Univerza na Primorskem Fakulteta za matematiko, naravoslovje in informacijske tehnologije

Ponedeljkov seminar računalništva in informatike - Arhiv

2022 2021 2020 2019 2018 2017 2016 2015 2014 2013 2012
1 2 3 4 5 6 7 8 9 10 11 12

torek, 16. avgust 2022 Arsen Matej GOLUBOVIKJ: Imputing Missing Answers In The World Values Survey

V torek, 16. avgusta 2022, bo ob 14.00 uri prek spletnih orodij na daljavo izvedeno
Oddelkov za Informacijske znanosti in tehnologije UP FAMNIT in UP IAM.

ČAS/PROSTOR: 16. avgust 2022 ob 14.00 na daljavo


Arsen Matej Golubovikj is a second-year student of Data Science at UP FAMNIT. He completed his Bachelor's degree in Computer Science at UP Famnit in September 2020, and is now working towards his Master's degree. He will be presenting the topic of his Master's thesis that he is completing under the mentorship of assist. prof. Branko Kavšek and prof. Marko Tkalčič.

NASLOV: Imputing Missing Answers In The World Values Survey


Many areas of science, in particular social sciences, use questionnaires to gather data. The process of collecting data through a questionnaire, called a survey, is often the prime means of gathering data directly from participants, however, it's prone to missing data. In order to keep the full survey sample, researchers must often use imputation to deal with this problem. Methods for imputation can sometimes offer reasonable estimates for the missing data, however, in the case of the survey: (i) imputation can add high noise to the data, which influences the inference, (ii) imputation becomes unreliable when more than 40% of the data is missing.
This thesis attempts to address these issues by evaluating if the usage of methods stemming from collaborative filtering (CF) in recommender systems can yield more accurate imputations of missing values in survey data. The rationale for the usage of these methods is (i) the similarity between the problem framing, methods and data representation used in CF and questionnaire imputation; (ii) the effectiveness of CF-based methods in recommender systems, especially in cases where much of the data is unavailable.
We use data from the World Values Survey, a valuable dataset in social science of high volume and veracity, to compare (i) one simple approach to imputation, (ii) two established imputation approaches (iii) two matrix completion techniques stemming from collaborative filtering.The results show that our chosen matrix completion techniques stemming from collaborative filtering perform comparable, but not better than existing imputation techniques in the case of the survey. The right technique for imputation often depends on the problem, these results beckon the consideration of CF-based techniques in future research on survey imputation.

Seminar bo tokrat izjemoma v torek od 14 uri in bo potekal v angleškem jeziku v Zoom "klepetalnici" na naslovu: