n order to develop a model to determine influence and trust of people by opinions of friends and objective metrics derived from the social network structure, the general rating behavior of My Secret Insights users is investigated. To find how ratings differ among different users, the standard deviation of ratings is determined. Finally, the relation between subcategories and the ratings for influence and trust is investigated.
Fig. 6 shows the cumulative distribution of the fraction of friends rated with a specific (a) influence and (b) trust score. Fig. 6(a) shows the cumulative distribution of influence ratings. The rating given is coded in the color and reaches from 1 (low influence) to 5 (high influence). The fraction of friends rated with low influence is highest on average. More than 10% of users pretend that none of their friends has an influence on them. For higher influence ratings the fraction of friends rated decreases. 90% of the users rate less than 10% of their friends with high influence, and 70% of the users rate none of their friends having a high influence. The results show that users pretend that they are not influenced by their friends. Most users assign only low ratings on the influence of friends. If higher influence ratings are assigned, they are only given to very few of their friends. The reason for this behavior could be that the users don’t want to admit that they are influenced. This behavior is clarified in Fig. 7(a), which shows the percentiles of the fraction of users rated dependent on the influence rating given. All percentiles decrease with higher influence ratings. Fig. 6(b) shows the cumulative distribution of trust ratings and Fig. 7(b) shows the corresponding plot for the percentiles dependent on the trust rating. On average the highest fraction of friends is rated with a medium trust score of 3. The fraction of friends rated with lower or higher trust scores decreases. On average least friends are rated with the lowest trust score of 1 which can be expected since friends should be able to trust each other. It is still remarkable that 60% of the users do not trust 10% of their friends. However, the percentiles are skewed towards higher trust ratings, which shows that users tend to trust their friends. The results can be explained that either users do not want to admit that they don’t trust their friends or that there is a base trust among friends in social networks.
Cumulative distribution of the fraction of friends rated with (a) influence and ...
Fig. 6.
Cumulative distribution of the fraction of friends rated with (a) influence and (b) trust rating 1 (low) to 5 (high).
Figure options
Percentiles for the fraction of friends rated for (a) influence and (b) trust.
Fig. 7.
Percentiles for the fraction of friends rated for (a) influence and (b) trust.
Figure options
In order to investigate if there are users that have a high influence, or that enjoy more trust than others, we study the ratings of users that received at least 10 ratings. Fig. 8(a) shows the distribution of influence ratings for the three users that received the most influence ratings. The users are coded in the different colors of the bars. The distribution of the blue and green user is similar; both are rated with a medium influence on average. The red user seems to have more influence, since it received influence score 4 most frequently and since it also has the highest share of score 5 ratings. This result shows that users have different influence. Fig. 8(b) shows the distribution of trust ratings for the 6 users who received most trust ratings. The 6 users can be divided in two groups. The dark blue, the turquoise, the orange and red user are highly trusted, receiving trust score 5 most frequently. The light blue and yellow user are less trusted, receiving mainly trust score 4 and also trust score 3 more frequently. Hence, there are users that enjoy more trust than others.
Ratings received of friends rated for (a) influence and (b) trust.
Fig. 8.
Ratings received of friends rated for (a) influence and (b) trust.
Figure options
To evaluate to what degree the ratings assigned and received by an individual user vary, the standard deviation of ratings is studied. Fig. 9(a) shows the cumulative distribution function of the standard deviation of ratings for influence and trust assigned and received. More than 10% of users assigned always the same rating for influence or trust resulting in a standard deviation of 0. The standard deviation of ratings assigned for trust tends to be higher than for received ratings. This might depend on the fact that users generally assigned low ratings for influence. The standard deviation of ratings received was evaluated for users that received at least 2 ratings. The standard deviation is 0 in more than 80% of the cases. This could depend on the fact the users agree on the trust and influence of a person. However the reason could also be the small sample size. Fig. 9(b