Inspiration
We drew inspiration form previous projects we've individually worked on.
What it does
Have the ever found looking for potential restraunts somewhat dull? Make your restuant browsing experience fun with adaptive review summaries and interpretative visualizations of your favoriate restraunts!!
How we built it
We used a simple algorithm that uses word counts to find the most important or relevant sentences in a pice of text (reviews), and uses that to create a summary. The data for our project was scraped from Google maps, and is specific to Northside Chicago.
Challenges we ran into
One of our goals was utilizing the ChatGPT API to help summarize the collection of reviews, rather than using other traditional methods for summarization. The issues were however, that GPT3.5 only accepts a maximum of 4096 tokens, and is capped at 20 requests per minute. The workouts for this would be too feed the reviews of locations in batches. A restaurant may have up to 500 reviews, so we would break that into batches small enough to be below the token limit. Theoretically if the gpt3.5 api allowed for more tokens, and more requests per minute, our original approach would theoretically work
Accomplishments that we're proud of
We are proud that we got most of our code working. We did have to leave some feature behind but in the end it still workout pretty good.
What we learned
We learned data preprocessing is sometimes more important than the model or algorithm we are using. Most of our issues involved preparing our data.
What's next for Restaurant Review Summaries
Next step would be to finish ChatGPT integration to help produce better summaries
Log in or sign up for Devpost to join the conversation.