Amalgamation of General and Domain Specific Word Embeddings for Improved Hierarchical Aspect Aggregation

摘要

Aspect aggregation or grouping similar aspects is a vital phase in review summarization. Hierarchical aspect aggregation is needed due to the multi-granular nature of aspects. This paper presents a novel approach for hierarchical aspect aggregation employing an amalgamation of domain specific and domain independent word embeddings along with agglomerative clustering to output a multi-granular structure of aspects. We evaluate our output using both internal and external measures. Our results outperform the state of the art approach in multi-granular aspect aggregation.