Recommender Systems: Task

This week’s task is to analyze an existing recommender system that appears interesting and judge them by following criteria: 1. Perform a Scenario Design analysis as described below. Consider whether it makes sense for your selected recommender system to perform scenario design twice, once for the organization (e.g. Amazon.com) and once for the organization’s customers. 2. Attempt to reverse engineer what you can about the site, from the site interface and any available information that you can find on the Internet or elsewhere. 3. Include specific recommendations about how to improve the site’s recommendation capabilities going forward. 4. Create your report using an R Markdown file and create a discussion thread with a link to the GitHub repo where your Markdown file notebook resides. You are not expected to need to write code for this discussion assignment.

Artfinder.com, the art recommender system:

https://www.artfinder.com/#/ is a London based website, which brings recommendations and image recognition to the world of art. Recommendation engine has revolutionized the way one thinks about appreciating and purchasing art. For recommendations on books, one can go to Amazon.com or Barnesandnoble.com. For music, one can go to Spotify.com. But, in the art world, Artfinder is an up and coming site that is giving competition to its competitors.

1a) Scenario Design Analysis for customers:

  1. Who are the target users? The target users are art consumers, who visit the website, probably with a goal of buy originals or prints. That includes not only individual buyers, but art institutes, museums, corporations, which want to decorate their walls with paintings.

  2. What are their goals? Artfinder.com has multiple goals, including creating awareness of art in public mind, reminding the people that “Art Exists” and sales promotion. Many other sites do that much, but what Artfinder brings to the table is recommendations. If a buyer is looking for the art of Rene Magritte, Artfinder.com not only pulls up paintings by Magritte, but other surrealists of earlier and later times, which stirs the imagination of the buyer.

  3. How can you I can help them accomplish their goals? I am fond of art and want art to play in the public imagination. So, firstly, I would spread the word to my friends. Secondly, I would see how best to improve their recommender engine. I would like to transport the data on big data, think of improving pattern recognition algorithms.

1b) Scenario Design Analysis for the organization:

In order to run the site, Artfinder.com has to do internal brain storming on what art to display. Many of the artworks haven’t yet been digitized, especially the up and coming, but talented artists. This is investment of time and money. So, running this kind of business not only requires a scenario analysis for end customers, but also the organization.

  1. Who are the target users? The target users are internal stake holders, which includes art critiques, who are on board, investors etc.

  2. What are their goals? In internal scenario, their goal is to convince a business proposal.

  3. How can you I can help them accomplish their goals? I don’t think I could contribute here, because it’s completely internal to their organization. However, if I come up with an improved recommender algorithm, I would approach them, with a business proposal.

2) Reverse engineering:

I will try to hazard a guess about how they engineered the recommender system. Probably, they created a vector data structure, with paintings that share similar characteristics (genre), e.g. renaissance or impressionism. Then added more characteristics, to form a table, from the individual rows. The table is being used to answer various classification related questions. Then used a filtering mechanism to remove the paintings that didn’t belong to a genre. The paintings are ranked according to a bunch of criteria, like year of production (simple linear distance), color scheme, theme, e.g. romanticism, nature, religious etc. They might have calculated similarity, using below function, or simply Euclidean distance:

similarity (P1, P2) = cos (P1, P2) = (P1 . P2) / (mod(P1) . mod(P2))

This is just a very rough sketch. A closer look and experience would help sharpen the picture.

3) Room for improvement:

I observed room for improvement in their recommendation. Here’s an example. They have broad categories, like surrealism, renaissance art, or impressionism, but not narrower categories, like pointillism. Consequently, when I select painter Pissarro, it selects other impressionists, like Monet, Cezanne, Degas etc. I would expect the recommender to pull up other pointillists, like Georges Seurat, Signac, Cross etc. With robust pattern recognition, they can improve this aspect of their recommendation.

Marker: 607-11