Methods Of Context Based And Social Media Oriented Information Recommendation

With the development of internet, user-generated data has been growing tremendously in Web 2.0 era. Facing such a big volume of resources, people need a method of fast exploration and indexing to find their demanded data. Traditional information retrieval technologies satisfy users’need to a great extent. However, for their all-purpose characteristics, they cannot satisfy any query from the different background, with the different intention and at the different time. Context-aware recommender systems, aiming to further improve performance accuracy and user satisfaction by fully utilizing contextual information, have recently become one of the hottest topics in the domain of recommender systems. The main contributions of this thesis are as follows:1. We propose a dominating set discovering method for the verbal context models to prune the irrelevant contextual factors and keep the major characteristics at the same time. And we present the verbal contextual graph to model verbal context in the collaborative tagging systems (folksonomy) to capture the user intention and facilitate personalized search.2. Context-aware movie recommender system is implemented by using the Collaborative Fil-tering (CF) framework that is integrated with a factorization machine (FM) approach as illustrated in the generic model. To select suitable features, we propose to choose the one with the lowest Root Mean Square Error (RMSE) first, then make use of the lowest one to pair-up with another remaining contextual features to find out which pair is able to achieve the lowest RMSE as illustrated in this Section 4. Using the above strategy, the amount of contextual features required can be easing found and all irrelevant contextual features can be removed. It is because each time only one contextual variable is added to evaluate the result of RMSE, which means the lowest one must be added. Once an irrelevant feature is used to do a rating prediction, the RMSE is starting to become large, making it simple to identify unsuitable feature.3. We focus on building a social emotion detection system for online news. The system is built based on the modules of document selection, Part-of-speech (POS) tagging, and social e-motion lexicon generation algorithm. We evaluate the system on an online news collection containing 40,897 articles gathered from the Sina society channel. It has 2,083,818 ratings distributed over 8 kinds of social emotions. Experimental results show that the proposed system can effectively choose a well-formed training set, and generate meaningful social e-motion lexicon with POS information. We also conduct qualitative investigation on samples of the social emotion lexicon. The result shows that the lexicon not only reflects explicit emotion words, but also implicit words that convey emotions potentially. The POS of each word is useful to detect emotion ambiguity of words and the context dependence of their sentiment orientations.4. We present a new term weighting scheme which models the Local element, Global element and Topical association of each story (LGT scheme), in addition to two nonparametric feature reduction strategies for New Event Detection (NED). The main conclusions are drawn as follows:Firstly, the local element, which represents the uniqueness of each story, has a significant impact on the performance both for events under the same subject and for diverse events in NED. Secondly, the topical association exploiting LDA is powerful in modeling multiple events on the same subject (polysemous words), and evolution of events (synonymous words). Finally, a nonparametric feature reduction strategy based on the boxplot of logarithmic word document frequency reduces redundant features, and performs a lot better on diverse events and relatively stable on events under the same subject.