Abstract:
The opinions on online platforms like Amazon, Goibibo, TripAdvisor for products
or services are widely used by customers or users for their decision making in recent
years. The products or services which are highest rated attract maximum attention
of users and are most likely to get purchased. Looking this trend on e-commerce sites,
spammers deceive users intentionally by giving dishonest reviews of products to give
undue promotion for their products and demote the products of their competitors.
The existing state-of-the-art techniques has done behavioral analysis on the features,
graphical analysis on review or reviewer or product relationships, other supervised
learning approaches to identify spam reviews. There is still a lot scope to work on
the temporal and semantically similar behavior among reviews.
This thesis work has been taken to explore the temporal behavior and the se-
mantic similarity of reviews and identify the unusual high deviation patterns. Some
active zones which spammers adopt are identi ed which further depend on average
truthful ratings of the product. Similarity analysis reveals the existence of a simi-
larity range which spam reviews show and can be used to identify reviews as spam
or genuine. Many ways to capture the similarity are tried and checked if this can
help reduce false positives. A hybridization of both these analysis again proves the
existence of this behavior of spammers.
