University of Toronto: Using Vertex AI to reveal the words that make online reviews most helpful

A deep dive into how we use language

As a Professor of Marketing and Psychology at the University of Toronto's Rotman School of Management, Sam Maglio works at the intersection of computational linguistics, cognitive psychology, and behavioral science: he wants to know how the words people use affect actions in the real world. To this end, he has studied the impact of specific words in online reviews and examined the use of passive versus active voice in sentences. As the capabilities of AI and natural language processing expand, he can ask bigger questions and answer them using larger data sets that require heavier computational power. His most recent research paper, co-authored with Ph.D. student David Fang and published in the Journal of Experimental Social Psychology, examines how verb tense in online Amazon reviews affects their impact.

Analysis of real-world datasets, at scale

Social scientists used to have few options for conducting research: they recruited participants for lab experiments or manually coded data for analysis. Now, there is a bounty of naturalistic online data that is readily available to harvest. In their study, Maglio and Fang collected scraped datasets of two million online Amazon reviews across different product domains and then analyzed them using a generalized linear model that they ran through Google Cloud. "Without cloud computing," Maglio says, "it would be simply too cumbersome to process, build, and run computational models for social scientists. Cloud computing democratizes this process and accelerates results. Google Cloud's modular-based structure provides a very intuitive system that is both flexible and integrative."

Reviews in present tense most helpful

In two studies, Maglio and Fang leveraged Google Cloud tools for a thorough analysis of nearly a quarter of a million words in the Amazon reviews. They used the Natural Language AI for its advanced text analysis capabilities, extracting entities, assessing sentiment, and performing syntactic analysis on the extensive data corpus to extract covariates to add to their regression model. To handle the large volume of data efficiently, they utilized BigQuery for its powerful data warehousing and analytics capabilities. BigQuery facilitated fast and scalable data querying, storage, and analysis, enabling the researchers to manage the vast datasets of millions of Amazon reviews with ease. In the final step of their analysis, Maglio and Fang performed data analysis using Vertex AI Notebooks integrated with BigQuery, leveraging Google Cloud’s scalable and secure environment to run econometric analyses.

In three further experimental studies, they compared their computational linguistic analysis with reviews scored for helpfulness by 1,259 human respondents. The results confirmed that reviews written in the present tense proved more helpful and persuasive than those written in the past or future tense. They speculated that the present tense fosters psychological closeness and concrete mental representations that make the messages more effective for readers. In a related lab experiment, Maglio and Fang pitched a charity with different wording and measured how much money people donated. The pitch written in the present tense garnered the most donations.

In 2022, Maglio participated in the Google Cloud Research Innovator program, in which global researchers get access to Google experts, technical resources, a professional network for collaboration, and additional research credits. When presenting this work, he says, "the group offered critical feedback that helped to refine the project, especially with respect to expanding its scope."

The future of social science research

Maglio sees enormous potential in the use of AI in social science research: "for too long, the social sciences have been outpaced by the natural sciences in their adoption of big data analytics to provide answers to pressing questions. I see this research as an attempt to tip the scales back in the direction of balance. Recognizing that present tense verbs can make messages more persuasive by increasing their concreteness provides a valuable tool for anyone aiming to deliver impactful and influential communications."

Read on to learn more about the Google Cloud Research Innovator program or apply for Google Cloud research credits. To find out how you can get started with generative AI for higher education, download the new 10-step public sector guide. With domain-specific use cases and customer stories from the city of Memphis, the state of Minnesota, the U.S. Department of Defense, and more, it offers a comprehensive guide to kickstart your gen AI journey.