.. title: Generating Recommendations with GPT-3
.. slug: generating-recommendations-with-gpt-3
.. date: 2020-08-04 12:57:27 UTC-07:00
.. tags:
.. catego ry:
.. link:
.. description:
.. type: text
`Serendipity `_ is a natural language recommendation engine powered by GPT-3. As opposed to traditional recommendation algorithm that generate recommendations based on implicit or explicit user feedback (such as ratings or items clicked), Serendipity works through a natural language search interface: you just tell it what you want. There are pros and cons to a system like this:
**Pros**
- Recommendations are context aware: you may be in the mood for a thriller one night but a comedy the next
- We are able to avoid the cold start problem -- recommendations can be generated even for users with no data
- You can use this system to find very specific items based on somewhat esoteric clues (examples below)
**Cons**
- Recommendations will be poor if the query is poorly worded
- It's a challenge to augment this using traditional ratings or engagement data
In this blog post, we'll showcase some interesting results GPT-3 is able to find.
Very specific queries
---------------------
- `romantic comedies startting Matt Damon but not Ben Affleck `_
- The results aren't perfect -- there are a couple where Ben Affleck slips through or neither of them are in the movie or the genre isn't right, but for the most part the results are relevant
Foreign language queries
------------------------
Soon after I create the search interface, some people from Japan found this website and started using it. Although I didn't code any special logic for foreign languages, it kind of just works. Here are some example queries:
- `日本の戦国時代あたりが舞台で、たたり神に呪いをかけられて村を出ていくことになり狼に育てられた娘に会うなどして最終的に呪いが解けるアニメ `_
- Google translates this as "Animated in the Japanese Warring States era, an animation that can be finally solved by cursing God and going out of the village and meeting a daughter raised by a wolf". I didn't realize until I saw the results that this is describing the plot to Princess Mononoke, which is the first result (note: this depends on which ranking algorithm AB test you're in).
- `魔法使い 少年 ファンタジー 映画 `_
- Google translates this as "witch/wizard boy fantasy". Results include Harry Potter, Narnia, The Wizard of Oz, and Howl's Moving Castle
I tried a couple other languages and the results seem reasonable:
- http://serendipityrecs.com/gpt3recs/movies/films%20qui%20montrent%20la%20joie%20de%20vivre/
- http://serendipityrecs.com/gpt3recs/movies/%E9%9F%93%E5%9C%8B%E9%9B%BB%E5%BD%B1/
Movies/Books where you forgot the name but it's on your mind
------------------------------------------------------------
- I started reading a book and then couldn't find it and didn't remember the name, so based on what I remembered of what I'd read so far I tried searching `scifi novels where the guy tries to make clones in computer simulation and they always kill themselves `_
- the book turned out to be Permutation City, which again is the first result (depending on ranking variant)
Anti-recommendations
--------------------
- `movies that are so dumb that nobody in their right mind would watch `_
- `worst movies of the 90s `_