Survey paper for Topic Models

The following paper provides a very detailed review of topic models. It provides a chronological review with classification of models which is of great help.

Knowledge discovery through directed probabilistic topic models. a Survey

http://www.iiu.edu.pk/wp-content/uploads/downloads/faculties/fbas/cs/ali_daud/Ali_RP/2010-FCS%20Journal-Knowledge%20discovery%20through%20directed%20probabilistic%20topic%20Models.%20A%20Survey.pdf

Another paper related to topic models (hierarchies of topics) would probably be interesting: nCRP

I think this is really necessary, but unfortunately I don't think there is one. Someone should try to do a NIPS or ICML tutorial on topic models, covering thinks like:

1. pachinko allocation,
2. HDP-LDA (which is not formally described by itself in any paper, I think),
3. LDA+ngrams,
4. LDA+hmm,
5. Barzilay & Lee content models (which kind of are like topic models on sentences),
6. held-out-likelihood evaluation (and the "reading tea leaves" evaluation as well),
7. inference techniques (variational as in the original lda paper, collapsed gibbs as in griffiths and steyvers' finding scientific topics, uncollapsed gibbs for completeness, em, collapsed variational as in teh and welling, and [online variational]. (http://www.cs.princeton.edu/~blei/papers/HoffmanBleiBach2010b.pdf) as in hoffman, blei and bach). Some connections have been already explored in "On smoothing and inference in topic models", by Asuncion et al.
8. the interface between topic models and supervised classifiers (blei et al supervised topic models; sauper, haghighi and barzilay "incorporating content structure in text analysis applications")
9. Miscellaneous applications, like summarization, Fei-Fei Li's image work with lda (a sample here), and some interesting applications from the 2009 NIPS workshop, like financial topic models
10. the connection (equivalence) between LDA and multinomial PCA

I hope this list helps you find solutions to your problem; most significant LDA work that I can remember off the top of my head is here, and I welcome additions/expansions (which is why this answer is marked as community wiki, which I've never quite figured out what is).

You might try to look into Blei's tutorial in Cambridge's summer school 2009, it is online in free lectures

I found the following paper to be very enlightening:

G. Heinrich, Parameter estimation for text analysis, Technical report, Fraunhofer IGD, 15 September 2009. PDF

It shows a lot of details from the ground up...

Ivan

Steyvers and Griffiths have a nice tutorial paper on topic models: http://psiexp.ss.uci.edu/research/papers/SteyversGriffithsLSABookFormatted.pdf

I'm not aware of other survey papers but David Mimno has created a nice bibliography on topic models (and I hope it will evolve with time).