Which methods can be used to analyze similarities between a reference set and a request set of paragraphs for classification purposes that considers not only word frequencies but also the meaning (i.e., syntax, semantic, …) of the sentences under investigation?
Check the Short Text Similarity task 2012/2013 participants for methods that perform matching beyond keywords:
http://www.aclweb.org/anthology/S12-1051
http://ixa2.si.ehu.es/sts/
https://dynamicecology.wordpress.com/2014/12/15/plagiarism-and-text-matching-software/
maybe you can use recuisive neural network models for sentence modelling.
To take into consideration the meaning of sentences in a classification purpose, you could use:
- Latent semantic analysis for modeling documents followed by any vectorial similarity measure;
- Or WordNet similarity.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts