Bing has announced new AI-based improvements in its search engine, especially in the image search section. The new improvements will allow multi-granularity matches, better understanding of user queries, images, webpages and the link between them.
Somewhat like Google, Microsoft now integrated BERT (Bidirectional Encoder Representations from Transformers), leveraging:
1. The text information is interpreted better through pre-trained knowledge
2. Images and webpages will be embedded with better awareness of each other. It is to ensure that the document embedded describes the salient features of image and the webpage’s key points.
A certain set of object attributes are extracted by Microsoft both from query and the document of candidate and use the characteristics for matching.
The query is them matched with the image content’s characteristics and the text around it. If the query and the document share same characterizes, they can be considered as “precise match”. This would be helpful for users when they will be able to find precise searches about anything with multiple features.
Metadata of documents and images is automatically generated that enable better search matches. BRQ “Best Representative Query” is one of the most popular types of metadata. BRQ summarizes the major topics on webpage and image content. The BRQs generation for Bing images mainly depends on the latest deep learning techniques.
According to Microsoft, the techniques has a great impact on their search engine as it improved the accuracy of results.
Also, the Bing Image Search provide better semantic understanding of queries from the users, says Microsoft.
Read next: Is Microsoft really more ethical than Facebook, Google, and Apple?
Somewhat like Google, Microsoft now integrated BERT (Bidirectional Encoder Representations from Transformers), leveraging:
1. The text information is interpreted better through pre-trained knowledge
2. Images and webpages will be embedded with better awareness of each other. It is to ensure that the document embedded describes the salient features of image and the webpage’s key points.
A certain set of object attributes are extracted by Microsoft both from query and the document of candidate and use the characteristics for matching.
The query is them matched with the image content’s characteristics and the text around it. If the query and the document share same characterizes, they can be considered as “precise match”. This would be helpful for users when they will be able to find precise searches about anything with multiple features.
Metadata of documents and images is automatically generated that enable better search matches. BRQ “Best Representative Query” is one of the most popular types of metadata. BRQ summarizes the major topics on webpage and image content. The BRQs generation for Bing images mainly depends on the latest deep learning techniques.
According to Microsoft, the techniques has a great impact on their search engine as it improved the accuracy of results.
Also, the Bing Image Search provide better semantic understanding of queries from the users, says Microsoft.
Read next: Is Microsoft really more ethical than Facebook, Google, and Apple?
