The news industry generates a substantial amount of content daily but listeners, viewers, and readers often struggle to find the most relevant and relatable news. Using Amazon Personalize, companies can address this issue by offering personalized news recommendations tailored to individual preferences, thus enhancing engagement.
A Fortune 500 media company implemented this solution in H1 2023. The recommendation system relies on analyzing user preferences to provide relevant and trending news. However, there are certain challenges in the process, such as capturing diverse user interests, lack of reader history, timeliness of information, evolving reader interests, and balancing personalization with discovery of popular content.
Amazon Personalize has been instrumental in overcoming these difficulties. Its User Personalization recipe analyzes user preferences based on their content engagement over time, while its Trending Now recipe determines rising trends and popular news stories.
This method, however, has its limitations. It is often challenging to provide personalized recommendations for fresh articles. Also, Amazon Personalize has a fixed number of interactions and items dataset features available for model training. Additionally, Amazon Personalize currently does not provide recommendation explanations at the user level.
For this news recommendation system to work, historical and real-time user click data and news article metadata are needed. Data ingestion occurs either through batch ingestion or real-time ingestion, supported by AWS Glue and Amazon Kinesis Data Streams respectively.
In addressing the “cold items” issue—just-published stories without historical interaction data—the team implemented a feature to randomly insert these articles into the final recommendation output. This allows for the presentation of new content as soon as it’s published.
The post also discussed utilizing AWS Step Functions workflows to generate batch recommendations for daily news round-ups, keeping users up-to-date without making them search for news. To scale the recommender system, Amazon Personalize model endpoints auto-scale to handle increased traffic.
In conclusion, Amazon Personalize provides an effective solution for news recommendation systems, striking a balance between user personalization and timeliness.
