Studies on User Intent Analysis and Minin

Predicting the goals of users can be extremely useful in e-commerce, online entertainment, information retrieval, and many other online services and applications. In this thesis, we study the task of user intent understanding, trying to bridge the gap between user expressions to online services and their goals behind it.;As far as we know, most of the existing user intent studies are focusing on web search and social media domain. Studies on other areas are not enough. For example, as people more and more rely our daily life on cellphone, our information needs expressing to mobile devices and related services are increasing dramatically. Studies of user intent mining on mobile devices are not much. And the intentions of using mobile devices are different from the ones we use web search engine or social network. So we cannot directly apply the existing user intention to this area. Besides, user's intents are not stable but changing over time. And different interests will impact each other. Modeling such kind of dynamic user interests can help accurately understand and predict user's intent. But there're few existing works in this area. Moreover, user intent could be explicitly or implicitly expressed by users. The implicit intent expression is more close to human's natural language and also have great value to recognize and mine.;To make further studies of these challenges, we first try to answer the question of ''What is the user intent?''. By referring amount of previous studies, we give our definition of user intent as ``User intent is a task-specific, predefined or latent concept, topic or knowledge-base that is under an expression from a user who is trying to express his goal of information or service need''.;Then, we focus on the driving scenario when a user using cellphone and study the user intent in this domain. As far as we know, it is the first time of user intent analysis and categorization in this domain. And we also build a dataset of user input and related intent category and attributes by crowdsourcing and carefully handcraft. With the user intent taxonomy and dataset in hand, we conduct a user intent classification and user intent attribute recognition by supervised machine learning models. To classify the user intent for a user intent query, we use a convolutional neural network model to build a multi-class classifier. And then we use a sequential labeling method to recognize the intent attribute in the query. The experiment results show that our proposed method outperorms several baseline models in precision, recall, and F-score.;In addition, we study the implicit user intent mining method through web search log data. By using a Restricted Boltzmann Machine, we make use of the correlation of query and click information to learn the latent intent behind a user web search.;We propose a user intent prediction model on online discussion forum using Multivariate Hawkes Process. It dynamically models user intentions change and interact over time.The method models both of the internal and external factors of user's online forum response motivations, and also integrated the time decay fact of user's interests.;We also present a data visualization method, using an enriched domain ontology to highlight the domain-specific words and entity relations within an article.