Ngram-LSTM Open Rate Prediction Model (NLORP) and Error_accuracy@C metric: Simple effective, and easy to implement approach to predict open rates for marketing email

TL;DR

This paper introduces NLORPM, a simple and effective Ngram-LSTM model for predicting email open rates, especially useful with sparse data, along with a new metric 'Error_accuracy@C' for evaluation.

Contribution

The paper presents a novel Ngram-LSTM approach for open rate prediction that is easy to implement and performs well with limited data, along with a new evaluation metric.

Findings

NLORPM outperforms existing models in sparse data scenarios.

The 'Error_accuracy@C' metric is intuitive and aligns well with marketing needs.

Model implementation is straightforward and has low latency.

Abstract

Our generation has seen an exponential increase in digital tools adoption. One of the unique areas where digital tools have made an exponential foray is in the sphere of digital marketing, where goods and services have been extensively promoted through the use of digital advertisements. Following this growth, multiple companies have leveraged multiple apps and channels to display their brand identities to a significantly larger user base. This has resulted in products, worth billions of dollars to be sold online. Emails and push notifications have become critical channels to publish advertisement content, to proactively engage with their contacts. Several marketing tools provide a user interface for marketers to design Email and Push messages for digital marketing campaigns. Marketers are also given a predicted open rate for the entered subject line. For enabling marketers generate targeted subject lines, multiple machine learning techniques have been used in the recent past. In particular, deep learning techniques that have established good effectiveness and efficiency. However, these techniques require a sizable amount of labelled training data in order to get good results. The creation of such datasets, particularly those with subject lines that have a specific theme, is a challenging and time-consuming task. In this paper, we propose a novel Ngram and LSTM-based modeling approach (NLORPM) to predict open rates of entered subject lines that is easier to implement, has low prediction latency, and performs extremely well for sparse data. To assess the performance of this model, we also devise a new metric called 'Error_accuracy@C' which is simple to grasp and fully comprehensible to marketers.