Building an AI-Powered Writing Assistant for WPS Office

This post details the technical implementation of an AI-powered writing assistant for WPS Office. It describes the process of making sense of unstructured textual data, using TFIDF for feature extraction, employing a bi-directional LSTM-CNNs-CRF deep learning model for named entity recognition, and creating sentence embeddings with Infersent. Milvus is used as the similarity search engine to store and query these embeddings, utilizing the IVF_FLAT index and partitioning for performance optimization. The system runs on Kubernetes clusters and uses MySQL for metadata management.