Pastle on paper
Tuesday, March 31, 2009
Effective review and its depiction:
I was thinking what makes a review effective and I put my ideas below using an example of book-review.
A book review in the context of this discussion (what is an effective book review) has two generic types – one in which the reviewer provides his background along with how he likes or dislikes the book and the other in which the reviewer provides same information on the book without mentioning his background.
Arguments can be used to deduce that the reviews containing reviewer’s background information are more useful for new users in making a decision (such as to read or to buy) compared to the reviews which do no contain reviewer’s background information.
A typical argument could be explained by considering a review of a book on a subject like artificial intelligence done by two reviewers. Since the book uses good amount of statistics we collect the background information of the two reviewers focusing their exposures to statistics. We assume for the clarity of this explanation that one reviewer say R1 is expert in statistics while other say R2 does not have any exposure to statistics. Now we can see various effects of the reviews on a user’s decision from the table below. For ease of understanding we also assume there are two users who use the reviews to make their decision and one of the users say U1 has an expert level understanding on statistics while other say U2 has no exposure to statistics at all.
We can see that when R1's and R2’s backgrounds are exposed to users, U1 finds R1 having similar background to him and that R2 doesn’t not have his background. This causes U1 to be influenced by R1’s reviews and to be unaffected by R2’s review. Similarly U2 gets influenced by R2 and stays unaffected by R1. Under this argument we see that a decision is made in all situations when reviewer’s background is exposed (see 4th column of the table above).
The assumption that similar background has higher influence and the argument provided above reveals few interesting information.
1) A review can be more effective if it contains reviewers’ background information.
2) A close relation to Bayesian Classification can be established.
Pr (buy background) = Pr (background buy) x Pr (buy) / Pr (background)
A more effective review summary depiction would be one that contains reviewer’s background.
A prediction system would work better if it includes reviewer’s background information.
A book review in the context of this discussion (what is an effective book review) has two generic types – one in which the reviewer provides his background along with how he likes or dislikes the book and the other in which the reviewer provides same information on the book without mentioning his background.
Arguments can be used to deduce that the reviews containing reviewer’s background information are more useful for new users in making a decision (such as to read or to buy) compared to the reviews which do no contain reviewer’s background information.
A typical argument could be explained by considering a review of a book on a subject like artificial intelligence done by two reviewers. Since the book uses good amount of statistics we collect the background information of the two reviewers focusing their exposures to statistics. We assume for the clarity of this explanation that one reviewer say R1 is expert in statistics while other say R2 does not have any exposure to statistics. Now we can see various effects of the reviews on a user’s decision from the table below. For ease of understanding we also assume there are two users who use the reviews to make their decision and one of the users say U1 has an expert level understanding on statistics while other say U2 has no exposure to statistics at all.
| R1’s rating | R2’s rating | Without reviewers’ background | With reviewers’ background |
|---|---|---|---|
| High | High | U1 decides to buy | U1 decides to buy |
| High | High | U2 decides to buy | U2 decides to buy |
| High | Low | U1 confused | U1 decides to buy |
| High | Low | U2 confused | U2 decides NOT to buy |
| Low | High | U1 confused | U1 decides NOT to buy |
| Low | High | U2 confused | U2 decides to buy |
| Low | Low | U1 decides NOT to buy | U1 decides NOT to buy |
| Low | Low | U2 decides NOT to buy | U2 decides NOT to buy |
We can see that when R1's and R2’s backgrounds are exposed to users, U1 finds R1 having similar background to him and that R2 doesn’t not have his background. This causes U1 to be influenced by R1’s reviews and to be unaffected by R2’s review. Similarly U2 gets influenced by R2 and stays unaffected by R1. Under this argument we see that a decision is made in all situations when reviewer’s background is exposed (see 4th column of the table above).
The assumption that similar background has higher influence and the argument provided above reveals few interesting information.
1) A review can be more effective if it contains reviewers’ background information.
2) A close relation to Bayesian Classification can be established.
Pr (buy background) = Pr (background buy) x Pr (buy) / Pr (background)
A more effective review summary depiction would be one that contains reviewer’s background.
A prediction system would work better if it includes reviewer’s background information.
Subscribe to:
Posts (Atom)
