Shopper Intent Prediction and Causal Inference: Connecting the Dots

Shopper Intent Prediction and Causal Inference: Connecting the Dots

Introduction:
In recent years, there has been a growing interest in understanding shopper intent in the e-commerce space. One study titled "Shopper Intent Prediction from Clickstream E-commerce Data with Minimal Browsing Information" explores the prediction of purchase events based on clickstream data. Another study, "Causal Inference The Mixtape - 10 Synthetic Control," discusses the use of synthetic control to create a counterfactual for causal inference. While these studies may seem unrelated at first glance, there are common points that can be connected to provide a deeper understanding of shopper intent prediction and its implications.

Understanding Shopper Intent Prediction:
The study on shopper intent prediction focuses on two specific tasks: the classification of arbitrarily long click sequences and the early prediction of limited-length sequences. The researchers explore two algorithmic approaches: hand-crafted feature-based classification and deep learning-based classification. They find that the use of k-gram statistics with visibility graph motifs can produce fast and accurate classifications, even for extremely short observation windows. This highlights the reliability of purchase prediction based on minimal browsing information.

Insights from Clickstream Data:
Analyzing the clickstream data, the researchers find that the relative abundance of each symbol (i.e., action) is not a fully discriminative feature of each class. Therefore, higher-order statistics are required to accurately predict shopper intent. Additionally, they observe that view and detail events are often overrepresented in very short sequences. This suggests that these actions may not be strong indicators of purchase intent on their own.

The Role of Horizontal Visibility Graph Motifs:
To better understand the structure and dynamics of time series data, the researchers introduce the concept of horizontal visibility graph motifs (HVGm). These motifs describe the sequential patterns within the clickstream data. By analyzing the distribution of these motifs, they are able to compute the entropy of the HVGm profile. A more evenly represented distribution of motifs indicates a higher predictive value for shopper intent.

Applying Actionable Advice:
Based on the findings from the shopper intent prediction study, here are three actionable advice for businesses looking to improve their understanding of shopper intent:

1. Consider the combination of actions: Instead of focusing on individual actions, businesses should look at the combination of actions to gain a better understanding of shopper intent. For example, the sequence of actions "add-view" may indicate a higher likelihood of purchase intent compared to just "view" or "add" actions.

2. Implement personalized strategies: The study suggests that customers who end up making a purchase tend to have a prior idea of what they want. Businesses can take advantage of this by implementing personalized strategies such as similar product recommendations. By understanding the preferences and intent of individual shoppers, businesses can enhance the shopping experience and increase conversion rates.

3. Use deep learning-based models: The study finds that deep learning-based models outperform hand-crafted feature-based models in shopper intent prediction. Businesses can leverage the power of deep learning algorithms to analyze clickstream data and make more accurate predictions. By training models on large datasets and incorporating higher-order statistics, businesses can improve their ability to predict shopper intent.

Connecting to Causal Inference:
Now let's connect the insights from shopper intent prediction to the concept of causal inference using synthetic control. Synthetic control is a method used to create a counterfactual for a treatment group when a suitable control group is not available. The study on causal inference discusses the advantages of synthetic control over regression-based methods.

The synthetic control method uses a combination of comparison units to model the counterfactual for the treated unit. This approach considers the weighted average of units in the donor pool, ensuring that the synthetic control closely replicates the characteristics of the treated unit. By using interpolation instead of regression, the method allows for the construction of the counterfactual without requiring access to post-treatment outcomes during the design phase of the study.

Applying Synthetic Control in Causal Inference:
To apply synthetic control in causal inference, researchers select matching variables that are predictors of post-intervention outcomes and remain unaffected by the intervention. The weights assigned to the comparison units reflect the predictive value of these matching variables. Through a falsification exercise, researchers can test the validity of the estimator and calculate exact p-values based on root mean squared prediction error (RMSPE) values.

By connecting the insights from shopper intent prediction to the concept of causal inference using synthetic control, businesses can gain a more comprehensive understanding of their customers and make informed decisions. The ability to predict shopper intent can help businesses optimize their marketing strategies, personalize the shopping experience, and improve conversion rates.

Conclusion:
In conclusion, the study on shopper intent prediction from clickstream e-commerce data and the study on causal inference using synthetic control may seem unrelated at first. However, by connecting the common points and insights, we can see the broader implications for businesses. Understanding shopper intent is crucial for businesses to optimize their marketing strategies and improve conversion rates. By leveraging the power of deep learning-based models and synthetic control, businesses can make more accurate predictions and draw causal inferences. The combination of these approaches can lead to actionable insights and ultimately drive success in the e-commerce space.