Recommendation systems

On the 8 of April on the regular meeting of the Research Group “Machine Learning and Social Computing” members told about recommender systems and practical problems, which these systems are aimed to solve. Students, guests from different universities and professors of various faculties found this event very interesting.

At the beginning, the head of the research group Sirotkin A.V. introduced the base theory and simple examples of the recommender systems. He also mentioned the data collecting processes and creating similar systems as study projects.

The second part of the seminar was dedicated to contest ACM RecSys Challenge 2017. Some students started to solve the problem and shared their results and thoughts with colleagues. At first, Stanislav Kozlov told about the problem itself and the baseline solution, provided by the organizers. He also mentioned the problems the team faced trying to execute the script in Python environment. Then Ivan Mishalkin told about xgboost -the algorithm used in the baseline solution. Next Viktor Karepin and Anna Golovchenko (Laboratory of Sociology in Education and Science) told about their translation of Python script into the R code. Moreover, they succeeded in writing the evaluation metric used for the leaderboard, which is described in the problem description. This helped to evaluate the significance of new variables, that were generated by the participants.

Finally, the audience shared the ideas about the improvement of the base solution.