Today, the World Wide Web has become the main source and medium for people to access, share, and manage information. Since user expectations towards all three types of functionalities are high and information volumes are growing very fast, modern web applications are exposed to new challenges by supporting the users in their daily and long-term interactions on the web. In this thesis, we contribute to the following core challenges related to the aforementioned functionalities. Diversification for improving information access - in Web search engines the user can access information by submitting a query that returns a set of search results. Web search queries often contain only a few terms, and can be ambiguous, which is a core issue for retrieval systems. For instance, modern search engines extract a large amount of additional features for building a sophisticated ranking model. Further, recent studies on web search results diversification show that retrieval effectiveness for ambiguous queries can be considerably improved by diversifying the search results. In this thesis, we present two approaches for improving retrieval effectiveness and efficiency. First, we present an efficient and scalable algorithm for web search results diversification for large-scale retrieval systems. Second, we present an approach for feature selection in learning-to-rank. Privacy issues and communication practices through information sharing - social networks allow the user to share information to a wider audience or communicate within specific groups. Understanding the users' motivation and behavior in social networks is crucial for supporting the users' needs, e.g. by suggesting relevant resources or creating new services. In recent years, the increasing amount of personal information shared in social networks has exposed users to risks of endangering their privacy. Popular social networks often allow the user to manually control the privacy settings of social content before it is shared. However, existing functionalities for privacy settings are often restricted and very time consuming for the user. In this thesis, we present an approach for predicting privacy settings of the user. Furthermore, we present an in-depth study of social and professional networks for identifying communication practices for different types of users with different skills and expertise. Personalized and long-term information management for social content - the information flood in social media makes it is nearly impossible for users to manually manage their social media posts over several years. Approaches for summarizing and aggregating of social media postings face the challenge to identify information from the past that is still relevant in the future, i.e., for reminiscence or inclusion into a summary. In this thesis, we conduct user evaluation studies to better capture the users' expectation towards information retention. Next, we extract various of features from social media posts, profile and network of the users. Finally, we build general and personalized ranking models for retention, and present a set of seed features which perform best of identifying memorable posts. The approaches in this thesis are compared to existing baselines and state of the art approaches from related work.