Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Discovering user intent In E-commerce clickstreams

Sheil, Humphrey 2019. Discovering user intent In E-commerce clickstreams. PhD Thesis, Cardiff University.
Item availability restricted.

[thumbnail of 2019sheilhphd.pdf]
PDF - Accepted Post-Print Version
Download (3MB) | Preview
[thumbnail of sheilh.pdf] PDF - Supplemental Material
Restricted to Repository staff only

Download (2MB)


E-commerce has revolutionised how we browse and purchase products and services globally. However, with revolution comes disruption as retailers and users struggle to keep up with the pace of change. Retailers are increasingly using a varied number of machine learning techniques in areas such as information retrieval, user interface design, product catalogue curation and sentiment analysis, all of which must operate at scale and in near real-time. Understanding user purchase intent is important for a number of reasons. Buyers typically represent <5% of all e-commerce users, but contribute virtually all of the retailer profit. Merchants can cost-effectively target measures such as discounting, special offers or enhanced advertising at a buyer cohort - something that would be cost prohibitive if applied to all users. We used supervised classic machine learning and deep learning models to infer user purchase intent from their clickstreams. Our contribution is three-fold: first we conducted a detailed analysis of explicit features showing that four broad feature classes enable a classic model to infer user intent. Second, we constructed a deep learning model which recovers over 98% of the predictive power of a state-of-the-art approach. Last, we show that a standard word language deep model is not optimal for e-commerce clickstream analysis and propose a combined sampling and hidden state management strategy to improve the performance of deep models in the e-commerce domain. We also propose future work in order to build on the results obtained.

Item Type: Thesis (PhD)
Date Type: Completion
Status: Unpublished
Schools: Computer Science & Informatics
Date of First Compliant Deposit: 12 March 2019
Last Modified: 04 Aug 2022 01:54

Actions (repository staff only)

Edit Item Edit Item


Downloads per month over past year

View more statistics