Content Models with Attitude

Christina Sauper	Aria Haghighi	Regina Barzilay
[email protected]	[email protected]	[email protected]

Abstract

We present a probabilistic topic model for jointly identifying properties and attributes of social media review snippets. Our model simultaneously learns a set of properties of a product and captures aggregate user sentiments towards these properties. This approach directly enables discovery of highly rated or inconsistent properties of a product. Our model admits an efficient variational mean-field inference algorithm which can be parallelized and run on large snippet collections. We evaluate our model on a large corpus of snippets from Yelp reviews to assess property and attribute prediction. We demonstrate that it outperforms applicable baselines by a considerable margin.

Full Text: http://groups.csail.mit.edu/rbg/code/content_attitude/sauper-acl-11.pdf

Code

This code is available for research use only.

Running

All main code is in variational.rb. To run, you need a configuration file (sample provided in params). Then, simply run with:

ruby variational.rb params

Data

Each line of each data file should contain one snippet, optionally with parts of speech (can be automatically tagged). For example:

Their_PRP$ bread_NN basket_NN was_VBD very_RB good_JJ

Annotation

Sample annotated files for testing are provided in annotation/.

aspect.json
Labels for aspect identification, represented as a clustering over snippets.

words.json
Per-word labels for type of word:

• 0 -- aspect
• 1 -- sentiment
• 2 -- background

Collected via Amazon Mechanical Turk.

sentiment_{train,test}
Annotation of snippet sentiments, positive or negative. All neutral or ambiguous snippets have been removed from this set.