The Structured Organization Of Social Media Contents And Its Applications

As one of the most important information sources in the era of Web 2.0, social media provides real world applications in information retrieval and data mining with large amounts of multi-source data, which could lead them improvements. However, the noisy,diversified and timely nature of social media contents also poses great challenges to the state-of-the-art techniques in both fields. In this thesis, we propose a topic hierarchy based approach on the structured organization and application of social media contents.Specifically, we first introduce a topic hierarchy to organize multi-source social media contents using a hierarchical topic structure and design an algorithm for topic hierarchy construction. Next, we propose a hierarchical information retrieval model to meet users’ various information needs on social media contents. Finally, we incorporate the information within topic hierarchies into recommendation systems to recommend products as well as its relevant social media contents to fit the users’ potential interests.Generally, our study focuses on three social media based applications, which are listed as follows:? For information organization, we propose to organize social media contents using automatically generated topic hierarchies, which employ the topics and topic relations embedded in social media corpus to construct the organization structure.Moreover, we further use information from di?erent social media sources to complement each other, resulting in significant improvements on the system performance.? For information retrieval, we propose a framework for hierarchical information retrieval of social media contents. We first utilize the semantic relation within the search results to organize them in a hierarchical perspective. Next, we design source-relevant features to enhance the document ranking function, which enable our system to obtain better retrieval results on multi-source social media contents.? For information recommendation, we propose to enhance latent factor based recommendation model with side information from items’ topic hierarchies. To this end, we convert topic hierarchies into topic-item ratings. Then we learn topic weights from user-topic and topic-topic relevance to optimize the topic-item ratings’ impact on the recommendation results. Finally, we further employ the social media contents on topic hierarchies as recommendation explanations.