Our team was inspired by the fact that there are so many social websites that provide services like dating, chatting, finding jobs, connecting with friends, listening to music etc and for each of those there is one particular service that is exceptionally popular. However, there is no one service when it comes to events. There are several sites that offer a range of event suggestions but hosts either have to pay, be popular enough or sell tickets to them. The only platform that hosts events for free without any specific conditions is facebook. However, there is no functions to recommend events in your area. If you are looking for something specific, facebook is convenient, but not if you need inspiration on a friday night!
What it does
Our app takes voice input from the user, like “show me events in london” and suggests events nearby sorted by categories like “concerts”, “food markets”, “seasonal”, along with time etc.
How we built it
With the aim of using voice recognition to find local events for users The build consisted of a two-prong approach, involving an Android front-end and a Java based backend. The Android front end component involved multiple API's, including Google's Speech Recognition, Facebook's Graph API for JSON retrieval of event data, as well as MonkeyLearn API for natural language processing elements. The backend is written in Java using the Spring Boot framework for accessible REST calls. It is calling the Facebook Graph Event API to pull a list of events in the area, which is then filtering using calls to a trained NLP model on MonkeyLearn. The Endpoint is then exposed using ngrok.
Challenges we ran into
-Multiple rewrites and builds were executed due to numerous compatibility issues, but all were analyzd and resolved -- Natural Language Processing: it is very hard to categorise the event names and descriptions for some events on facebook and put them into appropriate categories. Hard coding words is not an efficient approach therefore we used machine learning that uses a bag of words method. This method, however, is still not error-prone and relies on labelled data.
Accomplishments that we're proud of
-Excellent work distribution involving everyone's own strengths, with particular focus on redundancies incase an approach failed. -- We are very proud of the team spirit, we quickly found each other, decided on the topic and divided the tasks. There were no major conflict of opinions and everybody contributed significant knowledge and expertise to this project and everything went smoothly until the end.
What we learned
- 36 hours is not enough time to create a commercial product. -Stuff like this can ONLY be done in teams and when everyone shows commitment it will ALWAYS WORK ## What's next for Event5
- Scaling up involving other APIs