Deep Learning Feedback Navigation With Hierarchical Issue Themes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing application reviews, particularly in the context of privacy issues, fail to provide structured summaries and require significant manual effort to identify themes and issues, lacking the ability to automatically generate high-level themes and fine-grained issues from unstructured feedback texts.
Innovation Solution
A deep learning system that utilizes machine-learned models, such as T5 models, to classify, generate, and summarize application feedback, creating a hierarchical structure of themes and issues, enabling developers to navigate feedback at multiple levels of abstraction, including emotion classification and trend analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual analysis of application reviews is performed, then developers can understand user feedback, but significant manual effort and time are required
Solution Approach 1:
The system enables automatic self-analysis of application reviews through deep learning models that autonomously classify feedback texts, generate issues, and create themes without requiring manual developer intervention for each review analysis task
Solution Approach 2:
Manual mechanical analysis of reviews is replaced by automated deep learning-based natural language processing systems that perform classification, issue generation, and theme creation through computational models instead of human effort
2Quantity of substance
If unstructured feedback texts are analyzed directly, then all user comments are captured, but the feedback is challenging to understand at scale
Solution Approach 1:
The system segments unstructured feedback texts into structured hierarchical components including individual issues, themed categories, and aggregated insights, making large volumes of feedback organized and easily navigable through a graphical user interface
Solution Approach 2:
Deep learning models serve as intermediaries that translate raw unstructured feedback texts into structured, interpretable formats with automatic generation of issues and themes, bridging the gap between voluminous unstructured data and comprehensible insights
3Productivity
If deep learning models are used to automatically generate structured feedback summaries, then manual effort is reduced, but system complexity increases
Solution Approach 1:
The deep learning system performs multiple functions including classification of feedback texts, generation of issues, creation of themes, and visualization through GUI using a unified automated framework, reducing the need for separate manual processes despite increased system capabilities
Data Source
AI summary
A method using a computing system is described that classifies each feedback text from a plurality of feedback texts into one or more categories. The method dynamically generates, using the plurality of feedback texts, a set of feedback text issues. The set of feedback text issues includes one or more issues associated with each feedback text from the plurality of feedback texts. The method dynamically generates, using the set of feedback text issues, one or more themes associated with the plurality of feedback texts. Each of the one or more themes is associated with a respective subset of feedback text issues from the set of feedback text issues. The method outputs a graphical user interface that includes one or more from the group consisting of at least one feedback text issue from the set of feedback text issues, and at least on theme from the one or more themes.