ElectricDeel Data Collec*on And Product Recommenda*on Engine For Consumer Electronics Senior Project Poster Day 2011 – Department of Computer and Informa*on Science – University of Pennsylvania The Web Crawler Distributed File System Data Processing Pipeline Database Web Application User User User User Recommendation Algorithm Architecture Design Goals 1. Accuracy • Recommendation algorithm must accurately reflect user preferences • Data must be comprehensive and errorfree 2. Ease of Use • Must be powerful enough to be worth using • Must be simple enough for Grandma 3. Transparency • Recommendations should not be a black box – users need to understand why products were recommended 4. Scalability and Modularity • It should be easy to add new product types and data sources • Today: TV recommendations, Newegg data • Tomorrow: Laptop and cell phone recommendations, Amazon and BestBuy data Eddie Siegel, Vikas Shanbhogue, Bennett Blazei Advisor: Zachary Ives User Interface 1 1 2 3 Questions are simple and jargonfree. 2 The priorities that are determined by the recommendation algorithm are shown. If the user disagrees with the results, they can click and drag to rearrange them. 3 Results are shown as a score from 0100 along with the cheapest price available. Clicking on a result provides more info. Recommenda8on System Crawler Mo8va8on • Shopping for consumer electronics is hard • Shoppers do the same research repeatedly • No simple way to find the perfect product BAD “What resolution should it have?” GOOD “What are you going to use it for?” Overview ElectricDeel makes it easier to shop for consumer electronics by asking consumers everyday, nontechnical questions instead of forcing them to decipher a complex list of technical specifications. User Feature Preferences Similar Feature Tradeoff • User input is used to build a set of weights • Weights describe the relative importance of product attributes for a particular user • Weights are used to calculate a score for each product