Skip to content

Latest commit

 

History

History
58 lines (34 loc) · 3.07 KB

Defense.md

File metadata and controls

58 lines (34 loc) · 3.07 KB
Raw

Defense

Instructions

The exam is oral and each group will have 30 minutes (sharp) to present their project and 30 minutes to reply to the instructors' questions. All the members of a group must be present during the whole presentation of their group. The presentation can be delivered by multiple people.

The presentation should explain what does the project do (what did you build? What problem does it solve?), the design of the project (e.g., how did you organize your code? What modules did you create and what is their relationship?), and show a live demo of the project (it is ok to fix bugs in your project after the submission on the 18th -- just do not add new features). In the 30 minutes slot for questions the instructors will ask various question about the project (e.g., how did you managed the communication in the project?).

Main organisation

I-What is our project (about -4 minutes)

What problem does it solve?What did we build?

II-Project architecture: each part will be explained (about –20 minutes)

  1. Crawler
  2. Database
  3. Server and API
  4. Recommendation

III-Live demo of the project (about 3-minutes)

I- Where we started (Pitch presentation) , to where we ended (Final project's description)

Initial goal:

Create a recommender engine of brands for customers depending on fashion brands. We wanted to extract data from social media: Twitter, Instagram, Pinterest, about the actual fashion trends (using #hastags, descriptions, numbers of like, of citations, number of followers). Using that data we would have been able to create bipartite graph of customers and brands (there would have been a link between a user and a customer if there has been an interaction of any type between the brand's account and the customer account on a social media. Using clustering theory we would have created some groups of customers which have similar preferences and interactions. From this we would have use a neighbours-based system to recommand new brands to customers. One other main aim would have been to keep our graph updated , adding new fashion brands (which are being more and more trendy).

Issues:

Difficulties to use social networks' API (Twitter, Instagram,...): Restrictions to access some data, limited amount of data.

Solution:

We decided to change the orientation of the project: changing the subject/ type of data on which we were going to work, keeping the final aim/algorithms with the same final aim: creating a recommendation engine which could be updated based on some specific methods: data extraction (crawler), creation of a database, matrix factorization, clustering and K-nearest-neighbours algorithm for predictions and recommendation.

Our final project is the following:

Tinder for researchers: Recommendation system of papers for researchers

Description of the project:

II- Architecture of the project WHY? (why did you do that) HOW? (How did you do?)

III- Live Demo of the project

TO BE DONE:

Powerpoint Each part to be completed by the designated person Global presentation of the project (Clémence//Loan)