To validate the effectiveness of the proposed social context endorsement mechanism for online advertising, we used the CTR of advertisements and user feedback as the performance indicators. The former is the most popular practical statistical measure for evaluating advertising effectiveness, and the latter allowed us to evaluate the improvement in the impressions made by the advertisements the target users received.
For the purpose of performance comparison, in Experiment 1, besides the proposed social context endorsement advertising strategy (denoted by Ads+SC), we chose three other types of endorsement approaches as bench-marks: bare banner advertisements (denoted by AdsOnly), advertisements with the number of fans on the fan page (denoted by Ads+#fan), and advertise-ments with the names of friends who were also fans (denoted by Ads+name).
The first one is the most common type of advertisement, and the others are usually used by Facebook. In Experiment 2, we further compared the perfor-mance of our proposed social context endorsement approach with three other types of endorser-finding approaches (random, content-based, in-degree).
Click-Through Rate
The CTR of online advertisements can be calculated as the ratio of the number of users who clicked on an advertisement to the number of times the adver-tisement was delivered. This is defined as
Downloaded by [National Chiao Tung University ] at 02:19 21 July 2015
CTR Total # of clicks Total # of ads delivered
= . (11)
The results of the CTR with respect to various endorsement approaches (Experiment 1) are shown in Figure 7. We observed that by adding social information to the advertisements, people become more willing to click on the advertisements (advertisements with number of fans, advertisements with names of friends, advertisements with social contexts). The advertisements with number of fans and those with names of friends obtained a similar CTR.
The results indicate that the effect of knowing why someone loves a product is stronger than only knowing who loves the product.
A paired sample t-test was used to statistically verify the significance of the difference in the advertising results (see Table 2). At the 95 percent signifi-cance level, all the test results showed that the strategy “social context” was significantly different, at 0.05, in relation to the other strategies. Therefore, this proves that our proposed strategy outperforms other strategies.
The CTR results in Experiment 2 are shown in Figure 8. These show that our proposed social context endorsement discovery mechanism outperformed other benchmarks, including the random approach, content-based approach, and in-degree approach.
Table 3 shows the statistical verification of the results, which confirms that our proposed mechanism outperforms other benchmark approaches at a significant level. The meaning of the token SC in Table 2 is the same as that of Social Context in Table 3. In Table 2, we consider the pattern of advertise-ment: Ads+SC is the advertising pattern we proposed. In Table 3, we consider the comparison between different approaches used to discover the fittest endorsement.
Our category tree comprises three main categories, and each main category has six leaf categories. Hence, 36 (6 × 6) advertisements were shown for each main category (e.g., entertainment and living, electronics and computers, con-sumer products). The results of the CTR with category of concon-sumer products, electronics and computers, and entertainment and living are 3.721, 3.987, and 3.896, respectively. It can clearly be seen that the social context endorsement mechanism has a better result in the electronics and computers category Figure 7. CTR Performances of Different Advertisement Patterns
Downloaded by [National Chiao Tung University ] at 02:19 21 July 2015
than in other categories. This is mainly because mobile devices, such as the smartphone and tablet, have been so popularized that they are often held to be indispensable to daily life. The mobile environment and pervasive comput-ing technology promote the development of products in the electronics and computers category. According to a report by Nielsen [30], over 54.9 percent of U.S. mobile subscribers now own smartphones, and approximately 70 percent of Americans who acquire new phones choose smartphones instead of feature phones; people are more interested in and pay more attention to these new products.
Table 2. Statistical Verification of Results of Experiment 1 Based on CTR Measurement.
Paired group Mean Standard deviation
Standard error
mean t Sig.
(two-tailed)
Ads+SC AdsOnly −3.2547 0.53566 0.03679 −8.847 0
Ads+#fan −0.20283 0.51645 0.03547 −5.718 0
Ads+name −0.13208 0.51656 0.03548 −3.723 0
Figure 8. CTR Performance for Different Endorsement Discovery Strategies
Table 3. Statistical Verification of Results of Experiment 2 Based on CTR Measurement.
Paired group Mean Standard deviation
Standard error
mean t Sig.
(two-tailed) Social
context
Random −3.13494 0.52186 0.03278 −7.517 0
Content-based
−0.31011 0.51684 0.03348 −4.524 0
In-degree −0.1347 0.51677 0.03179 −3.781 0
Downloaded by [National Chiao Tung University ] at 02:19 21 July 2015
Advertisements
Even though people are interested in the advertisements they receive, it is still quite likely that they will not click on the advertisements because of concerns about privacy and potential fraud [4]. Nonetheless, it is still of value to companies to increase the impression made by advertisements. Therefore, we also recorded the impressions of users in relation to different versions of advertisements and used the impression feedback of users as an evalua-tion measurement. The initial impression is the impression evaluaevalua-tion of the original ad (i.e., AdsOnly). In Experiment 1, we recorded the impressions of participants about the received advertisements expressed in four different formats. Those users who received the ads were asked to score their impression of the ads before and after watching three kinds of different attachments. The average impression scores of the various advertisement formats are shown in Figure 9. We can see that, with social context, the impression score of the advertisements increased to 4.2398 from the initial 3.1319 (social context–free advertisements).
The results for impression with category of consumer products, electron-ics and computers, and entertainment and living are 4.2539, 4.427, and 4.391, respectively. It is clearly found that the social context endorsement mechanism achieved a better result in electronics and computers than in other categories.
The advent of social media has created a fundamental shift in human behavior:
Sociable Labs [35] has shown that 38 percent of online shoppers have shared comments with friends about products that they have purchased, and 62 per-cent of have read product comments shared by their friends on Facebook. Our experimental results show that the social context endorsement mechanism has a better CTR in the electronics and computers category than in other categories.
This outcome is consistent with the Nielsen Global Consumer Confidence Survey (2013) findings that the Internet is an important influence on consum-ers interested in buying new products, and the effects on the purchasing of electronics (81 percent) is the greatest.
We further separated the performances according to the initial scores to see the trend in relation to improved impressions, as shown in Figure 10. The results show that for those who initially disliked the advertisements, adding social context could enhance their impression of the product. The advantage of Figure 9. Product Impressions for Different Advertisement Patterns
Downloaded by [National Chiao Tung University ] at 02:19 21 July 2015
social context is particularly significant for an advertisement with an initially low impression.
In Experiment 2, we undertook the same evaluation of impression with respect to different social context endorsement discovery methods. As shown in Figure 11, social context–endorsed advertisements received a better impres-sion score than other methods. The results also indicate that the “endorser”
factor is more influential than the “opinion” factor, as the in-degree approach outperformed the content-based approach.
As Figure 12 shows, we found that the social context endorsement approach based on our proposed model obtained feedback with the highest level of improvement in the impression of the product. One interesting phenomenon is that the in-degree approach outperformed the content-based approach when the advertisements were of less interest to the target users. However, the opposite result was found when the advertisements were of greater interest to the target users. This coincides with the conjecture that if a person is not interested in the advertisements, the person who introduces them is more important than what is said. Yet, for those advertisements, the endorser is not as influential as the content of the opinions and advertisements.
Figure 10. Increase in Ratio of Impressions in Experiment 1
Figure 11. Advertisement Impressions for Different Advertisement Patterns
Downloaded by [National Chiao Tung University ] at 02:19 21 July 2015
Conclusions
The power of electronic word-of-mouth and viral marketing is inducing mar-keters to move away from their traditional marketing behaviors into social media, not only for the purpose of getting closer to their customers but also to increase their business exposure and consumer buying desires. There are two strategies commonly used to launch an advertising campaign on social media—social advertising and targeted advertising. Social advertising aims to leverage the influence of social opinion leaders in distributing information quickly and effectively along the networks.
With the overwhelming amount of content generated daily in social media, it is almost impossible for people to read each message that is created and distributed. However, a targeted advertising approach displays advertise-ments to potential customers according to the content they are viewing or the actions they perform; still, people can only receive relevant information and willingness to purchase cannot be enhanced significantly.
Besides, people tend to think of advertisements as spam and refuse to accept them even if an advertisement is nicely matched with user preference.
A more sensitive advertising system should exploit the advantages of both social and targeted advertising approaches. In this research, utilizing the power of social influence and context embellishment, we developed a social context endorsement model to discover a close acquaintance who had posted positive comments on the product advertised, which were then leveraged as a social context endorsement for the advertisements. Our experimental results show that the proposed advertising approach of social context endorsement can effectively enhance the attention of users and increase both the CTR and the impression of the product.