YPIE Scientist: Mircely Rodriguez

Research: Recommending People Potential Friends with Similar Interests using Vector Space Similarity Metrics

Mentor: Sol Vitkin

Research Location: Yonkers, NY

Awards: Participant in Somers Science Fair (2024), Participant in Westchester Science and

Engineering Fair (2025 & 2026)

Abstract:

According to Heinrich and Gullone (2006), 70% of teenagers face recurring loneliness

by age 18. This often stems from difficulties in making new friends with whom they feel a

genuine connection. These challenges can be due to many reasons, such as feeling

misunderstood, different energy levels, communication barriers, etc. Luckily, the internet has

become a tool for meeting new people. While it allows individuals to easily meet people online, the question arises: how can individuals find potential friends with whom they would form a genuine connection? Collaborative filtering can be a valuable tool in solving this problem. Other researchers have attempted to address this issue by using collaborative filtering techniques to recommend their participants’ possible friends on social media. They used data from participants’ profiles on social networking sites and went based on that to recommend friends. This paper differs from that as it seeks to recommend participants’ potential friends, specifically utilizing vector space similarity metrics, a type of collaborative filtering. In addition, the input for this experiment came directly from the participants themselves, which ensured more personalized recommendations to each individual. This approach conducted two self-report surveys, one about participants’ interests and the other about their sense of humor. Participants provided 2-3 sentences of their opinion on the given recreational activity or meme (n = 6 respondents). The data was then used to calculate cosine similarity between participants. Each respondent was informed of their nearest vector and farthest vector, without knowing which one was which. They were then set up to have a virtual conversation, and both completed a post-conversation survey. In that survey, individuals completed a self-report 10-point Likert scale survey assessing the conversations they had, and a short open-ended response to their experience meeting that person. This survey asked questions like “Did you feel comfortable?”, “Was there a connection?”, “Would you meet again?”, “Do you see yourself becoming friends with this

person?”. A linear regression model was fit to the data, modeling the relationship between cosine similarity and the number that individuals rated on the post-conversation survey for “Do you see yourself becoming friends with this person?”. When modeled, there was a weak relationship between these two variables (slope = 1.394103635 and R = 0.06463899352). These results do 2 not support the hypothesis that a higher cosine similarity increases the likelihood of individuals becoming friends. In conclusion, vector space similarity metrics were useful in uniting individuals with similar interests but not for making new friends.

About this Scientist:   

Mircely Rodriguez is a senior who is currently pursuing her IB Diploma at Yonkers High School. She is interested in learning new things, taking on academic challenges, and making a difference in people’s lives. She will be attending Lehman College as a Speech Pathology and Audiology major in the fall to help others communicate more effectively and confidently.

Presentation Slides