In today's world, publishers aim to share the best content with the widest audience possible. It's no secret that it’s all about sending the right message to the right audience at the right time. In fact, there's a delicate balance here—trying to align reach, relevance, and measurement to make everything work cohesively. That’s why we’re proud to release our enhanced contextual video recommendation engine, which is set to redefine the way you create and deliver video recommendations to your audience across various domains.
Our recommendation engine has recently undergone a significant facelift to enhance its ability to quickly match videos with articles, even for newly published content. Its primary goal is to ensure the delivery of the most relevant and up-to-date content in real-time, with speed as a secondary consideration. As they say in boxing, "timing beats speed," emphasizing the importance for publishers to keep their content timely and engaging. This ensures that publishers can seamlessly incorporate videos on every page, thereby enhancing content experiences and boosting viewer engagement. Publishers will also be pleased to discover that they retain full control over setting up and refining their video content bank according to their specific preferences and needs, ensuring it aligns with their requirements.
How does it work?
Recommendations are likely a familiar concept to you, whether you realize it or not. You probably engage with recommendation engines every day for things like shopping suggestions, movie recommendations, and dating app matches.
Our contextual recommendation engine leverages open-source large learning models (LLMs) for contextual matching purposes. Additionally, we utilize machine learning models based on vectorization aka vector embedding, to convert data such as text, images, audio, or other content types into numerical representations.
Through text vectorization, similarity calculation, and result ranking, we can match videos that are most relevant to each article. This ensures that digital publishers can provide audiences with the most relevant videos from their video content bank in real-time, enabling scalable video integration across an entire website without the need to produce specific content for each article or manually match articles with content.
Playlist Application
Our contextual engine applies to EX.CO playlists by actively scanning each new article that is loaded. Each time an article loads, the engine springs into action, generating a contextual recommendation playlist tailored to that specific article. This automated process ensures that the displayed content perfectly aligns with the context of the current page the user is viewing.
This process unfolds instantly with each new player load and then refreshes periodically, every 30 minutes, to update the playlist with newer, even more relevant videos. This ensures that the content remains consistently fresh and engaging, providing users with a seamless and delightful viewing experience.
Contextual relevance scoring
Our quest to provide relevant video recommendations relies on how our engine assigns a relevancy score to each video. This score indicates how closely the suggested videos match the content viewed on the page. It ensures that the most relevant videos are selected for display, which leads to higher user engagement and longer time spent on the page.
Furthermore, the engine can effortlessly craft tailored playlists for each article based on factors such as relevancy and the best performing videos. This capability ensures that users embark on a personalized and engaging video journey that aligns perfectly with their interests and the content they are consuming.
Dynamic content selection based on recency
In a landscape saturated with information, the freshness of your content is crucial. Ask any buyer or seller about their preferences for targeting, and contextual is top of mind. To make a lasting impression, you need up-to-date content and must remove any old or irrelevant assets.
That's why our online video platform (OVP) allows users to refine the recommended content pool curated from your video content bank or our video marketplace. Editors and content managers can create dynamic playlists based on video keywords, media categories, and titles, adjusting settings like video recency and length. For example, they can specify recency settings to recommend videos from the last available in the playlist, such as the last 50 to 500 videos. Users can also choose specific time frames like today, the last 48 hours, or the last 7, 14, or 30 days to further fine-tune their selections. Based on these preferences, the engine curates a tailored playlist for each article.
Wrapping up
We are already witnessing the impact of our upgraded contextual recommendation engine on our partners who have implemented it. Our top-tier publishers are achieving an 80% relevancy match rate, driving engagement rates with the video player that are four times higher than the industry benchmark. Furthermore, average negative interactions with the video player have decreased by 30-40%, resulting in an increase in dwell and revenue.
Although our technology currently optimizes video recommendations by adjusting criteria including media category, title, recency, sentiment, keywords, and length, we plan to expand on this by rolling out ChatGPT-like functionality soon, allowing publishers to refine video recommendations with prompts. This will help train the engine to deliver more relevant recommendations for specific sections and articles.
Have questions about our contextual recommendation engine Submit the form below to reach out.