Enrichment of Knowledge Base of Conversational Agents by Active Question Targeting
Publication Date: 2017-Aug-10
The IP.com Prior Art Database
Enrichment of Knowledge Base of Conversational Agents by Active Question Targeting Abstract:
Disclosed is a system and method for enriching the knowledge base of conversational agents by targeting questions to users on the chat interface based user’s expertise and current conversation. Introduction:
Conversational bots are the intelligent software designed for making human’s feel as if they are talking to a real person. Conversational bots are also sometimes referred to as Chabot’s, these bots can automate human mundane tasks such as repeatedly answering the same question to different customers, and can automate the process of copy pasting the same email, or over chat.
These Conversational agents are replacing humans in technical and non-technical support roles in several industries such as banking, healthcare, IT support, etc. These agents are driven by knowledge store (e.g., knowledge graph). This knowledge store is either created automatically from existing knowledge, such as structured and unstructured documents, previous tickets, or generated manually by subject matter experts (SME). Gaps in knowledge base limits the capabilities of these agents to perform at par with humans. Bots fail to answer simple questions as the Knowledge Base is missing that knowledge
One of the solutions to fill the knowledge gap is by handing over the conversations to a SME whenever agents cannot respond. Furthermore, the agent learns actively from the conversations between the user and SME. This solution comes with a cost, SMEs are expensive resources in most of the companies. More over Ratio of users to SMEs is large which affects user response time and hence pace of data curation. Other way of enriching knowledge is Crowd-sourced way where end users only will share their expertise to help other users in need. Crowd sourcing platforms such as “Stackoverflow”1 and “Quora”2 has seen a huge success as p2p knowledge sharing platforms.
On the other hand, users’ involvement in enhancing the capabilities of conversational agents is limited to providing feedback on whether the solution is helpful or not. So, there is an immediate need for filling the content gaps in the knowledge base.
Here, we use users’ diverse expertise to enrich the knowledge base of conversational
agents by active question targeting System and Method:
This creation proposes a system and method for enriching the knowledge base of conversational agents by targeting questions to users on the chat interface with the combination of the following features:
• Identifying knowledge gaps based on earlier conversations 1. Analyzing conversations of the agent with users to identify unsuccessful
1 https://stackoverflow.com/ 2 https://www.quora.com/
2. Clustering unsuccessful conversations based on mapping user questions to a formal representation of knowledge base of the agent
3. Forming natural language anchor question per cluster of unsuccessful conversations
• Opportunistic (karmic) questio...