#	Pagina
attuale pagina	/open-h2020/projects/210539/results.html

Report

Teaser, summary, work performed and final results

Periodic Reporting for period 1 - MDDS (Mechanism Design for Data Science)

Teaser

Summary

The project has 4 central themes: re-vising search and information retrieval, competition in machine learning (segmentation / regression), re-visiting on-line learning and explore & exploit mechanisms, and incentive compatible diffusion / influencer selection in social networks. The common theme is a game-theoretic / mechanism design approach to the most basic challenges in data science.

A common theme established is around a game-theoretic approach to (broadly speaking) recommendation systems. The different works deal with incentives of the different participants in the recommendation systems, the content owners, the users, and the platform itself when competing with other platforms.

Good examples for uses of our novel game theoretic approach: to address the publisher incentives in search, to propose a regression algorithm for prediction in competitive context, to present algorithms for incentive competitive diffusion in social networks, to reduce recommendation systems to mediated facility location problems in service of dealing with participants\' incentives.

1.Search and information retrieval: relevance ranking of published information, providing users with effective information retrieval for their needs.

2. Wisdom of the crowd and on-line recommendations:exploiting usersâ€™ inputs and experience for optimization of the overall performance of recommendation systems.

3. Targeted data mining and segmentation: clusteringof users with the aim of providing them with targeted offers and campaigns.

4. Exploiting social networks: analysis of socialnetworks with the aim of effective information diffusion using e.g. selected influential users.

Work performed

The first 18 months of the projects have been instrumental to deploy all aspects of the project. Roughly speaking the progress made was amazing given the high bar. We stablished a group of 3 PhD students and 3 MSc student. One of the MSc. students then transferred to the PhD program. Moreover, during the establishment of the group we managed to attract quite a few faculty members to engage and collaborate with us on the topics. This created a second to none group, which its structure and achievements are detailed in mdds.net.technion.ac.il We also had one monthly visitor. A post doc joined on 1.1.2019.

Significant process has been obtained in all directions. Among the 12 publications that have been published, let me emphasize that 8 publications appeared in data science / AI inclined forum (2 NIPS, 2 WWW, 2 AAAI, 1 SIGIR 1 JAIR), and 5 at game theory / OR / Econ inclined outlets (2 MOR, 1 GEB, 1 WINE, 1 TEAC).

Final results

We are excited about the progress made. The next 18 months are planned to harness the initial success. We are now in the beginning of generating a setup for validating our approaches, mainly in the context of search and information retrieval. We plan to extend our work on competition in ML to full equilibria analysis. We plan to publish results on incentive compatible explore in on-line learning. We aim at improving upon the algorithms for incentive compatible diffusion. We also plan to dig more into different user models and conduct a study on how to influence a user while collaborating with NLP and behavioural economics researchers. If the above is accomplished, we will be in remarkable position in bridging game theory and data science; it is definitely the case the first 18 months were more than instrumental in this regard.