Databases Reference
In-Depth Information
relationships between the different factors. In particular, we will analyze the impact of eciency, effec-
tiveness, and transparency on user satisfaction since satisfaction is often used as a dependent variable in
recommender system research [Cremonesi et al., 2011].
6.4.5 Relationships between variables
The user-centric evaluation of the causal relationships between different evaluation factors is a recent
topic within recommender system research [Pu et al., 2012]. We conducted a
path analysis
to analyze the
relationships between the different quality factors for recommender system explanations. SPSS AMOS
20 was used for model building and path analysis. Path analysis was done using regression analysis,
which is a standard approach in social sciences and in educational research [Tuijnman and Keeves, 1994]
to study of potential causal relationships.
Figure 6.7: The SPSS AMOS path analysis model which describes the dependencies among the variables.
Figure 6.7 shows our model which we used as input for the SPSS AMOS path analysis tool. Each
quality factor is represented by an independent or dependent variable in the model. We see satisfaction
as the dependent variable which depends on the quality factors eciency, effectiveness, and transparency,
which are consequently modeled as independent variables. The edges between the independent variables
represent the covariance parameters to estimate. We conducted a path analysis for each explanation
interface separately. The results of each path analysis run are shown in Table 6.7 and Table 6.8.
transparency eciency
effectiveness
R
2
↓
satisfaction
clusteredbarchart
0.760
-0.003
0.108
0.350
barchart
0.773
-0.005
0.144
0.467
neighborsrating
0.547
0.001
0.112
0.187
confidence
0.509
-0.002
0.066
0.254
0.473
0.008
0.171
0.467
neighborscount
rated
4+
0.674
0.016
0.302
0.469
0.482
0.010
0.235
0.339
average
0.888
0.003
0.237
0.510
tagcloud
0.883
0.002
0.238
0.505
perstagcloud
0.705
-0.011
0.317
0.330
piechart
∅
0.669
0.002
0.193
0.388
Table 6.7: Maximum likelihood estimates of the regression weights. Bold figures indicate weights with a
significant effect (
p<
0
.
001,
N
= 291).
