Information Technology Reference
In-Depth Information
Results
Right now, we have 6 subjects doing 100 trials. Each trial includes making ternary
choice from 80 pieces of garments from the Database. Then we do the model fitting.
Finally we use the parameters to predict the preference of the user for 80 garments
generated by computer system. Experiments' results show that our model has a higher
mean accurate prediction rate (90%) of the user's choice than the general IGA ones
(86%).
4 Discussions and Conclusion
The model proposed in this paper incorporates both objective feature space searching
and evaluation and subjective feeling space deliberating and evaluation. The case study
on fashion decision making shows that it outperforms the generally used IGA. Through
careful study of the decision process and the groups of garments for preference choice,
we can find there are two reasons lead to the relatively poor performance of IGA.
GA is based on a global search and global evaluation idea, while due to both the
limitation of the human brain process and the computer screens, we only offer 3
each time, which means choice alternatives are mainly compared within 3 with
some left memory of the earlier comparison result added to the final preference.
The existence of those kinds of 'psychological effects' generally happened in mul-
ti-alternative human decision making, such as similarity effect, attraction effect and
compromise effect, especially the similarity effect and compromise effect. Whe-
rever these two kinds of effects exist in groups, the results will lead to the conflict
with the IGA result, while MDFT is designed to explain and predict these effects.
To sum up, by the integration of genetic searching algorithm with MDFT, we can
successfully account for both the decision time and accuracy of the computer based
personal decision making process in the mean time.
Compared with the general decision making system, by incorporating the subjec-
tive feeling part using MDFT, we can:
Clear identify the subjective factors—better understanding of the decision process,
and will lead to a scalable and adjustable decision making model.
Account for several classical irrational effects.-similarity effect, attraction effect
and compromise effect, thus lead to a decision result that is more human like and
can predict more accurately.
Less interaction. -after model fitting, the proposed architecture needs less human
interaction, than the IGA based intelligent fashion design system.
Compared with the mere cognitive dynamic decision making models for multi-
alternative decision making, like MDFT, instead of using simple heuristics like elimi-
nate by aspects, we introduced the optimized searching strategy to help quickly find
the optimal solution in a larger objective feature space.
Search WWH ::




Custom Search