Review: Using ODP Metadata to Personalize Search
Baoning Wu (09/28/05)

This paper describes the method of using Open Directory Project to
personalize search. The results show that personalized search with ODP
can generate better results than ODP or search engine such as google
alone. Besides, the author also show that the extension of ODP
metadata to the whole Web is feasible via topic-sensitive biasing of
PageRank values.

The method of using ODP data to achieve personalized search is simple
and intuitive. The results are promising. So, the method proposed in
this paper is both feasible and useful to the real search engines. The
authors also designed effective ways to show that biased PageRank will
generate a different enough rankings.

Some weaknesses about this paper is:

1. There are several personalized method existing, although some of
them are expensive to calculate, the authors don't provide any
comparison with these methods. So, we are not sure whether the author
proposed a better personalized search method.

2. It seems to me that their idea can improve the precision based on
the results from popular search engines. But whether recall is also
important to personalized search is kept undiscussed. I wonder if the
method in this paper can improve recall with some small revisions.

3. At the end of section 4.1, they mentioned that Figure 4 till Figure
7 are generated with a crawl of 3M, and they also got similar results
by using stanford webbase crawl. I wonder why they just generate these
figures with stanford webbase crawl since it is bigger and then more
convincible. Another question is that even stanford webbase has
limited size compared to google's repository, I was still not
convinced that the results here can also be applied to the whole Web
respository. My point is that at least they need to mention the size
and date of the stanford crawl they used.

Generally speaking, the paper is well-written and worth-reading. An
effective and promising idea for personalized search is studied. Also the
analysis of biasing pagerank can generate different ranking results is
also interesting and helpful.