This chapter introduces conditional random fields (CRFs) for relational learning, which are a type of graphical model that captures statistical dependencies between entities. The authors discuss the advantages of CRFs over traditional graphical models, such as hidden Markov models (HMMs), in handling rich local features in relational data. They provide a tutorial on training and inference techniques for CRFs, including linear-chain CRFs and general CRFs with arbitrary graphical structures. The chapter also covers practical implementation techniques and an example application in information extraction, where CRFs are used to automatically build a relational database from unstructured text. The authors highlight the ability of CRFs to capture long-distance dependencies, which is particularly useful for tasks like named-entity recognition. The chapter concludes with a discussion of the relationship between generative and discriminative models, emphasizing the benefits of discriminative modeling in handling rich, overlapping features.This chapter introduces conditional random fields (CRFs) for relational learning, which are a type of graphical model that captures statistical dependencies between entities. The authors discuss the advantages of CRFs over traditional graphical models, such as hidden Markov models (HMMs), in handling rich local features in relational data. They provide a tutorial on training and inference techniques for CRFs, including linear-chain CRFs and general CRFs with arbitrary graphical structures. The chapter also covers practical implementation techniques and an example application in information extraction, where CRFs are used to automatically build a relational database from unstructured text. The authors highlight the ability of CRFs to capture long-distance dependencies, which is particularly useful for tasks like named-entity recognition. The chapter concludes with a discussion of the relationship between generative and discriminative models, emphasizing the benefits of discriminative modeling in handling rich, overlapping features.