LSI Vector Similarity for Semantic Technical Support
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing customer service solutions in the software development sector face challenges in providing effective technical support due to transient and probing queries, which require significant time and knowledge, and declarative semantics-based solutions often fail to fetch relevant results.
Innovation Solution
A system and method utilizing Latent Semantics Indexing (LSI) to extract keywords from knowledge and bug content, generate a keyword-content matrix, decompose it into sub-matrices, and compute cosine similarity measures to identify relevant knowledge content for user queries, providing semantics-based technical support.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If declarative semantics-based solutions are used to fetch relevant results, then the system can provide structured query handling, but it fails to retrieve relevant knowledge content for transient and probing queries in software development sector
Solution Approach 1:
The patent transforms the query processing approach by changing the semantic representation parameters from traditional declarative semantics to Latent Semantic Indexing (LSI) vectors. This allows the system to capture nuanced meanings and relationships in transient queries by mapping them to dimensional space where semantic similarity can be computed effectively, thus improving both reliability and adaptability
Solution Approach 2:
The patent replaces the mechanical keyword-matching mechanism of declarative semantics with a mathematical vector-based similarity computation system. By substituting the rigid keyword comparison with cosine similarity calculations on LSI-transformed vectors, the system achieves more flexible and accurate retrieval for diverse query types
2Reliability
If customer executives provide manual technical support for persistent and probing queries, then adequate support can be provided, but it requires considerable amount of time and knowledge
Solution Approach 1:
The patent implements a self-service technical support system where the LSI-based retrieval mechanism automatically processes queries and retrieves relevant knowledge content without human intervention. The system serves itself by computing vector similarities and ranking results, eliminating the need for customer executives to manually analyze each query while maintaining high-quality support
Solution Approach 2:
The patent performs preliminary action by pre-computing LSI vectors for all knowledge base documents and storing them in an indexed format. When a query arrives, the system leverages these pre-computed vectors for rapid similarity computation, significantly reducing the time required to process queries while maintaining comprehensive analysis
3Productivity
If LSI and low-rank approximation methods are used to process queries in real-time, then relevant knowledge content can be retrieved efficiently, but the system complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-computing the LSI transformation matrices and storing the transformed vectors for all knowledge base documents. This offline preprocessing step converts the complex LSI computation into simple vector retrieval and similarity calculation operations that can be executed rapidly in real-time, thus improving productivity while managing system complexity
Solution Approach 2:
The patent segments the query processing task into distinct phases: (1) offline LSI transformation of knowledge base documents, (2) online query vector transformation and similarity computation, and (3) result ranking and retrieval. This segmentation allows the complex LSI operations to be performed once during preprocessing, while real-time processing uses simpler operations, thereby improving speed without overwhelming system complexity
Data Source
AI summary
Disclosed is a system for providing semantics based technical support to a user. A keyword extraction module extracts a set of keywords from a plurality of knowledge content and a plurality of bug content. A matrix generation module generates a keyword-content matrix comprising a similarity score computed for each keyword corresponding to each knowledge content. The matrix generation module further decomposes the keyword-content matrix into one or more sub-matrices. A vector coordinates identification module determines a first set of vector coordinates for each knowledge content of the plurality of knowledge content and a second set of vector coordinates for a bug-query received from a user in real-time. A similarity measure module computes a cosine similarity measure of the bug-query with each knowledge content to identify at least one knowledge content relevant to the bug-query. The knowledge identification module further provides the at least one knowledge content to the user.