Can you suggest me some papers to read about deep learning models that find patterns/similarities between different texts?

What I have is a set of reviews with the following categories for each review: Rating, Review title, Review body, date, and helpful votes. What I would like to do is understand whether there exist similarities among the reviews. For instance, there exists a cluster of customers that complains about a specific aspect of a product. Or, as another example, to see if there exists a problem related to the season e.g. a product has many bad reviews in summer because it does not work well with high temperatures.

Thank you for your help.

  • $\begingroup$ Imho this is not feasible as such (I might be wrong). Because basically this would mean an unsupervised way to focus on a specific kind of similarity which is not predefined, and I think there are too many unknowns in this problem. You and I have some understanding of what kind of semantic information is intereesting, but the unsupervised model doesn't: for instanceit might consider it a similarity that a group of reviews contain the word 'right' or contain N times the letter 'x', or contain exactly 274 characters... How would it know what is a relevant similarity? $\endgroup$
    – Erwan
    Commented Dec 8, 2022 at 14:21
  • 1
    $\begingroup$ You may be interested in the Aspect Based Sentiment Analysis literature. I think there are approaches for automatic detection of Aspects, probably a pre-trained transformer. $\endgroup$
    – 20-roso
    Commented Dec 9, 2022 at 13:26

1 Answer 1


The problem is called by several different names: multi-aspect, semi-supervised, or hierarchical topic modeling.

One way to approach the problem is to start with a seed or anchor words (e.g., "winter" and "summer") and form clusters around those words.

It can be solved with a variety of methods, the most commonly used methods are variations of Latent Dirichlet Allocation (LDA).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.