ONLINE REVIEW ASSESSMENT USING MULTIPLE SOURCES

• US 20160070709A1
• Filed: 09/09/2015
• Published: 03/10/2016
• Est. Priority Date: 09/09/2014
• Status: Active Grant

• Current Assignee: UNM Rainforest Innovations (f/k/a STC.UNM) (The University of New Mexico)
• Original Assignee: UNM Rainforest Innovations (f/k/a STC.UNM) (The University of New Mexico)
• Inventors: Luan, Shuang, Faloutsos, Michalis, Mueen, Abdullah, Minnich, Amanda J.
• Granted Patent: US 10,089,660 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06Q 30/0282 Rating or review of business operators or products

Abstract

Multiple sources of reviews for the same product or service (e.g. hotels, restaurants, clinics, hair saloon, etc.) are utilized to provide a trustworthiness score. Such a score can clearly identify hotels with evidence of review manipulation, omission and fakery and provide the user with a comprehensive understanding of the reviews of a product or establishment. Three types of information are used in computing the score: spatial, temporal and network or graph-based. The information is blended to produce a representative set of features that can reliably produce the trustworthiness score. The invention is self-adapting to new reviews and sites. The invention also includes a validation mechanism by crowd-sourcing and fake review generation to ensure reliability and trustworthiness of the scoring.

Claims

1. A method for assessing the trustworthiness of a plurality of reviews directed to a provider, comprising the steps:

generating a database of reviewer-centric, provider-centric, and review-centric features, each of the features capturing temporal, spatial, contextual and graphical characteristics from two or more different review hosting sites;

determining outlier scores for the provider based on the features;

normalizing the outlier scores to conform to a scale;

producing an outlier probability value;

converting the outlier probability value to a trustworthiness score indicating a level of suspiciousness of the provider; and

outputting the trustworthiness score.

2. The method according to claim 1 further comprising the steps of:

averaging two or more trustworthiness scores;

producing an overall trustworthiness score indicating a level of suspiciousness activity of the provider; and

outputting the overall trustworthiness score.

3. The method according to claim 1, wherein the reviewer-centric features are based on inconsistent and unlikely reviews and comprise a maximum number of reviews within a particular location, a maximum number of reviews within a particular time period, and a maximum number of reviews in common with each other based on the provider reviewed, dates for each review, and a clique size.

4. The method according to claim 1, wherein the provider-centric features comprise one or more selected from the group of:

bursts, temporal bursts, oscillations, correlation of ratings between hosting sites, and correlation within the same hosting site.

5. The method according to claim 1, wherein the review-centric features comprise count of empty reviews and two or more reviews with a proportion of same text.

6. The method according to claim 1, wherein the outlier scores are determined by using one or more approaches selected from the group comprising:

global density-based approach, a local outlier factor approach, and a hierarchical cluster-based approach.

7. The method according to claim 6, wherein the global density-based approach computes an outlier score by locating one or more points outside a core connected cluster.

8. The method according to claim 6, wherein the local outlier factor approach computes an outlier score by averaging ratios of distances to one or more points and to each of its nearest neighbors.

9. The method according to claim 6, wherein the hierarchical cluster-based approach computes an outlier score by identifying one or more points outside the clusters at a certain level of the hierarchy.

10. The method according to claim 1, wherein the normalizing step is performed using logarithmic inversion and gaussian scaling.

11. The method according to claim 1, wherein the converting step is performed using linear inversion.

12. The method according to claim 1, wherein the generating step further comprises disambiguation of providers according to the steps of:

13. The method according to claim 1 further comprising the steps of:

comparing provider names using the number of letters in common to determine a similarity value;

comparing the physical location of the providers using geocoding to determine a distance value; and

combining the similarity value and the distance value to match providers.

14. The method according to claim 13, wherein the similarity value is computed according to the step of dividing the number of common letters by the number of letters in the longest name.

15. The method according to claim 13, wherein the distance value is computed according to the steps of:

translating a provider location into a latitude coordinate and a longitude coordinate;

converting the latitude coordinate and the longitude coordinate into spherical coordinates in radians;

computing an arc length from the spherical coordinates in radians;

multiplying the arc length by a radius of earth in miles; and

providing a distance value in miles between the two addresses.