Code and data used in my dissertation and in articles that spawned from it.

This file contains code and data employed in the article "Effect of item representation and item comparison models on metrics for surprise in Recommender Systems", presented at ICEIS 2019. The raw data was generated by an offline experiment performed on an extended version on the MovieLens-1M dataset that includes textual description of movies (file movie_descriptions.dat). The scripts in R (files anova_2by6.R and anova_4by2.R) were employed to analyse the target effects using a two-way, repeated measures ANOVA on the data obtained from a factorial experiment. The experiment comprises 56 configs, which vary in recommendation algorithm, item representation and item comparison.

It should be noted that the raw data is contained in the dataframe_m1_data.dat. However, the experiment follows a factorial design, and thus the data was split into groups. The splitted data was also made made available in files named as dataframe_m1_g<group#>.dat. Each of these files has also been made available as a dataframe that can be loaded both in R and Python (files named as dataframe_m1_g<group#>.pkl). The images represent interaction diagrams generated for each group and contrast (files named as interaction-<2by6 or 4x2>-l<contrast#>-g<group#>.png). Please refer to the article for details about the experiment design, the statistical analysis that was applied, or the final results and findings.