NER: Named Entity Recognition is an essential tool in information extraction, question answering systems, syntactic analysis, machine translation, and other application fields. It plays a crucial role in making natural language processing technology practical. Generally, the task of named entity recognition is to identify three major categories (entity, time, and numerical) and seven subcategories (person names, organization names, location names, time, date, currency, and percentage) of named entities in the text to be processed.

Disambiguate: Word sense disambiguation refers to identifying the correct meaning of an ambiguous word in a specific context.

Query Rewriting: The application method of query rewriting is to expand the original query into rewritten terms that are highly relevant to user needs. Multiple rewritten terms are used together with the user's search terms for retrieval, thus improving RAG's effectiveness through better expression.

Answer Generation: The model used in the answer generation phase of RAG, focusing on generating answers based on given context.

Embedding: A method of representing textual data (words, sentences, documents) as vectors.

PPO: Proximal Policy Optimization, a type of Reinforcement Learning from Human Feedback (RLHF) that uses human preferences and guidance to train and improve machine learning models.

DPO: Direct Preference Optimization, another type of Reinforcement Learning from Human Feedback. DPO is a relatively new method that directly optimizes user or expert preferences rather than traditional cumulative rewards.

Knowledge Editing: Directly editing LLM's knowledge through fine-tuning neurons. https://github.com/zjunlp/EasyEdit