A Latent Dirichlet Allocation Model for Entity Resolution Export

@techreport{citeulike:1465209, author = {Bhattacharya, Indrajit and Getoor, Lise}, citeulike-article-id = {1465209}, citeulike-linkout-0 = {http://www.cs.umd.edu/~indrajit/DOCS/cs-tr-4740.pdf}, comment = {Uses a lda-like generative model, where \# a reference (mention) is generated by a Noise model perturbing the author's correct name -- p(reference | authornames, authorentity, noise model) \# for each reference an author entity a is generated from a author-mixture of frequently co-authoring groups z -- p(a | group z, author-mixture phi, dirichlet parameter beta) \# for each reference an latent group is generated from the document's groupmixture -- p(z|d, groupmixture theta) \# for each document the groupmixture theta is sampled from a dirichlet with prior alpha The noise model incorporates for each of the tokens in a name reference \# dropping the token \# retaining the token (and perturbing it with string edits insert/delete/replace) \# intaling the token (using only the first letter) The parameters for the noise model are updated every X iterations. ---=note-separator=--- The conference version is here: http://www.citeulike.org/user/ldietz/article/1467629}, day = {1}, keywords = {authorattribution, authorcoreference, entitytopicmodels}, month = {May}, posted-at = {2007-07-18 17:02:10}, priority = {0}, title = {A Latent Dirichlet Allocation Model for Entity Resolution}, url = {http://www.cs.umd.edu/~indrajit/DOCS/cs-tr-4740.pdf}, year = {2005} }

TY - RPRT ID - citeulike:1465209 L3 - citeulike-article-id:1465209 TI - A Latent Dirichlet Allocation Model for Entity Resolution KW - authorattribution KW - authorcoreference KW - entitytopicmodels N1 - Uses a lda-like generative model, where # a reference (mention) is generated by a Noise model perturbing the author's correct name -- p(reference | authornames, authorentity, noise model) # for each reference an author entity a is generated from a author-mixture of frequently co-authoring groups z -- p(a | group z, author-mixture phi, dirichlet parameter beta) # for each reference an latent group is generated from the document's groupmixture -- p(z|d, groupmixture theta) # for each document the groupmixture theta is sampled from a dirichlet with prior alpha The noise model incorporates for each of the tokens in a name reference # dropping the token # retaining the token (and perturbing it with string edits insert/delete/replace) # intaling the token (using only the first letter) The parameters for the noise model are updated every X iterations. N1 - The conference version is here: http://www.citeulike.org/user/ldietz/article/1467629 AU - Bhattacharya, Indrajit AU - Getoor, Lise PY - 2005/05/01/ UR - http://www.cs.umd.edu/~indrajit/DOCS/cs-tr-4740.pdf ER -