Chatbot Database Generation from FAQ Triplets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Programming a chatbot to provide responses from a corpus of documentation is complex and time-consuming, and existing systems produce overly complex concept graphs that are difficult to use for conversational flows, requiring extensive training and not efficiently handling unforeseen user inputs.
Innovation Solution
A system that processes a question and answer document to generate a tree of base relation triplets, allowing a chatbot to answer questions about focus entities with minimal training, using entity extraction and natural language processing to create a database that can efficiently map user inputs to correct answers without needing example question phrasings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional systems use entire document corpus to train chatbot, then comprehensive knowledge coverage is achieved, but system complexity and training time increase significantly
Solution Approach 1:
The patent segments the entire document corpus into discrete question-answer pairs, which are then processed into individual triplets. This segmentation transforms the monolithic training corpus into manageable units that can be efficiently stored and queried, reducing overall system complexity while maintaining comprehensive knowledge coverage.
Solution Approach 2:
The patent extracts only the essential elements (questions, answers, entities, and relationships) from the document corpus, discarding redundant information. By taking out only the necessary components and representing them as structured triplets, the system achieves comprehensive knowledge coverage with reduced complexity.
2Loss of information
If conventional systems create concept graphs from documentation, then knowledge representation is achieved, but the graphs become overly complex and difficult to use for conversational flows
Solution Approach 1:
The patent segments complex concept graphs into simple triplets consisting of subject, predicate, and object. Each triplet represents a single factual relationship, making the knowledge representation much simpler and more usable for conversational flows while preserving all essential information.
Solution Approach 2:
The patent changes the structural parameters of knowledge representation from hierarchical concept graphs to flat triplet structures. This parameter change simplifies the data structure, making it easier to query and use in chatbot conversations while maintaining complete knowledge representation.
3Measurement precision
If chatbot is trained with extensive example question phrasings, then response accuracy improves, but training time and computational resources increase
Solution Approach 1:
The patent enables the chatbot to derive its own training data directly from the question-answer pairs without requiring external example phrasings. The system self-generates training data by extracting entities and relationships from the provided Q&A content, achieving accurate responses without extensive manual training examples.
Solution Approach 2:
The triplet structure serves multiple functions simultaneously: it represents factual knowledge, enables flexible querying, and provides training data for the chatbot. This multi-functionality eliminates the need for separate example question phrasings, reducing training time while maintaining response accuracy.
4Quantity of substance
If conventional systems use complex concept graphs, then comprehensive knowledge storage is achieved, but handling unforeseen user inputs becomes difficult
Solution Approach 1:
The patent creates a dynamic knowledge representation system where triplets can be easily queried and combined in different ways to handle various user inputs. The flexible triplet structure allows the system to adapt to unforeseen questions by combining relevant triplets, maintaining high adaptability while storing comprehensive knowledge.
Data Source
AI summary
A chatbot answer database can be generated from a question and answer document. A question/answer processor can receive a document that includes questions and answers. A set of entities can be determined from the answers in the document. A set of triplets can be constructed from the answers, and stored in a chatbot database. For each answer, the question/answer processor can determine text for slots of the triplet based on the entities extracted from the answer. The triplet can be stored as a tree structure in a database of tree structures associated with the document. A chatbot can receive questions, and provide answers based on the database of tree structures associated with the document.