In recent years, probabilistic topic modeling is gaining significant momentum in both academia and industry. Many probabilistic topic models have been proposed by researchers and have demonstrated good performance in real-word applications such as text mining and location based services. However, with the effectiveness of probabilistic topic modeling, applying it in web search scenarios has rarely been studied in literature. In this thesis, we discuss how to adapt probabilistic topic modeling to three major functionalities of contemporary search engines: Web Search Query Log Analytics, Web Search Query Suggestion and Web Search Query Processing.

In Web Search Query Log Analytics, we first discuss the temporal topic models, which discover the dynamics of web search and profil...[ Read more ]

In recent years, probabilistic topic modeling is gaining significant momentum in both academia and industry. Many probabilistic topic models have been proposed by researchers and have demonstrated good performance in real-word applications such as text mining and location based services. However, with the effectiveness of probabilistic topic modeling, applying it in web search scenarios has rarely been studied in literature. In this thesis, we discuss how to adapt probabilistic topic modeling to three major functionalities of contemporary search engines: Web Search Query Log Analytics, Web Search Query Suggestion and Web Search Query Processing.

In Web Search Query Log Analytics, we first discuss the temporal topic models, which discover the dynamics of web search and profile each search engine user. Then we discuss the spatial topic models that capture the latent relations between query terms, URLs and geographical locations. In Web Search Query Suggestion, we present our approach of utilizing probabilistic topic modeling to capture search engine users’ preferences, with the focus of integrating both diversification and personalization into the procedure of generating search query suggestion lists. In Web Search Query Processing, we first discuss two paradigms of incorporating probabilistic topic information into inverted index, then we present efficient query processing algorithms for top-k document retrieval with both TF-IDF and probabilistic topic information. We further demonstrate how to apply the proposed inverted indices and query processing algorithms to the scenario of mobile application search. Finally, we discuss possible future work of applying probabilistic topic modeling in web search scenarios.