Call Context Vectors for Similar Call Search Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing call center systems struggle to search for similar calls based on the content of conversations, relying solely on metadata such as date, time, and speaker identity, without considering the actual content of the calls.
Innovation Solution
A system that generates a transcript of call content, categorizes it into predefined topics with relevance scores, and uses cosine similarity to identify contextually similar calls based on multi-dimensional vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional metadata-based search is used, then search speed and system simplicity are improved, but search accuracy and content relevance deteriorate
Solution Approach 1:
The patent segments the contact data into multiple dimensions: metadata attributes (date, time, duration) and content attributes (transcript, topics, context). This allows the system to perform fast metadata filtering while simultaneously conducting accurate content-based similarity search using vector representations and cosine similarity calculations, thereby resolving the contradiction between search speed and accuracy.
Solution Approach 2:
The patent transforms content-based search from traditional keyword matching to a multi-dimensional vector space representation. By converting contact content into vectors with multiple dimensions representing different topics and contexts, the system achieves both fast computation through mathematical operations and high accuracy through holistic content analysis, eliminating the trade-off between speed and precision.
2Measurement precision
If content-based search is implemented, then search accuracy and contextual relevance are improved, but system complexity and processing requirements worsen
Solution Approach 1:
The patent performs preliminary processing by automatically generating transcripts from audio recordings and pre-computing topic classifications and context vectors for all contacts before search operations. This advance preparation stores processed data in optimized formats, allowing the actual search to proceed quickly without complex real-time processing, thus reducing operational system complexity while maintaining high search accuracy.
Solution Approach 2:
The patent introduces an intermediary layer of topic models and context vectors that mediate between raw contact data and search queries. These intermediaries simplify the search process by transforming complex content comparisons into straightforward vector similarity calculations, reducing the computational complexity required for accurate content-based search while improving scalability.
3Reliability
If holistic content analysis is performed, then contextual similarity and training quality are improved, but processing time and computational resources worsen
Solution Approach 1:
The patent applies partial action by focusing content analysis on the most relevant dimensions for training purposes, such as topic classification and contextual vectors, while using metadata for quick filtering. This selective deep analysis approach ensures sufficient training quality from the most impactful content features while avoiding the time cost of processing all possible data attributes in full detail, thereby optimizing the balance between training quality and processing time.
Data Source
AI summary
Systems and methods are provided for determining calls and contacts associated with calls that are contextually similar based on content of the calls. Data associated with a call include one or more utterances made by speakers and a set of values indicating relevance between content of the call and topic categories of the call. The disclosed technology generates a topic vector associated with a call and/or respective speakers of the call. The topic vector includes a multi-dimensional vector where each dimension corresponds to a topic category. The disclosed technology determines calls that are contextually similar by comparing angular distances between topic vectors. A search query receiver receives a search query that queries contacts and calls that are contextually similar to a given call and/or a speaker. The disclosed technology identifies calls with topic vectors that are within a predetermined angular distance.