Low-Dimensional Embedding for Web Sequence Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Representing and analyzing the interactions of users with large websites, such as www.paypal.com, is challenging due to the sheer number of webpages, which hinders the application of machine learning algorithms and identification of friction points in the shopping process that lead to abandoned purchases.
Innovation Solution
Applying the word2vec algorithm to network traffic history data to create a low-dimensional embedding of webpage sequences, allowing for the monitoring of browsing activity and prediction of user behavior, thereby flagging sessions that may result in abandoned purchases and enabling remedial actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to represent and analyze user interactions with large websites, then complete analysis of all webpages is possible, but the complexity and computational burden become unmanageable due to the sheer number of webpages
Solution Approach 1:
The patent extracts only the essential and relevant webpages from the complete website into a reduced subset. This subset is carefully selected to represent the most important user interaction paths and friction points, eliminating unnecessary complexity while preserving analytical completeness for the most critical analysis objectives.
Solution Approach 2:
The patent segments the large website into multiple components: a reduced subset of critical webpages, user interaction sequences, and friction point categories. This segmentation allows the analysis system to focus computational resources on the most important segments rather than processing every single webpage uniformly.
2Loss of information
If all webpages are analyzed in detail, then comprehensive understanding of user friction is achieved, but the time and computational resources required become prohibitive
Solution Approach 1:
The patent extracts and focuses analysis on a reduced subset of webpages that contain the most valuable friction point information. By identifying and analyzing only these critical pages, the system achieves comprehensive understanding of user friction without the prohibitive time cost of analyzing every single webpage on the site.
Solution Approach 2:
The patent applies different levels of analysis depth to different webpages based on their importance. Critical webpages in the reduced subset receive detailed analysis, while less important pages are either excluded or analyzed at a lower level, optimizing the balance between information completeness and analysis time.
3Reliability
If machine learning algorithms are applied to the complete website structure, then accurate prediction of user behavior is possible, but the computational burden and processing time become unmanageable
Solution Approach 1:
The patent extracts a reduced subset of webpages that contain the most predictive information for user behavior. By training machine learning models on this condensed dataset rather than the complete website structure, the system achieves accurate predictions while significantly reducing computational burden and energy consumption.
Solution Approach 2:
The patent segments the website data into user interaction sequences based on the reduced subset of critical webpages. This segmentation creates manageable input structures for machine learning algorithms, allowing accurate behavior prediction without processing the overwhelming complexity of the complete site structure.
Data Source
AI summary
Methods and systems for creating and analyzing low-dimensional representation of webpage sequences are described. Network traffic history data associated with a particular website is retrieved and a word embedding algorithm is applied to the network traffic history data to produce a low dimensional embedding. A prediction model is created based on the low-dimensional embedding. Browsing activity on the particular website is monitored. A set of sessions in the current browsing activity is flagged based on a result of applying the prediction model to the monitored browsing activity.