Home

Shared Perspectives for Query by Example in e-Commerce Systems

Welcome to the shared-perspectives wiki! In this, I'll address the theory behind Shared Perspectives, as well as the basic workflow of the project and how can it be reproduced. The concepts and design space are for Shared Perspectives, and Query by Example. The project basically consists of Processing and Application, in the first one we turn reviews into useful representations of a movie, and in the second we use those representations to find similar movies.

For the processing, the main steps were the Perceptual Feature Extraction (getting vectors from reviews), and the Construction of Shared Perspectives(clustering said vectors). The whole thing is supported with a Relational Database, so we those vectors are called tuples, since they are the encoded representation of the review: the values for those Perceptual Features.

Then for the application, we use the Shared Perspectives Tuples (the center of each Shared Perspective) to find movies that were perceived in a similar way by the users.

---

Concepts and Design Space:

• Shared Perspectives
• Query by Example

---

Processing:

• Perceptual Feature Extraction
• Constructing the Shared Perspectives

---

Evaluation:

• Evaluation: User Simulation
• Evaluation: Semantic Quality of Shared Perspectives

---

Applications:

• Prototype

---

Disclaimer: The code is not optimal, many for loops could be lambdas, some lists would be better as some special dictionary, etc. But during the processing and implementation I wasn't feeling very adventurous (like using a calculator to do 2+2 in a test, just to be sure). I'll be working on this continuously.