Page 101 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)

Thanh Trung Nguyen, Hai Bang Truong 85

appropriate to calculate a mean or median as these measures are typically used for
numerical data. Instead, the mode can be used to identify the most common category
or word in the dataset.
The mode works well as an aggregation method when individual predictions con-
sist of one word and the crowd performs well [26], [27]. However, when it comes to
ranking a larger number of items, other methods may be more suitable, such as using
a scoring system or conducting pairwise comparisons. In their article [28], the authors
reported that their Bayesian version of a Thurstonian model, which aggregates
orderings across individuals, is more effectively (Rank = 87.0) than the mode
(Rank=68.2) and other aggregation techniques (Fig. 4).

Fig. 4. Performance of the four heuristic models and the Thurstonian model [25]

4.6 Other aggregation methods

As data becomes more complex and diverse, simple aggregation methods like taking the
mode or mean of responses may not be suitable or may lead to biased or inaccurate
predictions. In such cases, more sophisticated aggregation methods may be needed to
handle the complexity of the data and make more accurate predictions. However, these
methods are more context-specific and not as commonly employed [29].

5 Conclusions and Future Works

This article presents a summary of the various aggregation methods used in the field
of Wisdom of Crowds. Despite the growing number of studies involving large and
complex data, the conventional aggregation methods, including arithmetic, geometric,
weighted aggregation, and mode, remain prevalent in the Wisdom of Crowds domain.
Overall, the choice of aggregation method depends on the nature of the data and
the goals of the study.
Our plan for future research is to investigate how social influence and aggregation
methods relate to each other, with the aim of identifying the optimal range of

ISBN: 978-604-80-8083-9 CITA 2023