Presentation is loading. Please wait.

Presentation is loading. Please wait.

Red Opal: Product-Feature Scoring from Reviews Christopher Scaffidi Kevin Bierhoff Eric Chang Mikhael Felker Herman Ng Chun Jin School of Computer Science.

Similar presentations


Presentation on theme: "Red Opal: Product-Feature Scoring from Reviews Christopher Scaffidi Kevin Bierhoff Eric Chang Mikhael Felker Herman Ng Chun Jin School of Computer Science."— Presentation transcript:

1 Red Opal: Product-Feature Scoring from Reviews Christopher Scaffidi Kevin Bierhoff Eric Chang Mikhael Felker Herman Ng Chun Jin School of Computer Science Carnegie Mellon University ACM EC 2007, San Diego, CA Note: Chun Jim presented this talk, which I am reposting with her permission.

2 2 Motivation Searching for quality digital camera Picture quality Auto mode Battery Shutter Memory Searching by product feature: Red Opal

3 3 Red Opal Overview Feature Extraction Product scoring User Interface Evaluation

4 Feature extraction Prior work identified technical terms as mostly nouns. For a given product category, if a certain noun occurs in reviews far more frequently than in generic English text, then that word is likely to be a product feature. A Poisson distribution was previously used to extract technical terms from texts on physics & politics.

5 Feature extraction For each product category, 1. Retrieve Amazon reviews for products in this category 2. Tag part-of-speech to text (e.g. “games”  NN/“game”) 3. Compute the count of each noun and compound noun 4. Compute their probability 5. Sort the most common nouns and compound nouns together according to probability, yielding the feature list

6 6 Product scoring for features Input: Product category, feature of interest Output: Score (1-5) Scores are specific to feature ProductMemoryPicture quality … CameraA4.53.0 CameraB3.74.1… CameraC2.14.8 CameraD4.33.9…

7 7 Compute scores and confidences based on review ratings Red Opal scoring algorithm Based on reviewer-assigned ratings Weighted average rating of reviews that mention feature Weigh reviews higher if feature mentioned more often Assumption: Review ratings consistent with features discussed Future work to identify the consistent rating with a feature Compute confidence in scores Measures uniformity of relevant reviewer ratings

8 8 User Interface: Search Screen http://redopal.ntelligentsolutions.net

9 9 User Interface: Search Results 3 min(RedOpal) vs 10-15min(general)

10 10 Evaluation: Feature Extraction Time complexity: O(n), 0.9s per review.

11 11 Evaluation: Scoring Precision

12 12 Summary Red Opal: searching on-line catalog by feature Future Work: Integrating manufacturer product description Finding opinion words User-specified features Multi-feature search

13 13 Acknowledgements Norman M. Sadeh, Mary Shaw, Jonathan Aldrich, and Jaime Carbonell Fall 2005 “Web Commerce, Security and Privacy” class (questionnaire subjects)

14 14 Thank you! Questions and Comments?


Download ppt "Red Opal: Product-Feature Scoring from Reviews Christopher Scaffidi Kevin Bierhoff Eric Chang Mikhael Felker Herman Ng Chun Jin School of Computer Science."

Similar presentations


Ads by Google