A document representation framework with interpretable features using pre-trained word embeddings

摘要

We propose an improved framework for document representation using word embeddings. The existing models represent the document as a position vector in the same word embedding space. As a result, they are unable to capture the multiple aspects as well as the broad context in the document. Also, due to their low representational power, existing approaches perform poorly at document classification. Furthermore, the document vectors obtained using such methods have uninterpretable features. In this paper, we propose an improved document representation framework which captures multiple aspects of the document with interpretable features. In this framework, a document is represented in a different feature space by representing each dimension with a potential feature word with relatively high discriminating power. A given document is modeled as the distances between the feature words and the document. To represent a document, we have proposed two criteria for the selection of potential feature words and a distance function to measure the distance between the feature word and the document. Experimental results on multiple datasets show that the proposed model consistently performs better at document classification over the baseline methods. The proposed approach is simple and represents the document with interpretable word features. Overall, the proposed model provides an alternative framework to represent the larger text units with word embeddings and provides the scope to develop new approaches to improve the performance of document representation and its applications.