PKBC: A Product-Specific Knowledge base Taxonomy Framework
Abstract: Recently, knowledge base construction (KBC) has become a hot and in-time topic with the increasing needs of applications for large-scale knowledge bases (KBs), such as semantic search and question answering (QA) systems.Existing KBC systems mainly focus on the encoding of general concepts in the world, e.g., people, dates, and places, which appear in our daily conversations regularly. However, less attention has been paid to the knowledge of commercial products. Commercial products take a vital part in our daily lives, since they are frequently searched in the QA session and used for recommendation of the online shopping websites. The construction of product-oriented KB will benefit the organization and query of these products.Nevertheless, general KBC systems mainly focus on the knowledge extraction from long-text corpus. Since the relations between products are likely to appear in different regions and forums with various formats, the general KBC methods cannot be directly applied to construct product-oriented KBs.In this paper, we propose a framework for constructing a product-oriented KB which help to describe and organize commercial products, by leveraging the product specifications as well as the general KBs. We first propose a model which extracts and links the entities discovered from the description of the product items to general KBs. We then utilize these mappings and the product information to automatically build a product taxonomy.We conduct a range of pilot experiments over Amazon product datasets and real KB. The experimental results confirm the effectiveness and generality of our proposed framework.
Loading