Application of latent Dirichlet allocation and autoencoder to real estate datasets

At present, there are too many types and numbers of real estate features in the real estate market, and it is difficult to effectively recommend real estate to customers with more complicated needs. The datasets of real estate often encompass a wide array of features, including categories, numerical values, and textual descriptions, which complicates the process of delivering precise and satisfactory recommendations. To address this issue, a cluster-based hybrid method combining latent Dirichlet allocation and autoencoder is proposed in this paper. It can effectively improve the early input of clustering by the two-stage feature extraction of the data through the topic model and the autoencoder. Experimental findings demonstrate that our hybrid method surpasses standalone LDA and autoencoder, as well as other conventional clustering algorithms. The combined results of Silhouette coefficient, DB index, and CH index are significantly better than the results of traditional clustering. Moreover, we conduct a comprehensive analysis of the results, emphasizing the benefits of our approach in managing multi-type features, encompassing categorical, numerical, and string data. It is anticipated that this research will contribute valuable insights into the field of recommendation systems.