Anime Recommendation System

• 

  About us ;)

Inspiration

I have been recently looking for interesting anime. Asking friends for recommendations was not working as well as I hoped. I thought that perhaps with more data, I could come up with something cool.

What it does

It uses data obtained from MyAnimeList to build a graph that then uses LangChain & LangGraph to create AQL queries on the graph to get info about reccomendations.

How we built it

We spent a lot of time cleaning and restructuring the data to fit into the graph data structure. Further we defined edge lists and some similarity indices. We use cosine similarity indices to define the similarity between the anime. We define different collections as node types and edge collections as node edge lists.

Challenges we ran into

Working with NetworkX in LangChain/LangGraph. Some LLM models do not generate good AQL queries. Further, some LLMs do not support tool use.

Accomplishments that we're proud of

Data conversion to fit the graph data structure took a lot of planning. We are proud of how well we accomplished that

What we learned

Building Graphs on Arango. Working with LangChain/LangGraph. Building tools. Team planning

What's next for Anime Recommendation System

Working with the manga dataset to enhance recommendations for Manga as well as Anime.

Built With

• arangodb
• langchain
• langgraph
• networkx
• python
• sambanova