and finally softly Supervised System which found to be best. Researches demonstrate better outcomes Keywords: Clustering, Supervised, Weakly Supervised, Search Engine, Information retrieval.
Cite this article:	Pratima Kadam, Prof.Sandeep B.Vanjale , " Inferring User Search Goals with Weakly Supervised Methodology " , International Journal of Emerging Trends & Technology in Computer Science (IJETTCS), Volume 5, Issue 2, March - April 2016 , pp. 153-160 , ISSN 2278-6856.
Full Text [PDF] Home