Introduction to Deep Learning Business Applications for Developers by Armando Vieira & Bernardete Ribeiro

Author:Armando Vieira & Bernardete Ribeiro
Language: eng
Format: epub
Publisher: Apress, Berkeley, CA

7.1 Online User Behavior

Predicting user intentionality (the desire to buy a given product or service), based on previous interactions within a web site, is critical for e-commerce and ad display networks, in particular retargeting. By keeping track of the search patterns of the consumers , online merchants can have a deep understanding of their behaviors and intentions.

In mobile e-commerce, a rich set of data is available, and potential consumers search for product information before making purchasing decisions, thus reflecting a consumer’s purchase intentions. Users show different search patterns (i.e., time spent per item, search frequency, and returning visits).

Clickstream data can be used to quantify search behavior using machine learning techniques, mostly focused on purchase records. While purchasing indicates a consumer’s final preferences in the same category, search is also an essential component to measuring intentionality toward a specific category. You can use a probabilistic generative process to model user exploratory and purchase history, in which the latent context variable is introduced to capture the simultaneous influence from both time and location. By identifying the search patterns of consumers, you can predict their click decisions in specific contexts and recommend the right products.

Modern search engines use machine learning approaches to predict user activity within web content. Popular models include logistic regression (LR) and boosted decision trees . Neural networks have an advantage over LR because they are able to capture nonlinear relationship between the input features and because their “deeper” architecture has inherently greater modeling strength. Decision trees—albeit popular in this domain—face additional challenges with high-dimensional and sparse data. The advantage of probabilistic generative models inspired by deep neural networks is that they can mimic the process of a consumer’s purchase behavior and capture the latent variables to explain the data.

In my 2016 paper, I proposed ( https://arxiv.org/pdf/1511.06247.pdf ) an algorithm based on auto-encoders to identify the activity patterns of certain users that led to buy sessions and then extrapolated as templates to predict high probabilities of purchase in related web sites. The data used consists of about 1 million sessions containing the click data of users. However, only 3 percent of the training data consists of buy sessions, making it a very unbalanced dataset. To handle this, I used an under-sampling technique (i.e., selecting only a fraction of negative examples).

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.