Abstract
(type = abstract)
With the advent of mobile, Internet, and sensing technologies, large-scale urban and mobile data are available and are linked with locations near real properties. These data can be a source of rich intelligence for classifying high-rated residential locations, developing livable communities, and enhancing urban planning in big cities. In this dissertation, we aim to address the unique challenges of real estate ranking, especially (i) how to build an effective ranking system by exploiting heterogeneous mobile data and modeling geographic dependencies; (ii) what are the underlying drivers for livable and sustainable communities.
Along these lines, I first introduced a method for ranking residential complexes based on invest- ment ratings by mining users opinions about residential complexes from online user reviews and offline moving behaviors (e.g., taxi traces, smart card transactions, check-ins). While a variety of features could be extracted from these data, these features are intercorrelated and redundant. Thus, selecting good features and integrating the feature selection into the fitting of a ranking model are essential. To this end, I first strategically mined the fine-grained discriminative features from user reviews and moving behaviors. Then, I proposed a Sparse Pairwise Ranking method by combining a pairwise ranking objective and a sparsity regularization in a unified probabilistic framework.
In addition, with the development of new ways to collect estate-related mobile data, there is a potential to leverage geographic dependencies of residential complexes for enhancing real estate evaluation. Indeed, the geographic dependencies of the value of a residential complex can be from the characteristics of its own neighborhood (individual), the values of its nearby residential complexes (peer), and the prosperity of the affiliated latent business area (zone). To this end, I proposed an enhanced method, named ClusRanking, for real estate evaluation by leveraging the mutual enforcement of ranking and clustering power. In ClusRanking, three influential factors (i.e., geographic utility, neighborhood popularity, and influence of business areas) are constructed and extracted for predicting real estate investment ratings. An estate-specific ranking objective is also proposed to jointly model individual, peer and zone dependencies.
Moreover, mixed land use refers to the effort of putting residential, commercial and recreational uses in close proximity to one another. This can contribute economic benefits, support viable public transit, and enhance the perceived security of an area. It is naturally promising to investigate how to rank residential complexes from the viewpoint of diverse mixed land use, which can be reflected by the portfolio of community functions in the observed area. To that end, I further developed a geographical function ranking method, named FuncDivRank, by incorporating the functional diversity of communities into real estate evaluation. In FunDivRank, a mix-land use latent model is developed to learn latent community functions and the corresponding portfolios. Also, a real estate ranking indicator is learned by simultaneously maximizing ranking consistency and functional diversity.
Finally, we present experimental results to demonstrate the effectiveness of our methods.