Dynamic Support Recommendations via Supervised Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users face difficulties in finding relevant support topics within large and complex online-support web pages or websites, leading to increased user frustration and higher operating expenses for service providers due to the inefficiencies in navigating through extensive topic hierarchies and limited vocabulary.
Innovation Solution
A system that monitors user actions during interactions with support documents and utilizes a supervised-learning model trained on multiple user sessions to provide dynamic recommendations for relevant support topics, ranking them based on estimated relevance, without requiring explicit user identification or search queries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If users browse through a large number of articles arranged in a wide and deep topic hierarchy, then they may find comprehensive support information, but the navigation becomes difficult and time-consuming
Solution Approach 1:
The patent segments the large corpus of support articles into smaller, manageable groups based on user behavior patterns. Instead of presenting the entire wide and deep topic hierarchy at once, the system divides articles into relevant subsets that match specific user actions or problem types, making navigation easier while maintaining access to comprehensive information.
Solution Approach 2:
The system performs preliminary analysis of user behavior (clicks, scrolls, time spent) before presenting support articles. By pre-processing user interaction data and pre-organizing articles based on predicted user needs, the system prepares personalized article groups in advance, reducing the navigation burden before the user even begins searching.
2Adaptability or versatility
If the topic hierarchy is expanded to cover more support areas, then more comprehensive support is provided, but the complexity of the hierarchy increases
Solution Approach 1:
The patent implements a dynamic topic hierarchy that adapts to user behavior. Rather than a static, pre-defined wide and deep structure, the system dynamically reorganizes and presents topic hierarchies based on real-time user interactions, making the complexity adaptive to user needs rather than fixed.
Solution Approach 2:
The system uses unsupervised learning to automatically organize and structure support articles based on user behavior patterns without requiring manual hierarchical classification. The articles self-organize into relevant groups based on how users actually interact with them, reducing the complexity of manual hierarchy maintenance while maintaining comprehensive coverage.
3Measurement precision
If users are required to provide detailed search queries or answer diagnostic questions, then more accurate support articles can be found, but user effort increases
Solution Approach 1:
The system performs preliminary analysis of user behavior (clicks, scrolls, time spent on pages) to infer user needs before the user even submits a search query. By pre-processing behavioral data to predict user problems, the system reduces the need for detailed user input while maintaining accurate article matching.
Solution Approach 2:
The system continuously monitors user interactions with support articles and uses this feedback to refine future recommendations. By observing which articles users click on, how long they spend reading them, and what they click within those articles, the system learns and improves its matching precision over time without requiring explicit user feedback or diagnostic questions.
4Reliability
If the vocabulary in support articles is limited to maintain consistency, then article quality is improved, but search engine performance deteriorates due to synonym issues
Solution Approach 1:
The patent introduces an intermediary layer (the unsupervised learning system) between the user's search query and the support articles. Instead of requiring exact vocabulary matching between user queries and article content, the learning system acts as a mediator that translates user behavior patterns and query intent into relevant article recommendations, bridging the vocabulary gap while maintaining article consistency.
Data Source
AI summary
Embodiments of a system (such as a computer system), a method, and a computer-program product (e.g., software) for use with the computer system are described. These embodiments may be used to provide one or more recommendations to a user based at least in part on monitored user actions while the user interacts with a support document (e.g., a website or a web page) that includes content for multiple support topics, such as help instructions associated with an application or a service. In particular, one or more support topics are associated with the monitored user actions using a supervised-learning model, and these support topics are recommended to the user. This supervised-learning model may be trained on portions of the content accessed by multiple users during multiple sessions. Moreover, in the supervised-learning model, content accessed by one of the users during a given session may be more relevant than content accessed by this user during the multiple sessions.