Construction and application of industrial knowledge atlas (3): commodity knowledge fusion

This chapter first introduces the definition of knowledge fusion 、 Difficulties and processes , Then, it briefly introduces the common methods of ontology alignment and entity alignment , Finally, it introduces the method and evaluation system of knowledge fusion in the practice of industrial field .

3.1 Overview of knowledge fusion

3.1.1 The main difficulties of knowledge fusion

For many fields that need to build knowledge maps , Because there are many practitioners , Naturally, a large number of databases or knowledge bases will precipitate , For example, in the film and television sub field , Baidu Encyclopedia 、 Wikipedia 、 Douban is a good knowledge base , Therefore, in the process of building the knowledge map, the knowledge of multiple knowledge bases is often merged , In order to obtain the most comprehensive knowledge , This process is called knowledge fusion . Besides , Because the knowledge in the merged knowledge base will inevitably have errors , Therefore, knowledge fusion will cross verify the knowledge of multiple knowledge bases , The knowledge obtained after knowledge fusion will be more accurate than the single knowledge base before fusion .

Due to the openness of natural language , The task of knowledge fusion in various fields often faces the following four difficulties .

（1） Heterogeneous problems . Different knowledge bases have great differences in the expression of the same entity or knowledge , Specifically, there are a large number of aliases for the same entity . Pictured 3-1（a） Shown , Different data sources for the same entity “ Huawei P30” There are different descriptions .

（2） Ambiguity . Different entities use the same or similar text expression . Pictured 3-1（b） Shown , In the field of e-commerce ,“ Apple ” It can be used as a category name , It can also be a brand name , It can even appear as a color or pattern name . It is easy to cause the decline of accuracy in engineering practice .

（3） Data noise problem . In the practice of knowledge map in industry , The problem of data noise is widespread , The fusion of multiple knowledge bases will magnify the noise problem , It is easy to cause a decline in accuracy . Pictured 3-1（c） Shown , Many attribute values of the commodity entity are “ An unknown