Voice of the Customers: Mining Online Customer Reviews for Product Feature-Based Ranking

Size: px

Start display at page:

Download "Voice of the Customers: Mining Online Customer Reviews for Product Feature-Based Ranking"

Ami Barnett
7 years ago
Views:

1 Voice of the Customers: Mining Online Customer Reviews for Product Feature-Based Ranking Kunpeng Zhang, Ramanathan Narayanan, Alok Choudhary Dept. of Electrical Engineering and Computer Science Center for Ultra-Scale Computing and Security Northwestern University 3 rd Workshop on Online Social Networks Boston, MA, June

of Electrical Engineering and Computer Science Center for Ultra-Scale Computing and Security

2 Customer Reviews More consumers are shopping online than ever before Online retailers allow consumers to add reviews of products purchased Customer reviews are more unbiased, honest than product descriptions provided by sellers 2

reviews of products purchased Customer reviews are more

3 Customers Shop Online and Read Reviews 3

4 4

5 Product Ranking Workflow Online Data Crawling Information Extraction Sentence Splitter POS Tagging Comparative Sentence Identification Comparative Sentence Refinement Subjective Sentence Identification Graph Construction Product Ranking 5

Sentence Identification Comparative Sentence Refinement

6 Product Features Lens Resolution Memory flash Burst compression Battery Focus 6

7 Identifying Feature Sentences Example: features from consumer reports 7

8 Sentence Labeling Comparative sentence (CS) A sentence which indirectly express an opinion by performing a comparison between two products Rules to identify general CS KW: 126 keywords (outperform, exceed ) POS tags (JJR, RBR, JJS, RBS) Predefined patterns (as <word> as ) Refine CS A sentence contains one or more than one different product names 8

KW: 126 keywords (outperform, exceed ) POS tags (JJR, RBR, JJS, RBS) Predefined

9 Sentence Labeling Subjective sentence (SS) A sentence expressing directed praise or deprecation about a product Rules to identify SS If a sentence contains subjective words (positive, negative), it is classified to be a subjective sentence 9

Rules to identify SS If a sentence contains subjective words

10 McCormick Sentiment Identification Use the keyword strategy {MPQA[1] + our own words 1974 positive words negative words + 28 negation words} Positive subjective sentences (PS) Negative subjective sentences (NS) Positive comparative sentences (PC) Negative comparative sentences (NC) Positive Subjective(PS) This camera has great picture quality and conveniently priced. Negative Subjective(NS) The picture quality of this camera is really bad. Positive Comparative(PC) This camera has superior shutter speed when compared to the Nikon P40. Negative Comparative(NC) This is the worst camera I have seen so far. [1]. 10

camera has great picture quality and conveniently priced. Negative Subjective(NS) The picture quality of this camera is really bad.

11 11

12 Constructing the Product Graph G = { V, E } V is the set of vertices. Each vertex represents a product E is the set of edges. The edge weight represents the comparative relationship between products Weight of a node is determined by the number of positive/negative subjective sentences (PS/NS) ex. A has excellent picture quality Weight of an edge is determined by the number of positive and negative comparative sentences ex. B is better than C. B is worse than C 12

The edge weight represents the comparative relationship between products Weight of a node is determined by the

13 Ranking Products (prank Algorithm) For each graph (G f ) related to feature f, we could evaluate the relative importance of each product by using the prank algorithm Pj Pi P Pk 13

14 Ranking Products (Example) Rank Product Score 1 B D C A 0.04 Example: PS(A)=1, PS(B)=2, PS(C)=3, PS(D)=4, NS(A)=3, PC(B,A)=3, PC(B,C)=7, PC(B,D)=3, PC(A,C)=2, NC(B,C)=2 14

15 Experiments Data Amazon.com (Digital camera) Precision and recall to identify feature sentences: and

16 Experiments Data Amazon.com (TV) 16

17 Experiments Results for Camera and TV 17

18 Top 10 Products for Each Feature and Overall 18

19 Comparing to Expert Opinion Digital Camera and TV 19

20 Comparing prank to ARS (average rating score), Salesrank for Camera 20

21 Comparing prank to ARS (average rating score), Salesrank for TV 21

22 Related Work 1. Sentiment analysis [B. Liu, 2010; B. Pang, 2002] 2. Extracting product features [M. Hu, 2004; A. Popescu, 2005] 3. Review summarization [M. Hu, 2004, 2006] Differences from our work: Keyword matching strategy to identify and tag product features in sentences Different strategies to assign sentiment orientation to sentences Using our ranking algorithm on the product graph to rank products 22

23 Summary Scalable technique to mine millions of online customer reviews to rank products Online Data Crawling Information Extraction Sentence Splitter POS Tagging Comparative Sentence Identification Comparative Sentence Refinement Subjective Sentence Identification Graph Construction Product Ranking 23

24 Thank You Dept. of Electrical Engineering and Computer Science Center for Ultra-Scale Computing and Security Northwestern University 3 rd Workshop on Online Social Networks Boston, MA, June

25 Thank You Dept. of Electrical Engineering and Computer Science Center for Ultra-Scale Computing and Security Northwestern University Workshop on Online Social Networks Boston, MA, June

Search and Information Retrieval

Search and Information Retrieval Search on the Web 1 is a daily activity for many people throughout the world Search and communication are most popular uses of the computer Applications involving search