Contextually Inferred Review Integrity CS 224 W Project Final Report Autumn 2015

Online reviews are an increasingly important resource for both consumer education and business profits. However reviews created by people are prone to bias, inconsistency, and inaccuracy. Current attempts to mitigate this problem have been limited in scope and have primarily involved methods such as analyzing text or user relationships. We will build upon these efforts by using user-business and business-business networks built atop the Yelp review network to more accurately rank the integrity of users’ reviews. We consider factors related to the breadth, consistency, and reputability of a user’s reviews across multiple businesses and the similarities between those businesses. Using this model, we can better rank reviews and draw novel conclusions regarding Yelp’s underlying network structure. Ultimately we conclude that the the context in which a user makes a review (as affected by that user’s other reviews) directly correlates with the integrity with which others perceive that review.