AI News, Building a Recommendation System Using Neural Network Embeddings

Building a Recommendation System Using Neural Network Embeddings

To learn meaningful embeddings, our neural network must be trained to accomplish an objective.

Working from the guiding assumption of the project — that similar books link to similar Wikipedia pages — we can formulate the problem as follows: given a (book title, wikilink) pair, determine if the wikilink is present in the book’s article.

We give the network some true examples — actually present in the dataset — and some false examples, and eventually it learns embeddings to distinguish when a wikilink is on a book’s page.

To generate the false examples — done later — we’ll simply pick a link index and book index at random, make sure it’s not in the pairs, and then use it as a negative observation.

Note about Training / Testing Sets While using a separate validation and testing set is a must for a normal supervised machine learning task, in this case, our primary objective is not to make the most accurate model, but to generate embeddings.

Neural Network Embeddings Explained

For example, if we have a vocabulary of 50,000 words used in a collection of movie reviews, we could learn 100-dimensional embeddings for each word using an embedding neural network trained to predict the sentimentality of the reviews.

The network I used has two parallel embedding layers that map the book and wikilink to separate 50-dimensional vectors and a dot product layer that combines the embeddings into a single number for a prediction.

In Keras code, this looks like the following (don’t worry if you don’t completely understand the code, just skip to the images): Although in a supervised machine learning task the goal is usually to train a model to make predictions on new data, in this embedding model, the predictions can be just a means to an end.

The Putin Files: John Brennan

Watch former CIA director John Brennan's candid, full interview on Putin and allegations of Russian interference in the 2016 U.S. election – part of FRONTLINE's ...

Mea Maxima Culpa : Silence in the House of God

Alex Gibney explores the charged issue of pedophilia in the Catholic Church, following a trail from the first known protest against clerical sexual abuse in the ...



SPAN 2017 Pittsburgh - Livestream Day 1

SPAN is a conference exploring the ways in which design and technology shape our everyday lives. Use the timestamps below to jump to individual talks: ...

Robert David Steele an ex-C.I.A. Agent On The Dangerous C.I.A. and Political Pedophilia

Brendon D. Murphy, Amy Devlin. (Apr. 17, 2017). Episode 65 GFM Media Robert David Steele An Ex CIA Agent On The Political Pedophilia Epidemic. Global ...