Deep Neural Networks (DNNs)
Multiple models in Personal Marketing Microservice use deep neural networks as part of their architecture. DNNs can process high volumes of data and adapt their calculations depending on the training datasets they receive. The same model can be re-used at a member level, as it can adapt to the input data without any additional manual corrections.
Product Embeddings
One of the more exciting Machine Learning techniques is a method initially developed for Natural Language Processing [NLP]. Its goal is to detect the latent meaning of a word by analyzing other words that are frequently used in close association or similar context. From the marketing perspective, this approach allows the program to recognize related products by examining what items are commonly purchased together. By and large, it relies on the assumption that if products A and B are purchased with products C, D and E, then it is highly probable that there is some relation between those products – whether it be that they belong to the same category or are purchased by similar customers.
Deep Collaborative Filtering
Deep Collaborative Filtering assumes that customers with a similar purchase history are likely to purchase similar products in the future. This straightforward idea powers a vast majority of commercially used recommender systems.
Dynamic Time Warping
DTW is a technique used to assess distance and similarity between two different temporal sequences or time-series, which may vary in speed or frequency. In PMM, it is one of the methods used for pattern identification and detection for both supervised and unsupervised learning approaches.
Association Rule Mining
Association rule mining is a category of Machine Learning techniques that is used to identify relations between variables in huge datasets. Frequent Pattern Mining within Personal Marketing Microservice utilizes the FP-Growth and Apriori association rule mining algorithms to detect sets of products frequently purchased together in a single transaction. FPM also follows the continuous learning approach, meaning that the model keeps evolving over time and adapts to new transactions coming through a wide range of supported data streams.
