View source for Cohn et al, Advances in Neural Information Processing Systems 2001

From Cohen Courses

← Cohn et al, Advances in Neural Information Processing Systems 2001

You do not have permission to edit this page, for the following reason:

---

You can view and copy the source of this page.

== Citation == Cohn et al. The Missing Link - A Probabilistic Model of Document Content and Hypertext Connectivity. Advances in Neural Information Processing Systems. == Online version == [http://www.cs.cmu.edu/~cohn/papers/nips00.pdf NIPS] == Summary == This [[Category::paper]] presents an interesting approach of jointly modeling the document and link generation. Potential applications include [[AddressesProblem:: Community Detection]] . The basic ideas are: * Use a corpus of articles which have links between them. Examples of such articles are webpages with hyperlinks, scientific articles with citations etc. * Build a [[UsesMethod:: Topic Model]] which could jointly model the documents along with the citations between the documents. Both the words and citations in a document are dependent on the topic proportion present in the document. * Use [[UsesMethod:: Expectation Maximization]] to compute the desired posterior probabilities. == Brief description of the method == The paper describes a method in which the document generation and link generation can be combined by using already known probabilistic version of [[UsesMethod:: Latent semantic indexing]] and [[UsesMethod:: HITS]] algorithm. More specifically both the terms in a document and the links present in the document are generated over a document-specific mixing proportion of factors. For all practical purposes these factors can be considered as topics which are multinomials over the entire vocabulary as in [[UsesMethod:: Latent Dirichlet Allocation]]. The standard method used is to evaluate an expression for the joint likelihood of the corpus and then use [[UsesMethod:: Expectation Maximization]] to compute the topic conditional distribution and the mixing proportions of the document. == Experimental Result == The author used external tasks to verify the usability of the joint model. The first evaluation task was that of classification of web-pages [[UsesDataset:: Web KB dataset]] and abstracts from [[UsesDataset:: Cora network]]. The classification was done using a nearest neighbor method where the proximity was computed using [UsesMethod:: Cosine Similarity]. The joint model shows higher accuracy than either of the model in isolation however, no statistical significance testing was carried out. The second evaluation task was to predict a quantity called reference flow which could be used to predict link between a source and target document. In comparison to a placebo link detector the joint model performs significantly better. == Related papers == An interesting related paper is [[RelatedPaper::Cohn, D. ICML 2000]] which proposes a latent variable model for citation.

Return to Cohn et al, Advances in Neural Information Processing Systems 2001.