How to get topic vector of new documents and compa

问题:

I'm trying to somehow compare a sole document's topic distribution (using LDA) with, other files and their topic distributions within a previously created topic model, using MALLET.

I know that this can be done through MALLET commands in terminal but I'm having problems in finding a way to implement this in Java.

To give a gist of what the functionality of my program is:

The already created topic model was created with a large corpus of texts. I want to use this to compare topic distributions with a tweet that contains a certain hashtag and to then pull out the file most similar to the tweet from the corpus.

Ive read through Mallet's Java API docs but they seem very confusing and not really explanatory.

If anyone could give me a few tips I'd appreciate it

回答1:

First, take a look at these:

• Developer's guide
• Tutorial slides after slide 97
• Code examples in the source directory: src/cc/mallet/examples

Now, these examples show the basic functionality, but they don't show how to save and load the model if you need to separate training from testing. Basically what you need is to save both the model and the instances after training (since you need to train and test with the same pipeline), and load them before testing.

Save model and pipeline after training:

model.write(new File("model.dat"));
instances.save(new File("pipeline.dat"));

Load model and pipeline before testing:

ParallelTopicModel model = ParallelTopicModel.read(new File("model.dat"));
InstanceList instances = InstanceList.load(new File("pipeline.dat"));

Hope this helps.