Multilayer Neural Network for Complaint Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to efficiently and accurately identify and classify customer complaints in unstructured computer text, often missing indirect complaints and failing to consider context, nuance, and sentiment, while also posing security risks due to reliance on cloud-based data processing.
Innovation Solution
A system using a multilayer neural network to analyze unstructured computer text, generating word vectors and tensors, filtering and classifying messages for potential complaints, and encrypting data for on-premises processing to prevent data exfiltration, incorporating natural language processing and ensemble classification models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If cloud-based data processing is used for analyzing unstructured text, then analysis capability is improved, but data security deteriorates
Solution Approach 1:
The system segments the data processing workflow into two distinct parts: (1) local preprocessing and vectorization of unstructured text data using BERT embeddings on-premises, and (2) cloud-based execution of pre-trained neural network models for complaint detection. This segmentation allows sensitive raw text to remain locally encrypted while only processed features are transmitted to the cloud, resolving the contradiction between cloud computing power and data security.
2Object-affected harmful factors
If manual interpretation of unstructured text is used, then data security is maintained, but processing efficiency deteriorates
Solution Approach 1:
The system introduces an intermediary layer of automated preprocessing and feature extraction that operates locally on encrypted data. The BERT-based word vectorization and complaint probability models act as intermediaries between raw unstructured text and final complaint identification, enabling automated analysis without requiring manual decryption or handling of sensitive text data.
3Extent of automation
If existing AI-based text analysis is used, then automation is improved, but accuracy in detecting nuanced complaints deteriorates
Solution Approach 1:
The system transforms the input parameters for complaint detection by converting raw text into BERT-based word vectors and contextual embeddings. This parameter transformation enables the neural network models to capture nuanced semantic meanings, sarcasm, and indirect complaints that traditional keyword-based or simple sentiment analysis methods miss, thereby improving detection accuracy while maintaining full automation.
Data Source
AI summary
Methods and apparatuses are described in which unstructured computer text is analyzed for identification and classification of complaint-specific user interactions. A data store receives unstructured computer text corresponding to current user interactions. A server filters the unstructured computer text to identify messages that comprise a potential complaint. For each message, the server: parses the message into sentences with words, converts each of the words in each sentence into a word vector, executes message classification models using the word vectors as input to generate a complaint probability value, combines the complaint probability values from each of the message classification models into an overall complaint probability value, classifies the message as a complaint when the overall complaint value exceeds a predetermined threshold value, and for each message classified as a complaint, routes the message to a client computing device that categorizes the message based upon one or more context features.