Determining a consensus opinion on a product sold online is no longer easy, because assessments have become more and more numerous on the Internet. To address this problem, researchers have used various approaches, such as looking for feelings expressed in the documents and exploring the appearance and syntax of reviews. Aspect-based evaluation is the most important aspect of opinion mining, and researchers are becoming more interested in product aspect extraction; however, more complex algorithms are needed to address this issue precisely with large data sets. This paper introduces a method to extract and summarize product aspects and corresponding opinions from a large number of product reviews in a specic domain. We maximize the accuracy and usefulness of the review summaries by leveraging knowledge about product aspect extraction and providing both an appropriate level of detail and rich representation capabilities. The results show that the proposed systemachieves F1-scores of 0.714 for camera reviews and 0.774 for laptop reviews.
We focus on product reviews, which are highly focused on relevant information, such as opinions about a product. As discussed in the terminology, a product consists of a set of components and attributes called aspects. Moreover, the product itself is also an aspect.
For example, a camera includes a set of components (e.g., screen, battery, lens), and a set of properties or attributes (e.g., image quality, options, weight). A screen also has its own set of attributes (e.g., screen quality, screen resolution, screen size).
At this stage, we present a potential application that can use opinion-aspect relationship knowledge, such as in a product aspect inference or sentiment extraction. The system takes a new product review as input and treats the review in the same way as in the first stage. However, instead of per-forming knowledge induction, the system performs opinion extraction. First, it dissects and performs a multidimensional analysis by applying the preprocessing and window extrac-tion process steps.
Second, it uses the opinion-aspect relation knowledge to extract opinions about product aspects using opinion words andmodifiers such as degree, transitional, and negation words. Third, it combines those data to produce an aspect-based summary of the review.
Determining a consensus opinion on a product sold online is no longer easy, because assessments have become more and more numerous on the Internet. To address this problem, researchers have used various approaches, such as looking for feelings expressed in the documents and exploring the appearance and syntax of reviews.
Aspect-based evaluation is the most important aspect of opinion mining, and researchers are becoming more interested in product aspect extraction; however, more complex algorithms are needed to address this issue precisely with large data sets.
Opinionmining at the document-level is themost widely used method for categorizing a whole-opinion review. Sentence-level sentiment analysis focuses on finding sub-jective sentences. In fact, most research has shown a close relationship between sentence- and document-level sentiment analyses. At both the document-level and the sentence-level, estimated opinion values are indirectly related to the topics (i.e., products or aspects of products) expressed in the text.
They are useful, but they are too coarse for most applications because they do not identify the opinion targets. Without knowing what people liked and disliked, sentiment analysis is too narrow to be useful. In contrast, the aspect-based sentiment analyses found in recent sur-veys use more information from the review
This paper addresses product-aspect-extraction-based knowl-edge in product reviews. We introduce a system that works in two main stages: knowledge extraction and sentiment analysis. First, the system automatically extracts broad syntac-tic knowledge and infers opinion-aspect relationships using the DP, CR, and NER NLP tools. The knowledge creation process isolates subtrees, extracts dependency relations, and detects additional annotations, such as co-reference chains, named entity annotations, and syntactic features.
Second, that knowledge is used to analyze new reviews and generate a feature-based summary. Product aspect extraction was performed and achieved satisfactory experimental results, espe-cially for the Laptop domain. This strategy could offer a new approach to product aspect extraction in large datasets and potentially be applied to challenging tasks such as implicit opinion inferences, sarcastic statements, and the opinion-behavior model