Masters Thesis

Extracting Psychological Factors from Amazon Reviews to Predict Perceived Informational Value Using Data-driven and Closed-vocabulary Approaches

Digital interactions can provide rich and novel information for psychological research. An influential form of digital communication is online reviews. Individuals supply critical judgments and opinions to facilitate the decision-making process of their fellow consumers. Psychological factors (emotional, social, and cognitive) extracted from review language may elucidate important direct or indirect signals of informational value during the decision-making process. Psychological semantics were extracted using latent Dirichlet allocation (a dimension reduction technique) and using the Linguistic Inquiry and Word Count (LIWC) dictionary from 3,000 and 50,000 randomly selected Amazon book reviews, respectively, to predict perceived informational value. The final regression model using the LIWC factors demonstrate that negative emotion and inhibitory language negatively influence perceived informational value, while social and analytic language positively influence perceived informational value F(6, 49985) = 14445.19, p < .001. The data-driven approach demonstrated that topical information in book reviews can be unhelpful, while impressions or thoughts about features of the book can be helpful. Both methods produced unique but complimentary results important in distinguish which emotional, social, and cognitive cues within language can impact helpfulness during the decision-making process.


In Collection: