A data analysis pipeline which recommends a list of books based on a Goodreads user's reviews, his or her friend's, and the books' average ratings.
Goodreads is a website in which users can sign up and register books to generate library catalogs and reading lists.
The Book Recommender suggests which books a Goodreads 's user could read next!
To demonstrate, I used my own profile information and selected the profiles of some of my friends and people I follow. The Recommender now performs the following steps:
- Extracts the user information using Goodread's Web API.
- Prepares data for cleaning and analysis.
- Cleans and formats book data.
- Recommends a list of books.
According to the analysis, these are the books I should read next:
Not bad. These recommendations are based partly on the correlations I have with some of my contacts. The idea is to find contacts with similar tastes to mine. Such contacts probably liked books that I will also like. Conversely, contacts with a negative correlation probably liked books that I won't like.
| Contact | Taste correlation |
|---|---|
| Vanessa | 1.000000 |
| Nicolas | 0.361499 |
| Andrea | 0.316139 |
| Mario | 0.000000 |
| Cova | 0.000000 |
| Eduardo | -0.069824 |
| Fernando | -0.191663 |
| Stefan | -0.244444 |
| Maria | -0.291748 |
| Bill Gates | -0.555556 |
The Book Recommender also considers how high are the mean ratings of my contacs. This should make sense: if a contact rates most books with two stars, her five-star ratings should count more than those of a contact that rates most books with five stars.
| Contact | Mean rating | Standard deviation |
|---|---|---|
| Cova | 4.857143 | 0.478091 |
| Eduardo | 4.101796 | 1.015748 |
| Mario | 3.966102 | 0.927847 |
| Fernando | 3.931818 | 0.868285 |
| Nicolas | 3.819149 | 0.983332 |
| Vanessa | 3.810811 | 1.081294 |
| Andrea | 3.757282 | 0.856815 |
| Stefan | 3.732394 | 0.815762 |
| Bill Gates | 3.538153 | 0.933014 |
| Maria | 3.340909 | 0.914927 |
In the Notebook main.ipynb you can find all the steps of the pipeline.
To communicate with its Web API, Goodreads requires developers to get a key. I have hidden mine in the notebook, but it should work with yours. Just change the value of the CONSUMER_KEY variable.
Then, use the extract_info function for each of the users whose information you want to get. If your request was successful, the cell should print "Status code: 200".
You'll have to replace some of the variable names, depending on the users you chose. Besides that, you should be able to execute most of the cells without any problem.
The clean_title function eliminates duplicate books (which, for example, might only vary slightly in their wording). I believe it works for most books, but you might consider playing with it if you still get duplicates.
Execute all the cell to get your final recommendations. You can tweak the relative weight of correlations or ratings in the correlation_weighted and weighted_rating functions, respectively.
Happy reading!