Advances in Understanding, Mining, and Using People-Tags

Abstract

People-tagging involves the process of adding non-hierarchical metadata to users of a system. Such metadata facilitates organising contacts and building user profiles in a collaborative fashion. People-tag-based user profiles can be used in use cases such as finding people with relevant expertise and filtering information. This thesis contributes to the initiative of people-tagging in three areas: a) a better understanding of how people-tags are used; b) automatic extraction, ranking and assigning people-tags to knowledge workers; and c) using tag-based profiles for an information propagation use case.

Due to a lack of sufficient studies on people-tagging behaviour in online social platforms, initially, we studied how users of social media and in particular social blogs tag each other. We extracted people-tags from such websites and classified them into several categories. Our analysis suggests that people-tagging in public online social platforms is highly subjective and this may lead to interoperability drawbacks between systems that operate on top of people-tags. Building domain-specific vocabularies as well as ranking tags are approaches that we considered to eliminate subjectivity of people-tags.

Current practices of tagging knowledge workers are manual processes that offer several disadvantages, such as increasing cognitive overhead for taggers and cold-start problem of people-tag-based systems. To address these issues, we developed approaches to (semi-) automatically extract, rank, and assign people-tags to knowledge workers. To this end, we extract metadata from collaborative platforms used by knowledge workers such as question-answering (Q-A) forums. We rank and assign such metadata to knowledge workers based on their contribution and collaboration history within collaborative platforms (e.g., solving an issue or providing helpful answers).

We use tag-based profiles for an information propagation use case. We developed an access control and in particular an information propagation model which enables end users to define information sharing policies on top of people-tags and a numeric value called distance. The distance value determines the propagation depth of a resource in a network of connected users. As users may need help in drafting appropriate policies for a given resource, we further equipped our model with a policy advisor component to assist users for sharing items such as URLs and community-related announcements. The main goal of the policy advisor is to eliminate information overload and information shortage within a network of connected users. Given an item and tag-based user profiles as input, our policy advisor is capable of analysing the item and recommending topic-sensitive hubs who may propagate information in the network, in order to eliminate information overload and information shortage.

All of our approaches are supported by prototypes that helped us to evaluate them with real-world data such as micro-blog posts and technical Q-A forums. The evaluation showed that our approaches help users to tag each other and to use tag-based profiles for a more user-centric information propagation model in social and collaborative platforms.