Generative Topic Models for Community AnalysisObjectivesOutlineIntroduction to Topic ModelsSlide 5Slide 6Slide 7Slide 8Slide 9Slide 10Slide 11Slide 12Slide 13Slide 14Slide 15Slide 16Slide 17Slide 18Slide 19Hyperlink modeling using PLSAHyperlink modeling using PLSA [Cohn and Hoffman, NIPS, 2001]Hyperlink modeling using PLSA [Cohn and Hoffman, NIPS, 2001]Slide 23Slide 24Hyperlink modeling using LDAHyperlink modeling using LDA [Erosheva, Fienberg, Lafferty, PNAS, 2004]Slide 27Author-Topic Model for Scientific LiteratureAuthor-Topic Model for Scientific Literature [Rozen-Zvi, Griffiths, Steyvers, Smyth UAI, 2004]Author-Topic Model for Scientific Literature [Rozen-Zvi, Griffiths, Steyvers, Smyth UAI, 2004]Slide 31Author-Topic-Recipient model for email data [McCallum, Corrada-Emmanuel,Wang, ICJAI’05]Slide 33Slide 34Slide 35Slide 36Modeling Citation InfluencesModeling Citation Influences [Dietz, Bickel, Scheffer, ICML 2007]Slide 39Slide 40Link-PLSA-LDA: Topic Influence in Blogs (ICWSM 2008)Slide 42Generative Topic Models for Community Analysis Pilfered from: Ramesh Nallapatihttp://www.cs.cmu.edu/~wcohen/10-802/lda-sep-18.ppt2 / 57Objectives•Cultural literacy for ML: –Q: What are “topic models”?–A1: popular indoor sport for machine learning researchers–A2: a particular way of applying unsupervised learning of Bayes nets to text•Quick historical survey of some sample papers in the area3 / 57Outline•Part I: Introduction to Topic Models–Naive Bayes model–Mixture Models•Expectation Maximization–PLSA–LDA•Variational EM•Gibbs Sampling•Part II: Topic Models for Community Analysis–Citation modeling with PLSA–Citation Modeling with LDA–Author Topic Model–Author Topic Recipient Model–Modeling influence of Citations–Mixed membership Stochastic Block Model4 / 57Introduction to Topic Models•Multinomial Naïve BayesCW1W2W3…..WNM• For each document d = 1,, M• Generate Cd ~ Mult( ¢ | )• For each position n = 1,, Nd• Generate wn ~ Mult(¢|,Cd)5 / 57Introduction to Topic Models•Naïve Bayes Model: Compact representationCW1W2W3…..WNCWNMM6 / 57Introduction to Topic Models•Mixture model: unsupervised naïve Bayes modelCWNM• Joint probability of words and classes:• But classes are not visible:Z7 / 57Introduction to Topic Models8 / 57Introduction to Topic Models•Probabilistic Latent Semantic Analysis ModeldzwM• Select document d ~ Mult()• For each position n = 1,, Nd• generate zn ~ Mult( ¢ | d)• generate wn ~ Mult( ¢ | zn)dNTopic distribution9 / 57Introduction to Topic Models•Probabilistic Latent Semantic Analysis Model–Learning using EM–Not a complete generative model •Has a distribution over the training set of documents: no new document can be generated!–Nevertheless, more realistic than mixture model•Documents can discuss multiple topics!10 / 57Introduction to Topic Models•PLSA topics (TDT-1 corpus)11 / 57Introduction to Topic Models12 / 57Introduction to Topic Models•Latent Dirichlet AllocationzwMN• For each document d = 1,,M• Generate d ~ Dir(¢ | )• For each position n = 1,, Nd• generate zn ~ Mult( ¢ | d)• generate wn ~ Mult( ¢ | zn)13 / 57Introduction to Topic Models•Latent Dirichlet Allocation–Overcomes the issues with PLSA•Can generate any random document–Parameter learning:•Variational EM–Numerical approximation using lower-bounds–Results in biased solutions–Convergence has numerical guarantees•Gibbs Sampling –Stochastic simulation–unbiased solutions–Stochastic convergence14 / 57Introduction to Topic Models•Variational EM for LDA–Approximate the posterior by a simpler distribution• A convex function in each parameter!15 / 57Introduction to Topic Models•Gibbs sampling–Applicable when joint distribution is hard to evaluate but conditional distribution is known–Sequence of samples comprises a Markov Chain–Stationary distribution of the chain is the joint distribution16 / 57Introduction to Topic Models•LDA topics17 / 57Introduction to Topic Models•LDA’s view of a document18 / 57Introduction to Topic Models•Perplexity comparison of various models UnigramMixture modelPLSALDALower is better19 / 57Outline•Part I: Introduction to Topic Models–Naive Bayes model–Mixture Models•Expectation Maximization–PLSA–LDA•Variational EM•Gibbs Sampling•Part II: Topic Models for Community Analysis–Citation modeling with PLSA–Citation Modeling with LDA–Author Topic Model–Author Topic Recipient Model–Modeling influence of Citations–Mixed membership Stochastic Block Model20 / 57Hyperlink modeling using PLSA21 / 57Hyperlink modeling using PLSA[Cohn and Hoffman, NIPS, 2001] dzwMdNzc• Select document d ~ Mult()• For each position n = 1,, Nd• generate zn ~ Mult( ¢ | d)• generate wn ~ Mult( ¢ | zn)• For each citation j = 1,, Ld • generate zj ~ Mult( ¢ | d)• generate cj ~ Mult( ¢ | zj)L22 / 57Hyperlink modeling using PLSA[Cohn and Hoffman, NIPS, 2001]dzwMdNzcLPLSA likelihood:New likelihood: Learning using EM23 / 57Hyperlink modeling using PLSA[Cohn and Hoffman, NIPS, 2001]Heuristic: 0 · · 1 determines the relative importance of content and hyperlinks (1-)24 / 57Hyperlink modeling using PLSA[Cohn and Hoffman, NIPS, 2001]•Classification performanceHyperlink contentHyperlinkcontent25 / 57Hyperlink modeling using LDA26 / 57Hyperlink modeling using LDA[Erosheva, Fienberg, Lafferty, PNAS, 2004]zwMN• For each document d = 1, ,M• Generate d ~ Dir(¢ | )• For each position n = 1,, Nd• generate zn ~ Mult( ¢ | d)• generate wn ~ Mult( ¢ | zn)•For each citation j = 1,, Ld • generate zj ~ Mult( . | d)• generate cj ~ Mult( . | zj)zcLLearning using variational EM27 / 57Hyperlink modeling using LDA[Erosheva, Fienberg, Lafferty, PNAS, 2004]28 / 57Author-Topic Model for Scientific Literature29 / 57Author-Topic Model for Scientific Literature[Rozen-Zvi, Griffiths, Steyvers, Smyth UAI, 2004]zwMN• For each author a = 1,,A• Generate a ~ Dir(¢ | )• For each topic k = 1,,K• Generate k ~ Dir( ¢ | )•For each document d = 1,,M• For each position n = 1,, Nd•Generate author x ~ Unif(¢ | ad)• generate zn ~ Mult( ¢ | a)• generate wn ~ Mult( ¢ | zn)xaAPK30 / 57Author-Topic Model for Scientific
View Full Document