Eight ways AI applications are already making conventional EC smarter – from chatbots to creative AI
There’s no doubt that artificial intelligence (AI) will fundamentally change the world over the next few decades. What many do not realize however, is that in some fields, it has already become a large part of the status quo. One such example is e-commerce (EC).
At Rakuten, we have been working with big data for several years purposefully searching for new and better ways to empower Rakuten Ichiba (EC) marketplace merchants and customers through the boundless abilities of AI.
Below are a few examples of how we are leveraging various applications of AI, specifically machine learning, to advance our EC business.
It’s an increasingly popular application for conversational interactions between an EC service and its customers. Chatbots can be applied to the wide range of interactive communications both in B2C and B2B domains. It also provides the ability to navigate the customer to their destination smoothly. Rakuten already released chatbots to many businesses with AI technologies such as NLP, machine learning techniques and others (cf. Rakuten’s First Personalized Bot & Beyond). One of them provides the delivery status of shipments that the customer purchased on Rakuten Ichiba. Customers enjoy the convenience of accessing the delivery status of purchased products.
2. Predicting sales
Chatbots are obviously functioning as part of the UX for customers. However, other things are running behind the scenes. At Rakuten Institute of Technology we leverage “supervised” machine learning to predict product sales. Supervised machine learning is a form of AI in which “the machine,” or algorithm, is given sample data from the past that helps train it to process the data of the future. With 250 million products being traded on Rakuten Ichiba at any given time, supervised machine learning algorithms allow us to use historical sales data to forecast the sales volume of products to a high degree of accuracy and make surprising discoveries in a far more efficient way than a team of humans ever could.
3. Marketing to the new groups
We also make use of so-called unsupervised learning algorithms when segmenting customer groups for marketing campaigns. Traditionally, marketers have defined market segments in ways that appeared to make sense to them by age or gender, for example. So, supervised learning techniques have basically been utilized for that. But AI is also demonstrating that those are not always the most effective approaches. In some recent cases, the past defined features and the past training data hindered accurate marketing when customers’ behavior or profiles change or evolve. Then an unsupervised learning algorithm, working from raw real-time data only, might identify alternative means of segmentation, such as online behavior or preferences, that can serve as a more accurate predictor of interests or tastes.
4. Classifying products
As mentioned above, there are more than 250 million products being traded on Rakuten Ichiba, covering many different genres from food to clothes, electrical goods, digital content, sports equipment, cars and so on. This can make categorizing a challenge. To solve the problem, we utilize a semi-supervised learning algorithm, which repeatedly resamples data until the algorithm learns how to process it in the most efficient way. This helps us to classify products on Ichiba accurately so customers can always find what they need.
5. Analyzing ratings and reviews
Understanding user ratings and reviews is important, but it is also time-consuming. Applying a combination of NLP techniques and structural machine learning algorithms, a method commonly used in the study of the structure and formation of words (morphology), we can efficiently collect and analyze product review text, both positive or negative. In addition, structural machine learning can help us mine valuable data from product explanations and reviews.
6. Improving recommendation and search
Recommendation and search are typical applications of EC. However, we can advance these through our use of reinforcement learning algorithms to process data on customer reactions in response to products they are shown, for example, whether users clicked on a product when it was served to them in search results or in a recommendation. Similar to an A/B test, reinforcement learning algorithms notice how much “reward” (positive reaction from users) is obtained when different products are displayed in response to certain circumstances (a particular search query, or a user’s browsing history, for example). Combining knowledge of past customer reactions in response to particular circumstances, the algorithm can determine the most efficient course of action when those circumstances reoccur. And with each action and reaction, the algorithm becomes smarter.
7. Image recognition
On a customer-to-customer platform like PriceMinister-Rakuten, deep learning algorithms can be effectively used for the purpose of image recognition. Inspired by the structure and function of the brain, deep-learning algorithms develop the ability to recognize an object in a photo and then automatically categorize it, making it easier for users to post products for sale. Besides, it also works for finding fraudulent products.
8. Creative AI
This is a kind of new concept. However, it’s gaining more and more momentum. Creative AI means, AI applications that are able to do non-routine work based on expert knowledge and which generate valuable content. Works of art generated by AI are a good and easy-to-understand example, like novels, paintings, music, film and so on. It is not only art – journalism is also in the scope of use of creative AI. You can see many cases of applying creative AI to services. Some artists tried to have AI compose music and based on that they created a music video. Some projects had AI write a screen play and based on that they made a short sci-fi film. EC too, is not outside the scope of creative AI. We have more potential if we use creative AI for EC. Rakuten has tons of data from merchants and users – supply side and demand side. Combining both kinds of data, we can propose new creative products, new creative genres, new creative trends and so on with the support of creative AI. Using creative AI, we conducted a project named “Fashion Style Extraction”. LDA, one method of topic models in NLP or “unsupervised learning” was utilized for analyzing user behavior data in order to extract hidden customer needs about what kind of style they seek. This system extracted 32 totally new possible fashion styles, and based on it we created and proposed new fashion styles as the emerging fashion trend to customers.
E-commerce may not be the first thing that comes to mind when people hear the words “artificial intelligence,” but there is no denying the impact it is already having on the way we buy and sell online. Through the “superhuman” abilities of machine learning, we hope to continue empowering both customers and merchants in our EC businesses.