| dc.description.abstract | Over the past decade, vast number of online consumer reviews have made a
significant presence on the Internet. These reviews play a vital role in consumer
awareness about the products and deeply impact the consumer's decision-making
process. On one hand, websites like Amazon, Yelp provide huge collections of crowd-
sourced reviews, which are written by consumers themselves having experience in
using that product. Many researchers argue about the credibility and bias of these
reviews. These factors, coupled with the sheer plethora of reviews for each product,
it can become tiring to form a perspective about the product. On other hand,
websites like Wirecutter, Thesweetsetup provide hand-made highly curated detailed
guides on products across various categories. Although these reviews are unbiased
expert opinions, they require vigorous reporting, interviewing, and testing by various
journalists, scientists, and researchers. Thus making them hard to scale.
Our aim is to study the possible correlations between the crowd-sourced noisy
domain reviews and the curated reviews. We take into account meta-features of re-
views, context-based textual features of reviews and word-embedding based features
of words from reviews. In addition to this, we identify “good reviews", defined as
those noisy domain reviews that align with the curated ones, and use this to propose
a general purpose, extremely streamlined recommender that can provide value to the
general public without any personalized inputs. This research will contribute significantly towards identifying unbiased crowd-sourced reviews that align with curated
reviews, across different categories of products, thereby linking the curated and noisy
domains. Our research will also contribute significantly towards understanding the
intricacies of good product reviews across different categories. | en |
