Research: Intelligent Question
Answering Module for Product Manuals

Question Answering (Q&A) is a well-researched NLP problem. The ability to query information in a range of organized and unstructured formats is a standard requirement for generic Q&A systems.

Neuron7 wanted to build a domain-specific Q&A system for document parsing, indexing and retrieval (identifying relevant documents), and machine comprehension (extracting spans of correct answers from the context). 

Intelligent Question Answering Module for Product Manuals

Data researchers at Neuron7 authored a study examining question answering in a constrained situation, using the high-tech domain for the purpose of this paper. They created a system of question answering based on structured and unstructured documents including product manuals, images, and product user guides.

In the paper, authors Abinaya Govindan, Gyan Ranjan, and Amit Verma,  demonstrate how to overlay bespoke domain knowledge on top of current and traditional systems to deliver contextualized outputs that help users quickly identify solutions to their problems.

Traditional Search vs. Intelligent Search

Several product manuals were used as the data source, where an agent attempts to discover a technical response to a client query by perusing the manuals. With multiple editions or versions of many products, product manuals are a constantly evolving source of information. Product manuals are intended for humans to understand rather than machines, which makes automatic parsing more challenging. 

Traditional Enterprise Search is a Three-Step Process

  1. Find the right manual (from hundreds or thousands of manuals in some cases)
  2. Find the right section/page in the manual (from hundreds of pages per manual)
  3. Find the right answer

In this traditional search model, queries may point to the right content title, but the user must then search through the document to find the right section, the right page, and the right answer. Many customer service documents (like product manuals and policies) are long and complex. Service teams and customers looking to self-solve need answers quickly and pointing to the content title is not enough.

Neuron7 Search Points to the Exact Answer

Neuron7’s search breaks from this tradition by attempting to find the right answer immediately using natural language understanding (NLU). Neuron7 was purpose-built for customer service, with search results that point to the exact answer. if there is any ambiguity, Neuron7 identifies the right section or page within the manual, getting customer service teams or customers much closer to the answer in seconds.

Neuron7’s search uses NLU to understand context that helps pinpoint the right answer. For example, if the user asks What is the expected time for device A’s battery to be charged?, Neuron7 recognizes that “device A” is specific context that directs the system to only look through manuals for device A.

Process Texts and Non-Text Information for any Domain

The use of product manuals for question answering involves the use of a document indexer engine in the question answering system, executed at scale to address queries in real-time. As a result, the system should be able to retrieve relevant sections of instructions from hundreds of acquired manuals for each user query. Because it can process both textual (paragraphs, summaries, etc.) and non-textual (tables, images, flow diagrams, etc.) information, this system can be extended to any domain as long as manuals, documents, or even books and articles are available.

In traditional question answering scenarios, a small chunk of text can be regarded as an answer to the question asked but this doesn’t work for customer service. If a user inquires, What are the steps for me to log in to a device?, the response can’t be provided in a short segment. Instead, they need to see an entire section titled How to use and setup?. The system should be able to decide whether the answer should be returned as a short sequence or as a portion of text in real-time. 

In the full paper, we show how various existing systems try to solve this domain based question answering by comparing their performances on standard business dataset. Our experiments show that this intelligent question answering system outperforms traditional question answering systems on standard business datasets. 

Let us show you how Neuron7’s AI-powered search delivers the right answer to resolve customer service issues in seconds. 

Contact Us

Resolve issues faster with AI that understands your service data and continually learns as you use it.