Research on Multi-channel Retrieve Mechanism Based on Heuristic.

Search system is a crucial component in information systems, that is to retrieve the relevant documents by the specific query. There are many search retrieve algorithms, almost all focus on a specific domain of retrieve algorithms, such as text matching, tag matching, semantic matching. However, in a real search system, the purpose of the retrieve module is to find the documents related to the specified query as fully as possible, so it requires that the search system should contain multiple forms of retrieve capability. In this paper, we propose a multi-channel retrieve model, including text channel, semantic channel and intention channel. In text channels, we propose heuristic two-stage query rewrite model, which can generate more semantically rich queries. In semantic channel, we present a novel sample dynamic construction method, which saves the manual annotation cost, and can train the model more fully. Finally, a brief general implementation of the retrieve methods based on intention channel is presented. The results show that the retrieve method based on multi-channel can significantly improve the retrieve number and accuracy of documents, and business indicators such as click-through rate are also significantly improved.