Two years ago, my partner and I went through a really lengthy and frustrating search for our dog, Lola Ray. We knew that we wanted a very particular dog for our lifestyle: her being active and constantly looking for a running partner, and me being endlessly lazy and wanting a snuggler. We knew it would be a hard match to find, but what we didn't expect was how hard it could be to understand the puppy-talk that dominates pet-descriptions on adoption websites.
I was ultimately motivated by this experience for my Insight project. So I came up with "Who's a Good Dog?": an app that provides a better way to search for adoptable dogs. Instead of hoping for the best when searching by breed, size, age, etc; you search for the personality traits that matter most to you!
Dogs listings are queried from the Petfinder API in real time to ensure that you're able to find dogs actually up for adoption.
Text-descriptions are processed to ensure that the text documents are actually describing the dog. This relies on a Naïve Bayes classifier to identify the subject content and ensure it's dog-focused. Summaries are generated from the pet-descriptions using Latent Semantic Analysis to highlight the important descriptors from the text. Both summaries and search queries are embedded into word vectors using Word2Vec pre-trained on the spaCy large, english web-corpus model.
For all returned pet-description a score is assigned using the average cosine-similarity between summaries and search word-vectors.