Assessing quality of user reviews

• US 8,990,124 B2
• Filed: 01/14/2010
• Issued: 03/24/2015
• Est. Priority Date: 01/14/2010
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Tsaparas, Panayiotis, Lu, Yue, Ntoulas, Alexandros, Polanyi, Livia
• Primary Examiner(s): KINSAUL, DANIEL W
• Application Number:
  US12/687,127
• Publication Number:
  US 20110173191A1
• Time in Patent Office: 1,895 Days
• Field of Search: None
• US Class Current: 705/347
• CPC Class Codes: G06F 16/313 Selection or weighting of terms for indexing
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

User generated reviews and scores associated with the reviews may be used to train a review scoring model with textual features of the reviews. The review scoring model may be used to predict scores for newly received reviews. One or more constraints based on social networking application data associated with an author of a review may be used to adjust the predicted score of the review.

Claims

1. A method comprising:

receiving a review at a computer device through a network, wherein the review was generated by an author;

receiving metadata associated with the author at the computer device through the network, wherein the metadata includes an average score given to reviews generated by the author and an average score given to reviews generated by authors who the author has provided an indicator of trust to, wherein the average score given to the reviews generated by the author is based on scores given to reviews generated by the author, wherein each score given to a review generated by the author is given by an author or user who is not the author of the review;

determining a feature vector for the review based on the text of the review by the computer device; and

determining a score for the review based on the determined feature vector and the metadata associated with the author by the computer device, wherein determining the score for the review comprises;

determining the score for the review using the feature vector and a review scoring model; and

applying a co-citation constraint to the score for the review by skewing the score for the review towards an average score given to reviews generated by authors who have provided an indicator of trust to the authors who the author has also provided an indicator of trust.

2. The method of claim 1, wherein determining a feature vector for the review comprises performing at least one of a text-statistics based analysis, a syntactic analysis, conformity based analysis, a sentiment based analysis, or a feature based analysis.

3. The method of claim 1, further comprising applying an author consistency constraint, a link consistency constraint, and a trust consistency constraint to the determined score for the review.

4. The method of claim 3, wherein the metadata comprises social networking data.

5. The method of claim 4, wherein the social networking data comprises a number of nodes that are connected to a node associated with the author.

6. The method of claim 1, further comprising associating the determined score with the review.

7. The method of claim 1, further comprising determining one or more features from the metadata.

8. A method comprising:

receiving a plurality of reviews through a network by a computer device, wherein each review was generated by an author and has an associated score;

receiving metadata associated with each author of a review from the plurality of reviews through the network by the computer device, wherein the metadata associated with each author includes an average score given to reviews generated by the author and an average score given to reviews generated by authors who the author has provided an indicator of trust to, wherein the average score given to the reviews generated by the author is based on scores given to reviews generated by the author, wherein each score given to a review generated by the author is given by an author or user who is not the author of the review;

determining a feature vector for each review from the plurality of reviews using the received metadata associated with the author of the review and the text of each review by the computer device;

training a review scoring model using the feature vector for each review from the plurality of reviews and the associated scores by the computer device;

receiving a review by the computer device, wherein the review is not a review from the plurality of reviews;

receiving metadata associated with the received review by the computer device;

determining a feature vector for the review by the computer device; and

generating a score for the review using the feature vector for the review and the review scoring model by the computer device; and

applying a co-citation constraint to the generated score for the review by skewing the score for the review towards an average score given to reviews generated by authors who have provided an indicator of trust to the authors who the author has also provided an indicator of trust.

9. The method of claim 8, further comprising applying an author consistency constraint, a link consistency constraint, and a trust consistency constraint to the generated score for the review.

10. The method of claim 9, wherein the metadata comprises social networking data.

11. The method of claim 8, wherein determining a feature vector for each review from the plurality of reviews comprises determining a value for at least one text feature based on the text of each review.

12. The method of claim 11, wherein the text features comprise one of text-statistics features, syntactic analysis features, conformity based features, sentiment based features, or product features.

13. A system comprising:

at least one computer device;

a review scoring model; and

a review scoring engine adapted to;

receive a plurality of reviews, wherein each review was generated by an author and has an associated score;

receive metadata associated with each author of a review from the plurality of reviews, wherein the metadata associated with each author includes an average score given to reviews generated by the author and an average score given to reviews generated by authors who the author has provided an indicator of trust to, wherein the average score given to the reviews generated by the author is based on scores given to reviews generated by the author, wherein each score given to a review generated by the author is given by an author or user who is not the author of the review;

determine a feature vector for each review from the plurality of reviews using the received metadata associated with the author of the review and the text of each review;

train the review scoring model using the feature vector for each review from the plurality of reviews and the associated scores;

receive a review that is not from the plurality of reviews;

receive metadata associated with the received review;

determine a feature vector for the review;

generate a score for the review using the feature vector for the review and the review scoring model; and

apply a co-citation constraint to the generated score by skewing the generated score towards an average score given to reviews generated by authors who have provided an indicator of trust to the authors who the author has also provided an indicator of trust.

14. The system of claim 13, further comprising applying a trust consistency constraint to the generated score, wherein applying the trust consistency constraint comprises skewing the generated score towards the average score given to reviews generated by the authors who the author has provided an indicator of trust to.

15. The system of claim 13, further comprising applying an author consistency constraint and a link consistency constraint to the generated score.