Natural Language Question Answering

Answer Server provides natural language question answering, using several different options including a database of facts, a list of FAQ-style common questions, and extracting passages from relevant documents. You can optionally use Answer Server with LLM (large language model) options to generate answers and perform RAG search (retrieval-augmented generation).

When you use Answer Server, you configure one or more answer systems, each of which uses a different method to provide answers when you send it a question. The following sections describe the available answer systems. Some answer systems use Named Entity Recognition grammars to parse questions, to help match the correct answers.

Answer Bank

The Answer Bank contains a store of reference questions and answers, which you can add and administer. When a user asks a question, the Answer Bank queries the store of existing questions for any that match, and returns the relevant answers. You can use an Answer Bank to maintain an FAQ list to answer simple questions.

Fact Bank

The Fact Bank contains a database of factual information in a database, to return simple factual answers. When a user asks a question, the Fact Bank parses the question to determine what fact is required, and queries the database for an answer.

Passage Extractor

The Passage Extractor links to a store of documents that contain information that Answer Server can use to answer questions. This system is useful for more general questions, as it can flexibly find the answers in your document store.

When you send a question, Passage Extractor queries the document store and attempts to extract a short statement, sentence, or paragraph that contains a relevant answer. You can use Passage Extractor to answer general questions that would be difficult to maintain as a database, or that do not have a straightforward answer.

You can use Passage Extractor with or without an LLM module. Without the LLM module, Answer Server extracts text directly from the documents in the store. With the LLM module, you can use an LLM model of your choosing to extract the answers from the text, or use a generative model to generate the answers based on each candidate document. You can also configure Answer Server and the data index it uses to perform a vector search for candidate answers before it generates an answer. However, in most cases where you might want to use the LLM passage extractor system, OpenText recommends that you use the RAG system instead.

RAG

RAG uses LLMs to generate answers from documents in your system. It is similar to Passage Extractor, and can answer similar questions, but it generates a single answer based on all your candidate documents, rather than an answer for each candidate.

In the RAG system, Answer Server retrieves candidate documents and sends them all to the LLM as context, with a standard prompt that contains the user question, and the LLM uses this input to generate the answer.

Conversation

The Conversation module runs a real-time conversation task with your end users, selecting text and options from a chat script that you create. This option allows you to set up an interactive virtual assistant to answer common user queries.