Discovering Missing Values in Semi-Structured Databases

We explore the problem of discovering multiple missing values in a semi-structured database. For this task, we formally develop Structured Relevance Model (SRM) built on one hypothetical generative model for semi-structured records. SRM is based on the idea that plausible values for a given field could be inferred from the context provided by the other fields in the record. Small-scale experiments on IMDb (Internet Movie Database) show that SRM matched three state-of-the-art relational learning approaches on the movie label prediction tasks. Large-scale experiments on a snapshot of the National Science Digital Library (NSDL) repository show that SRM is highly effective at discovering possible values for free-text fields even with quite modest amounts of training data, compared with state-of-the-art machine learning approaches.