Engineering challenges of building a neural search system
Now, you would have noticed that the most important building blocks of the neural search system are the encoder and indexer. The quality of encoding posts has a direct impact on the final search result, while the speed of the indexer determines the scalability of your neural search system.
Meanwhile, this is still not enough to make your neural search system ready to use. Many other topics need to be taken into consideration as well. The first question is: does your encoder (neural model) have the same distribution as your data? For new people coming into the neural search system world who are using a pretrained deep neural network, such as ResNet trained on ImageNet, it is trivial to quickly set up a search system. However, if your target is to build a neural search system on a specific domain, let’s say a fashion product image search, it is not going to produce satisfying results.
One important topic before we...