predictive.link.probability: Use the RTM to predict whether a link exists between two documents.

Description

This function takes a fitted LDA-type model (e.g., LDA or RTM) and makes predictions about the likelihood of a link existing between pairs of documents.

Usage

predictive.link.probability(edgelist, document_sums, alpha, beta)

Arguments

edgelist

A two-column integer matrix where each row represents an edge on which to make a prediction. An edge is expressed as a pair of integer indices (1-indexed) into the columns (i.e., documents) of document_sums (see below).

document_sums

A $K \times D$ matrix where each entry is a numeric proportional to the probability of seeing a topic (row) conditioned on document (column) (this entry is sometimes denoted $\theta_{d,k}$ in the literature, see details). The document_su

alpha

The value of the Dirichlet hyperparamter generating the distribution over document_sums. This, in effect, smooths the similarity between documents.

beta

A numeric vector of regression weights which is used to determine the similarity between two vectors (see details). Arguments will be recycled to create a vector of length dim(document_sums)[1].

Value

A numeric vector of length dim(edgelist)[1], representing the probability of a link existing between each pair of documents given in the edge list.

Details

Whether or not a link exists between two documents $i$ and $j$ is a function of the weighted inner product of the document_sums[,i] and document_sums[,j]. After normalizing document_sums column-wise, this inner product is weighted by beta.

This quantity is then passed to a link probability function. Like rtm.collapsed.gibbs.sampler in this package, only the exponential link probability function is supported. Note that quantities are automatically scaled to be between 0 and 1.

References

Chang, Jonathan and Blei, David M. Relational Topic Models for Document Networks. Artificial intelligence and statistics. 2009.

Examples

Run this code

## See demo.

demo(rtm)

Run the code above in your browser using DataLab