Applying Semantic Analyses to Content-based Recommendation and Document Clustering
1,259 views · Published 17 August 2016 · 43:16 · Indexed 14 November 2026
Channel: Microsoft Research · 2016 · Science & Technology
This talk will present the results of my research on feature generation techniques for unstructured data sources. We apply Probase, a Web-scale knowledge base developed by Microsoft Research Asia, which is generated from the Bing index, search query logs and other sources, to extract concepts from text. We compare the performance of features generated from Probase and two other forms of semantic analysis, Explicit Semantic Analysis using Wikipedia and Latent Dirichlet Allocation. We evaluate the semantic analysis techniques on two tasks, recommendation using Matchbox, which is a platform for probabilistic recommendations from Microsoft Research Cambridge, and clustering using K-Means.