Types of Analyses

We now discuss the types of analyses that we perform on the notification and browse logs, and the motivations for doing these analysis.

1.

Content analysis: We are interested in questions such as: (i) what are the most popular content categories, and (ii) what is the distribution of message sizes? We believe such questions are important to (i) content providers who need to understand better how to prioritize and use the system and network resources efficiently, and to (ii) web site developers who are interested in supporting fast access to popular content.

2.

Popularity analysis: We are interested in the popularity distribution of notification and browse documents. In particular, we are interested in comparing these accesses to the well-known Zipf-like distribution as reported in previous web studies [4,7,10,14,16], and in determining how concentrated are the number of requests/transmissions for popular documents. This has significant implication for the effectiveness of web caching and multicast delivery.

3.

User-behavior analysis: We are interested in classifying users according to their access patterns. This is useful for personalization, targeted advertising, prioritizing, and capacity planning. Specifically, we look at the following aspects of user behavior:

• Spatial Locality: whether users in the same geographical region tend to receive/request similar notification and browsing content.
• Temporal Stability: whether users are interested in browsing similar documents over time.
• User Load Distribution: how different users place load on the web site; for service providers, this distribution has implications on pricing.