Enhancing User Engagement by Profile-Matching-Algorithm in a Social Network Platform using AI


Traditionally, mentorships are provided based on titles, wisdom, hierarchy, and status. Tribute platform believes that wisdom is gained from the lived experience in and outside the workplace. 

Tribute provides a technically delivered, just in time mentorship platform that adds on continuous learning organization serving everyone in need.

Good mentor recommendation on such platforms not only saves time for those who are looking for guidance but also improves user experience and engagement. Tribute differentiates itself in this regard. Generally, mentor-mentee connection platforms want users with similar areas of expertise to engage and share knowledge. But Tribute Mentorship pushes this a step further to recommend mentors based on their similar stories and life experiences in addition to their areas of expertise. You can visit the Tribute platform HERE

We collaborated with the Tribute team to build a POC recommendation engine in 6 weeks.

The Problem: Finding the Right Mentor for the Mentee

User engagement is one of the biggest challenges for any platform. Moreover, when a platform connects people and has no good recommendations, it is a harder and daunting task for the user to search and find a good match. For Tribute, whenever a new mentee signs up on the platform, there are numerous mentors a mentee can reach out to. While it is an opportunity to search for mentors in the large pool, it is also a tedious process to find out the perfect mentor who has similar expertise and experiences the mentee is looking for. In addition to it, users might have a negative experience if they continuously fail to search for good mentors when having no good recommendations. 

Our Approach: Finding Top-K similar profiles using a language model

There were already rule-based filters implemented in tribute, like demography, gender, etc. Leapfrog’s AI team and Tribute collaborated and used different mentor/mentee features to develop an AI-based mentor recommendation system. The developed AI model analyzes and finds the best match with higher confidence.

When a new client joins the platform, certain fields are mandatory to fill up. Tribute platform asks the user their preferences like timezone, demography, gender, etc., before getting a recommendation from an AI model. Initially, we apply the rule-based filter to select the mentor who matches the preferences of the mentee. Features of selected candidates now go through the AI model and analyze the similarity between candidates with mentee life experiences. 

The Algorithm: Building a faster inference ML model

Because life experiences are textual data, we initially developed a larger language model to find the similarity between life experiences. Upon iterative training and tuning the parameters of the model and manually analyzing the prediction made by the AI model, we now have the model which predicts the similarity of life experience of users with higher confidence. Because we used the more complex and larger AI model to find similarities, the embedding representation of users was huge. To deal with this issue, we developed another AI model that could shrink those longer embeddings into a shorter one. We carefully designed and analyzed the recommendations we got after using the shrunk embedding. We used the cosine distance between the embedding representation of users’ experiences to find the similarity scores.

Results: Connecting similar life experiences for mentor/mentee

The AI-based recommendation engine is deployed in a Tribute platform. We can get the suggestion of mentors who are close to our experience and can connect with them. With similar life experiences, mentors and mentees can empathize with each other and push each other for a better future. 


Our AI platform depends upon the profile that users input in their bio. Some users might randomly enter incorrect information into their profile. In those cases, a recommended mentor might not be the one user is looking for. The other challenge is, if the number of candidates selected after using basic filters like time zone and gender preference is below the threshold, we don’t apply the AI model recommendation. In this case, you might be recommended with different life experiences mentors.

Author: Sushil Ghimire | Check him out on Linkedin and Github.

Sushil Ghimire

Sushil Ghimire is a Machine Learning Engineer at Leapfrog Technology. He is proficient at NLP, Recommendation system has equal interest in Data Pipelining, MLOps and data structures.

More in Blogs

Age of Personalized Healthcare Artificial Intelligence

Age of Personalized Healthcare

With the proliferation of technology, crucial components of healthcare are evolving. Suddenly, patients are turning into “Digital Consumers”, medicines are

Read more
Handling Cold Start Problems in Recommender Systems Artificial Intelligence

Handling Cold Start Problems in Recommender Systems

Wikipedia defines a cold start problem as a potential problem in computer-based information systems which involve a degree of automated

Read more
Optimal Vaccine Allocation Using Recommendation-as-a-Service Artificial IntelligenceInsights

Optimal Vaccine Allocation Using Recommendation-as-a-Service

Given the global nature of the Covid-19 crisis, a high degree of coordination is needed among global bodies such as

Read more