How is Alibaba Visual Search going? Well, visual search is now an important business tool. It changes the way people interact with eCommerce and social media platforms.
Rather than entering text for a search query, a visual search uses images to identify and locate items. Additionally, the customer’s desire to discover new products and brands fuels the rise of visual search.
Thus, Alibaba as one of the eCommerce platforms also developed its visual search engine.
Alibaba Visual Search – The Development Of Pailitao
Alibaba released the Pailitao app in 2014. “Pailitao” means shopping with a camera. This app enables users to search for items on Alibaba’s eCommerce platform by capturing the query object using their devices’ camera.
Alibaba’s team brought the concept of “ content-based image retrieval” to an area where a good photo can make a difference. Especially between success and failure. And that’s online shopping.
Therefore, Alibaba collaborates with the Machine Intelligence Technology Lab to develop the Pailitao.
The Challenges Researchers Faced
Researchers faced a set of unique challenges. Why? User queries on Pailitao are taken from real life. Unlike the carefully staged inventory photos on eCommerce sites.
Also, often, query images are of low quality. Others include busy background features, which may complicate object localization.
Moreover, most visual search solutions do not operate at the scale of Alibaba. Thus, because of the platform’s large image collection with many fine-grained categories, visual search technology needs massive data architecture.
And maintaining training data for deep models can be very expensive as well.
The Solution
Researchers proposed a scalable as well as an efficient visual search system. They used an effective category prediction method.
Well, this method uses a model and search-based fusion to reduce the search space. Thus, allowing for better scalability and better performance.
They also proposed a deep CNN model with branches for joint and feature learning. This minimizes the expense of human-labeled data.
Besides, the model simultaneously discovers the detention mask. As well as exact discriminative features without background interference. They then also instated a retrieval process.
That process uses a binary indexing engine and re-ranking. Thus, resulting in improved customer engagement.
How Does Alibaba’s Pailitao Work?
Alibaba’s Pailitao was averaging over 10 million daily unique visitors. Users can search with an image to find matching products from among billions of items on Taobao – Alibaba’s online shopping website.
And since Pailitao covers almost all Taobao product categories, users can now search for any items they wish. That includes fashion apparel, bags, shoes, food, makeup, furniture, etc. Additionally, Pailtao improves user experience by eliminating the need to describe the target item.
Also, it integrates AliExpress. It’s suitable because it eliminates the language barriers for international searches. AliExpress can worry less about the translation of product descriptions. These customers can quickly find what they search for by the use of photos.
Moreover, Alibaba partnered with Samsung. Thus integrating its technology into their smartphone cameras and virtual assistant Bixby. This allows users to search for products by pointing their phone’s camera at an object.
They also partnered with Youko, to allow their audiences to search and purchase products seen in videos.
