A Comparative Study of Opinion Summarization Techniques

In the Web 3.0 platforms, enormous amount of information is shared whereby individuals express their thoughts and opinions and learn from others' experiences. Many e-commerce websites provide service of posting opinionated reviews to allow consumers post their opinions using free text. Examples of these e-commerce websites include eBay, Amazon, and Yahoo shopping. Summarizing text is taken as an interesting task of Natural Language Processing (NLP). The proposed work presents a comparative study of different techniques used for opinion summarization. It covers both abstractive and extractive approaches where summary of sentences is achieved by considering aspects. This article highlights the gaps in the previous study by proposing a novel graph-based technique for generating abstractive summary of duplicate sentences. The method discusses the details by constructing graphs, ensuring the sentence correctness using some constraints, and finally scoring the sentences individually by fusing sentiments using SentiWordNet. Extractive approach uses the principle of principal component analysis (PCA). The work includes the application of PCA in summarization of text by reducing the number of dimensions in data (aspects) and relatively finding the summary of the reviews on ranking the most relevant ones, according to the prime aspects without any loss of information respective of a particular domain. The analysis is conducted on the standard Opinosis data set and comparison is made between both of the techniques to discuss which method generates more coherent and complete summary.