Conclusion
Although there are many wine recommenders, we created two unique ones. Most other recommender systems have bottle recognition systems where the user will take a photo of a bottle of wine and the recommender will suggest other bottles that are similar to this. Some of them give in depth detail about the wine with facts such as alcohol content, price, availability and where the grapes are grown and other winery information, etc. Most of them rely on user ratings (either on a five point or ten point scale for example) to find similar wines. This relies heavily on the user and allows for user bias (such as someone giving a wine a high rating simply because they like the winery or the way the label looks). Our concept is novel because our recommender works relies only on tasting notes, as a sommelier would, and uses cosine similarity to calculate the similarity between wines. Additionally, our recommender can recommend across wine varieties. For example, given a chardonnay it will recommend a wine similar to the chardonnay based on tasting notes, even if the recommended wine is a rose or pinot grigio.
If we were to continue with this project our next step would be to have people use our recommender and give feedback to see how well it recommends wines. From there we could update our code to reflect their input and give better recommendations.
We hope you enjoyed reading about our project!
Our code can be found at https://github.com/kwombach/WineApp