Entertainment and Media - Search Optimization

The ability to search efficiently through news content is of primary importance for end users of news sites. Often the sheer amount of information that is indexed, which can be updated at sub minute intervals, can be overwhelming.

 

Search is typically one of the first features a user will interact with on a news website. Results have to be relevant to the input query. Determining how well a result matches the query typically requires a scoring function that is difficult to define with rules and heuristics alone. Click feedback and analytics is required to ensure quality of results. Good quality search will drive engagement and time on site.

Optimizing search using Machine Learning and Natural Language Processing to improve relevancy and presentation of results.

Machine learning algorithms, trained on user inputs and click feedback, are commonly used to score candidate results. Manual annotation of result sets is often used to further optimize search relevancy. Natural Language Processing techniques are also often used to both analyze the input query before searching against the index and for organizing the results into related themes for the end user.

 

On the Bloomberg terminal, 17 news stories a second are published1. Bloomberg uses proprietary AI models to prevent users from becoming overwhelmed by results. Firstly, a model groups similar results together, while a second model then assigns a summary to each result cluster. To make navigation easier for the end user, instead of a long list of results, the Bloomberg user interface presents the clusters as “Key Themes” that can be further drilled into.

 

Nowadays many enterprise search solutions offer tunable search relevancy straight out of the box with no coding required.