| More and more health consumers are actively seeking healthcare information online in the past few years. The increasing demands from health consumers for healthcare information have boosted the emergence of Online Health Communities (OHCs), which established online communication platforms such as discussion forums and online social groups. Health consumers discuss medical conditions and treatments with peer health consumers through these platforms. OHCs have captured a vast amount of healthcare information and made the information easy to access for health consumers. However, finding the needed information is still no easier than finding a needle in a haystack for most health consumers due to the huge number of discussion threads and users as well as the wide variety of healthcare issues being discussed. This information overload issue would make it harder to keep users in OHCs. In order to encourage consumers to actively participate in OHCs, it is imperative to facilitate their access to demanded information, as well as connection with peers who are interested in similar healthcare topics. In this dissertation, we addressed three research problems to assist health consumers in accessing relevant information in a more efficient and effective way.;Firstly, we introduced methods for constructing a heterogeneous healthcare information network from healthcare social media data. We used external dictionaries to identify different types of entities from the data. However, it is long been recognized that consumers use very different language from health professional for even the same health-related concepts. Simply using professional vocabularies to extract entities from consumer-contributed data would not be effective. Therefore, we used Consumer Health Vocabularies (CHVs) to expand professional terms for entity extraction. After that, we used co-occurrence analysis to find relationship between entities.;Secondly, we explored the approaches for measuring similarity between different users in a heterogeneous information network. In an OHC, finding similar users and recommending them to health consumers not only provides a shortcut to relevant healthcare information, but also helps consumers to find emotional resonance. We proposed a content-based approach, and two structure-based approaches using local and global structural information of nodes in the network. The experiment results showed that content-based method performed better than structural methods for inactive user group, while structural methods yielded better performance for active users. To further investigate the underlying patterns, we used logical operator AND/OR to integrate different similarity measuring methods. We demonstrated that different methods could capture different aspects of user similarity, and we could get a more comprehensive measurement by combining different methods together.;Last but not least, we discussed thread recommendation in the previously constructed network. We proposed a novel approach for thread recommendation in OHCs in the context of heterogeneous information network. We extracted structure-based features from the network and used the features to train a binary classification model to predict users' preference in threads. The experiment showed that we could effectively predict users' preference for threads. We also demonstrated that we could boost the recommendation performance by integrating both thread-thread relationship and user-user relationship with basic network metrics. |
