Studo/Talto Job Matchmaking Platform

Jun 1, 2017
Studo Job Matchmaking Platform

Project Overview (June 2017 – September 2019)

At Know Center Research, we partnered with Studo, a widely-used student app in Germany and Austria, to develop a personalized recommender system for their job listings feature. The successful solution later evolved into the dedicated platform “Talto – Talents of Tomorrow,” significantly enhancing the student job application process through intelligent matchmaking.

The initial project phase (“Studo Matchmaking Recommender – Empfehlungs Plattform personalisierter Job Anzeigen für Studenten”) ran from June 2017 to May 2018. Following its success, we secured additional funding from the Austrian Research Promotion Agency (FFG) for the follow-up project (“FAT – Fair, Accountable & Transparent Matchmaking: Explainability, Sessions, Scoring and Biases for Passive Sourcing”), which took place from October 2018 to September 2019.

Technical Challenges

  • Defining meaningful and varied recommender use-cases for student-job matchmaking
  • Developing custom real-time recommender algorithms to handle diverse data and contexts
  • Ensuring scalability and seamless integration into the Studo app infrastructure
  • Addressing fairness, transparency, and explainability in recommendations

Technologies & Methods

  • Core Development: Java, Apache Solr, Microservices with ScaR recommender framework
  • Machine Learning & Algorithms: Python, Doc2Vec embeddings, Autoencoders, Vector Search
  • Deployment & Testing: Docker, Continuous Integration, Extensive Testing
  • Research & Documentation: Scientific publication writing, comprehensive documentation

Results & Impact

  • Improved user engagement with the Studo job marketplace (~20% increase)
  • Increased application rate to job postings (~18% increase)
  • Conducted pioneering research on matchmaking algorithms tailored for student and graduate job markets
  • Enabled high-quality personalization even for anonymous users and additional content recommendations (e.g., news articles)
  • Produced multiple impactful academic publications
  • Featured as an FFG success story (full story here)

Selected Publications

@article{lacic2017beyond,
  title={Beyond accuracy optimization: On the value of item embeddings for student job recommendations},
  author={Lacic, Emanuel and Kowald, Dominik and Reiter-Haas, Markus and Slawicek, Valentin and Lex, Elisabeth},
  journal={arXiv preprint arXiv:1711.07762},
  year={2017}
}

@inproceedings{reiter2017studo,
  title={Studo Jobs: Enriching Data With Predicted Job Labels},
  author={Reiter-Haas, Markus and Slawicek, Valentin and Lacic, Emanuel},
  booktitle={CEUR Workshop Proceedings},
  volume={2025},
  year={2017},
  organization={RWTH Aachen}
}

@inproceedings{lacic2019should,
  title={Should we embed? A study on the online performance of utilizing embeddings for real-time job recommendations},
  author={Lacic, Emanuel and Reiter-Haas, Markus and Duricic, Tomislav and Slawicek, Valentin and Lex, Elisabeth},
  booktitle={Proceedings of the 13th ACM conference on recommender systems},
  pages={496--500},
  year={2019}
}

@article{lacic2020using,
  title={Using autoencoders for session-based job recommendations},
  author={Lacic, Emanuel and Reiter-Haas, Markus and Kowald, Dominik and Reddy Dareddy, Manoj and Cho, Junghoo and Lex, Elisabeth},
  journal={User Modeling and User-Adapted Interaction},
  volume={30},
  pages={617--658},
  year={2020},
  publisher={Springer}
}

@inproceedings{reiter2020heterogeneous,
  title={On the heterogeneous information needs in the job domain: A unified platform for student career},
  author={Reiter-Haas, Markus and Wittenbrink, David and Lacic, Emanuel},
  booktitle={Proceedings of the 14th ACM Conference on Recommender Systems},
  pages={573--574},
  year={2020}
}

Personal Contribution

  • Managed a team of five, overseeing project execution, requirements engineering, and budget planning
  • Actively contributed as a Software Engineer, ML Engineer, and Data Scientist to the development and implementation of recommender algorithms
  • Coordinated deployment and continuous improvement of the platform
  • Authored academic publications and secured additional project funding from FFG
  • Co-advised academic work including one Master’s thesis and supported the onboarding of one PhD student
Recommender Systems Matchmaking Machine Learning Real-Time Recommendations Data Pipeline Vector Search Project Management User Modeling Embeddings
Tomislav Đuričić
Tomislav Đuričić
Researcher / Machine Learning Engineer / Software Engineer

My research interests include social-based recommender systems, graph neural networks and user modeling.

Publications

Should we embed? A study on the online performance of utilizing embeddings for real-time job recommendations
In this work, we present the findings of an online study, where we explore the impact of utilizing embeddings to recommend job postings under real-time constraints. On the Austrian job platform Studo Jobs, we evaluate two popular recommendation scenarios - (i) providing similar jobs and, (ii) personalizing the job postings that are shown on the homepage. Our results show that for recommending similar jobs, we achieve the best online performance in terms of Click-Through Rate when we employ embeddings based on the most recent interaction. To personalize the job postings shown on a user’s homepage, however, combining embeddings based on the frequency and recency with which a user interacts with job postings results in the best online performance.
Emanuel Lacić, Markus Reiter-Haas, Tomislav Đuričić, Valentin Slawicek, Elisabeth Lex
PDF Cite Project DOI
Should we embed? A study on the online performance of utilizing embeddings for real-time job recommendations