11.5 Opinion Spam

1
11.5 Opinion Spam

• Speaker Yu Wen,Hsu
• Advisor Dr. Koh, JiaLing

2
Objectives and Actions of Opinion Spamming

• two main objectives for writing spam reviews
• To promote some target objects
• To damage the reputation of some other target
  objects
• the spammer writes some irrelevant information or
  false information in order to annoy readers and
  to fool automated opinion mining systems.

3

• To achieve the above objectives, the spammer
  usually takes both or one of the actions below
• hype spam write undeserving positive reviews for
  the target objects in order to promote them.
• defaming spam write unfair or malicious negative
  reviews for the target objects to damagetheir
  reputation.

4
Types of Spam and Spammers

• 1.3.5 write by owners or manufacturers.
• 2.4.6 write by competititors
• 1.4 are not damaging, 2.3.5.6 are very harmful.
• spam detection techniques should focus on
  identifying reviews in these regions.

5
Types of Spam and Spammers

• Manual and Automated Spam
• Spam reviews may be manually written or
  automatically generated.
• Individual Spammers and Group Spammers
• A spammer may act individually or as a member of
  a group

6

• Individual spammers a spammer, who does not work
  with anyone else, writes spam reviews.
• Group spammers A group of spammers works
  collaboratively to promote a target object and/or
  to damage the reputation of another object.

7
Hiding Techniques

• An Individual Spammer
• The spammer builds up reputation by reviewing
  other products in the same or different
  categories/brands that he/she does not care about
  and give them agreeable ratings and reasonable
  reviews. Then, he/she becomes a trustworthy
  reviewer.

8

• The spammer registers multiple times at a site
  using different user-ids and write multiple spam
  reviews under these user-ids so that their
  reviews or ratings will not appear as outliers.
  The spammer may even use different machines to
  avoid being detected by server log based
  detection methods that can compare IP addresses
  of reviewers (discussed below).

9

• The spammer gives a reasonably high rating but
  write a critical (negative) review. This may fool
  detection methods that find outliers based on
  ratings alone. Yet, automated review mining
  systems will pick up all the negative sentiments
  in the actual review content.

10

• Spammers write either only positive reviews on
  his/her own products or only negative reviews on
  the products of his/her competitors, but not
  both. This is to hide from spam detection methods
  that compare ones reviews on competing products
  from different brands.

11

• A Group of Spammers
• reviews the same product to lower the rating
  deviation
• writes a review roughly at the time when the
  product is launched
• reviews at random or irregular intervals to hide
  spikes.
• divided into sub-groups so that each sub-group
  can spam at different web sites

12
Spam Detection

• Review Centric Spam Detection spam detection is
  based only on reviews. A review has two main
  parts rating and content.

13
- Review Centric Spam Detection -

• Compare content similarity
• Detect rating and content outliers
• Compare average ratings from multiple sites
• Detect rating spikes

14

• Reviewer Centric Spam Detection unusual
  behaviors of reviewers are exploited for spam
  detection. It is assumed that all the reviews of
  each reviewer at a particular site are known.
  Most review sites provide such information, or
  such information can be found by matching
  user-ids.

15
-Reviewer Centric Spam Detection-

• Watch early reviewers
• Detect early remedial actions
• Compare review ratings of the same reviewer on
  products from different brands
• Compare review times

16

• Server centric spam detection The server log at
  the review site can be helpful in spam detection
  as well. If a single person registers multiple
  times at a Web site having the same IP address,
  and the person also writes multiple reviews for
  the same product or even different products using
  different user-ids, it is fairly certain that the
  person is a spammer.

17
Conclusion

• To ensure the quality of information provided by
  an opinion mining and/or search system, spam
  detection is a critical task.
• Without effective detection, opinions on the Web
  may become useless.
• This may just be the beginning of a long journey
  of the arms race between spam and detection of
  spam.