acl acl2013 acl2013-lda knowledge-graph by maker-knowledge-mining
topicId: wordWeight*wordName (topN-words)
topic #0: 0.009*similarity + 0.006*inference + 0.005*wordnet + 0.005*monolingual + 0.004*prediction + 0.004*text + 0.004*probability + 0.004*parsers + 0.004*measures + 0.004*lexical
topic #1: 0.001*boundary + 0.001*hrg + 0.001*autoencoder + 0.001*rae + 0.001*treewidth + 0.001*combinatory + 0.001*subgraphs + 0.001*bn + 0.001*hyperedge + 0.001*autoencoders
topic #2: 0.004*adjunction + 0.004*tsg + 0.003*cvg + 0.002*plsa + 0.002*fst + 0.002*adjunctions + 0.001*auxiliary + 0.001*eigenmaps + 0.001*kin + 0.001*laplacian
topic #3: 0.002*msd + 0.002*sl + 0.002*sign + 0.002*msds + 0.001*articulator + 0.001*neurons + 0.001*pdlsl + 0.001*romanian + 0.001*layer + 0.001*propositional
topic #4: 0.006*unlabeled + 0.003*ssl + 0.003*embeddings + 0.002*mtr + 0.002*disagreement + 0.002*mukherjee + 0.002*paraphrasing + 0.002*embedding + 0.002*ndcg + 0.002*ranking
topic #5: 0.003*cws + 0.003*similes + 0.002*simplification + 0.002*simile + 0.002*sanskrit + 0.001*cw + 0.001*primitives + 0.001*metaphor + 0.001*figurative + 0.001*schank
topic #6: 0.006*annotation + 0.006*alignment + 0.006*summarization + 0.005*alignments + 0.005*state + 0.005*quality + 0.005*parse + 0.004*summaries + 0.004*annotators + 0.004*summary
topic #7: 0.003*flaw + 0.002*metaphor + 0.002*liwc + 0.002*flaws + 0.001*valence + 0.001*hunalign + 0.001*date + 0.001*farsi + 0.001*cleanup + 0.001*dct
topic #8: 0.001*distant + 0.001*subjectivity + 0.001*decipherment + 0.001*sr + 0.001*passage + 0.001*subjective + 0.001*distantly + 0.001*xf + 0.001*quadratic + 0.001*cut
topic #9: 0.002*itg + 0.001*meetings + 0.001*abstracts + 0.001*slotvp + 0.001*swap + 0.001*mono + 0.001*slotnp + 0.001*ami + 0.001*icsi + 0.001*neubig
topic #10: 0.002*polish + 0.001*kennedy + 0.001*oswald + 0.001*slavonic + 0.001*przepi + 0.001*fuzziness + 0.001*lucene + 0.001*znany + 0.001*malarz + 0.001*zabi
topic #11: 0.008*chinese + 0.006*parser + 0.005*relations + 0.005*relation + 0.005*verb + 0.005*pairs + 0.004*pair + 0.004*algorithm + 0.004*parsing + 0.004*term
topic #12: 0.002*malay + 0.002*katakana + 0.001*iban + 0.001*lookup + 0.001*ws + 0.001*monosemic + 0.001*morin + 0.001*polysemous + 0.001*chiao + 0.001*prochasson
topic #13: 0.002*lvcs + 0.002*causal + 0.002*pj + 0.001*lix + 0.001*hungarian + 0.001*vincze + 0.001*biomedical + 0.001*ppl + 0.001*dsim + 0.001*lvc
topic #14: 0.004*cfg + 0.004*constituent + 0.003*median + 0.003*penn + 0.003*coarse + 0.003*ecs + 0.003*palmer + 0.003*formalism + 0.002*derivations + 0.002*vp
topic #15: 0.005*ranking + 0.004*bias + 0.003*aspects + 0.003*overlap + 0.003*network + 0.002*field + 0.002*subjectivity + 0.002*der + 0.002*influence + 0.002*observations
topic #16: 0.004*segmentation + 0.003*gslda + 0.002*punctuations + 0.002*compressive + 0.002*extractive + 0.002*chinese + 0.001*yue + 0.001*sighan + 0.001*woodsend + 0.001*ctb
topic #17: 0.002*scf + 0.002*scfs + 0.002*dpp + 0.001*caption + 0.001*captions + 0.001*sps + 0.001*berg + 0.001*ordonez + 0.001*visually + 0.001*korhonen
topic #18: 0.003*fv + 0.002*tolerance + 0.001*intolerant + 0.001*edus + 0.001*tipping + 0.001*korhonen + 0.001*tolerant + 0.001*joanis + 0.001*parisien + 0.001*rhetorical
topic #19: 0.003*toponym + 0.002*ostag + 0.002*qa + 0.002*toponyms + 0.002*moschitti + 0.002*keyword + 0.001*eda + 0.001*ptk + 0.001*yeah + 0.001*xx
topic #20: 0.003*affix + 0.002*lattice + 0.002*rating + 0.001*review + 0.001*coreg + 0.001*agglutinative + 0.001*regressor + 0.001*confidently + 0.001*qian + 0.001*uyghur
topic #21: 0.003*zt + 0.002*factoids + 0.002*freebase + 0.002*emotion + 0.002*addressee + 0.002*hoffmann + 0.002*crossword + 0.001*factoid + 0.001*detections + 0.001*kb
topic #22: 0.003*normal + 0.002*grained + 0.002*consonants + 0.001*simplification + 0.001*vowel + 0.001*consonant + 0.001*coarse + 0.001*vowels + 0.001*transfering + 0.001*nasal
topic #23: 0.003*emotion + 0.002*dispersion + 0.002*stacking + 0.002*quest + 0.002*tabular + 0.001*ensemble + 0.001*writer + 0.001*geology + 0.001*devset + 0.001*poon
topic #24: 0.007*bilingual + 0.007*document + 0.006*user + 0.006*segmentation + 0.006*discourse + 0.006*documents + 0.004*lda + 0.004*distribution + 0.004*representation + 0.004*corpora
topic #25: 0.003*causal + 0.002*mtus + 0.002*friendfeed + 0.002*osm + 0.002*tsm + 0.001*reddit + 0.001*urls + 0.001*tsunamis + 0.001*microblogging + 0.001*grammaticality
topic #26: 0.011*sentiment + 0.007*tree + 0.006*tweets + 0.006*feature + 0.006*training + 0.005*languages + 0.005*sentences + 0.005*news + 0.005*social + 0.005*twitter
topic #27: 0.003*edgar + 0.002*icsi + 0.001*bigrams + 0.001*ilp + 0.001*nlu + 0.001*monserrate + 0.001*tac + 0.001*bigram + 0.001*virtual + 0.001*meinedo
topic #28: 0.003*unlabeled + 0.003*orientation + 0.002*community + 0.002*tracking + 0.002*dialog + 0.002*mos + 0.002*deceptive + 0.002*diverse + 0.002*union + 0.002*preference
topic #29: 0.003*td + 0.002*recurrent + 0.002*deceptive + 0.002*factored + 0.002*truthful + 0.002*propminer + 0.002*born + 0.001*rnnlm + 0.001*mikolov + 0.001*flm
topic #30: 0.003*verbnet + 0.002*wsd + 0.002*mace + 0.002*underspecified + 0.002*metonymic + 0.002*semlink + 0.001*inventories + 0.001*vp + 0.001*literal + 0.001*eng
topic #31: 0.003*ucca + 0.003*pivot + 0.003*wordnets + 0.002*lingpy + 0.002*historical + 0.002*cognate + 0.002*wiktionary + 0.001*ccgs + 0.001*scene + 0.001*cc
topic #32: 0.002*supertags + 0.002*ptb + 0.001*sb + 0.001*cd + 0.001*malt + 0.001*mst + 0.000*deepbank + 0.000*ivanova + 0.000*bohnet + 0.000*vrelid
topic #33: 0.002*uncertainty + 0.002*monsters + 0.002*pro + 0.001*svmn + 0.001*gram + 0.001*uncertain + 0.001*szarvas + 0.001*orgy + 0.001*acceptance + 0.001*gy
topic #34: 0.002*duc + 0.002*docsys + 0.002*docbase + 0.002*docorig + 0.001*biography + 0.001*fudannlp + 0.001*generations + 0.001*mds + 0.001*weather + 0.001*cuid
topic #35: 0.006*entity + 0.005*semantic + 0.005*topic + 0.005*query + 0.005*distributional + 0.005*topics + 0.004*proceedings + 0.004*entities + 0.004*grammar + 0.004*models
topic #36: 0.002*icarus + 0.002*quantisation + 0.001*hyperplanes + 0.001*lsh + 0.001*political + 0.001*bits + 0.001*schrodt + 0.001*bit + 0.001*neighbours + 0.001*nearest
topic #37: 0.001*tad + 0.001*ontologies + 0.000*alexe + 0.000*rajasekar + 0.000*krishnamurthy + 0.000*films + 0.000*brave + 0.000*stanoi + 0.000*ioana + 0.000*modules
topic #38: 0.002*connor + 0.001*fes + 0.001*crisis + 0.001*zielinski + 0.001*bollen + 0.001*ritter + 0.001*brendan + 0.001*castillo + 0.001*sakaki + 0.001*chamberlain
topic #39: 0.002*harmony + 0.002*vowels + 0.002*vowel + 0.001*inflection + 0.001*phonmatrix + 0.001*scripts + 0.001*consonants + 0.001*gen + 0.001*vcv + 0.001*diffs
topic #40: 0.003*specification + 0.002*lms + 0.002*relaxation + 0.002*specifications + 0.002*conjuncts + 0.002*reorderings + 0.002*rlt + 0.002*embedding + 0.002*css + 0.002*adnn
topic #41: 0.002*att + 0.002*flat + 0.002*nested + 0.001*businesses + 0.001*iaa + 0.001*dialog + 0.001*recommendation + 0.001*pagerank + 0.001*dialogs + 0.001*mcrae
topic #42: 0.008*dependency + 0.007*phrase + 0.006*sentence + 0.005*reordering + 0.005*clustering + 0.005*decoding + 0.005*translation + 0.005*association + 0.005*source + 0.004*clusters
topic #43: 0.002*padding + 0.001*idle + 0.001*clu + 0.001*bldl + 0.001*dh + 0.001*sagae + 0.001*bldr + 0.000*xh + 0.000*blml + 0.000*tldr
topic #44: 0.002*qsd + 0.002*pet + 0.002*scoping + 0.001*bow + 0.001*burger + 0.001*bootstrapped + 0.001*manshadi + 0.001*durme + 0.001*outscoping + 0.001*scopal
topic #45: 0.004*jokes + 0.003*derivational + 0.003*zv + 0.003*dissect + 0.002*zu + 0.002*compo + 0.002*funny + 0.002*joke + 0.001*lk + 0.001*zk
topic #46: 0.003*gusp + 0.002*lis + 0.002*italian + 0.002*sign + 0.001*oovs + 0.001*flight + 0.001*connectives + 0.001*aewlis + 0.001*cloud + 0.001*atis
topic #47: 0.002*vsem + 0.002*punctuation + 0.002*crawl + 0.002*commoncrawl + 0.001*upp + 0.001*cpp + 0.001*transcribed + 0.001*bovw + 0.001*punctuations + 0.001*ract
topic #48: 0.005*vectors + 0.005*pos + 0.005*arguments + 0.004*predicate + 0.004*words + 0.004*learning + 0.004*vector + 0.004*classes + 0.004*class + 0.004*meaning
topic #49: 0.002*qa + 0.002*mbrar + 0.001*mbr + 0.001*jeopardy + 0.001*answer + 0.000*ghm + 0.000*aar + 0.000*aqx + 0.000*qai + 0.000*passages
topic #50: 0.002*mcdc + 0.001*misclassified + 0.001*pebl + 0.001*signature + 0.001*reuters + 0.001*msrpar + 0.001*elkan + 0.001*svs + 0.001*yk + 0.001*sts
topic #51: 0.003*tm + 0.003*vsm + 0.002*lms + 0.002*neural + 0.001*ntt + 0.001*price + 0.001*semtree + 0.001*stock + 0.001*subcorpora + 0.001*duh
topic #52: 0.003*fcg + 0.002*derivational + 0.002*turkish + 0.002*temporal + 0.002*steels + 0.001*signal + 0.001*fluid + 0.001*typing + 0.001*luc + 0.001*derivationally
topic #53: 0.002*pargram + 0.002*glyph + 0.001*pargrambank + 0.001*bailey + 0.001*lfg + 0.001*ink + 0.001*pixel + 0.001*typesetting + 0.001*tracy + 0.001*pixels
topic #54: 0.003*cohesion + 0.001*seg + 0.001*angeli + 0.001*mira + 0.001*devices + 0.001*week + 0.001*kleine + 0.001*reiteration + 0.001*nil + 0.001*friday
topic #55: 0.003*chronic + 0.002*typesetting + 0.002*critics + 0.002*breaks + 0.002*break + 0.001*critic + 0.001*typeset + 0.001*anticipation + 0.001*anct + 0.001*raggedness
topic #56: 0.002*headline + 0.002*ciphers + 0.002*khapra + 0.002*listener + 0.002*brainsup + 0.002*headlines + 0.002*literal + 0.001*creative + 0.001*heady + 0.001*agents
topic #57: 0.003*dialog + 0.002*paralex + 0.002*factoids + 0.001*qazvinian + 0.001*surveys + 0.001*chelba + 0.001*hik + 0.001*wikianswers + 0.001*radev + 0.001*ngrams
topic #58: 0.003*taboo + 0.002*humor + 0.002*dshmm + 0.002*hk + 0.001*atc + 0.001*humorous + 0.001*dk + 0.001*contextualized + 0.001*clinical + 0.001*cdt
topic #59: 0.003*webanno + 0.002*apposition + 0.002*appositions + 0.001*polls + 0.001*bilinear + 0.001*intention + 0.001*lbp + 0.001*ww + 0.001*brat + 0.001*uu
topic #60: 0.004*rouge + 0.002*deception + 0.002*parma + 0.002*conroy + 0.001*rf + 0.001*tac + 0.001*aesop + 0.001*responsiveness + 0.001*mtc + 0.001*rankel
topic #61: 0.002*pyramid + 0.002*scus + 0.002*scu + 0.002*ordering + 0.001*haji + 0.001*lvc + 0.001*zeman + 0.001*swedish + 0.001*nilsson + 0.001*genetic
topic #62: 0.003*argo + 0.003*metaphor + 0.002*lmf + 0.002*sumo + 0.001*metaphors + 0.001*literal + 0.001*metaphorical + 0.001*rdf + 0.001*indonet + 0.001*ili
topic #63: 0.003*stance + 0.003*debate + 0.002*debates + 0.002*polarities + 0.001*polarity + 0.001*reconstruction + 0.001*coders + 0.001*cj + 0.001*weed + 0.001*wd
topic #64: 0.004*crowdsourcing + 0.004*hypernym + 0.003*hypergraph + 0.003*definitional + 0.003*chang + 0.002*bipartite + 0.002*supervision + 0.002*mooney + 0.002*paths + 0.002*workers
topic #65: 0.002*hamlet + 0.002*reducible + 0.001*dir + 0.001*shakespeare + 0.001*gertrude + 0.001*psdtmopv + 0.001*ch + 0.001*othello + 0.001*psetsotp + 0.001*reducibility
topic #66: 0.004*catenae + 0.002*pivot + 0.002*pivoting + 0.001*persian + 0.001*subjectivity + 0.001*lb + 0.001*subjective + 0.001*libter + 0.001*connectivity + 0.001*bootstrapped
topic #67: 0.005*neural + 0.003*weimar + 0.003*network + 0.002*layer + 0.002*embeddings + 0.002*mnb + 0.002*collobert + 0.002*valency + 0.002*potthast + 0.002*atemporal
topic #68: 0.004*pas + 0.003*annomarket + 0.002*gate + 0.001*cloud + 0.001*cunningham + 0.001*services + 0.001*equivalences + 0.001*paas + 0.001*upload + 0.001*ambiguities
topic #69: 0.003*pal + 0.002*variational + 0.001*chatterbot + 0.001*approximation + 0.001*fellowship + 0.001*chatterbots + 0.001*field + 0.001*ftoiorn + 0.001*counseling + 0.001*bp
topic #70: 0.006*question + 0.005*questions + 0.005*trees + 0.005*graph + 0.004*parsing + 0.004*visual + 0.004*character + 0.004*temporal + 0.004*zhang + 0.004*event
topic #71: 0.003*anaphora + 0.003*ctb + 0.002*charniak + 0.002*bengali + 0.002*guitar + 0.002*pronoun + 0.002*modifications + 0.002*curran + 0.002*hyena + 0.002*essays
topic #72: 0.004*politeness + 0.003*binarization + 0.003*submodular + 0.002*britannica + 0.002*subtrees + 0.001*subtree + 0.001*requests + 0.001*gi + 0.001*irtg + 0.001*compressed
topic #73: 0.003*matcher + 0.001*freebase + 0.001*kite + 0.001*ubl + 0.001*satellite + 0.001*ecological + 0.001*behaviours + 0.001*lextender + 0.001*schema + 0.001*conservation
topic #74: 0.003*morph + 0.002*morphs + 0.001*weibo + 0.001*xilai + 0.001*hsiung + 0.001*censorship + 0.001*uncensored + 0.000*alias + 0.000*buhou + 0.000*chongqing
topic #75: 0.003*linggle + 0.003*docent + 0.002*quotes + 0.002*quote + 0.001*hardmeier + 0.001*beach + 0.001*keefe + 0.001*carbon + 0.001*ovix + 0.001*tax
topic #76: 0.002*elementary + 0.002*mrl + 0.002*sister + 0.002*mrs + 0.002*stock + 0.001*geoquery + 0.001*dpm + 0.001*lag + 0.001*nl + 0.001*kwiatkowski
topic #77: 0.002*cws + 0.001*proceeding + 0.001*namt + 0.001*informal + 0.001*fcrf + 0.001*pinyin + 0.001*levenberg + 0.001*lclr + 0.001*ck + 0.001*itg
topic #78: 0.002*bericht + 0.001*subcategorization + 0.001*mitarbeiter + 0.001*multir + 0.001*npnom + 0.001*inflected + 0.001*npdat + 0.001*kollege + 0.001*npacc + 0.001*kollegen
topic #79: 0.002*empowerment + 0.001*minority + 0.001*competition + 0.001*preference + 0.001*irt + 0.001*ros + 0.001*carolyn + 0.001*penstein + 0.001*trk + 0.001*mayfield
topic #80: 0.003*quantum + 0.002*connotation + 0.002*lemmatisation + 0.002*cutting + 0.002*prc + 0.002*operator + 0.001*qc + 0.001*gramtype + 0.001*assessors + 0.001*nom
topic #81: 0.002*ehdp + 0.002*timeline + 0.001*epoch + 0.001*evolutionary + 0.001*hdp + 0.001*dish + 0.001*epochs + 0.001*yan + 0.001*mtk + 0.001*rouge
topic #82: 0.003*geolocation + 0.002*acoustic + 0.001*facial + 0.001*plausibility + 0.001*modalities + 0.001*video + 0.001*geotagged + 0.001*livetest + 0.001*stacking + 0.001*afp
topic #83: 0.003*numerical + 0.002*realisation + 0.002*thesaurus + 0.001*restaurant + 0.001*sparky + 0.001*moby + 0.001*dethlefs + 0.001*manager + 0.001*vik + 0.001*ferret
topic #84: 0.003*rst + 0.002*sensespotting + 0.002*entailment + 0.002*facet + 0.001*thadani + 0.001*auc + 0.001*rel + 0.001*manli + 0.001*faceted + 0.001*muscles
topic #85: 0.003*smatch + 0.002*travatar + 0.002*tss + 0.002*wcl + 0.001*amr + 0.001*gf + 0.001*wcls + 0.001*sieve + 0.001*reconcile + 0.001*lattices
topic #86: 0.003*caseframes + 0.002*agreements + 0.002*peer + 0.001*subgroups + 0.001*signed + 0.001*caseframe + 0.001*discussant + 0.001*perceptrons + 0.001*discussants + 0.001*summarizers
topic #87: 0.002*weibo + 0.001*collective + 0.001*aggregators + 0.001*spans + 0.001*endriss + 0.001*stymne + 0.001*microblog + 0.001*aggregator + 0.000*parenthetical + 0.000*integrity
topic #88: 0.008*sense + 0.007*classification + 0.007*projection + 0.006*coreference + 0.005*users + 0.005*polarity + 0.005*classifier + 0.004*disambiguation + 0.004*resolution + 0.004*mentions
topic #89: 0.001*ci + 0.001*edus + 0.001*beverage + 0.001*intension + 0.001*hyponymy + 0.001*sandy + 0.001*mcnally + 0.001*hammer + 0.000*taboada + 0.000*dkl
topic #90: 0.006*rules + 0.005*koehn + 0.005*score + 0.004*rule + 0.004*mention + 0.004*moses + 0.004*rank + 0.004*decoder + 0.004*average + 0.003*mean
topic #91: 0.002*dtdp + 0.002*clinical + 0.001*dual + 0.001*learner + 0.001*opposite + 0.001*instantiation + 0.001*suggestion + 0.001*collocation + 0.001*probable + 0.001*mainichi
topic #92: 0.002*excitation + 0.002*nuclei + 0.002*aid + 0.002*hbms + 0.002*cdpmm + 0.001*bbam + 0.001*hbm + 0.001*overgeneralization + 0.001*dod + 0.001*nucleus
topic #93: 0.003*thwarting + 0.003*universals + 0.002*suggestion + 0.001*kj + 0.001*greenberg + 0.001*learner + 0.001*dryer + 0.001*pgm + 0.001*thwarted + 0.001*campbell
topic #94: 0.002*phone + 0.002*telugu + 0.001*tts + 0.001*synthesized + 0.001*phones + 0.001*epsilon + 0.001*polyglot + 0.001*perceptual + 0.001*synthesis + 0.001*phoneset
topic #95: 0.007*translation + 0.007*target + 0.006*smt + 0.005*parallel + 0.005*features + 0.005*bleu + 0.005*translations + 0.004*mt + 0.004*system + 0.004*senses
topic #96: 0.001*testimony + 0.001*weighed + 0.001*broaden + 0.001*visweswariah + 0.000*constituent + 0.000*mccord + 0.000*permutations + 0.000*preordering + 0.000*headwords + 0.000*watson
topic #97: 0.002*mdbn + 0.002*clir + 0.001*equivalency + 0.001*rbm + 0.001*munteanu + 0.001*pruning + 0.001*dbpp + 0.001*alternation + 0.001*plex + 0.001*cells
topic #98: 0.002*amortized + 0.002*subjective + 0.001*amortization + 0.001*eurovoc + 0.001*cognate + 0.001*titov + 0.001*jakob + 0.001*srikumar + 0.001*span + 0.001*klementiev
topic #99: 0.002*gen + 0.002*aggregation + 0.002*impar + 0.002*deps + 0.002*facts + 0.001*gp + 0.001*filler + 0.001*ilpnlg + 0.001*gps + 0.001*owl
Latent Dirichlet allocation
In natural language processing, latent Dirichlet allocation (LDA) is a generative model that allows sets of observations to be explained by unobserved groups that explain why some parts of the data are similar. For example, if observations are words collected into documents, it posits that each document is a mixture of a small number of topics and that each word's creation is attributable to one of the document's topics. LDA is an example of a topic model and was first presented as a graphical model for topic discovery by David Blei, Andrew Ng, and Michael Jordan in 2003.
Topics in LDA
In LDA, each document may be viewed as a mixture of various topics. This is similar to probabilistic latent semantic analysis (pLSA), except that in LDA the topic distribution is assumed to have a Dirichlet prior. In practice, this results in more reasonable mixtures of topics in a document. It has been noted, however, that the pLSA model is equivalent to the LDA model under a uniform Dirichlet prior distribution.
For example, an LDA model might have topics that can be classified as CAT_related and DOG_related. A topic has probabilities of generating various words, such as milk, meow, and kitten, which can be classified and interpreted by the viewer as "CAT_related". Naturally, the word cat itself will have high probability given this topic. The DOG_related topic likewise has probabilities of generating each word: puppy, bark, and bone might have high probability. Words without special relevance, such as the (see function word), will have roughly even probability between classes (or can be placed into a separate category). A topic is not strongly defined, neither semantically nor epistemologically. It is identified on the basis of supervised labeling and (manual) pruning on the basis of their likelihood of co-occurrence. A lexical word may occur in several topics with a different probability, however, with a different typical set of neighboring words in each topic.
Each document is assumed to be characterized by a particular set of topics. This is akin to the standard bag of words model assumption, and makes the individual words exchangeable.
Model
With plate notation, the dependencies among the many variables can be captured concisely. The boxes are “plates” representing replicates. The outer plate represents documents, while the inner plate represents the repeated choice of topics and words within a document. M denotes the number of documents, N the number of words in a document. Thus:
α is the parameter of the Dirichlet prior on the per-document topic distributions,
β is the parameter of the Dirichlet prior on the per-topic word distribution,
\theta_i is the topic distribution for document i,
\phi_k is the word distribution for topic k,
z_{ij} is the topic for the jth word in document i, and
w_{ij} is the specific word.
The w_{ij} are the only observable variables, and the other variables are latent variables. Mostly, the basic LDA model will be extended to a smoothed version to gain better results .[citation needed] The plate notation is shown on the right, where K denotes the number of topics considered in the model and:
\phi is a K*V (V is the dimension of the vocabulary) Markov matrix each row of which denotes the word distribution of a topic.
The generative process behind is that documents are represented as random mixtures over latent topics, where each topic is characterized by a distribution over words. LDA assumes the following generative process for a corpus D consisting of M documents each of length N_i:
1. Choose \theta_i \, \sim \, \mathrm{Dir}(\alpha) , where i \in \{ 1,\dots,M \} and \mathrm{Dir}(\alpha) is the Dirichlet distribution for parameter \alpha
2. Choose \phi_k \, \sim \, \mathrm{Dir}(\beta) , where k \in \{ 1,\dots,K \}
3. For each of the word positions i, j, where j \in \{ 1,\dots,N_i \} , and i \in \{ 1,\dots,M \}
(a) Choose a topic z_{i,j} \,\sim\, \mathrm{Multinomial}(\theta_i).
(b) Choose a word w_{i,j} \,\sim\, \mathrm{Multinomial}( \phi_{z_{i,j}}) .
(Note that the Multinomial distribution here refers to the Multinomial with only one trial. It is formally equivalent to the categorical distribution.)
The lengths N_i are treated as independent of all the other data generating variables (q and z). The subscript is often dropped, as in the plate diagrams shown here.
Inference
Learning the various distributions (the set of topics, their associated word probabilities, the topic of each word, and the particular topic mixture of each document) is a problem of Bayesian inference. The original paper used a variational Bayes approximation of the posterior distribution;[1] alternative inference techniques use Gibbs sampling and expectation propagation.
Following is the derivation of the equations for collapsed Gibbs sampling, which means \varphis and \thetas will be integrated out. For simplicity, in this derivation the documents are all assumed to have the same length N_{}. The derivation is equally valid if the document lengths vary.
Applications, extensions and similar techniques
Topic modeling is a classic problem in information retrieval. Related models and techniques are, among others, latent semantic indexing, independent component analysis, probabilistic latent semantic indexing, non-negative matrix factorization, and Gamma-Poisson distribution.
The LDA model is highly modular and can therefore be easily extended. The main field of interest is modeling relations between topics. This is achieved by using another distribution on the simplex instead of the Dirichlet. The Correlated Topic Model follows this approach, inducing a correlation structure between topics by using the logistic normal distribution instead of the Dirichlet. Another extension is the hierarchical LDA (hLDA), where topics are joined together in a hierarchy by using the nested Chinese restaurant process. LDA can also be extended to a corpus in which a document includes two types of information (e.g., words and names), as in the LDA-dual model. Nonparametric extensions of LDA include the Hierarchical Dirichlet process mixture model, which allows the number of topics to be unbounded and learnt from data and the Nested Chinese Restaurant Process which allows topics to be arranged in a hierarchy whose structure is learnt from data.
As noted earlier, PLSA is similar to LDA. The LDA model is essentially the Bayesian version of PLSA model. The Bayesian formulation tends to perform better on small datasets because Bayesian methods can avoid overfitting the data. In a very large dataset, the results are probably the same. One difference is that PLSA uses a variable d to represent a document in the training set. So in PLSA, when presented with a document the model hasn't seen before, we fix \Pr(w \mid z)—the probability of words under topics—to be that learned from the training set and use the same EM algorithm to infer \Pr(z \mid d)—the topic distribution under d. Blei argues that this step is cheating because you are essentially refitting the model to the new data.
Variations on LDA have been used to automatically put natural images into categories, such as "bedroom" or "forest", by treating an image as a document, and small patches of the image as words; one of the variations is called Spatial Latent Dirichlet Allocation.
from wiki http://en.wikipedia.org/wiki/Latent_Dirichlet_allocation