Machine Learning for E-Commerce: OXID e-Shop

Just as Amazon knows which products you may be interested in buying, based on your past purchase-history, it should be possible for independent shop-owners to offer a more intelligent shopping-experience to their users.

As a E-Commerce technology service-provider, our mission is to bring sustainable technology-trends to merchants and E-Commerce agencies in Germany and Switzerland.

– In 2015, we added Shopware to our portfolio.
– In 2016, we worked hard on building a Multichannel E-Commerce Framework, including ERP capabilities based on odoo.
– In 2017, we are looking forward to adding Machine Learning capabilities for E-Commerce to our technology portfolio. This will immediately enhance our HyperSearch® capabilities, and give us new a business-spaces to play in.

Our efforts will primarily be structured around these problems:

Ranking Top Sellers

Based on the sum of order-data available in a database, we would like to derive the top-sellers on specific web-shops. Such a result can of course be derived using SQL database queries. But we intend to solve the same problem using Machine Learning techniques.

User Profiling

Based on user-behavior while on a specific web-shop, we’d like to profile the user, without the user having to log in. For example, we’d like to determine if a user is male or female without he/she telling us. Similarly, we’d like to guess the approximate spending capacity of the user, and promote products in that price-segment.

  • Gender
  • Spending-power
  • Product-category

Trend List

Based on click-pattern of a user, we’d like to guess which type of products the user is interested in, and load those with a higher priority into the shop-interfaces. In other words, utilise ML to provide a highly relevant user-experience.

Demand Forecasting

Based on daily sales data, predict demand, to help the online-seller with purchase decisions. While analysing sales data, consider:

  • Seasonality: Correlation between date of the year the sales happened and quantity
  • Volumes of product sold (without correlations)

Search Results

Search features, especially “enriched” results like those from HyperSearch® can be enriched even more by learning from past click-patterns (matched with search keywords).

Dynamic Pricing

As we perfect our odoo based Omnichannel E-Commerce solution at Simplify-ERP®, we look for more and more ways AI can help our retailer (B2C) and wholesale (B2B) customers derive ever more economic value from their data.

Liquidating inventory is a well known trader’s strategy to maintain high liquidity, to invest cash in new products, and hence offer a better selection to buyers. ML can help sellers determine price-propositions for products, correlated to costs and gross-margins, and help him “push” the product on specific online-sales-channels. Very useful for this learning would of course be competitive price-data on specific channels. So a sub-problem to the Dynamic Pricing problem would be to learn the market-price (on specific channel/market, ex. Amazon) for a specific product.


HyperSearch® becomes Smarter

Our E-Commerce Search solution should get smarter by training the algorithm to learn keywords that humans type-in and the search-results they have selected to follow. Based on these, suggestion can be pre-loaded (below search-field).

The above learning can be correlated to purchase-data of articles, such that those that have been known to have been bought in the past can be promoted to the top of the SERP.


OXID E-Shop

Of course, we are most keen to apply our solutions to our favorite E-Commerce product, the OXID E-Shop!

We invite friendly agencies to participate as a stakeholder in this research and reap the benefits from an early stage.

Resources

  1. Google Cloud ML Platform (Documentation)
  2. Google Prediction API
  3. Learning Works, As Explained By Google
  4. Machine Learning 101 – Supervised Learning
  5. Learning Machines 101: A Gentle Introduction to Artificial Intelligence and Machine Learning
  6. Machine Learning is Fun!
  7. BIG ML- Machine Learning 101
  8. Essentials of Machine Learning Algorithms (with Python and R Codes)
  9. Simple Sample Program. After understanding basics of theory, try this code.
  10. 50 Useful Machine Learning & Prediction APIs
  11. Amazon Machine Learning – Developer Guide
  12. Quora answers on Applications of Machine Learning in E-Commerce. 
  13. Stanford Univ. MOOC by Prof. Andrew Ng – Machine Learning
  14. Building Machine Learning Predictive Model – Data-Sailors.com
  15. Classifying E-Commerce Products based on Images and Text – Christopher Bonnet

Important Concepts

  • Model, Parameters & Learner
  • Gradient Descent, Learning-rate
  • Neural Networks (NN)
  • Constitutional Neural Networks (CNN)
  • Binary, multiclass, regression classifications
  • Algorithms
  • Validation dataset
  • Supervised vs. Unsupervised learning
  • Reinforced learning
  • Softmax Classification Layer
  • Deep Learning
  • Visual Geometry Group (VGG), VGGNet
  • Image-Net
  • Fine-tuning
  • Convolution Filters
  • Convolutional Computer Vision
  • Annotation
  • Logistic Regularisation
  • L1, L2 Regularisation
  • Random Search
  • Random Forest
  • Hyperparameter Optimisation
  • Logistic Regression
  • Transfer Learning
  • Bag of Words Model

Deep Learning Libraries

  • Keras
  • TensorFlow