Time-Decayed Clickstream Embeddings for Intent Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Customer service systems lack predictive capabilities to understand customer intent, leading to inefficient call routing and resource management due to the instability of one-hot encoding in clickstream data analysis, which results in increased call hold times and inadequate resource planning.
Innovation Solution
A system that uses time-based featurization of clickstream data by converting URLs into tokens, generating frequency matrices, and applying non-negative matrix factorization to create latent feature vectors, which are then combined with time-decayed weights to generate clickstream embeddings for predicting future user actions using machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If one-hot encoding is used to convert webpages into vectors for prediction, then the system can process clickstream data, but the encoding becomes unstable when websites change URLs or page names
Solution Approach 1:
The patent transforms the encoding approach from one-hot encoding (which is sensitive to URL changes) to time-based featurization using TF-IDF weighting. This changes the parameter representation from binary vectors to continuous weighted vectors that capture the temporal patterns and importance of different URLs, making the encoding stable even when websites change their URL structures.
Solution Approach 2:
The patent introduces time-decayed weighting that dynamically adjusts the importance of historical clickstream events based on their recency. More recent events receive higher weights while older events receive lower weights, allowing the system to adapt to changing user behaviors and website structures over time while maintaining stability through the dynamic nature of the weighting scheme.
2Productivity
If traditional customer service systems route calls without predictive understanding, then all calls are handled in FIFO manner, but this increases average call hold time and reduces support resolution efficiency
Solution Approach 1:
The patent performs preliminary analysis of clickstream data to predict customer intent and service needs before the customer actually contacts support. By pre-processing the clickstream data and generating predictions about what services the customer is likely to need, the system can prepare appropriate routing decisions in advance, reducing both hold time and improving productivity.
Solution Approach 2:
The patent implements a feedback mechanism where predicted customer intent from clickstream analysis is continuously fed into the call routing system. This feedback loop allows the routing system to adjust its decisions based on predicted needs, creating a closed-loop system that continuously improves routing efficiency while minimizing hold times through informed decision-making.
3Ease of operation
If IVR systems use predefined broad options to gather caller information, then the system can structure the interaction, but it takes several minutes for customers to navigate menus and the CSR still must verify information
Solution Approach 1:
The patent performs preliminary extraction of customer intent and relevant information from clickstream data before the customer even reaches the IVR system. By pre-analyzing the customer's browsing behavior, search queries, and interaction patterns, the system already has insights into what the customer needs, eliminating the need for lengthy IVR menu navigation and subsequent verification by the CSR.
Solution Approach 2:
The patent introduces clickstream analysis as an intermediary between the customer and the IVR/CSR system. This intermediary pre-processes the customer's digital footprint to extract meaningful intent signals, which then inform and streamline the subsequent customer service interaction, reducing both the time customers spend navigating menus and the verification time required by CSRs.
4Adaptability or versatility
If customer service systems lack predictive capabilities, then resource planning is reactive, but this results in inadequate resource allocation and inability to anticipate incoming demand
Solution Approach 1:
The patent performs preliminary prediction of customer service demands by analyzing clickstream data patterns before the actual service requests occur. By pre-processing historical and real-time clickstream data to identify trends and predict future service needs, the system gains advance knowledge of incoming demand, enabling proactive resource allocation and planning rather than reactive responses.
Data Source
AI summary
Methods and apparatuses are described for determining future user actions using time-based featurization of clickstream data. A server captures clickstream data corresponding to web browsing sessions and converts the clickstream data into tokens by identifying each unique URL and parsing each unique URL into tokens. The server generates a frequency matrix based upon the tokens, and generates a latent feature vector for each URL in the session based upon the frequency matrix. The server merges the latent feature vectors and the clickstream data into an aggregate clickstream vector set for a user. The server assigns time-decayed weight values to each latent feature vector in the aggregate clickstream vector set. The server combines the time-decayed latent feature vectors to generate a clickstream embedding for the user, and executes a machine learning model using the clickstream embedding to generate one or more predicted actions of the user.