Suchergebnisse

Evropski parlament je politično telo, ki že dolgo časa razdvaja javnost. Njegovi nasprotniki po eni strani trdijo, da je tako politično in kulturno heterogen organ nezmožen učinkovitega delovanja, po drugi strani pa naj bi evropski poslanci hitro izgubili stik s svojimi volivci ter glasovali predvsem tako, kot jim to naročijo vodje skupin. Zanimalo nas je, če evropski poslanci glasujejo dovolj avtomatično in predvidljivo, da bi lahko rezultate glasovanj uspešno napovedovali z modelom strojnega učenja. Da smo zmanjšali časovno kompleksnost smo se odločili za napovedovaje rezultatov glasovanj po političnih strankah. Najprej smo implementirali spletne pajke, s katerimi smo pridobili čim več rezultatov glasovanj in z njimi povezanih podatkov. Iz teh smo nato z interdisciplinarno kombinacijo metod podatkovnega rudarjenja ter strokovnega geopolitičnega znanja izluščili značilke in zgradili model. Rezultati so pokazali približno 80% uspešnost napovedovanja rezultatov glasovanj (uteženi oceni f1 in roc-auc). Napovedovanje je bilo bistveno bolj uspešno pri strankah s proevropsko,liberalno in globalistično politično usmeritvijo kot to velja za evroskeptične, ekonomsko socialne in nacionalistične stranke. S tem smo dodatno podprli tezo o obstoju in pomembnosti nove nacionalistično-globalistične politične delitve. Poleg tega predstavlja problem tudi razred vzdržanih glasovanj, ki ga je težko napovedati že s pomočjo človeške inteligence. ; European parliament is the major political legislative body of the EU that causes divides in public opinion since its beginning. While some of its opponents usually point out its political and culutural heterogenity as a major weakness and cause of inability to function eciently, others claim that MEPs often quickly lose their connection with voters and vote mainly as it is directed to them by the leaders of their politival groups. We wanted to nd out if MEPs voting patterns are predictable enough to be successfully predicted with machine-learning based computer model. To reduce time complexity of the problem we rather focused on joint votes of (national) political parties than individual MEPs. At rst we implemented web crawlers that we used to extract as many roll-call voting oriented data as we can. Than we combined data mining with expert geopolitcal approach to extract the features and build a model for voting prediction. Our predictions were overall nearly 80% successful (weighted f1, roc-auc), however results vary greatly between political groups. It became clear that we could easily predict votes of coallition parties with liberal-globalist political orientation while eurosceptic, economic social and nationalist parties seemed to be much more unpredictable. With that information we further backed the importance of the new nationalist-globalist political cleavage. A challenge to the model presents also a class of votes of abstention, which is hard to successfully predict even with expert human-knowledge.