HAMLET to promote an efficient and precise reuse of shared metadata in dynamic where tags are scarce [1]. J. Gelernter(2008)offers a method of evaluating user fidic= n,d:t,∈d tag preference and the relative strength of social tag LCSH string retrieval performance. Choochaiwattana (2009)examined the use of social annotations to improve the quality of web searches [3]. Li (2008) use the self fdf,=n×log ∑n,°(d:t1∈d organizing characteristics of SoM neural networks to classify the popular tags in"Del icio us"website [101 tfid/, lo (3) IIL. FRAMEWORK FOR SOCIAL TAGGING BASED RESEARCH n t.∈d In this section, we discuss the experimental design and Let ny be the number of occurrences of the considered evaluation method. The experiment was divided into five term in document d, I T I is total number of"Tag Only steps follow to Fig. I documents in the corpus, I TA I is total number of“" Tag and Abstract"documents in the corpus, and I TTA I is total number of"Tag, Title with Abstract"documents in the corpus TTA. d: t,edy is number of documents where the term t, appears(that is ny +0). If the term is not in the corpus, Searchpaper WWw this will lead to a division-by-zero. It is therefore common Cite C. Research Paper Searching Three search engines based on the three indexers were Research paper search developed. Fig. 2 shows an interface of the first search engine. Fig 3 shows an interface of search result. Subjects can see: titled of the document. title name that can link for Crawler ind link for obtaining data from CiteULike Fl Edt wew Farvurtes Toots Hak Index ⊙=0沿P如m合,回,回 Result tser Result Welcome to Research Paper Search Engine Fig 1. A framework for social tagging based research paper searching 20 Resuts Per Page [Search J A. Research Paper Crawling Copyright 2009 Research papers were crawled from CiteULike between March and May 2009. The final set consisted of 62. 192 research papers from 103 research communities related Fig 2 Shows Research Paper Search Engine web page. computer science. To compare a query with research paper indexes, we B. Research Paper Indexing used a cosine similarity measurement to retrieve and rank In our experiment, three different indexers were research paper search results developed. The equation(1),(2), and (3)show a modified D. Experimental Setting Term Frequency/Inverse Document Frequency (tf/idf Fifteen subjects who were lecturers and Ph. D. students formula for the different indexers, where T is"Tag only", from Chulalongkorn University were recruited to be TA is"Title with Abstract, and TTA is"Tag. Title with participants In the experiments, each subject was assigned abstract to find research papers using our search engines. EachHAMLET to promote an efficient and precise reuse of shared metadata in highly dynamic where tags are scarce [1]. J. Gelernter (2008) offers a method of evaluating user tag preference and the relative strength of social tag vs. LCSH string retrieval performance. Choochaiwattana (2009) examined the use of social annotations to improve the quality of web searches [3]. LI (2008) use the self organizing characteristics of SOM neural networks to classify the popular tags in "Del.icio.us" website [10]. III. FRAMEWORK FOR SOCIAL TAGGING BASED RESEARCH PAPER SEARCHING In this section, we discuss the experimental design and evaluation method. The experiment was divided into five steps follow to Fig.1. Fig.1. A framework for social tagging based research paper searching A. Research Paper Crawling Research papers were crawled from CiteULike between March and May 2009. The final set consisted of 62,192 research papers from 103 research communities related to computer science. B. Research Paper Indexing In our experiment, three different indexers were developed. The equation (1), (2), and (3) show a modified Term Frequency/Inverse Document Frequency (tf/idf) formula for the different indexers, where T is “Tag only”, TA is “Title with Abstract”, and TTA is “Tag, Title with Abstract”: ^ ` d t d T n n tfidf k k j i i j i j  u ¦ : log , , , ^ ` d t d TA n n tfidf k k j i i j i j  u ¦ : log , , , ^ ` d t d TTA n n tfidf k k j i i j i j  u ¦ : log , , , Let ni,j be the number of occurrences of the considered term in document dj , | T | is total number of “Tag Only” documents in the corpus, | TA | is total number of “Tag and Abstract” documents in the corpus, and | TTA | is total number of “Tag, Title with Abstract” documents in the corpus TTA. ^ ` d t d : i  is number of documents where the term ti appears (that is nij z 0 ). If the term is not in the corpus, this will lead to a division-by-zero. It is therefore common to use ^ ` d t d 1 : i  . C. Research Paper Searching Three search engines based on the three indexers were developed. Fig. 2 shows an interface of the first search engine. Fig. 3 shows an interface of search result. Subjects can see: titleID of the document, title name that can link for download full paper and link for obtaining data from CiteULike. Fig.2. Shows Research Paper Search Engine web page. To compare a query with research paper indexes, we used a cosine similarity measurement to retrieve and rank research paper search results. D. Experimental Setting Fifteen subjects who were lecturers and Ph.D. students from Chulalongkorn University were recruited to be participants. In the experiments, each subject was assigned to find research papers using our search engines. Each (3) (2) (1) 153