The unsolicited reviews people post on the internet are feedback gold to companies looking to enhance customer satisfaction and retention.

A team of researchers in computer science professor Hossein Fani’s lab combed social media sites looking for these reviews. However, they hit a roadblock when some reviews failed to explicitly mention the products or services being reviewed.

These types of posts are referred to as implicit, or latent, reviews. The LADy team, which derived its name from “Latent Aspect Detection,” had a trick up its sleeve: applying a social science solution to a computer science challenge.

The team used the art of language translation to improve their machine learning Artificial Intelligence coding.

Master’s student Farinam Hemmatizadeh leads computer science undergraduate Christine Wong and high school student Alice Yu as the LADy team.

“I really like the combination of these two areas and the way it makes it understandable for people with any level of technical knowledge,” says Hemmatizadeh.

“I’ve been working on natural language processing, going deeper into the linguistic part and features behind these languages to reach grammatical features that are behind these languages.”

She explains that existing methods fall short in teasing out the information. To bridge this gap the team used a method called natural language back-translation to augment the data. Essentially, they translated the text from English into another language and then back into English.

“Specifically, we presume that back-translation can reveal latent aspects by uncovering social knowledge between languages, generating context-sensitive synonymous aspects, and clarify semantic contexts of terms and sentences,” Hemmatizadeh says.

“We used 10-plus languages, including some forgotten languages, and found that some can have beneficial and positive impact on languages like English. If we translate to that language and come back it will be a richer review and come up with more words.”

She says humans would understand these latent reviews but machines will not pick up on it, which is why they employed back-translation to enrich the reviews.

Third-year undergraduate Wong will present the team’s findings at the annual conference of the Association for Computing Machinery Special Interest Group in Information Retrieval, July 14 to 18 in Washington, D.C.

She joined the project in 2023 and says she had no experience with natural language processing prior to joining the lab.

“I really like the interdisciplinary method of using computer science, which we often think of as a more rigid, binary sort of science, and combining it with linguistics, which we think of as more human, more social and often is a lot more complex to understand,” says Wong.

Hemmatizadeh says working with Yu, a Grade 12 student at Vincent Massey Secondary School, allowed for different perspectives and points of view. She says she contributed a lot to the team.

“I got involved in due to my passion for programming and interest in natural language processing,” says Yu. “Additionally, my background in digital art translated well to UX/UI design, enhancing the user-friendliness of the websites.”

The team plans to publish its findings after presenting in Washington.