Expectation-Experience Misalignment in Marketplace Text
Aspect-space and teacher-student pipeline for measuring listing-review mismatch.
Overview
This UCSD CSE 261 project measures expectation-experience misalignment in marketplace text. The project compares listing descriptions with user reviews, builds an aspect space, and estimates mismatch scores that can be used in downstream classification and regression analysis.
The pipeline covers preprocessing, aspect-space construction, teacher scoring, student-model training and inference, mismatch-score computation, and edge-case analysis.
Project Components
- Collected and cleaned marketplace text data.
- Built an aspect-space pipeline for extracting interpretable dimensions of listing-review mismatch.
- Trained a student model to approximate teacher scores over review-description pairs.
- Computed mismatch proxy scores and evaluated their predictive value in downstream analyses.