Assignments Flashcards
- Give examples of how association rule mining can be used in other than the market-basket use case. Explain in detail.
Another idea that came to mind would be education, for many years it’s really been me teaching myself.
Some courses that I have taken will have amazing professors, but I still find myself teaching myself. In
which it partially makes sense to learn the material. But if there was a way in which the material would
adapt its delivery that would be a game changer for our education. The things that would need to be
looked at would be performance data of the students and overall class, and many more that I can’t think of
now. To be fair there might be some software that can do this but I doubt it’s fully structured for the
student specifically.
- While building a recommendation system with collaborative filtering, why do we use Pearson correlation coefficient and why does Jaccard or Cosine similarity not work well? You may use an example to explain your reasoning.
Unlike Jaccard or Cosine, Pearson can correlate between the variables it is given. In a
recommendation system we would use ratings, all three of these would work but only Pearson
correlation would be the most effective. For example, Jaccard wouldn’t focus on rating values,
instead it would focus on if the film/product got a rating. Cosine similarity will assume that any
empty reviews as “negative” reviews in which affects our recommendation system. Pearson
correlation on the other hand can capture the similarity even if rates differ from one another
Netflix is interested in building a movie recommendation engine for
watchlist generation. You have been asked to provide a list of movie features that will be
useful for generating movie profiles. Describe three features that you think would be
important as attributes for movie profiles. Justify your choices using concepts from similarity
and recommender systems.
Well, if I were to create a recommendation engine I would add Genre, Ratings/Reviews, and
Directors/Cast. Let’s start with why I chose these three, when I look for a movie the Director/Cast
plays a huge role. For example, Guillermo del Toro is a very popular filmmaker, the genre might be
action/horror/fantasy, and the cast has many reputable actors. Each single category/feature will
make me want to watch the film more and more.
Earlier in the assignment I spoke about collaborative filtering, and Pearson correlation. Our system
will work best if we use this approach, due to the use of ratings and other variables. Each rating will
be different from one another because everyone has their own opinion on a film. And not every film
that features a great director/cast will meet a user’s expectations.
- In the analysis of spam farms, as discussed in the applications of page rank, we analyzed the spam-farm where every supporting page links back to the target page. Repeat the analysis for a spam farm where:
a. Each supporting page links to itself instead of to the target page.
b. Each supporting page links nowhere
c. Each supporting page links both to itself and the target page.
a. If the supporting page links to itself the it wouldn’t receive any page rank from any of the
supporting pages. It would basically be a self loop. Since it has no link structure it wouldn’t
be effective for page rank.
b. If the support page links don’t point to themselves or the target page it wouldn’t have a page
rank. Causing it to be ineffective in spam-farms.
c. With each supporting page linking to itself and the target page it would gain page rank. But
we would run into some issues since each supporting page is linking itself. For a spam farm
it would be effective but you might see some ineffective use since it’s also linking to itself.
Not the best for spam farms and gaining page rank but it would still see a significant amount
of page rank.
- Explain how Pinterest used the concept of page rank to generate
recommendations to users.
As a pinterest user myself I get the concept of page rank (I believe) from when I pin something to my
boards. I think of page rank as popularity to an extent, the more users that pin a post the more
popular it gets.
For example, in lecture we spoke about smoothies. If I were to pin a smoothie recipe it would
basically gain a page link to smoothies. If I were to refresh my pinterest page, I would be
recommended other smoothies to pin to my board. And if I keep pinning smoothies to my board,
more and more will appear on my page to add.
