Authors: Tomasz Dryjanski
This paper proposes an alternative to the Paragraph Vector algorithm, generating fixed-length vectors of human-readable features for natural language corpora. It extends word2vec retaining its other advantages like speed and accuracy, hence its proposed name is doc2feat. Extracted features are presented as lists of words with their proximity to the particular feature, allowing interpretation and manual annotation. By parameter tuning focus can be made on grammatical aspects of the corpus language, making it useful for linguistic applications. The algorithm can run on variable-length pieces of texts, and provides insight into what features are relevant for text classification or sentiment analysis. The corpus does not have to, and in specific cases should not be, preprocessed with stemming or stop-words removal.
Comments: 4 Pages.
[v1] 2017-03-07 09:36:39
Unique-IP document downloads: 28 times
Add your own feedback and questions here:
You are equally welcome to be positive or negative about any paper but please be polite. If you are being critical you must mention at least one specific error, otherwise your comment will be deleted as unhelpful.