high_scalability high_scalability-lda knowledge-graph by maker-knowledge-mining
topicId: wordWeight*wordName (topN-words)
topic #0: 0.004*nuodb + 0.004*berkeley + 0.004*lore + 0.004*pleasestart + 0.004*tachyon + 0.003*django + 0.002*viddler + 0.002*curve + 0.002*science + 0.002*backward
topic #1: 0.007*cloud + 0.006*monitoring + 0.006*servers + 0.006*performance + 0.006*applications + 0.006*web + 0.005*application + 0.005*free + 0.005*java + 0.004*looking
topic #2: 0.005*system + 0.005*load + 0.005*cache + 0.005*database + 0.005*memory + 0.004*mysql + 0.004*architecture + 0.004*data + 0.004*much + 0.004*like
topic #3: 0.005*filters + 0.003*kent + 0.003*subjected + 0.003*synchronization + 0.003*langley + 0.003*absolute + 0.003*xml + 0.003*lew + 0.003*con + 0.002*sun
topic #4: 0.003*resizable + 0.003*oculus + 0.003*netmap + 0.003*tag + 0.003*henderson + 0.003*clustrix + 0.003*cal + 0.002*vr + 0.002*primer + 0.002*heterogeneity
topic #5: 0.004*consists + 0.003*jim + 0.003*moore + 0.003*friendster + 0.003*commutative + 0.003*sensors + 0.003*gigaom + 0.003*phones + 0.003*arrays + 0.003*peg
topic #6: 0.005*slick + 0.004*lease + 0.004*infinisql + 0.004*administrator + 0.003*draws + 0.003*federated + 0.002*paxos + 0.002*storyhere + 0.002*pretty + 0.002*concentrates
topic #7: 0.003*drills + 0.003*zynga + 0.003*activity + 0.002*hyperic + 0.002*typepad + 0.002*mongrel + 0.002*hq + 0.002*todd + 0.002*realtime + 0.002*litespeed
topic #8: 0.005*zookeeper + 0.004*cloudcamp + 0.004*training + 0.003*voltdb + 0.003*dryad + 0.003*resisters + 0.003*fai + 0.003*july + 0.002*supporters + 0.002*fc
topic #9: 0.005*mar + 0.004*vedis + 0.003*reducing + 0.003*percolator + 0.003*pointers + 0.003*abound + 0.003*jurriaan + 0.002*cq + 0.002*flux + 0.002*dataflow
topic #10: 0.007*instances + 0.006*version + 0.006*expensive + 0.006*resource + 0.006*ebs + 0.005*random + 0.005*needed + 0.005*articles + 0.005*instance + 0.004*else
topic #11: 0.004*wikipedia + 0.003*socket + 0.003*masstree + 0.003*continental + 0.002*mituzas + 0.002*gslb + 0.002*bizo + 0.002*amanda + 0.002*zmanda + 0.002*tree
topic #12: 0.005*livejournal + 0.003*french + 0.003*registrar + 0.003*gce + 0.003*glusterfs + 0.002*hp + 0.002*divides + 0.002*servlets + 0.002*screencasts + 0.002*seamless
topic #13: 0.003*parascale + 0.003*sans + 0.003*armstrong + 0.002*fbi + 0.002*downside + 0.002*cuckoo + 0.002*weather + 0.002*dedicate + 0.002*roblox + 0.002*crops
topic #14: 0.003*leaf + 0.003*datacenterknowledge + 0.003*pnfs + 0.003*replacement + 0.003*parent + 0.002*crawling + 0.002*blekko + 0.002*postdoc + 0.002*wit + 0.002*london
topic #15: 0.004*couchconf + 0.004*millwheel + 0.003*terabyte + 0.003*lectures + 0.002*php + 0.002*spread + 0.002*templating + 0.002*dell + 0.002*acquainted + 0.002*wikipedia
topic #16: 0.004*agents + 0.003*monkeys + 0.003*knows + 0.003*batching + 0.002*anyone + 0.002*summarized + 0.002*thereis + 0.002*behind + 0.002*upstream + 0.002*encouraged
topic #17: 0.004*slime + 0.004*accesses + 0.004*insert + 0.004*edges + 0.004*king + 0.003*loading + 0.003*mold + 0.003*training + 0.003*sun + 0.003*pregel
topic #18: 0.004*tunnels + 0.003*slas + 0.003*ria + 0.003*quickbooks + 0.002*dynamo + 0.002*vp + 0.002*contents + 0.002*kettle + 0.002*pixel + 0.001*asolution
topic #19: 0.007*cloudkick + 0.005*membase + 0.005*utilities + 0.005*joyent + 0.005*appirio + 0.005*kabam + 0.005*analyst + 0.004*position + 0.004*icloud + 0.003*vancouver
topic #20: 0.003*sensible + 0.002*squarespace + 0.002*talkon + 0.002*httpd + 0.002*nas + 0.002*stated + 0.002*replies + 0.002*statements + 0.002*aker + 0.002*nov
topic #21: 0.005*forums + 0.003*elastra + 0.003*ldb + 0.003*dispatcher + 0.003*amplify + 0.003*subdivide + 0.003*espn + 0.003*cslid + 0.003*gartner + 0.002*highway
topic #22: 0.005*farmville + 0.003*telegraph + 0.003*olio + 0.003*print + 0.003*gearmand + 0.002*weighted + 0.002*veriscale + 0.002*gearman + 0.002*alot + 0.002*jobs
topic #23: 0.015*stateserver + 0.009*cloudsigma + 0.007*european + 0.006*discover + 0.005*ads + 0.005*subscribe + 0.004*tailored + 0.004*europe + 0.004*cloudera + 0.004*impressions
topic #24: 0.004*func + 0.004*scheduler + 0.003*dido + 0.003*puppet + 0.002*workers + 0.002*airbnb + 0.002*card + 0.002*santiago + 0.002*jobs + 0.002*basecamp
topic #25: 0.005*tokyo + 0.004*repository + 0.003*tags + 0.003*megastore + 0.003*entered + 0.003*enter + 0.003*calculating + 0.003*neither + 0.003*sequences + 0.003*teraflops
topic #26: 0.004*sleep + 0.004*linearly + 0.004*linkedin + 0.004*curve + 0.003*ea + 0.003*exhibit + 0.003*ddd + 0.003*thethe + 0.003*organizational + 0.003*beta
topic #27: 0.005*triggers + 0.004*measuring + 0.003*improving + 0.003*parallelization + 0.003*partially + 0.003*separated + 0.003*metrocluster + 0.003*codes + 0.003*cutting + 0.003*conjecture
topic #28: 0.003*kngine + 0.003*aditya + 0.003*mailinator + 0.003*ddg + 0.003*ml + 0.003*suit + 0.002*parse + 0.002*counters + 0.002*imdb + 0.002*ing
topic #29: 0.004*gemfire + 0.003*hadoopsummit + 0.003*facebook + 0.003*hpc + 0.002*weblog + 0.002*ace + 0.002*duckduckgo + 0.002*faces + 0.002*snabb + 0.002*worthy
topic #30: 0.007*gb + 0.007*image + 0.006*images + 0.005*cached + 0.005*hour + 0.004*nfs + 0.004*book + 0.004*command + 0.004*percent + 0.004*beyond
topic #31: 0.004*jose + 0.003*wuala + 0.003*european + 0.003*wiredrive + 0.003*august + 0.003*cloudsigma + 0.003*trafficked + 0.003*home + 0.003*deviantart + 0.003*whitepaper
topic #32: 0.009*guidelines + 0.004*advertise + 0.003*discounts + 0.003*details + 0.003*cookie + 0.003*contact + 0.003*registrations + 0.002*blue + 0.002*scalability + 0.002*scala
topic #33: 0.003*drupal + 0.003*coherence + 0.003*mistress + 0.002*dmz + 0.002*tigase + 0.002*starters + 0.002*incubator + 0.002*zoosk + 0.002*newbie + 0.002*bangdb
topic #34: 0.006*rackspace + 0.004*front + 0.004*valued + 0.004*dreams + 0.004*usability + 0.003*wackamole + 0.003*udfs + 0.003*admins + 0.003*moment + 0.003*joel
topic #35: 0.003*dynamodb + 0.003*bitcask + 0.002*whirling + 0.002*rum + 0.002*terracotta + 0.002*causal + 0.002*applicaitons + 0.002*fucntionality + 0.002*compact + 0.002*median
topic #36: 0.004*icq + 0.004*quantum + 0.003*parallelism + 0.003*dating + 0.002*imdb + 0.002*edward + 0.002*slide + 0.002*firm + 0.002*capriolo + 0.002*presents
topic #37: 0.005*tungsten + 0.005*netadvantage + 0.004*austin + 0.004*administration + 0.003*outage + 0.003*administering + 0.003*referred + 0.003*kfs + 0.003*pinboard + 0.003*nginx
topic #38: 0.003*timeout + 0.003*frangipani + 0.003*lsi + 0.003*sata + 0.002*federation + 0.002*spike + 0.002*certified + 0.002*mail + 0.002*circles + 0.002*spinlocks
topic #39: 0.005*components + 0.003*raft + 0.003*datasift + 0.003*akiban + 0.003*hypertable + 0.003*hadron + 0.003*green + 0.003*artists + 0.003*headers + 0.003*collider
topic #40: 0.007*processors + 0.006*hashing + 0.006*processor + 0.005*utilization + 0.005*pattern + 0.005*culture + 0.004*tests + 0.004*sizes + 0.004*aware + 0.004*latencies
topic #41: 0.004*closure + 0.003*applicant + 0.003*widsets + 0.002*varnish + 0.002*illusions + 0.002*memcachedb + 0.002*hs + 0.002*sub + 0.002*entity + 0.002*agency
topic #42: 0.006*social + 0.005*dart + 0.003*numa + 0.003*tape + 0.003*dryad + 0.003*detective + 0.002*investigating + 0.002*hurdles + 0.002*walled + 0.002*fluctuating
topic #43: 0.004*dram + 0.003*radical + 0.003*hyder + 0.003*visitor + 0.003*jpa + 0.003*abadi + 0.003*hi + 0.003*mullet + 0.003*memories + 0.003*activity
topic #44: 0.005*breakout + 0.004*lineup + 0.004*training + 0.004*dbshards + 0.003*aconex + 0.003*automates + 0.003*mathworks + 0.003*thrive + 0.003*hyatt + 0.003*regency
topic #45: 0.004*wolfram + 0.003*hibernate + 0.003*alpha + 0.003*adsense + 0.003*avg + 0.002*drupal + 0.002*oprah + 0.002*shards + 0.002*scans + 0.002*striping
topic #46: 0.004*voldemort + 0.003*cmss + 0.003*agnostic + 0.003*joomla + 0.003*gemfire + 0.002*apc + 0.002*gizzard + 0.002*acquisitions + 0.002*adrianco + 0.002*programmed
topic #47: 0.008*shards + 0.007*consistency + 0.006*rules + 0.005*served + 0.005*patterns + 0.005*mail + 0.004*entities + 0.004*decide + 0.004*split + 0.004*task
topic #48: 0.004*eucalyptus + 0.004*gridgain + 0.003*cheat + 0.002*refreshing + 0.002*vacation + 0.002*category + 0.002*title + 0.002*demoted + 0.002*hi + 0.002*posting
topic #49: 0.007*patterns + 0.004*accelerator + 0.004*joyent + 0.004*commerce + 0.004*alternatives + 0.003*premature + 0.003*forrester + 0.003*steven + 0.003*faststats + 0.003*memristors
topic #50: 0.004*linkedin + 0.002*slogans + 0.002*cried + 0.002*parrot + 0.002*cking + 0.002*whoever + 0.002*callaghan + 0.002*webserver + 0.002*ck + 0.002*car
topic #51: 0.008*consensus + 0.005*prefer + 0.005*helped + 0.004*steve + 0.004*entity + 0.004*disqus + 0.004*gpu + 0.004*marketplace + 0.004*photos + 0.004*realtime
topic #52: 0.004*prismatic + 0.003*hipchat + 0.003*puppet + 0.003*academy + 0.003*khan + 0.003*worker + 0.002*mumps + 0.002*publicly + 0.002*concerned + 0.002*yslow
topic #53: 0.004*juju + 0.003*tenzing + 0.003*pyshards + 0.003*zachman + 0.003*widget + 0.003*photos + 0.002*widgets + 0.002*shard + 0.002*moresql + 0.002*charms
topic #54: 0.003*myspace + 0.003*openspaces + 0.003*ramcloud + 0.003*volume + 0.002*conquer + 0.002*tempo + 0.002*contextual + 0.002*cio + 0.002*bigmemory + 0.002*prince
topic #55: 0.003*ringo + 0.003*scalingtip + 0.003*eli + 0.002*wsc + 0.002*oftentimes + 0.002*chapter + 0.002*wade + 0.002*chapters + 0.001*tornado + 0.001*summaryof
topic #56: 0.006*tree + 0.005*tour + 0.005*reporting + 0.005*businesses + 0.004*appliance + 0.004*isp + 0.004*fundamental + 0.004*coming + 0.004*trace + 0.003*puppet
topic #57: 0.005*portability + 0.004*ctl + 0.003*dbshards + 0.003*netcache + 0.003*entity + 0.003*databus + 0.003*reddit + 0.003*vagrant + 0.003*hrd + 0.003*ssh
topic #58: 0.003*fastbit + 0.003*buffer + 0.002*jobs + 0.002*bitmap + 0.002*augment + 0.002*etags + 0.002*gc + 0.002*rsync + 0.002*netty + 0.002*ltrim
topic #59: 0.005*affinity + 0.004*cachingepisode + 0.004*sharepoint + 0.004*kv + 0.004*cuda + 0.003*heap + 0.003*bursting + 0.003*gc + 0.003*hypothetical + 0.003*ml
topic #60: 0.005*buy + 0.005*wapt + 0.004*sqlite + 0.003*wave + 0.003*unlikely + 0.003*tenants + 0.003*puppet + 0.003*scenario + 0.002*sauce + 0.002*stratos
topic #61: 0.007*nosql + 0.006*search + 0.005*users + 0.005*part + 0.005*ram + 0.005*means + 0.004*app + 0.004*store + 0.004*easy + 0.004*framework
topic #62: 0.004*hey + 0.003*emr + 0.002*keystrokes + 0.002*overviews + 0.002*stewart + 0.002*worthwhile + 0.002*clark + 0.002*rage + 0.002*edified + 0.002*yields
topic #63: 0.004*coffee + 0.003*calvin + 0.003*hypertable + 0.003*cohesive + 0.003*ravelry + 0.002*cup + 0.002*placements + 0.002*enlisted + 0.002*cloudworks + 0.002*automattic
topic #64: 0.004*china + 0.004*wire + 0.003*nanoseconds + 0.003*nanosecond + 0.003*microseconds + 0.002*exim + 0.002*gmail + 0.002*hotmail + 0.002*bespoke + 0.002*sprinkle
topic #65: 0.003*dance + 0.003*ashort + 0.003*thumb + 0.003*protocol + 0.002*az + 0.002*reinvent + 0.002*beautiful + 0.002*gray + 0.002*examines + 0.002*presence
topic #66: 0.004*mailbox + 0.003*jesse + 0.002*platforms + 0.002*qps + 0.002*boys + 0.002*bjorn + 0.002*thrudb + 0.002*upgrading + 0.002*neutral + 0.002*median
topic #67: 0.003*disco + 0.003*ives + 0.003*hopper + 0.003*grace + 0.003*lesson + 0.002*symphony + 0.002*peoples + 0.002*crime + 0.002*secured + 0.002*conductors
topic #68: 0.003*causal + 0.003*overflow + 0.002*cops + 0.002*adserver + 0.002*alps + 0.002*gig + 0.002*taste + 0.001*sun + 0.001*hw + 0.001*smp
topic #69: 0.005*gce + 0.003*buddypoke + 0.003*schlossnagle + 0.003*memsql + 0.003*gis + 0.003*pownce + 0.003*gcutil + 0.003*spatially + 0.002*theo + 0.002*ofscalable
topic #70: 0.004*openflow + 0.003*intercloud + 0.003*advertisements + 0.003*sysadmin + 0.003*movable + 0.002*consequently + 0.002*feast + 0.002*supervisor + 0.002*demands + 0.002*lest
topic #71: 0.003*anonymous + 0.002*loaded + 0.002*smarty + 0.002*edit + 0.002*oversell + 0.002*posts + 0.001*military + 0.001*seems + 0.001*though + 0.001*invented
topic #72: 0.005*memsql + 0.003*kos + 0.003*kelly + 0.003*scribe + 0.003*colm + 0.002*followed + 0.002*requirement + 0.002*gavethis + 0.002*bluekai + 0.002*char
topic #73: 0.005*webinar + 0.004*collector + 0.004*berkeley + 0.003*approximately + 0.003*microsystems + 0.003*demonstrate + 0.003*erasure + 0.003*codes + 0.003*ubuntu + 0.003*robustness
topic #74: 0.003*soap + 0.003*xap + 0.003*gigaspaces + 0.003*summaryof + 0.003*spreading + 0.002*ashort + 0.002*pdf + 0.002*rental + 0.002*dough + 0.002*fall
topic #75: 0.004*squarespace + 0.004*tempest + 0.003*lustre + 0.003*peregrine + 0.003*tornado + 0.002*friendfeed + 0.002*conclusions + 0.002*zettabytes + 0.002*celeron + 0.002*alsohosts
topic #76: 0.004*coast + 0.004*denormalization + 0.004*opensolaris + 0.003*ben + 0.003*attend + 0.003*concentrated + 0.003*ef + 0.002*licensing + 0.002*authorization + 0.002*spdy
topic #77: 0.006*jobs + 0.005*task + 0.005*jeff + 0.005*era + 0.005*thousand + 0.005*success + 0.005*magic + 0.004*switches + 0.004*plus + 0.004*nature
topic #78: 0.003*collectl + 0.003*study + 0.002*pathologies + 0.002*tales + 0.002*tyrant + 0.001*tokyo + 0.001*stone + 0.001*conducted + 0.001*races + 0.001*sar
topic #79: 0.009*google + 0.005*platform + 0.005*machines + 0.005*sql + 0.005*hardware + 0.005*network + 0.005*problem + 0.005*would + 0.005*internet + 0.004*data
topic #80: 0.004*mono + 0.004*velocityconf + 0.002*water + 0.002*youporn + 0.002*lake + 0.002*imdg + 0.002*filers + 0.002*overlooked + 0.002*aqueduct + 0.002*aztecs
topic #81: 0.005*playfish + 0.004*mollom + 0.004*gravatar + 0.003*meetups + 0.003*rush + 0.002*arts + 0.002*networked + 0.002*variants + 0.002*cisco + 0.002*geoserver
topic #82: 0.005*condor + 0.003*ning + 0.003*pic + 0.003*behindgoogle + 0.002*acquired + 0.002*sms + 0.002*plant + 0.002*galaxy + 0.002*cgi + 0.002*refers
topic #83: 0.004*dean + 0.004*profiling + 0.003*eucalyptus + 0.002*inventory + 0.002*crdt + 0.002*claims + 0.002*josh + 0.002*ensuring + 0.002*holds + 0.002*established
topic #84: 0.004*lustre + 0.004*slab + 0.004*dryadlinq + 0.003*collectl + 0.003*focussed + 0.003*facelift + 0.002*aggregator + 0.002*yermo + 0.002*drive + 0.002*uploader
topic #85: 0.006*views + 0.005*device + 0.005*modern + 0.005*gets + 0.005*matter + 0.005*python + 0.005*erlang + 0.005*transactional + 0.004*though + 0.004*page
topic #86: 0.005*solaris + 0.003*subdomains + 0.003*doctors + 0.002*cosql + 0.002*nsread + 0.002*sequentially + 0.002*quantitatively + 0.002*monads + 0.002*nssend + 0.001*incorporates
topic #87: 0.005*megapipe + 0.004*jrockit + 0.004*tickets + 0.003*truths + 0.003*randy + 0.002*inputs + 0.002*bidding + 0.002*headed + 0.002*couchconf + 0.002*memo
topic #88: 0.004*okta + 0.004*teaser + 0.003*drbd + 0.003*spanner + 0.002*radio + 0.002*steve + 0.002*wozniak + 0.002*downloads + 0.002*functionally + 0.002*parallelization
topic #89: 0.003*photos + 0.003*websphere + 0.003*amundsen + 0.002*applicaitons + 0.002*scott + 0.002*hero + 0.002*south + 0.002*crawling + 0.002*tailrank + 0.002*pole
topic #90: 0.003*facestat + 0.003*stale + 0.002*keen + 0.002*thelarge + 0.002*neilk + 0.002*financial + 0.002*hiphop + 0.002*watt + 0.002*enomaly + 0.002*globus
topic #91: 0.003*cellular + 0.003*graphchi + 0.003*warehouses + 0.003*collectd + 0.003*authority + 0.003*bcube + 0.003*ramcube + 0.002*explode + 0.002*sometime + 0.002*randy
topic #92: 0.004*propagation + 0.004*ef + 0.003*lbpool + 0.003*ciency + 0.003*shrine + 0.003*lightcloud + 0.003*expressive + 0.002*abstractions + 0.002*additionally + 0.002*concern
topic #93: 0.003*sequencing + 0.003*aload + 0.003*genome + 0.003*providers + 0.003*iis + 0.002*rrdtool + 0.002*ip + 0.002*graphing + 0.002*hello + 0.002*atm
topic #94: 0.007*static + 0.007*bandwidth + 0.006*gae + 0.006*clusters + 0.005*facebook + 0.005*algorithm + 0.005*type + 0.005*ip + 0.005*billion + 0.004*object
topic #95: 0.003*portal + 0.003*nameserver + 0.003*mb + 0.003*dc + 0.002*unlimited + 0.002*grows + 0.002*complicates + 0.002*limit + 0.002*nodejs + 0.002*mind
topic #96: 0.005*instagram + 0.005*deduplication + 0.004*normalized + 0.004*negative + 0.003*newbie + 0.003*artist + 0.003*dathan + 0.003*album + 0.003*hammer + 0.003*pattishall
topic #97: 0.004*capistrano + 0.003*proxy + 0.003*continent + 0.002*sun + 0.002*sdb + 0.002*virginia + 0.002*coolthreads + 0.002*sa + 0.002*inserts + 0.002*sparc
topic #98: 0.004*institute + 0.003*institutes + 0.003*tokyo + 0.002*preferences + 0.002*asia + 0.002*supercomputer + 0.002*entertainment + 0.002*axis + 0.002*petabyte + 0.002*united
topic #99: 0.004*fishin + 0.004*varnish + 0.003*disappointing + 0.003*moves + 0.003*posting + 0.003*colin + 0.003*vacation + 0.003*sprouter + 0.002*botnet + 0.002*approve
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