Aug 28th 2017
6 months ago
Sep 11th 2017
5 months ago
The platform of the Ahoolee search engine consists of three main blocks: Parsing, Matching and Search.
In order to collect information from stores, we use a decentralized parsing method based on our own blockchain. The key tasks that we solve are keeping the information about prices up-to-date and providing the most detailed information about products. It is important not to create a DDoS attack on the store’s website; for this reason, the fact that the price is up-to-date is confirmed based on smart contracts by a limited number of miners.
Information about products that has been collected from stores is processed by machine learning algorithms in order to consolidate the information about the same product sold by different stores in a single product card. Training data are marked by assessors. In this way we solve the issue of the incompleteness of information on products sold in different online stores. Our system does not rely on product categories in its search, which is why the way a store builds its category tree makes no difference.
At the moment, the engine of the Ahoolee search engine is based on Elasticsearch, as well as some heuristic techniques designed to improve the quality of search results. As more and more statistical data will be accumulated from users’ search requests, we plan to rank search results using machine learning methods, such as gradient boosting and neural networks.