Correlation in popular content categories

We now look at the question whether users are interested in a similar set of content categories across the two services. To answer this we take the following approach: first, we classify notification messages and browsing accesses into different categories. (The details of categorizing notifications are described in Section 4.2, and the details of categorizing browse accesses are described in our earlier work [1].) Then for each individual user, we pick the top N content categories in browsing and top N content categories in notification (if the next few categories after the N^th category have the same frequency of access as the N^th category, we include those categories as well for the top N case). Figure 20 shows the percentage of users who have at least some overlap between their top N browse and notification categories. The degree of overlap is much higher when we consider wireless users only. For example, for the top 3 categories, the percentage of overlapped users is less than 10% when considering all the users, and around 50% when considering only the wireless users. On the other hand, even when considering wireless users only, the number of overlapped users is never more than 65%.

  

Figure 20: Number of users who have overlap between their top N browsing categories and top N notification categories.

We now compare the extent of the overlap by varying N from 1 to the total number of categories. The results are shown in Figure 21. The figure shows the average percentage of overlap between two categories, where the average overlap is computed as follows:

$$overlap_{high} = \frac{\sum_{i} \frac{\char93 categories \ o... ...apped \ for \ user_{i}}{min(N, min(BC, NC))}}{relevant \ users}$$

$$overlap_{low} = \frac{\sum_{i} \frac{\char93 categories \ ov... ...pped \ for \ user_{i}}{min(N, max(BC, NC)))}}{relevant \ users}$$

where BC denotes the number of browse categories, NC denotes the number of notification categories, and relevant users refers to those users that have at least one browse record and one notification record in the respective logs. We show the results for only the top 9 categories, since the values beyond that are stable.

  

Figure 21: Correlation between the number of browse requests and notifications of wireless users.

Essentially these ratios compute the percentage of overlap for each individual user, and then take the average of these percentages over all wireless users or all users. Since not all users have at least N browsing or notification categories, we compute overlap_high and overlap_low, where the former computes the percentage of overlap by using the minimum of BC and NC, and the latter uses the maximum of BC and NC. The figure shows that the amount of overlap is considerably higher when considering only wireless users. For example, for the top three categories, the overlap is less than 7% when considering all users. In comparison, for wireless users, the overlap_low and overlap_high values are 21% and 36%, respectively. We also observe that the effect of increasing N is small. Even when N is 8, the percentage of overlap is less than 50% for wireless users. The above results indicate that wireless users have moderate correlation in the way they use browse and notification services. In comparison, the correlation is much lower when considering all users. This is because the most popular browsing categories for desktop users are sign-up services, direction, and general help, whereas notification is usually not used to deliver these types of content. On the other hand, some wireless users are interested in both browsing and receiving notifications about emails, stock quotes, personalization, news and sports. However, the degree of correlation is limited, and service providers cannot solely rely on a user's notification profile to determine what content he/she may be interested in browsing.