Vincent Chen

Correcting Recency Bias in Retrospective Judgments Using Diachronic Word Embedding

Many problems in social science require knowing the perceptions of people in the past. Currently, the only way for scientists to obtain such data, when lacking longitudinal datasets, is through self-report retrospection of survey participants. In our study, we show that people mistakenly rely on their current impression when asked to predict past consumer perception toward brands. This recency bias thus poses a fundamental challenge for the study of social trends of any sort. In this project, we demonstrate how diachronic word embedding, a recently developed class of NLP technique, can overcome the recency bias and provide more accurate prediction of past perception.

Bio: