Text summarization produces a compressed version of the original document by selecting the most important contents. Text summarization is regarded as a generic technique because it did not provide any distribution of opinions and their expressed sentiments. Review summarization provides its user with the description of all product aspects or features with their sentiments or feelings, expressed in reviews which aid the online customer to judge the product or service and helps in decision making process. Review Summarization is considered as difficult task, due to unstructured behavior, review short length, and free style writing. This work proposes an aspect based abstractive summarization of customer reviews using encoder decoder architecture with attentions and pointer generator network. We have used Bi-directional Gated Recurrent Units (Bi-GRU) for encoder-decoder architecture to ensure that the adjoining words have influence on the resultant summaries. We have achieved ROUGE-I with 35.32 score, ROUGE-II with 38.46 score and ROUGE-L with 29.26 score on the Amazon reviews dataset. Empirical results indicate the efficacy and efficiency of our suggested model with respect to the base line systems.
展开▼
