Efficient Cross-Domain Asepct-based Sentiment Analysis and Knowledge Base Construction for Conventional Smart Devices

Market survey industry has a capacity of hundreds of millions a year, but the existing social network market survey systems are basically based on the mentions of brands and specific product names. Such volume statistics and correlation analysis between named entities are not enough for marketing and business strategies. In this project, we explore transfer learning and multi-task learning technologies for opinion mining and aspect-based sentiment analysis with respect to target products or services from different domain. Various online opinions can be forwarded to the corresponding department (customer service, public relations, product design or manufacturing department) for further processing. The goal here is to reduce the effort of training data preparation for new domain and improve the performance of aspect-based sentiment analysis.In addition, the conversational search mode required for smart devices is also the direction of KKLAB's product development. KKBOX possesses database of abundant songs, albums and singers. However, how these songs are related to drama, movie or other events require entertainment ontology to support request or searches by composer or lyric writer. We will exploit Web data extraction technologies on the construction of entertainment knowledge base to enhance the conversational search on smart speakers. Therefore, in the second year of the project, we will focus on the development of dialogue intent recognition and slot filling technology to build a dialogue understanding module for smart devices.