User Interest Profile Generation via Dwell Time Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for building user interest profiles rely on click or view data, which may not accurately represent user engagement, as they do not account for the duration of user interaction with content, leading to incomplete or inaccurate representations of user interests.
Innovation Solution
The method involves measuring article-level dwell time, applying weights based on logarithmic functions of dwell time, and using these weights to determine user interest scores for features, thereby generating a more accurate user interest profile that reflects true user engagement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If click or view data is used to build user interest profiles, then the implementation is simple, but the accuracy of user engagement representation is insufficient
Solution Approach 1:
The patent transforms the engagement measurement from a binary click/view event into a continuous parameter by introducing dwell time duration. This parameter change enables more granular differentiation of user engagement levels, where longer dwell times indicate higher engagement. The system calculates interest scores by aggregating dwell time data across multiple articles, creating a nuanced profile that reflects true user interest rather than mere exposure.
2Measurement precision
If dwell time measurement is implemented, then user engagement accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent segments the user engagement measurement into discrete, manageable components: individual article dwell times are measured separately, then aggregated into overall user interest scores. This segmentation approach breaks down the complex task of measuring overall engagement into simpler atomic units (per-article dwell times), making the data collection and processing more tractable while maintaining high accuracy.
3Loss of information
If traditional view-based profiling is used, then the system is easy to implement, but the user interest representation is incomplete
Solution Approach 1:
The patent implements a feedback mechanism where dwell time measurements continuously inform and refine user interest profiles. The system collects dwell time data, processes it into interest scores, uses these scores to personalize content delivery, and then collects additional dwell time data that further refines the profiles. This closed-loop feedback ensures that user interest representations become progressively more complete and accurate over time.
Data Source
AI summary
A method is provided for building a user interest profile. According to the method, features of each of a plurality of articles are identified. For a given user, views of one or more of the plurality of articles are logged. For each view, a corresponding dwell time for the view by the given user is measured. A weight is applied to each view based on the corresponding measured dwell time. User interest scores for features of the articles are determined based on the weighted views, and the user interest profile is generated for the given user based on the user interest scores.